From 2ca726e3af7c1ee5e5dd523727f5db2a05b9f977 Mon Sep 17 00:00:00 2001 From: Nexoniarz Date: Sat, 3 Oct 2026 14:17:59 +0200 Subject: [PATCH 1/2] YAPF v1: fast byte-aligned format, SIMD, tools and plugins Format (version 1, replaces the earlier experimental versions): - per-tile YCoCg-R or subtract-green colour transform, 8-bit modular - 12 row filters per 8-row band (Left, Gradient, Up-right, Up-average, Up 1-8 rows), none with more than a one-add serial dependency - residuals in groups of 8 at the group's width (w bits x 8 = w bytes), 4-bit control codes with repeat-previous-width runs chosen by DP - constant and stored tiles, independent 64x64 tiles, mips, GPU hints - all sizes validated before decoding; corrupt input is rejected Library: - SIMD on by default: SSSE3 (runtime CPUID check; GCC, Clang, MSVC, MinGW) and NEON on ARM64; -DYAPF_NO_SIMD for plain C - yapf_load_memory, yapf_load_mt, yapf_load_memory_mt - 6-10x faster decoding than PNG (stb_image) on one core; -O2 and -O3 builds perform the same - fixes for macOS (sysconf) and the Windows import library exports Tools and integrations: - `yapf` CLI: convert PNG/JPG/TGA/BMP/PSD/GIF <-> YAPF, info, bench, thumbnail, mips - JavaScript and Python bindings (byte-identical to the C encoder) - GIMP 3 plugin: exports the flattened image, optional mipmaps - Blender extension: import (also as plane), export, drag and drop - VS Code preview, Figma plugin, Photoshop scripts, Linux MIME type and thumbnailer - Makefile, updated build.sh / build.bat, tools/build_release.sh - tests/test_yapf.c (784 checks) and bindings/js/test.js Co-Authored-By: Claude Opus 5.5 --- .gitignore | 17 + Makefile | 65 + README.md | 380 +- bindings/js/README.md | 22 + bindings/js/package.json | 11 + bindings/js/test.js | 57 + bindings/js/yapf.js | 463 + bindings/python/README.md | 23 + bindings/python/yapf.py | 312 + build.bat | 30 +- build.sh | 21 +- extensions/blender/README.md | 34 + extensions/blender/build.py | 59 + extensions/blender/yapf_image/__init__.py | 251 + .../blender/yapf_image/blender_manifest.toml | 16 + extensions/figma/README.md | 32 + extensions/figma/build.js | 9 + extensions/figma/code.js | 54 + extensions/figma/manifest.json | 14 + extensions/figma/src/png.js | 126 + extensions/figma/src/ui.src.html | 124 + extensions/figma/ui.html | 713 ++ extensions/gimp/README.md | 48 + extensions/gimp/howtocompile.txt | 40 - extensions/gimp/yapf-gimp.c | 173 +- extensions/linux/README.md | 15 + extensions/linux/install.sh | 18 + extensions/linux/yapf-mime.xml | 12 + extensions/linux/yapf.thumbnailer | 4 + extensions/photoshop/Open YAPF.jsx | 28 + extensions/photoshop/README.md | 33 + extensions/photoshop/Save as YAPF.jsx | 40 + extensions/photoshop/yapf_common.jsxinc | 41 + extensions/vscode/README.md | 27 + extensions/vscode/extension.js | 65 + extensions/vscode/media/icon.png | Bin 0 -> 12341 bytes extensions/vscode/media/preview.css | 26 + extensions/vscode/media/preview.js | 91 + extensions/vscode/package.json | 29 + extensions/vscode/scripts/copy-codec.js | 6 + other/YAPF.YAPF | Bin 445122 -> 91429 bytes tests/test_yapf.c | 180 + third_party/stb_image.h | 7988 +++++++++++++++++ third_party/stb_image_write.h | 1724 ++++ tools/build_release.sh | 58 + tools/yapf_cli.c | 345 + yapf.c | 1708 +++- yapf.h | 131 +- 48 files changed, 14929 insertions(+), 734 deletions(-) create mode 100644 .gitignore create mode 100644 Makefile create mode 100644 bindings/js/README.md create mode 100644 bindings/js/package.json create mode 100644 bindings/js/test.js create mode 100644 bindings/js/yapf.js create mode 100644 bindings/python/README.md create mode 100644 bindings/python/yapf.py mode change 100644 => 100755 build.sh create mode 100644 extensions/blender/README.md create mode 100644 extensions/blender/build.py create mode 100644 extensions/blender/yapf_image/__init__.py create mode 100644 extensions/blender/yapf_image/blender_manifest.toml create mode 100644 extensions/figma/README.md create mode 100644 extensions/figma/build.js create mode 100644 extensions/figma/code.js create mode 100644 extensions/figma/manifest.json create mode 100644 extensions/figma/src/png.js create mode 100644 extensions/figma/src/ui.src.html create mode 100644 extensions/figma/ui.html create mode 100644 extensions/gimp/README.md delete mode 100644 extensions/gimp/howtocompile.txt create mode 100644 extensions/linux/README.md create mode 100755 extensions/linux/install.sh create mode 100644 extensions/linux/yapf-mime.xml create mode 100644 extensions/linux/yapf.thumbnailer create mode 100644 extensions/photoshop/Open YAPF.jsx create mode 100644 extensions/photoshop/README.md create mode 100644 extensions/photoshop/Save as YAPF.jsx create mode 100644 extensions/photoshop/yapf_common.jsxinc create mode 100644 extensions/vscode/README.md create mode 100644 extensions/vscode/extension.js create mode 100644 extensions/vscode/media/icon.png create mode 100644 extensions/vscode/media/preview.css create mode 100644 extensions/vscode/media/preview.js create mode 100644 extensions/vscode/package.json create mode 100644 extensions/vscode/scripts/copy-codec.js create mode 100644 tests/test_yapf.c create mode 100644 third_party/stb_image.h create mode 100644 third_party/stb_image_write.h create mode 100755 tools/build_release.sh create mode 100644 tools/yapf_cli.c diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..d93af31 --- /dev/null +++ b/.gitignore @@ -0,0 +1,17 @@ +# build outputs +/build/ +/dist/ +*.o +*.a +*.so +*.dylib +*.dll +*.exe +*.vsix +test_yapf_tmp.yapf + +# generated: copied from bindings/js/yapf.js when packaging +/extensions/vscode/media/yapf.js +node_modules/ +__pycache__/ +*.pyc diff --git a/Makefile b/Makefile new file mode 100644 index 0000000..0c2322e --- /dev/null +++ b/Makefile @@ -0,0 +1,65 @@ +# YAPF — simple build for Linux, macOS and other Unix systems. +# +# make library (static + shared) and the `yapf` tool, in build/ +# make test run the C test suite (and the JS one if node is present) +# sudo make install install header, libraries and tool under $(PREFIX) +# make NO_SIMD=1 plain C only +# +# Windows: use build.bat (LLVM) or any MinGW / MSVC compiler on yapf.c. + +CC ?= cc +CFLAGS ?= -O2 +PREFIX ?= /usr/local +BUILD ?= build + +YAPF_CFLAGS = -std=c99 -Wall -Wextra $(CFLAGS) $(if $(NO_SIMD),-DYAPF_NO_SIMD) +LDLIBS = -lm -pthread + +ifeq ($(shell uname -s),Darwin) + SHARED = libyapf.dylib + SHARED_FLAGS = -dynamiclib -install_name $(PREFIX)/lib/libyapf.dylib +else + SHARED = libyapf.so + SHARED_FLAGS = -shared +endif + +all: $(BUILD)/libyapf.a $(BUILD)/$(SHARED) $(BUILD)/yapf + +$(BUILD): + mkdir -p $(BUILD) + +$(BUILD)/yapf.o: yapf.c yapf.h | $(BUILD) + $(CC) $(YAPF_CFLAGS) -fPIC -c yapf.c -o $@ + +$(BUILD)/libyapf.a: $(BUILD)/yapf.o + $(AR) rcs $@ $^ + +$(BUILD)/$(SHARED): $(BUILD)/yapf.o + $(CC) $(SHARED_FLAGS) $^ -o $@ -pthread + +$(BUILD)/yapf: tools/yapf_cli.c yapf.c yapf.h | $(BUILD) + $(CC) $(YAPF_CFLAGS) tools/yapf_cli.c yapf.c -o $@ $(LDLIBS) + +$(BUILD)/test_yapf: tests/test_yapf.c yapf.c yapf.h | $(BUILD) + $(CC) $(YAPF_CFLAGS) tests/test_yapf.c yapf.c -o $@ $(LDLIBS) + +test: $(BUILD)/test_yapf + $(BUILD)/test_yapf other/YAPF.YAPF + @if command -v node >/dev/null 2>&1; then node bindings/js/test.js; \ + else echo "(node not found: skipped the JavaScript tests)"; fi + +install: all + install -d $(DESTDIR)$(PREFIX)/include $(DESTDIR)$(PREFIX)/lib $(DESTDIR)$(PREFIX)/bin + install -m 644 yapf.h $(DESTDIR)$(PREFIX)/include/yapf.h + install -m 644 $(BUILD)/libyapf.a $(DESTDIR)$(PREFIX)/lib/libyapf.a + install -m 755 $(BUILD)/$(SHARED) $(DESTDIR)$(PREFIX)/lib/$(SHARED) + install -m 755 $(BUILD)/yapf $(DESTDIR)$(PREFIX)/bin/yapf + +uninstall: + rm -f $(DESTDIR)$(PREFIX)/include/yapf.h $(DESTDIR)$(PREFIX)/lib/libyapf.a \ + $(DESTDIR)$(PREFIX)/lib/$(SHARED) $(DESTDIR)$(PREFIX)/bin/yapf + +clean: + rm -rf $(BUILD) + +.PHONY: all test install uninstall clean diff --git a/README.md b/README.md index 3aa78d3..e9bf4a0 100644 --- a/README.md +++ b/README.md @@ -1,90 +1,127 @@ # YAPF — Yet Another Picture Format -**Fast. Small. Easy.** +**Lossless images that decode several times faster than PNG — on one core.** -YAPF is a lossless image codec built around a single principle: decoding should -be as close to a memory copy as possible. Instead of running a general-purpose -compressor like DEFLATE over pixel data, YAPF encodes images with five cheap -operations — a pixel cache lookup, two flavours of delta encoding, run-length -encoding, and verbatim pixel storage — producing files that are smaller than -uncompressed formats while decoding nearly as fast as `memcpy`. +YAPF is a lossless image format for photos, UI and game assets, built for +decode speed first. There is no entropy coder, no bit reader and no context +model: every structure is byte aligned, and the decoder runs short loops that +modern CPUs execute at memory speed — with SSE / NEON where available. The +encoder does the hard work, so the decoder only follows instructions. + +- **6–10× faster decoding than PNG** (stb_image), single-threaded; a 4K image + decodes in about 6 ms (5.6 GB/s). Multithreaded decoding is optional. +- **Smaller than PNG** on UI, icons and screenshots; 20–45 % larger on photos + and natural textures — the price of having no entropy coder. +- **Made for games and tools too**: mipmaps, GPU format hint, sRGB / + premultiplied flags, and random access to 64×64 tiles. +- **Tiny and portable**: two C99 files, no dependencies. Plus a CLI, Python, + JavaScript and plugins for the tools you already use. + +This is **YAPF version 1**, the first stable format. --- -## Why YAPF? +## Use it in… -Standard lossless formats like PNG compress well, but decompressing DEFLATE -streams is CPU-intensive. On large images (4K and above) or in -latency-sensitive applications (game engines, real-time pipelines, embedded -systems) this is a meaningful bottleneck. +| Where | What you get | Guide | Status | +| :---- | :----------- | :---- | :----- | +| **C / C++** | `yapf.h` + `yapf.c`, drop into any project | [below](#c--c) | ✅ tested on Linux x64, Windows x64, Linux ARM64 | +| **Command line** | convert PNG/JPG/TGA/BMP/PSD/GIF ↔ YAPF, `info`, `bench`, `thumbnail` | [below](#command-line) | ✅ tested on Linux, Windows | +| **GIMP 3** | open, export (with mipmaps) | [extensions/gimp](extensions/gimp) | ✅ tested in GIMP 3.0.8 | +| **Blender 4.2+** | import, export, drag & drop, image as plane | [extensions/blender](extensions/blender) | ✅ tested in Blender 5.1 | +| **VS Code** | image preview with zoom and mip picker | [extensions/vscode](extensions/vscode) | ✅ packaged, preview tested in Chromium | +| **Figma** | import `.yapf`, export selection as `.yapf` | [extensions/figma](extensions/figma) | ✅ plugin UI tested in Chromium | +| **Photoshop** | File → Scripts → Open YAPF / Save as YAPF | [extensions/photoshop](extensions/photoshop) | ⚠️ not yet tested inside Photoshop | +| **Linux desktop** | file type + thumbnails in GNOME Files, Nemo, Caja, Thunar | [extensions/linux](extensions/linux) | ⚠️ thumbnailer tested, not in a file manager | +| **Python** | `yapf.load()` / `yapf.save()` | [bindings/python](bindings/python) | ✅ tested | +| **JavaScript** | browsers, Node.js, workers | [bindings/js](bindings/js) | ✅ tested, byte-identical to C | -YAPF trades a modest compression ratio advantage for dramatically lower decode -complexity. There are no Huffman tables to build, no sliding windows to -maintain, and no bit-level unpacking. The decoder is a tight byte-dispatch -loop that any modern CPU handles at memory bandwidth. - -### Performance - -Benchmark: loading a 5K professional photograph (5120 × 2880 px). - -| Format | Load time | -| :----- | ----------: | -| **YAPF** | **0.111 s** | -| PNG | 0.194 s | - -YAPF decodes approximately **1.7× faster** than PNG on the same hardware -without SIMD or threading. - -### When YAPF excels - -- **Pixel art, sprites, UI textures** — large flat regions of identical colour - compress to single RUN bytes; palette-like images benefit heavily from the - cache. -- **Screenshots and diagrams** — solid backgrounds and repeated gradients hit - DIFF/LUMA and the cache constantly. -- **Game asset pipelines** — load textures at startup without paying a - decompression tax every frame. -- **Embedded / resource-constrained targets** — the entire codec fits in ~200 - lines of portable C with zero external dependencies. +Pre-built downloads (command-line tool and libraries for Windows, macOS and +Linux, plus the plugin packages) are published on the +[Releases](https://github.com/Nexoniarz/YAPF/releases) page; everything can +also be built from source as described below. --- -## Integration +## Command line -The core library is two files. Drop them into your project and compile: +```sh +yapf photo.png photo.yapf # encode (formats by file extension) +yapf texture.png texture.yapf --mips # with a full mip chain +yapf photo.yapf photo.png # decode to PNG / JPG / BMP / TGA +yapf texture.yapf mip2.png --mip 2 # export one mip level +yapf info photo.yapf # size, channels, mips, flags, ratio +yapf bench photo.yapf # decode speed, 1 thread and all cores +yapf thumbnail photo.yapf thumb.png 256 +``` + +Options when writing `.yapf`: `--mips`, `--linear` (pixels are not sRGB), +`--premult` (alpha is premultiplied). + +--- + +## Benchmark + +Single core, AMD Ryzen 7 5700G, `gcc -O2`. PNG is decoded with `stb_image`; +the PNG files are the ones shipped with the respective projects. "Plain C" +is the library built with `-DYAPF_NO_SIMD`; the default build uses SSE / NEON. + +| Image | Raw | PNG | **YAPF** | vs PNG | PNG decode | YAPF, plain C | **YAPF (default)** | +| :--------------------- | ------: | -----: | -------: | -----: | ---------: | ------------: | -----------------: | +| Wallpaper 3840×2160 | 32400 K | 1812 K | 2267 K | 125 % | 44.7 ms¹ | 10.3 ms | **5.9 ms** (5.6 GB/s) | +| UI screenshot 261×430 | 438 K | 176 K | 143 K | 81 % | 1.6 ms | 0.31 ms | **0.17 ms** | +| UI screenshot 305×529 | 630 K | 201 K | 185 K | 92 % | 2.1 ms | 0.45 ms | **0.24 ms** | +| App icon 128×128 | 64 K | 24 K | 23 K | 93 % | 0.26 ms | 0.06 ms | **0.03 ms** | +| Rock texture 512×512 | 768 K | 182 K | 221 K | 121 % | 1.4 ms | 0.35 ms | **0.16 ms** | +| Wood texture 512×512 | 768 K | 185 K | 234 K | 126 % | 1.4 ms | 0.46 ms | **0.16 ms** | +| Canvas texture 512×512 | 768 K | 171 K | 245 K | 143 % | 1.0 ms | 0.35 ms | **0.15 ms** | +| Pixel art 512×512 | 1024 K | — | 66 K | — | — | 0.74 ms | **0.32 ms** | +| Random noise 512×512 | 1024 K | — | 1024 K | — | — | 0.1 ms | 0.1 ms | + +¹ PNG decode of the wallpaper measured between 44.7 and 66 ms depending on +the harness; the table uses the fastest. `-O2` and `-O3` builds of YAPF +perform the same. + +--- + +## C / C++ + +The library is two files. Add them to your project: ``` -yapf.h — public API and format documentation -yapf.c — encoder / decoder implementation +yapf.h — public API and the complete format specification +yapf.c — encoder and decoder ``` ```c #include "yapf.h" -int main(void) { - /* Load */ - yapf_image_t *img = yapf_load("photo.yapf"); - if (!img) { /* file not found or not a valid YAPF stream */ } - - printf("%u x %u, %u channel(s)\n", - img->width, img->height, img->channels); - - /* Save */ - int rc = yapf_save("output.yapf", img); - if (rc != YAPF_OK) { /* write error */ } - +yapf_image_t *img = yapf_load("texture.yapf"); /* or yapf_load_memory(buf, len) */ +if (img) { + /* img->pixels: width × height × channels bytes, rows top to bottom */ + upload_texture(img->width, img->height, img->channels, img->pixels); yapf_free(img); - return 0; } + +yapf_image_t out = { .width = w, .height = h, .channels = 4, + .mip_levels = 1, .pixels = rgba, .flags = YAPF_FLAG_SRGB }; +yapf_save("out.yapf", &out); ``` -Compile with any C99-capable compiler — no flags required for the baseline: +Compile with any C99 compiler (`-pthread` on Unix for the `_mt` loaders): ```sh -gcc -O2 -std=c99 main.c yapf.c -o myapp +cc -O2 -std=c99 main.c yapf.c -o app -pthread ``` ---- +SIMD is on by default: SSSE3 on x86 / x64 (chosen at run time with CPUID, +so binaries still run on older CPUs) and NEON on ARM64, with GCC, Clang, +MSVC and MinGW. + +| Define | Effect | +| :------------------ | :----- | +| `-DYAPF_NO_SIMD` | plain C only | +| `-DYAPF_NO_THREADS` | no threads; the `_mt` loaders decode on the calling thread, and only the C standard library is needed | ## API Reference @@ -92,15 +129,20 @@ gcc -O2 -std=c99 main.c yapf.c -o myapp ```c typedef struct { - uint32_t width; /* pixels per row */ - uint32_t height; /* rows */ - uint8_t channels; /* 1 gray / 2 gray+alpha / 3 RGB / 4 RGBA */ - uint8_t *pixels; /* raw pixel buffer, malloc-owned */ + uint32_t width; /* pixels per row */ + uint32_t height; /* rows */ + uint8_t channels; /* 1 gray / 2 gray+alpha / 3 RGB / 4 RGBA */ + uint8_t gpu_format; /* YAPF_GPU_* hint */ + uint8_t flags; /* YAPF_FLAG_* (premultiplied alpha, sRGB) */ + uint8_t mip_levels; /* 1 = base only, up to 16 */ + uint8_t *pixels; /* base level, malloc-owned */ + uint8_t **mips; /* mips[0] == pixels, mips[1..] smaller levels */ } yapf_image_t; ``` `pixels` is a tightly packed, row-major byte array of `width × height × channels` -bytes. The layout per pixel matches the `channels` value: +bytes. Mip level *m* is `max(1, width >> m) × max(1, height >> m)`. The +layout per pixel matches the `channels` value: | `channels` | Layout per pixel | | :--------- | :--------------- | @@ -113,29 +155,40 @@ bytes. The layout per pixel matches the `channels` value: ```c yapf_image_t *yapf_load(const char *filename); +yapf_image_t *yapf_load_memory(const void *buffer, size_t size); ``` -Decodes a YAPF file from disk. Returns a heap-allocated `yapf_image_t` on -success, `NULL` on any failure (file not found, invalid magic, unsupported -version, out of memory). Free with `yapf_free()`. +Decode a YAPF file from disk or from memory (e.g. a game archive). Return a +heap-allocated image on success, `NULL` on any failure (file not found, +invalid or corrupt data, out of memory). Corrupt input is rejected, never +read out of bounds. + +```c +yapf_image_t *yapf_load_mt(const char *filename, int threads); +yapf_image_t *yapf_load_memory_mt(const void *buffer, size_t size, int threads); +``` +Same, decoding tiles on up to `threads` threads (`0` = one per CPU core). +The output is identical to the single-threaded loaders. ```c int yapf_save(const char *filename, const yapf_image_t *img); ``` -Encodes an image to disk. Returns `YAPF_OK` (0) on success or a negative -`YAPF_ERR_*` code on failure. +Encodes an image to disk. Writes `img->mip_levels` +levels; `img->mips` may be `NULL` when `mip_levels == 1`. Returns `YAPF_OK` +(0) on success or a negative `YAPF_ERR_*` code on failure. ```c void yapf_free(yapf_image_t *img); ``` -Releases memory returned by `yapf_load()`. Passing `NULL` is safe. +Releases memory returned by the loaders. Passing `NULL` is safe. ### Return codes -| Constant | Value | Meaning | -| :---------------- | ----: | :----------------------------------- | -| `YAPF_OK` | 0 | Success | -| `YAPF_ERR_INVALID`| –1 | NULL or logically invalid argument | -| `YAPF_ERR_IO` | –2 | File could not be opened or written | +| Constant | Value | Meaning | +| :----------------- | ----: | :----------------------------------- | +| `YAPF_OK` | 0 | Success | +| `YAPF_ERR_INVALID` | –1 | NULL or logically invalid argument | +| `YAPF_ERR_IO` | –2 | File could not be opened or written | +| `YAPF_ERR_OOM` | –3 | Out of memory | ### Channel constants @@ -148,115 +201,102 @@ YAPF_CHANNELS_RGBA 4 --- -## Building the libraries +## Building -The build scripts uses the [LLVM](https://releases.llvm.org/) compiler +| System | Command | Output | +| :----- | :------ | :----- | +| Linux, macOS, BSD | `make` · `make test` · `sudo make install` | `build/`: `libyapf.a`, `libyapf.so` / `.dylib`, `yapf` | +| Linux, macOS (LLVM) | `./build.sh` | `dist//` incl. macOS universal binaries | +| Windows (LLVM) | `build.bat` | `dist\windows-x64\`, `dist\windows-arm64\`: `yapf.dll`, `yapf.lib`, `libyapf.a`, `yapf.exe` | +| Windows (Visual Studio) | `cl /O2 /c yapf.c` and add `yapf.obj` to your project | | +| All platforms at once | `tools/build_release.sh` (needs [zig](https://ziglang.org)) | `dist/release/`: every download on the Releases page | + +The command-line tool is `tools/yapf_cli.c` + `yapf.c`; it uses the public +domain [stb](https://github.com/nothings/stb) headers in `third_party/`. + +--- + +## How it works + +1. **Two colour transforms** — per 64×64 tile, YCoCg-R or subtract-green + (`G, R−G, B−G`), whichever codes smaller. Both run mod 256, so every + plane stays one byte per sample. +2. **Twelve row filters** — per band of 8 rows: Left, Gradient (`L + A − D`), + Up-right, Up-average, or the row 1–8 rows above (catches repeating + textures and dither patterns). None depends on the previous sample by + more than one add, and the vertical ones decode as plain vector code. +3. **Groups of 8** — residuals are packed in groups of 8 at the group's bit + width: 8 values × *w* bits = exactly *w* bytes, so every group is byte + aligned and unpacks with one 64-bit load (or one SIMD shuffle). +4. **4-bit control codes** — one per group, or one per run of up to 64 groups + repeating the previous width; the encoder picks the cheapest sequence with + a small dynamic program. +5. **Validate first, then run** — all sizes in a tile are checked before the + hot loops start, so the loops need no bounds checks. Corrupt files are + rejected, never read out of bounds. +6. **Constant and stored tiles** — a plane that is constant in a tile costs + one byte; a tile that would not shrink is stored raw (noise never grows). +7. **Independent tiles** — any tile can be decoded on its own, in any order, + on any thread. + +The complete specification is the comment at the top of [`yapf.h`](yapf.h): + +``` +HEADER (20 bytes) "YAPF", version 1, channels, GPU hint, mip count, + flags, 3 reserved, width u32, height u32 +MIP INDEX per level: offset u32, size u32 +LEVEL DATA tile_size u32 × tile_count, then the tiles +TILE (64×64) header byte: stored raw, or constant-plane mask + colour transform + per plane: filter byte per 8 rows, control nibbles, + w bytes per group of 8 values +``` + +--- + +## Testing ```sh -# Unix (Linux / macOS) -chmod +x build.sh -./build.sh - -# Windows -build.bat +make test # C suite: 784 checks, plus JavaScript if node is installed +node bindings/js/test.js # JavaScript: round-trips, sample file, corrupt input ``` -Output goes to `dist/` with one subdirectory per platform: +What has been verified for this release: -``` -dist/ -├── windows-x64/ yapf.dll yapf.lib -├── windows-arm64/ yapf.dll yapf.lib -├── macos-x64/ libyapf.dylib -├── macos-arm64/ libyapf.dylib -├── linux-x64/ libyapf.so libyapf.a -└── linux-arm64/ libyapf.so libyapf.a -``` +- exact round-trips for 1–4 channels, odd sizes from 1×1 up, mip chains, + single- and multi-threaded decoding; +- the same tests pass on **Linux x64**, **Windows x64** (MinGW, run under + Wine), **Linux ARM64** (NEON, run under qemu), with and without SIMD, and + under AddressSanitizer, UndefinedBehaviorSanitizer and ThreadSanitizer; +- thousands of corrupted and truncated files are rejected without crashing; +- the JavaScript and Python versions produce byte-identical files to C; +- the GIMP plugin and the Blender extension in the real applications. -### Optional build flags - -| Flag | Effect | -| :---------------- | :-------------------------------------------------- | -| `-DYAPF_USE_SIMD` | Reserved for future SIMD acceleration paths | - -The interactive build script will prompt for these options. +The macOS builds are cross-compiled and not yet run on a Mac. --- -## Technical Specification - -### File header — 14 bytes - -| Offset | Size | Type | Description | -| :----- | ---: | :--------- | :-------------------------------------- | -| 0x00 | 4 | `char[4]` | Magic: `"YAPF"` | -| 0x04 | 1 | `uint8_t` | Format version (currently `1`) | -| 0x05 | 4 | `uint32_t` | Width in pixels, little-endian | -| 0x09 | 4 | `uint32_t` | Height in pixels, little-endian | -| 0x0D | 1 | `uint8_t` | Channel count: 1 / 2 / 3 / 4 | - -All multi-byte integers are always stored little-endian, regardless of host -architecture. - -### Byte-stream opcodes - -The payload begins immediately after the header and is decoded left-to-right. -A 64-slot pixel cache (zero-initialised) and a running "previous pixel" state -(`r=0, g=0, b=0, a=255`) are maintained throughout. - -| Byte range | Opcode | Size | Description | -| :---------- | :----- | ------: | :-------------------------------------------- | -| 0x00–0x3F | INDEX | 1 byte | Restore pixel from cache slot `byte & 0x3F` | -| 0x40–0x7F | DIFF | 1 byte | Per-channel deltas: dr = bits[5:4]–2, dg = bits[3:2]–2, db = bits[1:0]–2. Range –2..+1. Alpha unchanged. | -| 0x80–0xBF | LUMA | 2 bytes | Green delta dg = (byte & 0x3F)–32. Second byte: dr = ((b1>>4)&0xF)–8+dg, db = (b1&0xF)–8+dg. Alpha unchanged. | -| 0xC0–0xFD | RUN | 1 byte | Repeat previous pixel (byte&0x3F)+1 times (1–62). Cache not updated. | -| 0xFE | RGB | 4 bytes | Next 3 bytes = R, G, B. Alpha unchanged. | -| 0xFF | RGBA | 5 bytes | Next 4 bytes = R, G, B, A. | - -After every non-RUN opcode, the cache slot `HASH(r,g,b,a)` is updated and the -pixel becomes the new previous pixel. +## Project layout ``` -HASH(r,g,b,a) = (r×3 + g×5 + b×7 + a×11) mod 64 +yapf.h, yapf.c the library (format spec in yapf.h) +tools/yapf_cli.c the `yapf` command-line tool +tools/build_release.sh builds every download for all platforms +tests/test_yapf.c C test suite +bindings/js/ JavaScript encoder / decoder +bindings/python/ Python module +extensions/gimp/ GIMP 3 plugin +extensions/blender/ Blender extension +extensions/vscode/ VS Code preview +extensions/figma/ Figma plugin +extensions/photoshop/ Photoshop scripts +extensions/linux/ MIME type and thumbnailer +third_party/ stb_image / stb_image_write (public domain, CLI only) +other/YAPF.YAPF sample image (the YAPF logo) ``` -The stream ends when exactly `width × height` pixels have been decoded; no -end-of-stream marker is required. - -### Pixel buffer layout - -Within the decoded buffer, each pixel occupies `channels` consecutive bytes in -the order shown in the API table above. Grayscale images store a single -luminance byte per pixel; the encoder internally promotes gray values to -`r = g = b = Y` for uniform opcode selection. - ---- - -## GIMP Plugin - -A plugin for GIMP is located in `extensions/gimp/`. See the -[build instructions](extensions/gimp/howtocompile.txt) for details. - -Supported colour modes: Gray, Gray+Alpha, RGB, RGBA. -File extension: `.yapf`. MIME type: `image/x-yapf`. - ---- - -## Platforms - -| Platform | Architectures | Library format | -| :------------- | :------------ | :------------------ | -| Windows | x64, ARM64 | `.dll` + `.lib` | -| macOS | ARM64 | `.dylib` | -| Linux (glibc) | x64, ARM64 | `.so` | -| Linux (musl) | x64, ARM64 | `.a` (static) | - -The core library (`yapf.c` / `yapf.h`) uses only the C99 standard library -(``, ``, ``, ``) and compiles cleanly -on any conforming C99 or C++ toolchain with no configuration required. - --- ## License -Apache License 2.0 — see [LICENSE](LICENSE). +Apache License 2.0 — see [LICENSE](LICENSE). `third_party/` contains +stb_image and stb_image_write by Sean Barrett (public domain / MIT). diff --git a/bindings/js/README.md b/bindings/js/README.md new file mode 100644 index 0000000..849f640 --- /dev/null +++ b/bindings/js/README.md @@ -0,0 +1,22 @@ +# yapf.js — YAPF in JavaScript + +Encoder and decoder for YAPF v1 in one dependency-free file. Works in +browsers, Node.js, web workers, Figma plugins and VS Code webviews, and +produces exactly the same bytes as the C library. + +```js +const YAPF = require('./yapf.js'); // or + + + + diff --git a/extensions/figma/ui.html b/extensions/figma/ui.html new file mode 100644 index 0000000..0b419ac --- /dev/null +++ b/extensions/figma/ui.html @@ -0,0 +1,713 @@ + + + + + + + +
+ + +
+ +
+

Import YAPF images

+
Drop .yapf files here
or click to choose
+ +
+ + + +
+ + + + + + diff --git a/extensions/gimp/README.md b/extensions/gimp/README.md new file mode 100644 index 0000000..d3dbae0 --- /dev/null +++ b/extensions/gimp/README.md @@ -0,0 +1,48 @@ +# YAPF for GIMP + +Open and export `.yapf` images in **GIMP 3.0 or later**. + +- **File → Open** reads `.yapf` (Gray, Gray + Alpha, RGB, RGBA). +- **File → Export As… → `name.yapf`** writes the visible image (layers are + flattened into a copy; your document is untouched), with an optional + **Store mipmaps** checkbox for game engines. + +## Install + +You need GIMP 3 with its development files (`gimptool`). + +| System | Development files | +| :-------------- | :-------------------------------------------------- | +| Arch / Manjaro | `sudo pacman -S gimp` (includes them) | +| Debian / Ubuntu | `sudo apt install libgimp-3.0-dev` (GIMP 3 packages) | +| Fedora | `sudo dnf install gimp-devel` | +| Windows | MSYS2: `pacman -S mingw-w64-ucrt-x86_64-gimp` | +| macOS | build GIMP 3 from source or use MacPorts/Homebrew | + +Then, from this folder: + +```sh +gimptool --install yapf-gimp.c +``` + +That compiles the plugin (it pulls in `../../yapf.c` itself) and installs it +into your personal GIMP plug-ins folder. Restart GIMP. + +### Manual build + +```sh +gcc -O2 $(pkg-config --cflags gimpui-3.0) yapf-gimp.c $(pkg-config --libs gimpui-3.0) -o yapf-gimp +``` + +Copy `yapf-gimp` to `/plug-ins/yapf-gimp/yapf-gimp`, where the +GIMP config folder is `~/.config/GIMP/3.0` on Linux, +`%APPDATA%\GIMP\3.0` on Windows and +`~/Library/Application Support/GIMP/3.0` on macOS. + +## Notes + +- Pixels are saved with the sRGB flag, because GIMP's 8-bit buffers are + gamma encoded. Files marked linear load as linear. +- When a file has mipmaps, GIMP opens the full-size image. +- Tested with GIMP 3.0.8 on Linux: opening matches the reference decoder + pixel for pixel, and export equals GIMP's own flattened PNG of the image. diff --git a/extensions/gimp/howtocompile.txt b/extensions/gimp/howtocompile.txt deleted file mode 100644 index 264a26c..0000000 --- a/extensions/gimp/howtocompile.txt +++ /dev/null @@ -1,40 +0,0 @@ -YAPF GIMP Plugin — Build Instructions -====================================== - -Requirements ------------- - - GIMP 3.0 or later (with development headers) - - gimptool (included with GIMP 3 dev packages) - - yapf.c and yapf.h must be in the SAME directory as yapf-gimp.c - (the plugin embeds the codec via a direct #include) - - On Arch/Manjaro: sudo pacman -S gimp - On Debian/Ubuntu: sudo apt install gimp libgmp3-dev - On Fedora: sudo dnf install gimp gimp-devel - -Installation (automatic) -------------------------- -Open a terminal in this directory and run: - - $ gimptool --install yapf-gimp.c - -This compiles the plugin and installs it into your GIMP plug-ins folder -automatically. Restart GIMP and .yapf files will appear in the "Open" and -"Export As" dialogs under "YAPF image (*.yapf)". - -Manual compilation ------------------- - $ gcc $(gimptool --cflags) -O2 yapf-gimp.c \ - $(gimptool --libs) -o yapf-gimp - $ install -Dm755 yapf-gimp \ - ~/.config/GIMP/3.(your version here)/plug-ins/yapf-gimp/yapf-gimp - -Notes ------ - - The plugin registers two procedures: file-yapf-load and file-yapf-save. - - Supported colour modes: Gray, Gray+Alpha, RGB, RGBA. - - The MIME type is image/x-yapf and the file extension is .yapf. - - RGB and RGBA images are saved with the sRGB GPU hint (YAPF_FLAG_SRGB) - because GIMP's native 8-bit buffers are always gamma-corrected sRGB. - - Mip levels are NOT generated on export — GIMP does not expose them. - Use the YAPF C API directly if you need a mip chain. diff --git a/extensions/gimp/yapf-gimp.c b/extensions/gimp/yapf-gimp.c index 869b735..7153f10 100644 --- a/extensions/gimp/yapf-gimp.c +++ b/extensions/gimp/yapf-gimp.c @@ -21,22 +21,24 @@ * * ------------------------------------------------------------------------- * - * Compilation: + * Compilation (from this folder): * gimptool --install yapf-gimp.c * - * yapf.c is embedded via #include below because gimptool compiles a single - * source file. Do NOT link separately — the include IS the build. + * yapf.c (two folders up) is embedded via #include below because gimptool + * compiles a single source file. Do NOT link it separately. */ #include +#include /* Pull in the entire encoder/decoder as a single translation unit. */ -#include "yapf.c" +#include "../../yapf.c" /* ── GIMP procedure names ────────────────────────────────────────────── */ #define LOAD_PROC "file-yapf-load" #define SAVE_PROC "file-yapf-save" +#define PLUG_IN_BINARY "yapf-gimp" /* ── Plugin type boilerplate ─────────────────────────────────────────── */ @@ -123,6 +125,14 @@ static GimpProcedure *yapf_plugin_create_procedure(GimpPlugIn *plug_in, "image/x-yapf"); gimp_file_procedure_set_extensions(GIMP_FILE_PROCEDURE(procedure), "yapf"); + /* GIMP flattens / converts a copy of the image for us as needed. */ + gimp_export_procedure_set_capabilities(GIMP_EXPORT_PROCEDURE(procedure), + GIMP_EXPORT_CAN_HANDLE_RGB | GIMP_EXPORT_CAN_HANDLE_GRAY | + GIMP_EXPORT_CAN_HANDLE_ALPHA, NULL, NULL, NULL); + gimp_procedure_add_boolean_argument(procedure, "mipmaps", + "Store _mipmaps", + "Also store a full mip chain (for game engines and GPUs)", + FALSE, G_PARAM_READWRITE); } return procedure; @@ -142,7 +152,7 @@ static GimpValueArray *yapf_load_run( (void)run_mode; (void)metadata; (void)flags; (void)config; (void)run_data; gchar *path = g_file_get_path(file); - yapf_image_t *img = yapf_load(path); + yapf_image_t *img = yapf_load_mt(path, 0); /* all cores */ g_free(path); if (!img) { @@ -164,12 +174,12 @@ static GimpValueArray *yapf_load_run( case YAPF_CHANNELS_GRAY: base_type = GIMP_GRAY; layer_type = GIMP_GRAY_IMAGE; - babl_fmt = "Y' u8"; + babl_fmt = is_srgb ? "Y' u8" : "Y u8"; break; case YAPF_CHANNELS_GRAY_ALPHA: base_type = GIMP_GRAY; layer_type = GIMP_GRAYA_IMAGE; - babl_fmt = "Y'A u8"; + babl_fmt = is_srgb ? "Y'A u8" : "YA u8"; break; case YAPF_CHANNELS_RGB: base_type = GIMP_RGB; @@ -204,6 +214,44 @@ static GimpValueArray *yapf_load_run( /* ── Save handler ────────────────────────────────────────────────────── */ +/* Full mip chain with a 2×2 box filter (edges clamp for odd sizes). + * Fills img->mips / img->mip_levels; mips[0] aliases img->pixels. */ +static gboolean yapf_build_mips(yapf_image_t *img) { + int levels = 1; + while (levels < (int)YAPF_MAX_MIPS && + ((img->width >> levels) > 0 || (img->height >> levels) > 0)) + levels++; + img->mips = (uint8_t **)calloc((size_t)levels, sizeof(uint8_t *)); + if (!img->mips) return FALSE; + img->mips[0] = img->pixels; + img->mip_levels = (uint8_t)levels; + + int ch = img->channels; + for (int m = 1; m < levels; m++) { + uint32_t pw = MAX(1u, img->width >> (m - 1)), ph = MAX(1u, img->height >> (m - 1)); + uint32_t w = MAX(1u, img->width >> m), h = MAX(1u, img->height >> m); + const uint8_t *src = img->mips[m - 1]; + uint8_t *dst = (uint8_t *)malloc((size_t)w * h * ch); + if (!dst) return FALSE; + for (uint32_t y = 0; y < h; y++) + for (uint32_t x = 0; x < w; x++) + for (int c = 0; c < ch; c++) { + uint32_t x0 = MIN(2 * x, pw - 1), x1 = MIN(2 * x + 1, pw - 1); + uint32_t y0 = MIN(2 * y, ph - 1), y1 = MIN(2 * y + 1, ph - 1); + unsigned sum = src[((size_t)y0 * pw + x0) * ch + c] + src[((size_t)y0 * pw + x1) * ch + c] + + src[((size_t)y1 * pw + x0) * ch + c] + src[((size_t)y1 * pw + x1) * ch + c]; + dst[((size_t)y * w + x) * ch + c] = (uint8_t)((sum + 2) / 4); + } + img->mips[m] = dst; + } + return TRUE; +} + +static GimpValueArray *yapf_error(GimpProcedure *procedure, const gchar *msg) { + return gimp_procedure_new_return_values(procedure, GIMP_PDB_EXECUTION_ERROR, + g_error_new_literal(GIMP_PLUG_IN_ERROR, 0, msg)); +} + static GimpValueArray *yapf_save_run( GimpProcedure *procedure, GimpRunMode run_mode, @@ -214,93 +262,88 @@ static GimpValueArray *yapf_save_run( GimpProcedureConfig *config, gpointer run_data) { - (void)run_mode; (void)options; (void)metadata; (void)config; (void)run_data; + (void)metadata; (void)run_data; - /* Retrieve the top-most layer. */ - GimpLayer **layers = gimp_image_get_layers(image); - gint n_layers = 0; - if (layers) { while (layers[n_layers]) n_layers++; } - - if (!layers || n_layers == 0) { - g_free(layers); - GError *err = g_error_new(GIMP_PLUG_IN_ERROR, 0, - "No layers to export."); - return gimp_procedure_new_return_values(procedure, - GIMP_PDB_EXECUTION_ERROR, err); + if (run_mode == GIMP_RUN_INTERACTIVE) { + gimp_ui_init(PLUG_IN_BINARY); + GtkWidget *dialog = gimp_export_procedure_dialog_new( + GIMP_EXPORT_PROCEDURE(procedure), config, image); + gimp_procedure_dialog_fill(GIMP_PROCEDURE_DIALOG(dialog), NULL); + gboolean ok = gimp_procedure_dialog_run(GIMP_PROCEDURE_DIALOG(dialog)); + gtk_widget_destroy(dialog); + if (!ok) + return gimp_procedure_new_return_values(procedure, GIMP_PDB_CANCEL, NULL); } - GimpDrawable *drawable = GIMP_DRAWABLE(layers[0]); - g_free(layers); + gboolean mipmaps = FALSE; + g_object_get(config, "mipmaps", &mipmaps, NULL); - /* Determine channel layout and GPU format hint. */ - gboolean has_alpha = gimp_drawable_has_alpha(drawable); - gboolean is_gray = gimp_drawable_is_gray(drawable); + /* A flattened copy when the image has several layers; the visible + * result is what gets exported. */ + GimpExportReturn export = gimp_export_options_get_image(options, &image); + GList *layers = gimp_image_list_layers(image); + if (!layers) { + if (export == GIMP_EXPORT_EXPORT) gimp_image_delete(image); + return yapf_error(procedure, "No layers to export."); + } + GimpDrawable *drawable = GIMP_DRAWABLE(layers->data); + g_list_free(layers); + + gboolean has_alpha = gimp_drawable_has_alpha(drawable); + gboolean is_gray = gimp_drawable_is_gray(drawable); const gchar *babl_fmt; - uint8_t ch; - uint8_t gpu_fmt; - uint8_t flags = YAPF_FLAG_SRGB; /* GIMP's native u8 formats are sRGB */ + uint8_t ch, gpu_fmt; + /* GIMP's 8-bit buffers are gamma encoded, so pixels are written as sRGB. */ if (is_gray) { - if (has_alpha) { - ch = YAPF_CHANNELS_GRAY_ALPHA; babl_fmt = "Y'A u8"; - gpu_fmt = YAPF_GPU_RG8; - flags = 0; /* grayscale has no sRGB GPU hint */ - } else { - ch = YAPF_CHANNELS_GRAY; babl_fmt = "Y' u8"; - gpu_fmt = YAPF_GPU_R8; - flags = 0; - } + ch = has_alpha ? YAPF_CHANNELS_GRAY_ALPHA : YAPF_CHANNELS_GRAY; + babl_fmt = has_alpha ? "Y'A u8" : "Y' u8"; + gpu_fmt = has_alpha ? YAPF_GPU_RG8 : YAPF_GPU_R8; } else { - if (has_alpha) { - ch = YAPF_CHANNELS_RGBA; babl_fmt = "R'G'B'A u8"; - gpu_fmt = YAPF_GPU_SRGB8_A8; - } else { - ch = YAPF_CHANNELS_RGB; babl_fmt = "R'G'B' u8"; - gpu_fmt = YAPF_GPU_SRGB8; - } + ch = has_alpha ? YAPF_CHANNELS_RGBA : YAPF_CHANNELS_RGB; + babl_fmt = has_alpha ? "R'G'B'A u8" : "R'G'B' u8"; + gpu_fmt = has_alpha ? YAPF_GPU_SRGB8_A8 : YAPF_GPU_SRGB8; } gint w = gimp_drawable_get_width(drawable); gint h = gimp_drawable_get_height(drawable); - - /* Read pixel data from GIMP's buffer. */ uint8_t *pixels = (uint8_t *)malloc((size_t)w * (size_t)h * ch); if (!pixels) { - GError *err = g_error_new(GIMP_PLUG_IN_ERROR, 0, "Out of memory."); - return gimp_procedure_new_return_values(procedure, - GIMP_PDB_EXECUTION_ERROR, err); + if (export == GIMP_EXPORT_EXPORT) gimp_image_delete(image); + return yapf_error(procedure, "Out of memory."); } GeglBuffer *buffer = gimp_drawable_get_buffer(drawable); - gegl_buffer_get(buffer, - GEGL_RECTANGLE(0, 0, w, h), - 1.0, babl_format(babl_fmt), - pixels, GEGL_AUTO_ROWSTRIDE, GEGL_ABYSS_NONE); + gegl_buffer_get(buffer, GEGL_RECTANGLE(0, 0, w, h), 1.0, + babl_format(babl_fmt), pixels, + GEGL_AUTO_ROWSTRIDE, GEGL_ABYSS_NONE); g_object_unref(buffer); + if (export == GIMP_EXPORT_EXPORT) gimp_image_delete(image); - /* Build yapf_image_t — base level only, no mip chain from GIMP. */ yapf_image_t img; + memset(&img, 0, sizeof(img)); img.width = (uint32_t)w; img.height = (uint32_t)h; img.channels = ch; img.gpu_format = gpu_fmt; - img.flags = flags; + img.flags = YAPF_FLAG_SRGB; img.mip_levels = 1; img.pixels = pixels; - img.mips = NULL; - gchar *path = g_file_get_path(file); - int result = yapf_save(path, &img); - g_free(path); + int result = YAPF_ERR_OOM; + if (!mipmaps || yapf_build_mips(&img)) { + gchar *path = g_file_get_path(file); + result = yapf_save(path, &img); + g_free(path); + } + if (img.mips) { + for (int m = 1; m < img.mip_levels; m++) free(img.mips[m]); + free(img.mips); + } free(pixels); - if (result != YAPF_OK) { - GError *err = g_error_new(GIMP_PLUG_IN_ERROR, 0, - "Failed to export YAPF file."); - return gimp_procedure_new_return_values(procedure, - GIMP_PDB_EXECUTION_ERROR, err); - } - + if (result != YAPF_OK) + return yapf_error(procedure, "Failed to export YAPF file."); return gimp_procedure_new_return_values(procedure, GIMP_PDB_SUCCESS, NULL); } diff --git a/extensions/linux/README.md b/extensions/linux/README.md new file mode 100644 index 0000000..04c08f2 --- /dev/null +++ b/extensions/linux/README.md @@ -0,0 +1,15 @@ +# Linux desktop integration + +Makes file managers recognise `.yapf` files (MIME type `image/x-yapf`) and +show **thumbnails** for them. + +```sh +sh extensions/linux/install.sh # for your user +sudo sh extensions/linux/install.sh --system +``` + +Thumbnails are made by `yapf thumbnail` (from the `yapf` tool, which must be +on your `PATH` — `sudo make install` puts it in `/usr/local/bin`). They work +in file managers that use freedesktop thumbnailers: GNOME Files (Nautilus), +Nemo, Caja, and Thunar (with tumbler). KDE Dolphin uses its own plugin +system and is not covered yet. diff --git a/extensions/linux/install.sh b/extensions/linux/install.sh new file mode 100755 index 0000000..426b3e8 --- /dev/null +++ b/extensions/linux/install.sh @@ -0,0 +1,18 @@ +#!/bin/sh +# Registers the YAPF MIME type and thumbnailer for the current user, so file +# managers that use freedesktop thumbnailers (GNOME Files, Nemo, Caja, +# Thunar with tumbler) show .yapf previews. Needs `yapf` on your PATH. +# sh extensions/linux/install.sh (current user) +# sudo sh extensions/linux/install.sh --system +set -e +here=$(cd "$(dirname "$0")" && pwd) +if [ "$1" = "--system" ]; then data=/usr/share; else data=${XDG_DATA_HOME:-$HOME/.local/share}; fi + +command -v yapf >/dev/null 2>&1 || echo "warning: 'yapf' is not on your PATH; thumbnails need it" >&2 + +mkdir -p "$data/mime/packages" "$data/thumbnailers" +cp "$here/yapf-mime.xml" "$data/mime/packages/yapf.xml" +cp "$here/yapf.thumbnailer" "$data/thumbnailers/yapf.thumbnailer" +update-mime-database "$data/mime" >/dev/null 2>&1 || true +echo "Installed the image/x-yapf MIME type and thumbnailer into $data" +echo "Restart your file manager (e.g. 'nautilus -q') and clear ~/.cache/thumbnails/fail if needed." diff --git a/extensions/linux/yapf-mime.xml b/extensions/linux/yapf-mime.xml new file mode 100644 index 0000000..5049dba --- /dev/null +++ b/extensions/linux/yapf-mime.xml @@ -0,0 +1,12 @@ + + + + + YAPF image + + + + + + + diff --git a/extensions/linux/yapf.thumbnailer b/extensions/linux/yapf.thumbnailer new file mode 100644 index 0000000..e911922 --- /dev/null +++ b/extensions/linux/yapf.thumbnailer @@ -0,0 +1,4 @@ +[Thumbnailer Entry] +TryExec=yapf +Exec=yapf thumbnail %i %o %s +MimeType=image/x-yapf; diff --git a/extensions/photoshop/Open YAPF.jsx b/extensions/photoshop/Open YAPF.jsx new file mode 100644 index 0000000..52860bf --- /dev/null +++ b/extensions/photoshop/Open YAPF.jsx @@ -0,0 +1,28 @@ +// Open YAPF — opens .yapf images in Photoshop. +// Copyright 2026 Nexoniarz — Apache License 2.0. +// Install: copy this folder's .jsx / .jsxinc files (and yapf / yapf.exe) +// into Photoshop's Presets/Scripts folder; then File > Scripts > Open YAPF. + +#target photoshop +#include "yapf_common.jsxinc" + +(function () { + var filter = YAPF_IS_WINDOWS ? "YAPF images:*.yapf,All files:*.*" + : function (f) { return f instanceof Folder || /\.yapf$/i.test(f.name); }; + var files = File.openDialog("Open YAPF image", filter, true); + if (!files) return; + if (!(files instanceof Array)) files = [files]; + + for (var i = 0; i < files.length; i++) { + var tmp = yapfTempFile(".png"); + if (!yapfRun([files[i].fsName, tmp.fsName], tmp)) { + alert("Could not open " + files[i].name + ".\n\n" + YAPF_TOOL_HELP); + return; + } + var doc = app.open(tmp); + // The temporary PNG is no longer needed once the document is open; + // use File > Scripts > Save as YAPF to write a .yapf again. + doc.info.caption = files[i].fsName; + tmp.remove(); + } +})(); diff --git a/extensions/photoshop/README.md b/extensions/photoshop/README.md new file mode 100644 index 0000000..7cea292 --- /dev/null +++ b/extensions/photoshop/README.md @@ -0,0 +1,33 @@ +# YAPF for Photoshop + +Two scripts that appear under **File → Scripts**: + +- **Open YAPF** — opens one or more `.yapf` files as documents. +- **Save as YAPF** — saves the visible image of the active document (a + flattened 8-bit copy; your document is not changed), keeping + transparency, with an optional mip chain. + +They convert through the `yapf` command-line tool, so Photoshop needs no +native plugin. + +## Install + +1. Download `yapf-.zip` from the + [Releases](https://github.com/Nexoniarz/YAPF/releases) (or build the tool: + `make` → `build/yapf`). +2. Copy `Open YAPF.jsx`, `Save as YAPF.jsx`, `yapf_common.jsxinc` **and** + `yapf` / `yapf.exe` into Photoshop's scripts folder: + - Windows: `C:\Program Files\Adobe\Adobe Photoshop \Presets\Scripts` + - macOS: `/Applications/Adobe Photoshop /Presets/Scripts` +3. Restart Photoshop. The scripts are under **File → Scripts**; you can give + them keyboard shortcuts in **Edit → Keyboard Shortcuts** or record them + in Actions. + +Instead of copying `yapf` next to the scripts you can put it on your `PATH`. + +## Status + +The scripts are written for Photoshop CC's ExtendScript and are not yet +tested inside Photoshop (the converter itself is fully tested). A native +format plugin (`.8bi` / `.plugin`, so `.yapf` shows up directly in the Open +and Save dialogs) needs Adobe's Photoshop SDK and is planned. diff --git a/extensions/photoshop/Save as YAPF.jsx b/extensions/photoshop/Save as YAPF.jsx new file mode 100644 index 0000000..1ca326c --- /dev/null +++ b/extensions/photoshop/Save as YAPF.jsx @@ -0,0 +1,40 @@ +// Save as YAPF — writes the active document as a .yapf image. +// Copyright 2026 Nexoniarz — Apache License 2.0. +// The visible result is saved: a flattened 8-bit RGB (or grayscale) copy, +// transparency kept. Your document is not modified. + +#target photoshop +#include "yapf_common.jsxinc" + +(function () { + if (!app.documents.length) { alert("Open a document first."); return; } + var doc = app.activeDocument; + + var base = doc.name.replace(/\.[^.]+$/, ""); + var start = new File((doc.saved && doc.path ? doc.path.fsName : Folder.desktop.fsName) + "/" + base + ".yapf"); + var out = start.saveDlg("Save as YAPF", YAPF_IS_WINDOWS ? "YAPF images:*.yapf" : undefined); + if (!out) return; + if (!/\.yapf$/i.test(out.name)) out = new File(out.fsName + ".yapf"); + var mips = confirm("Also store mipmaps (for game engines / GPUs)?", true, "YAPF"); + + var tmp = yapfTempFile(".png"); + var copy = doc.duplicate(base + " (YAPF export)", true); // merged copy + try { + if (copy.mode !== DocumentMode.RGB && copy.mode !== DocumentMode.GRAYSCALE) + copy.changeMode(ChangeMode.RGB); + if (copy.bitsPerChannel !== BitsPerChannelType.EIGHT) + copy.bitsPerChannel = BitsPerChannelType.EIGHT; + var opts = new PNGSaveOptions(); + opts.interlaced = false; + copy.saveAs(tmp, opts, true, Extension.LOWERCASE); + } finally { + copy.close(SaveOptions.DONOTSAVECHANGES); + } + + var args = [tmp.fsName, out.fsName]; + if (mips) args.push("--mips"); + var ok = yapfRun(args, out); + tmp.remove(); + if (!ok) { alert("Could not write " + out.name + ".\n\n" + YAPF_TOOL_HELP); return; } + alert("Saved " + out.fsName + " (" + Math.round(out.length / 1024) + " KB)"); +})(); diff --git a/extensions/photoshop/yapf_common.jsxinc b/extensions/photoshop/yapf_common.jsxinc new file mode 100644 index 0000000..2a77128 --- /dev/null +++ b/extensions/photoshop/yapf_common.jsxinc @@ -0,0 +1,41 @@ +// Shared helpers for the YAPF Photoshop scripts. +// Copyright 2026 Nexoniarz — Apache License 2.0. +// +// The scripts convert through the `yapf` command-line tool. It is looked +// up next to these scripts first (yapf.exe / yapf), then on the PATH. + +var YAPF_IS_WINDOWS = $.os.indexOf("Windows") !== -1; + +function yapfQuote(s) { + return '"' + String(s).replace(/"/g, '\\"') + '"'; +} + +function yapfTool() { + var here = new File($.fileName).parent; + var names = YAPF_IS_WINDOWS ? ["yapf.exe"] : ["yapf"]; + for (var i = 0; i < names.length; i++) { + var f = new File(here.fsName + "/" + names[i]); + if (f.exists) return f.fsName; + } + return names[0]; // rely on the PATH +} + +// Runs the tool; returns true when `expectOut` was produced. +function yapfRun(args, expectOut) { + var cmd = yapfQuote(yapfTool()); + for (var i = 0; i < args.length; i++) cmd += " " + (args[i].charAt(0) === "-" ? args[i] : yapfQuote(args[i])); + if (YAPF_IS_WINDOWS) cmd = '"' + cmd + '"'; // cmd.exe strips one pair of quotes + if (expectOut.exists) expectOut.remove(); + app.system(cmd); + return expectOut.exists; +} + +function yapfTempFile(ext) { + return new File(Folder.temp.fsName + "/yapf_" + new Date().getTime() + "_" + + Math.floor(Math.random() * 1e6) + ext); +} + +var YAPF_TOOL_HELP = + "The 'yapf' command-line tool was not found or failed.\n\n" + + "Put yapf" + (YAPF_IS_WINDOWS ? ".exe" : "") + " in the same folder as this script, " + + "or add it to your PATH.\nDownload or build it: https://github.com/Nexoniarz/YAPF"; diff --git a/extensions/vscode/README.md b/extensions/vscode/README.md new file mode 100644 index 0000000..d004122 --- /dev/null +++ b/extensions/vscode/README.md @@ -0,0 +1,27 @@ +# YAPF Image Preview for VS Code + +Click any `.yapf` file in VS Code to see the image. + +- Checkerboard behind transparent pixels. +- **Click** to switch between *fit to window* and *100 %*; **Ctrl + scroll** + to zoom (pixels stay sharp when enlarged). +- Picker for every **mip level** stored in the file. +- Info bar: size, channels, sRGB / linear, GPU format hint, file size and + compression ratio, decode time. +- Reloads automatically when the file changes on disk. + +Decoding runs in plain JavaScript inside the preview; no native code. + +## Install + +- From a release: download `yapf-preview.vsix`, then in VS Code run + **Extensions: Install from VSIX…** (or `code --install-extension yapf-preview.vsix`). +- From source (needs Node.js): + + ```sh + cd extensions/vscode + npm run package # → yapf-preview-1.0.0.vsix + code --install-extension yapf-preview-1.0.0.vsix + ``` + +Works in VS Code, VSCodium and other VS Code–based editors (1.75+). diff --git a/extensions/vscode/extension.js b/extensions/vscode/extension.js new file mode 100644 index 0000000..d2bdb19 --- /dev/null +++ b/extensions/vscode/extension.js @@ -0,0 +1,65 @@ +// YAPF image preview for VS Code — a read-only custom editor for *.yapf. +// Copyright 2026 Nexoniarz — Apache License 2.0. +const vscode = require('vscode'); + +class YapfPreviewProvider { + constructor(context) { this.context = context; } + + openCustomDocument(uri) { return { uri, dispose() {} }; } + + async resolveCustomEditor(document, panel) { + const media = vscode.Uri.joinPath(this.context.extensionUri, 'media'); + panel.webview.options = { enableScripts: true, localResourceRoots: [media] }; + panel.webview.html = this.html(panel.webview, media); + + const send = async () => { + try { + const bytes = await vscode.workspace.fs.readFile(document.uri); + panel.webview.postMessage({ type: 'file', name: document.uri.path.split('/').pop(), bytes }); + } catch (e) { + panel.webview.postMessage({ type: 'error', message: String(e.message || e) }); + } + }; + panel.webview.onDidReceiveMessage((m) => { if (m.type === 'ready') send(); }); + + // Reload when the file changes on disk. + const folder = vscode.Uri.joinPath(document.uri, '..'); + const name = document.uri.path.split('/').pop(); + const watcher = vscode.workspace.createFileSystemWatcher(new vscode.RelativePattern(folder, name)); + watcher.onDidChange(send); + panel.onDidDispose(() => watcher.dispose()); + } + + html(webview, media) { + const uri = (f) => webview.asWebviewUri(vscode.Uri.joinPath(media, f)); + const nonce = Math.random().toString(36).slice(2) + Date.now().toString(36); + return ` + + + + + + +YAPF Preview + + +
Loading…
+
+ + + + +`; + } +} + +function activate(context) { + context.subscriptions.push(vscode.window.registerCustomEditorProvider( + 'yapf.preview', new YapfPreviewProvider(context), + { webviewOptions: { retainContextWhenHidden: true }, supportsMultipleEditorsPerDocument: true })); +} + +function deactivate() {} + +module.exports = { activate, deactivate }; diff --git a/extensions/vscode/media/icon.png b/extensions/vscode/media/icon.png new file mode 100644 index 0000000000000000000000000000000000000000..1ff64e699995c4b7dd2c410c6062ecf45ad0fb17 GIT binary patch literal 12341 zcmaKTc|6qL_y21aGnSWKyv8vT^t6o zX6bWnqi9@U!3CLG&n)~yMhJs(%jzDC_zu40ha5t9{WJ(FAGIU~ZfWC7Lai?ooM|40 zz&;epTiMMGy&i?~^F*O^5>cr0wBZ`o^9iYx!I4s>%xHXH*8>u`vegD z$jA_BSVpTetr3VKTf+5VB*LXm3BlGPhSu{{l?~Ru0;iCq)&99H{c4TOJb=H~ z;Y+Ani}C87zO+kGcq6~i4F(?La@^0cw7Die{mvDZ!66-l@RK-Cb~FeMEdKRPfiO>i z#ffY_QrR1SBQB?!?Z44KoCQ@Pq}*&Cj~JYli4V_T!p2q7n!yYrPn z0FKxr{BFrteSU@CqTg-)P3#+Y+lq+nlsTo&(htSHPAoqo2V^7!(1wq}ew@88(5F`9 zd-2=Qdqi^=-gu)pj~Y0meAON56e5D%lHk^|h0pxX-s-sHSFGBzq`H2JH(bUTuBd=Q z_1pu&04XRaXGY3Vl>1Hx!D1w^cK3sv5G7_BUn6w@IETKKa@Fr(4!~SaXz9$zxWlqM zCLlU9QB~Kb@>n6t^w;xk)Jf6~ZPvAnjC#IXj_o~osLDwNK~VD3lt-&8{!>FB2B_PC z-4?KZ2-?k(qirCn_&dRz(p$Cf!a!X!HKYzrX{AqO`t+|}UcjXhE zVe|~XtX)x^R30G|mZ>0j--Ay=kh2B^>7rk}a(aO>=6%FqCN^E}c$;p?W`(HH6PRaA z4FR`9_P#)!_xS2XpQItEm~i8^-lx5$C!nsB7S4J7VZ2?hky9zu{N zIO9rrgd}EXTW-u{g=m^h{wx9N>M<+n7fDN6nvpcoNECA`31L=KW+{*zCOpf8{igw> zU2l~xwMoX8&vD2Pf-IM02BnM@l`ZMA#kMcUg?!%1RaOYq$6%Kc42a#*tLN6Dv(){x zv^=nbPl>}xqr$@GEY`{i_((pm0mdb$>W+J4D-97$cDvQOCKa1Pw6iRu=7En7Fp11Z7Lh|wB9BJ!G`z#n^{ zE+uV;p;I3BN7CXQ{ncU3aujUc+Uv#H$F;lJwMF}oKPDUG|mA{oAJ4CYA1tTe-#UVpf#O`D-8F?@y$U{vr`cge3TG z{3aqXS?Pzlfx?Zoy`?Ncfxi_622V#N#n6WK6dM{E0;KttixH{|Qgi6KP5St!E)%SFZR0R|VnL zqm*HtF?a=Ku!!}yalE|hS<$-v40!STv+_+3wyszuh+U5UT#BNFCrt|8F;J1Hqo|j& zAHd)TfuW-36M0T!q^e8L+eNSFnITkTz-@b@lK;}?j#VTor^l=8Si4BBDx2Nqo+8Fp z2D}1(s^L%fa>q^$?{z57%)29&Q~P|`WyFfoneu3X0Ygt&{7*)$NP*~^e=YjCc$pZo zuZKV#N62wNl+Ur+23F1&i9d`7RN)f&RRtkB!kwIdlm$Szp5s)Vz$`}+Q?jc`*{rh6 z{c<8Kz4mWPiN26=infTkH`d}2ZrC^$S$Q@d{wnO-78y5=kiRwY7E4Eik_H$hIZeDy z*7h=$gPUHkU^y&JamBwL+yO-!?#hOhSp5$5Z>lN`F&3`nEP@Ex)f-)eFn*t1Y)%rX zR~H4@4<2+$vNAJa_HVhd_FlR#G=Mv-_us2pFf1hS`kxl-nfH@=BC%iQL)hqh`9nQ# z_{?lm%8bA{OLbcb%vTL}hcP{(D*PT?VDQFW>A0K7W(Ez8K{qXH5$zYI5ZOwr2Qn}i zrl_9Ax^Y!(VF>6X2g)8f8N8}1ujEtHkXzOz+!=BSH zZItgSZk>N?wKF*L{9VR3jfljA@1c}q8D*2z5$Nf_i+^{9JHnPTzzopsw;LFf5o4Oh zg?&{@L>=L@@n7?b46;})o7c8_1N>r_p3k;a+i|UEE*jFPxb(Vg!;YZ=hJV($o)2@d&4|rx$IWy=@8O*L4h_D#FZ0|@ z6VhGe%$hmK4vI#au7$M@s3Oc-7?qVg@Xv~gfZdd*=2HPdSLIoXTSN(p+YRw1cK#J1 z>8tMA!+pC0eew7xddxM8hDts&*f9vcHRER+C8Pnyw-9&bm8^b8c897|Y}MvPn*;VP z*^MhQ#SZfd`~MR%^G3_IZt2iXxB(~l8shJyRFAh=5^JNRQSlbc7}$P^@&ia(}sMH)t1o=stm&1#{( z4U}S%3w=oaar@s$GN;HDhc3u$syRThtorJqXahPTJjhgor->CFu1CKQy;D}KkVgMt zeTE&1IRt)8JP=018LP??M-iJZKEpV(4x2x)e13heGtUr~VEjt8Av_#fMSRh>Qoig9 z)1w{qogJy-oI(TPVRK8=^nHy5W+%fqlIBc{oP&9(j3*~rI#?r*z71vXHkDA^FCqJw zs%EORA%Z&zK{);`-0($hc2ss$-Ry|IFiT1%s>53C;;dw#X=WO2J$pS{U1-G_f9HV= zHvwoD8SI+4i?Sb=(B>6$YYkZCY8rb!y2fEw>`|1scs&&&_T23%P{>o@4xtm}{1#&$$AV3wt?-I2xSCFi!xNE`@0& z!4IBPDWbxGdvBM+V>%>Sc0I8w(8QgMcqr(2n(%9uHgOezeW=~sg!J+tr8?DZXDqh^ zgE}Gsm#z5IxmV-EA@tfi+>PgY#w zUISaL-&OlB&UmOIPb>KLT>9+;E|0SJROMg_61fh($aYeD`D2)@yQ9YzD;Fc78Vd88 zbJ29a3cOhEDbm4tgUa4%EBw~Aq(u`2w9xw2ASMM*T7|T{U+(#K8N*45e8lAfY#Tmt*T8A3QOC%RZ=FosNzw`ek zxo@&1*s_6KK$s%Ms{G6-bme#9x`=Xi$5fiQLW(iZ6b5)QhvblEFAXXLDkCqFS-W;| zSZQMNcBssDGhC?}DBdw|?tE{4li7+%_X0O8uGv;=rTjS3wYx;iq1*TzMd1TLG zUyNr3t>e#8zew`sGx|JbLNr4-#scG<$$>G(gPrdYotj~4VRdifuhpsngMP*w6i1H# z$n6vuyv+jjzwCzakqp;*jzypP9n*8~ug%)X%14Lbn)4ap#owa`_PQFn;N{A6c3&fr z@Ke(3Cl$)>9kLf0$520xlMk)eF;c_uzPb&=u(hs$`OSSskfA>lO4E$s4AZDejHy{g zVSVWcWRv){RWeWhB|&@W-KRvVa?WkFhdTG{9hRbdLF`>FQT*lfkRnMG}3<1qz+s**4ID!k?6j-@Q?y|`AxR>?Bf z`l0AzLO26|XIoSy!30KqZbADhmEOT{uTI@KrN?izV9GB?(cm*Z_5R!s?E+Gl4@!I9Qj_PeH5^WNMmo=Z#;w>;1R zAFMnNotu&Bur<;(T;m*$Z<=KrwMht=3n%Laq0<+Z)d7{D1 zl_fzn&|cu9UM^SG)EmG(V``aoV^he}7r^0{BjDF!uerpus z=+i=7K|9Y3{^`h+c#*-$`;!7eeFMDdqEA=)FwbR(S%zv+s!_nkx8GLIr}z9qeGlb2 zobcZ;SH%?fjJwkxI?4UEpnr|LNihz@SRtA=RPA**3kTlMgWbMq@7nOx(<(Qb_?KlV z7pvNSS|VWuIkJ|4Mu|14>*RO=gvp-MDRLe%e;DyM_;04Qf~2tzElGuIt68x%hGZ){ zKcCODm&I+1j8xeRPvBB3*DJ^Q1nlrg+Tu&g1Q8kOr;v`+a@0aOMtT&CO@HhKjr)zk zjPAECxh}clR=-vn=6N3MmX8}sHBvS2`Q&7mXI`>Q&eEh*;r7*r%0rZ|Tf${TNTwlh zaW_?KXYn>&2V=>MG=L*)P`>_SXLNXtRSn{fm$7J%k(Ai)w0oATT1Pj$XrN-};ZVTx zpqzoL^r$X~tdFck#QW&};3=&)g;DgT6bd!p9;wF|&Z#brIA^R_diAB(6(By4xmuBt zsl*@2=fb3&!C=Zb+-JOsEF-5-G3Y=gs@y-i`05WoCaaegpN24kPVPOK$8(S>JUv^a`Zk}kT6(bW7wkb*2B33>qK4gj8Wu(l z&1I{O?ve2@FjeKBgIDAOIK*>I(zb1YhAkX(AhIKz7MUZ6>2X@>Y@EY>8(mnQB@2^= z2B`3+14Y+ZrRwi7IEL1yDPk3+w{7v56mwG-_@xe*--(>(ko-eCj&$I<%$04XgagXp zs|Nk3_?bmazYba#dC38#;oN;xc@99Iwu#?j&#_OBQv zw|DRN2k8cZA7Va$x6N*YquI(N*$5EuL<-f7?6Z~MUgwMWxL5o(iXqPoR_K>ZO#O9% zNITLdMQbg4W$^YYIt%_D9$*&CEnnEI{Ek0mOF3}h$A7{%y!Dx;#l~>Y)T{y457`4K z^a`V*dX`p3tK0X?@lWX7lvG(b6r3dvz5V;Pzuz~WEfDOSek#%}5_M+p8BX-eXpVlh zK8kuN8DldtrOSCBRlUApv>qE!$c~)Xm!9jO|LDSS?vbC6nE4#`dnW=^BH|^`br62T z?52q}CtZG<=nZXa3>?=WFUlQ?uwGx9>`=z;^mU8=#eCP`1u|F@cGA%(&e>pC-6VkU z=B&B((vo>w*2&zMV;yj`uK#5b8KX2XFL%&P{u`y*`O*QUzyTrhbV5JO8#l`^;G{U& zZ@+Nhg1DtiC-pIc2rDxIj@TUGoXMGjD1%iT{hEmO2D7NX<@2`wvLEs(K0ue$F{)1* zOOCG^X>QO0aiI2qJGMBK`Pd2Ks=@AdHX^UCchU#9X3@C{vPN)%NR5ONDY@c(VV~rf zJt3hRiLOS+O#nBu5A5v-HxA-SBGI*BLCC|Ss)GQ3nYzQV#F&@U*9_WOMXUlu`Rwu& z2DocR$lss3DHJNt5`{W(8^+zb32Xsf z;gf32yvrKwVi*1JxGdF_JzSKSOq2-`YR}=$npbH=E4!M-DQ^4`8b2|i%zlU}_r6v& zVuh$1MStSU{SLNbL@@8Q!X7Pro%<>}b!@Rt(eZ;s^bw(9%TnWi&N__Y5@1_k{9#J{ z8jYX43FZ*_5}U9x%V>29&WL`P0)+uQ%S5yKt1#{A(FIJ;M?-igdV@*`uNK*jHrETo zDRD0YQ0Hr=UZTIt5ILx5g?|;G?x`A_n!APt{YsZ3yYrK3n<9``FxuSP7hc#V0_iJQ ziENZ8N&g{2k#@}KSo|`0s90~|gzR6XZtcxfHkFC-?PRI}c4}Nybhi#G_nS{>-NMwm zx;a#)n5~Fd^L`Xz=aB}yd+R>qP^Sk|V;aHx!Zl9l3-(qBr-@0MWEAekY(RcG?@RoEE^NM%p4Khg)J5C)>=ST|maRc1tvDrhAs&}*EWN4|b!fVs2h zQHR<8{$cEbggVT-9e2T%q!s#ktTZvaRjWOUOS|AT%6!W70#j%15g!ukj-tQE_@B=R zg6poFDilvObb+Jo4OKUfMjH-lBb|HL8uL<&>|JY=F_dv5q)!dJe;XE}rB2s^(XZ(V zkD%bpVI)7g@Cs;lxE@nAuXe+RnFltkkVqnJn8WNST3GMi0A&pKQN6|nrqaZ{r<*$O z(UyOwHkUkp{upcxlf1qA7Q|(AJ=kBoAAIogIefi1Vo;{*x!@g}bKWUiSzq9y-x5Ki z&1L1MO9?v+t``72C!Tb4SNG@-gP-V6S;eL+Kv%|GC*`sl5Ep45{X$QGr0lol>eap2 z+qPcDidXjaxs-b!i%?shQdUPGF4`gDT7eu&ehKAT(D{`K3t{$`fsl)$X~u^xNx5qM zSjvXCPDYLC{v$ZPM=}c`l_i$x@1{$Q6t08PTjIvh0EZgEu%Gt*L zb{|GhN0JQ*BTd?6d*ZKjVjS%H0DnL!SxnBTwrDNm5;Mnx!p!F4kn zwpt-p$JrjG&m-7HbIs;^roz^`U+1_=ZXTzV&!^4Zl9oye0OFTaPu$%mL*VUgK*rXh zs;pZLYY9)?H%-u53{GUpS-Ea)FuQql4%FgqPKeZItIGv^~++vPoi<_K8wKg8v^C&q( z4>&(laM{3vmvFD_^Qx+Q5=go%pk7(hu=dj~WHz*R0L#7V7Go}A1WMeZ0+X+&B3cm2Z9Izua-#W?4~14RoRXk|s(&Lpiv zXNXK1N_mh^2cUR2J9Xnu+^4LZ&6WfiD}nR-=MnMkfKNu#nh7x8GcBO$o*eEal1sFG zwPi9TH4tUiJfckVlYGHOqRcg?7xdy&@osA*=)+HTvfnesk^&*fYtffwznUpy&0_1b zn^P+vtb%bHr*ioiDhBZDQBr$llA_tRT5E__Y8f?l{vyBnqzTEQ4ojLY6RZ9^BWJ~h zS>PODlC`5wUJF?Rt{MlE+rKT-LPUd3h5&Vy8H&Aaid7JfV5LyW`8lCvNf^xW>vqGP zWlt=%Yk?!BrvqOxsbR@?e zbOqGZn0GhT?*Ur%jwM!+JYuDkPgHRG`b2fCRl>5|e?6(7#=%0j_$oNM36*1aU zj#(GPDI}dyG0oX*RO@p%Wnqn@F<4nUpj5SMdJBY{DP%?Qhc7rDJ$d z${=ib$#h$X@a5BvG+PbIqQZ|SWTP*;*e;S0wUvlAE9cvKbvuZTIrVn`5F58dCf^U{ zYYFCtX_~@!f#G#c6hoXjj4q$Fu!H)@BRe@_Sb4D$+qBW*Wwq7PtJGlzQeD}9mLYNw zxwpQ?qVNP|qa28fm(%=@=i!=VSi$7DYg-d7JHjaUl%#~BloC@)B+ngB{Y@Bh=gd-C zDcg}@yVA{0rHRc>ssji%u{Cy%cuHZXdizLl2N=vOghe&N+|;61=E#S#U_ryrcSYdr zM?%sKhg&J9?DA!kJclmgV)Ub$@UosN(Q<$Fgd&>D+oW!tG6xyPct6$=wqt4GCxtJ}ma`n>!(M7AV^a_t#!GgWV!v_9unUFv1-X zmwaW~F3)Y3q5fqwd+`1#*A!8IGH?Gd5$Yc@0bu8C;$Qew^|>EI@LDjH8u7Eam&Igj z3-zO0DAVlrs!lG^zYi|MBt9_i)k1?CNCrt83R*@l_4OEDO&6w4tH!h*0lC$% zVSSmi=QDTqGW%r0r*Y0=?*@?1Y6Pm zydGTNBQb&(25JedQEIT`d-z3%`MGPIHsfjbYl=3w?uMBvdZ19 zgW3ify1k|?X%?YP*CsOeYO$wvjzdoQxv=+F>4n5R*#e?z^*T$YO=>*-9I@51-I94J z5rzS-Ua|?bf*-Sn9f4qa)0M5|tw4EoCyVbQr|rc1`7OZf3H90eX>|`N`UX4{`2i{DfA0o0qHj^H`B-8 zD|7C9U{D9m4LIw5%A>p^MaEF#X;gR35y_;P|4!v2kJ_u+h*f|=%#$6gXqZbf1?h6d z1H;ySrtR{?)s`(^BrkMLO9KPF0>Pl?L13`-yo?3za!59;3!n6ub6FwZs`BA~R(Y~{ zM5soG2AvZ99-b zRudOkEE_o>{MS^jA0&GUhEaJB^cy9sw-=3@fj*88Es3}_|5f>lyEA>j6OA$8!Z=!LmNCHgOk(iQ*R3btLz>{IN-O`QT8QX-{E6{bn7qC`=u!f?u}bHzNs4R?}<1U{Q?689oGxeeZ-9lLM^vSf%sERJ&Qk`ztY-quO${FMu zlh{pyB_zbbJb#yeO4ayOVW`P_92NWtDoG3pqa5={l~fUn8FQD#wsarN`l`(TZyLUbt1yK@Or$8YI||#xmo3 zbakPw*`SO;8%<-5PFEuF`7-E$8UC(}TI+O*Q_BBpd~5uC*xd47j`+o} zKFU7GiF25IpEdhlcr(ptpAm*)l32W$cnMaEm#zJ+Q#*l(OF~B*b^kYhi~no>$~@*r zLNVg#W%B$KGEmBXTzFn8)l0~+t$d{PbWu6(y+ZUIyHQltX2xIU7uk=}z$oCs@P!4X z^WrvTV^PiN{8(@RD%)IMz18|gvs8{#56*+V7$_QJg?o6W;u8k1^g@PIZ+SQAl?4Iu zotf=aIAqJlOMOt?jrtw2tet;o8Zet3ft#3HiEr~rIcnu^6y5eJZ?@4KR^{0(JVk_d zDWVjof+AmZ0KcYIY30W4L%A+9>mDb=1Kk8PWIrPAig-)C*gHW^GBy>A`VNQ`+m?QT zpoW_@3pu4%m$yWq>lu(2gK6H2S*J|xwWUQWbX#TOWHZk;4)3o0PT#H8C_v=lDo-8u zh-@kVXpw|7*rHE9S4v5N5ihp?u}_@dcAWuk zLu_CzFF)b87;K&*w>s_0sEp*fem=Yhg|b|QeaRE97e_aZqFS82-3uy38I$P4(1$|B3p2)yMHpE(zhE+!dPlcNgylgnj=V4?v0V*kuZri^K zMR&S7RU+O9c?NYtXuR4{KPzgS|{ZC8%Wg|O@CMfPZ6`xyTkc`}NL z_Biiu2roR`9c!MEHN2l{xGC1&%8qVNgsOCywrNQ=*GLos(q6S}w*vLyH)2CbUDMe# zvHFw=s<1@%gDhvg^uDO%@7AoCn1q;6EzD#1RN^LjXBT6W3G)dV8ZJHw*=;?UdK)Es z6YV9_8|I3+1V63xYX2nZxpbyLu>S&E5SYyl`JhUR9s5W|9XqejaXq(_GftVY#T`{B z2zZdwguM(OOip|bWbMBZs@?YpQ&D&W(zVlt zH2l{{}xi*kT6o0Llu&NO_nT$JJ-2%l%Aw`k5| zp*M(e{A@_P*R^z`ZDjn9UyWC|yWL;6ViDqy(9yfR`_?*jmz6iI9bOWaf}TE!k@WY2 z;qg=7v1}Y!nSCo8a+%710?L=`g4l-T5E^`ecW!JK$7;5L=)Zj(i9Rq9LWGs{AU2&3 zDA`eB;EGEQ87sI4@{ES>4c-Hz7{w95fj;C#J5&XLI-(1^rXZ{R2XU=WZj)tGv9LJg zQ$Yln_Ix!?WCVgC{UM<83Sa?~Ru>4;dedNeW2Fxrqq?Jdqo`4PqwFNJ02})d%T~Fr zBJ7p=2;|CspT_+C{w;YsFr7T1YS%N?B$9JL{k1||27bzk1oxLgGPzi2Z_9ii(v+#Y z9pDvq4hS2Pr6W!6!=RyF}%JwyLpTww%KjS0LlxQh{VnU4fp+ zydYMk*Gwu)fOQd;N}R=y6_E7+htxlRCJXj11+51jpE3#mrdPW|S3|&q>37dJwEnY> zw$pmojJURT?#F-*DoiPiqfYi_qPdgHQ(>f^Xz$?&sIoSrUwi05X1P(VXM?y8!a{ z-v^aQ8E<&46|MaC)^_^HfF5d%+Zxn1w{0luM#5`Td510RTtV)qT!Rezj3pbujKyr3 z*TekAm=hf$cqH6|_bsS?WBsSIW{AzD1`~gs9?GY8-tEW`;g?5_d3-@xBV^SOpC-_D zHpq2)uJMJ?%asM?7zE`Vy?V9s082?B178%1`4|_+87b(%4-on@+wWdT99A%2qD^;_I@eo4=*Mm;YwgTRjX=jgSZvD@Q%vbsWbP*|k2g|FY;$iM%rfc1~OKE4%7 z=s^$&b~%(U)s*aoqMx>i=;6Owbe&B*57?pb_^fRFMh3Hh^3JltGLE~s!_&)^su_@Y z*pHYRePNg8P~3K6%NPVLF?$E<@;ywKZRq0v4%%q9mcR z==;B4yrQ5f``~jc@WN}%F{EKDL zbPdgX0axgsa<#4;+ji37!&{lYA}XQ;eyIK;%So&6W<*O(G$J5*$P<~YCX1lY990&Q zI9ow0=I|O044JX}VSvYhVnNu-agU&$c>g=7_?a?%B+4k&wjf)6`$4Y1 zKl4^j8_en!p8kfSX*aY^ruax+-E1yv-eNF1b~>3wZ#hTXlBi>M=rqPIKK7EBC^`F) z4f-)dr4R~3t9W`YFCh+!377>^Mw=`OunRxfA$oCLS5q9PR3x%B)C_x7mAN~PW_F1 z#&k16{46;qq#xO-D!hP6$0V+HpUW|?94%Q|SRx4jWVS;q%P*U76-Qt5JIp=uzh+o( zGvizR^7e_cqpCF0Pq)kQbTDr6G+mXf69~VAU$+cNmB_s?bmlPq8OwP@)%?H;@p)ZM zl6WxLn4=YS)GC9*x8o&74n>4?XMhpisU0YNmt#Aw1b9ro*5SA>YLNniH(d~a|AYn! z5K&3Hp?y;{sdZB_;bI_`C{RhnrpQbO#$= s_M_eAih5f06>> level), h = Math.max(1, image.height >>> level); + canvas.width = w; canvas.height = h; + canvas.getContext('2d').putImageData(new ImageData(YAPF.toRGBA(image, level), w, h), 0, 0); + layout(); + } + + function layout() { + if (!image) return; + const w = canvas.width, h = canvas.height; + const stage = document.getElementById('stage'); + const scale = fit ? Math.min(1, (stage.clientWidth - 32) / w, (stage.clientHeight - 32) / h) : zoom; + canvas.style.width = Math.max(1, Math.round(w * scale)) + 'px'; + canvas.style.height = Math.max(1, Math.round(h * scale)) + 'px'; + canvas.classList.toggle('pixelated', scale >= 2); + document.body.classList.toggle('fit', fit); + stage.title = fit ? 'Click for 100%, Ctrl+scroll to zoom' : 'Click to fit'; + } + + function show(name, bytes) { + errorBox.hidden = true; + const t = performance.now(); + try { + image = YAPF.decode(bytes); + } catch (e) { + image = null; + errorBox.hidden = false; + errorBox.textContent = name + ': ' + e.message; + info.textContent = ''; + return; + } + const ms = performance.now() - t; + const raw = image.width * image.height * image.channels; + level = Math.min(level, image.mipLevels - 1); + info.textContent = [ + image.width + ' × ' + image.height, + CH_NAMES[image.channels], + (image.flags & 2 ? 'sRGB' : 'linear') + (image.flags & 1 ? ', premultiplied' : ''), + 'GPU ' + (GPU_NAMES[image.gpuFormat] || '?'), + kb(bytes.length) + ' (' + (100 * bytes.length / raw).toFixed(1) + '% of raw)', + 'decoded in ' + ms.toFixed(1) + ' ms', + ].join(' · '); + + controls.textContent = ''; + if (image.mipLevels > 1) { + const sel = document.createElement('select'); + for (let m = 0; m < image.mipLevels; m++) { + const o = document.createElement('option'); + o.value = m; + o.textContent = 'Mip ' + m + ' (' + Math.max(1, image.width >>> m) + '×' + Math.max(1, image.height >>> m) + ')'; + sel.appendChild(o); + } + sel.value = level; + sel.onchange = () => { level = +sel.value; draw(); }; + controls.appendChild(sel); + } + draw(); + } + + document.getElementById('stage').addEventListener('click', () => { fit = !fit; zoom = 1; layout(); }); + document.getElementById('stage').addEventListener('wheel', (e) => { + if (!e.ctrlKey || !image) return; + e.preventDefault(); + fit = false; + zoom = Math.min(64, Math.max(1 / 64, zoom * (e.deltaY < 0 ? 1.25 : 0.8))); + layout(); + }, { passive: false }); + window.addEventListener('resize', layout); + + window.addEventListener('message', (e) => { + const m = e.data; + if (m.type === 'file') show(m.name, m.bytes instanceof Uint8Array ? m.bytes : new Uint8Array(m.bytes)); + if (m.type === 'error') { errorBox.hidden = false; errorBox.textContent = m.message; } + }); + if (vscode) vscode.postMessage({ type: 'ready' }); + window.yapfPreviewShow = show; // used by the browser test page +})(); diff --git a/extensions/vscode/package.json b/extensions/vscode/package.json new file mode 100644 index 0000000..fd0070d --- /dev/null +++ b/extensions/vscode/package.json @@ -0,0 +1,29 @@ +{ + "name": "yapf-preview", + "displayName": "YAPF Image Preview", + "description": "View YAPF (.yapf) lossless images inside VS Code", + "version": "1.0.0", + "publisher": "Nexoniarz", + "license": "Apache-2.0", + "icon": "media/icon.png", + "repository": { "type": "git", "url": "https://github.com/Nexoniarz/YAPF.git" }, + "engines": { "vscode": "^1.75.0" }, + "categories": ["Visualization", "Other"], + "keywords": ["yapf", "image", "texture", "preview", "viewer"], + "main": "./extension.js", + "activationEvents": [], + "contributes": { + "customEditors": [ + { + "viewType": "yapf.preview", + "displayName": "YAPF Image Preview", + "selector": [{ "filenamePattern": "*.yapf" }], + "priority": "default" + } + ] + }, + "scripts": { + "vscode:prepublish": "node scripts/copy-codec.js", + "package": "node scripts/copy-codec.js && npx --yes @vscode/vsce package --allow-missing-repository" + } +} diff --git a/extensions/vscode/scripts/copy-codec.js b/extensions/vscode/scripts/copy-codec.js new file mode 100644 index 0000000..7f4ab2d --- /dev/null +++ b/extensions/vscode/scripts/copy-codec.js @@ -0,0 +1,6 @@ +// Copies the shared JavaScript codec into media/ so the extension is self-contained. +const fs = require('fs'); +const path = require('path'); +const src = path.join(__dirname, '..', '..', '..', 'bindings', 'js', 'yapf.js'); +fs.copyFileSync(src, path.join(__dirname, '..', 'media', 'yapf.js')); +console.log('copied', path.relative(process.cwd(), src), '-> media/yapf.js'); diff --git a/other/YAPF.YAPF b/other/YAPF.YAPF index 35e0272666d1311e4104e4d47d01b2fef3d6a72b..5b8fe1258649aa49c355543732501e409df26731 100644 GIT binary patch literal 91429 zcmZsE2VfM%_y60yy}eBcxCsG5(@jW1NDvRif~YK^NU>ZHFckGbdMu%rpiV-7P{acf z5kZflVpKe_5p0K|Ulh~`C?cRi#fAlJSdjcbZ)Okpe*eG8?aZ4uedf)ZH*aS)L%N50 zYlfle1j81dR&bI=YvhM}vqp4|kLX#r7sF-T*-r{Qex!K`&oYF!X1^C_A#jDj36p4j zfapF1ypcpS0Dg`$nagTKG`u0v)&@l55cUC{d5HTOB0LNCB%abVqGytc8Ug;^AUc4s z=E&oDJaPEHIEUy?q`wT$BMAEi`5p&8gl99-lmo9L&Oa$c4J;xPcrKrxWWfFK%)%4> zpZ1+YLQuM}rP1}R|JK)E5GjZp&|~c*JI{|dBtaIkb1c$i6rXLyPS)4o9Vv;FMTX%i zp{_fZd|`W zY&3D**hHmG%et|fen`W>i&s+&ovfc4XfW~Fj%~rK%%-lDcHu^b-KvhTodM`<5=q>* zG=y#Qu|!;oT*+8z;``x<1-k5B&DL#u>%sa9s6Z>wVtS3P=SK>tKxZ(PMFmmaVLv** zfOY~`)1JrSlw}LhtFVMyNED`O)FX<*lHm%tc_gAbT!(ZA7_y?2%tPhl9%|>yBUvyq_Q6|esG7y~ zq#uqSwRP0-j+2*Fav-^%ZQ1}QUjNZZi=Mszeuqhe&6DT1%TrLd0%4FC@ou3bn}*im zr-BYXp6fyJvVIJ0e>{uC*oB!YnC<$U{%1Tn2e}e78B7pA=3Zo^f&@t?H{XXC z>t%Y(Lsgcvy;4snQ%Q9~2NUVKjX49lr!p!F#H?5qx^)pP0$5|kW=KzoB|Cc&6sN{l}+uWE4@;Nvj?{OFg65wr8;GnOk>vN`Q3P$4n>5uT*oF7jKZ`y6Gkr z40e{_Hn#l(Gn<`1>NWUbe0U&0x{}dYHT8Ed3(U-#Y&nsNKx}fP%qp0(nF2Mwqi;la zcG!vs1gd8>aA@aQL`t=zY-0;0rMpMw3t7)gx>Mco6nzvK78!G`ZJp76-Ip}zDXveh zhS`%m_-P`aKcp&1#*QL#YPEy8mf42h1?V`h9FzS?#*?<)eqJIMRi?dzA9#?_)|gXcn~+e<9~`CT$NGB}0S97} zz%~lzgQ}Uk)=a0aff_4+0XT1e01_8y*N=lxbJ+6{A*)O>W@gX*0}0-eaG$^CdI{>f zwfQXYx|?PF!1p3!e`BiGweG*Z0&)?6olwv9iS!hWUB5`ht+@SfKOiCERS$DJ9V)IK ztAc=C@qG&k8%W z`n8KB#@@0U?iXI)FwHc>bMREXF1vdiwX|a8)G{xYY8908h?%h}kMorAc&NVw3gHIA zdC^pVWdWVK9-6@4tQ_>cq;pC?I(C9W-8A`->NV@&uq6src<;+zM{d1wibD0gvmUb*#yT~?E@W|$rITGMtkHXiRK@7vqpGeHoqVYboAG&A#gUTL|5)il z_e^YAq+Y8zsKpLyHLW8DV9`#7V6{qEF$sAHc3MVu(#C2STX$_k6H-XggzGw(nbzq! zIy5!rhcM=^$!GRO_YkWx1X0xluhRZ-qWT!zE^6rYKKSwZ}uFIk-p1v z=O)!ASrGNak3OO)7l18Jwu$yF!yM8wgq##RR`KM9iGRfY4j82vB8&Fu>QR}PbPX7d z$=W)}kuH2NyVPh>ghg#O_1Oyeo}D1lh*rqKq7Z};%VW>=ESSLjByH-V9HKwHl`jf} zn}fn2wR4dP76Czhj3m3TWG@ivQC^D5>lk~;ZO*nx1~*Bt21{JeIAxZp$aGA)E^;VF z058rB&J^UpdjZxC672v{y+>@n$uy!H<0w&SutD`>w}=D0*q;EbQ>AbhwLg%{vyQ6zBX2(pfr@dbCuV;DHk8YGXZD76>Pm}oI8-PXI zrm(5|Q(UDYStwnj9LCcR+0s$g+gG3Q1TT|~fUNyAknTVxTAJ~tjId)fJ6%M@o+b>`n7{;ZO$=rk%}k`BbC zQJKHpOebk3#`p!ZGgoJeS&{MD9u zf(Tf2(Lf1$&H0f79ZvtE2RcNr_Vw#!v>TKi8Kh6RnDlT?rZ&cjj)~}D(9CqLETU-v zO`93h92QKpmhd;XBhp2NF`YeH^B9mY&@l)?pYSe2a62p9?W~2Ny0lJ@W?VlZZ(`8b zGm5Qt+OWK}S08_R`QUy-SdPo4yn8ync=Y3DXajQ-a>|_@qxMZ5#7&Qg_|k``&#u1# zLxH6D2{bG&HvGN##s>(SO`5UsyVG`TV{GR+~zX)*+ORya?-|BiuBj!OyJ3nw6SJoemxenSEy z-M1H-;`sZ24jmZE-@bCd1f-%3!+(Cd(e+7dvH`m%U%%w`j&B=)V)xuQUqt6i@)8=) zF2S(-_cuvP&#mG*)0FXlRXrx#w(sx!>L+B>Eh$Hje}VKIbzReYo3&m4<6W~5a^z!) z9~)#8jEH{oUn%VOODdQs5TK4x**1M zs*Tl#v7#!&j#@2X+hk}hqLOl3F@e%Ne^LYf`|d5!-MDw(eK)z^zAqO2j;$Ju_UQ7a z2WCxvpE*0|&|3V$$8R*9cL*B8whN1!E;-C-sS$#1vF)d#jAL||ctO6< z$a1rpbf0aznZ-`;j|$$2{qP1h_8puA{T{f1`VXREy=YiYa@JtT4mHqPnXDFwWldif zVIm$&!z3-|J8hHuMvuOT!U(lwrmJ@sf~PWgzT&)0$P{mu!l-iR2g$_TU{fFa9cvCZ z&Y{__vKMMvCRX@C4c^<4lq0y+X*IKVDM?9i)pLtJ7NR3{4s77L}cc#B30rP|F4@*Ela{iAJ zpr3P;OFxo0qf~+hxLy(hXyOuZI4sD$eKO*;-5_nlSf6x|-4buptiMD=j^D2qJ`l+A zSxS&dzj!4#KT&lkJ`$NG+_?$ON1(>fx)A*J$I~?stB7SOS(j}C?6o4Mb1Hw$WR4Qc zGV-Cx^7>(CRv=&Zsfa8~$Fc(B^G)Bs_v4(TS6j{w9miFkea*W3jDc{DLYw zr&whddY-#Q5tH>ka=e4nz=)JY));qvwL(&NwR5IMiNIA}K#a3>B;z2?+CgFbGoNG| z#?Ih^V1B_PB6x0-UZzp1xWt16MFFA2?q2OpuOM9!E%HntR&o+9Mv#NPkp+@$l(UzM zk`s(bc2g9^Ay-ktX^8%kg<8TOdh<8c_WM+bxN6ElB(6|Ma`vpbQpUv`SW||~<-SBl z4gU6x>{zJm@4f9pE|9ArG-Aa~D*g^GksIqp2{ss6q@PyVB-Ap}HDmLIStfEI{JK+_s7XQ12->kyO5T_;wkvtzGZC>`s!)*s5% zOM6C*7$R@_Q>Mk1)$4xar9KMdtOrkGGxR^orPvo9aI{nb)=Nq)N4>GEN%Mx>RRG*` zrH6qdY7|A%tWQ1U@4Xd}zSFBn!=JhRM?Su0hibk(4X@Bo1}Y23w4G~Zau|1 z_nHpa)D!i+@ z6`OMol^(n|lnQ;N*4GmA-m6f;`y0x_>;SLLK$+-eH8;)&*nm=a3YE%L8Qr7P96WRq zLD=SUSM$0nC@yUTS+3)~0K^-%lR8+b`cukLC*?T* ze7ex^LX?$COiTy$*pKzyz^0&c^!_}ES{ku|G)#wjvI#-=q$!1_#_CEr9tIRo zU46g`3QI{pF3V1r1={Q2VUfNzm$9vh#9H`5tRM=z=%HvB=moqAh{dpAA}v&xQv7fs z``^HGu4&`lpb0fukQ@>{{NzH`Mn*j`GJ@74vjK<(g#ulVoFSdTf?3&ANcSrwg>mPq zms=2pGEjG&pLQwZr3qnk9o6RT5p+HaToyd99~ra!KoM8~h`zykUx+%QFsu$v@8_liKiJm9}s_@%8~z z3_L*aqr{OfFiFmyi#r&>`$-`oX(cBp$Min15Rh| z`S@YElE;7LK8&yQ!JbiR4_#!{janMx(i(ju1treT^kRMWaNu_FfF+t|_>qMR<+Z+60y zh9~{mAgap`o1nCn+XQg>Sx;kzle8kJRs5PZKBk$Rucvl7t!ZI}3vk%mw&1*^SJ&G) z5xoF=#F##$XJkMxLm%na$LBh>NF!~!vs7SvFx`$!^f#wwc4WHNyc9W!v$N4a6D0aO^{t_zI;cHxKzpj;fC}j=wa^ac;=n2 zawf&_=)O1)unEGC%O01z;IC6sz#5V>P~>6=qImbMQ!q#!Y|U|kWQ4=4LCTK1giUQ6 zlwb*LYXxee`!tOv!SPw9srT_wieV;ODJlLm%HVS}L{&jM0~u0lCl~}%iTJ`{WQo%Q zu8V93Q#|gv1PKHA#p4u06>%J81`_DH;B=&CbBoC^l9Ajp@ZtyRAt;C(7wRDwiVwBH zhW3+t#noN7Y)fqbT5VTu`2O8pi}Ru*EUAVPhgS^fcVyN6nwQI(MS=m@m&T-wKBK3; z-{l)te-SNgx2o>=ZTlYZbD&7SGAt@?x-@<8AZ{#$Lq*?B%iGD73VqZ~=#_K zwQXvx4=uj^#ZV7~>IOz(WaZ8~)8HZ9dSv)d5Td!i9W?j(DQf|7ix4Yw`JBFR6YsR3 zgq^8rv-ov$TW6?Lu&OghI2hnKGo{v5t@*Kuv~H}fn5DAU1@sA?mW!k0&(yP|x&XVe zPoXGdR$KGK82_^AgY}sK=);Qjg;rL?YNz8MM*s&l;P|Da$_)e~B^UMTbwPJjuP*jf zpN{0VUHB%#%<`L)`?HHtgvCFYsvGX*g+ZFoi2;(YuqRk$MkP5=-LsD#MvEu#3uO_$ z6gEuV;nA#G0b>$uNks;AeqXUcj4;W|j%XH%JBNqx0AyhjjF` z?nuzsq+U`;FJ;Uw0XCTN5{JdEK)J3X#n$)~UIfGhIeP@pI&A9SYvt7kLEuW=rK7A3 z)$D#U#RvOX9+z8JdOK`4T*i&z(j1CQX-;>zLSm5YhWlTob4T$`?b?5)o>?R#?Vj1BFUJ=S+X3+6+C2_1z;ct^Q_ek(F6#t7YY!^1r7I-uvn` zT#smVbuwLb;Ni|Ij*MUrXI^c3LqV5kF=OY>%IbILzaP&&ijKxG?c6+W^0dSIwO0LR zt*9Cvw74y7%+BM)tVaV3;w5r2wHi2Y;qW}^VH_22oN?*<5`yyR_m@W`UoA?5pdV>g z?!9%8eBF%n@$MgWZF2H%tUnOo*yMrxM9$*ynV`AmX&ho}EqRHt?YP`2CUo&s0Imk3 z{D5195J%WCOV>0!nRq7p@n~nBsamGi7Ed4zTK`l^<>L-bZKt#e9&o_0_mrk7c#xgu zGzHFkGMm_xnVy`9g|K_JrgcO{-8HS^5=|RXg^G;AvlJPkuEA{E?hddp)k!OQ_Uz@4 z=6Rg~Rtnn0A1hcpfJG+Z$(GC!{lz>IRIWmkq?~A>E1>U$w%)RNtAcqT8FoUU&wLaS zb=BJ`^G+J1QOfMUSEbugEKTN^DxTaSkz$Y-w?u9J+;R(x6imkN=T`FUESKvFP1=Jw zI~Qvw^}GP0Rt8mKew_l(#`v$7iHLZbCIW`puM~aE+$|WMuTg~a`xW%#cd@vG`7eYbVh!+={<7z?7 zK`e9*Wnw}IEOg~Y~jL6F>S&HdpAxl zZzTqzW+{tr8PaOwr-h;7QKe#8bY*2JDvhe~;-4y7-P#x)$m@sDh z!Gd2~06{N9%S_7vm}*A?0R8y`nnbdU=1yU>yiaT;4TH)R$4z@^`9`cTl1koYH&)2) z!BFVX$s6FPSikTju|vF=fi*ke%Wwn_AVykuFtoh3z~aQ--naR;4OARN)~VS^f4m*i zxpdBP|5KN6wCK?WtIm2^{y?i`O69>lNj)kG?{JCEd(D5qhuoh_`IfmY#6T{>IaoQo z7hF_=OV=DgIJyPueS60e5&WBpurxBE)!z-Ee-6zssV1h%Y(lpt)yOjh% zn-X!ObFwi&G|v!>cqwnR6j=PV;T*yX3p)Smy*}h1wCPRP%JtagXfRK1%|s26vtS?5 z--BBVIJOIm{2QwMhUK8Tc#j0#{!OO{5TbYN`~rfp7Kn#_!i3T_L&oVS`3iR_2BS)@ zlQJ<%p6PN7(vF=IzwFAw;lppeYvpTS{r=I@3r4h~)DD!H)2-WeH_v?Jt)suaUtQ79 z={#xE0m^S7&AEjcfj2SW_dT0unX5-qOVx`0=bCx@}w@wvB91GX-IoX5U&!gc6JItlU<{iic#I?{M1dmWb{ zjU@>PZ|=!Hcn%q&FZT?IFS*;K*&+uW4}a|wq^T@vTT zw`_T_y85xz%PS|18QABtJf31H^Lu~UeVZ6?Q|au5tDoQg+J=P_Kp8M`r*#t_-1>7I zjs9HBm4?B27&dC)sBTMjHNT=2J{~-B!g4%cqvVdp(7ohNrf{}}M0~Esvv|_ZGjV{1`u^zd)Nv`fD zmSVL`^Fl2b>%=-u<~$2L*?itA!m5W+HXQqXM?B zVVBIw9V8BOE8?2kbo=}cr$%zLLXuWfO-o(o)ZUbnG;(lsMRclJ7U(Yz==<`g$13;` zzC5GPuS+63>+8?CSoWD~ms=HK%p2APM1P)jbj=GZtZUQIKiFtwh;4fwK2p3Q=dBSd z=N2_>nG`{Pr+j(x{lEURE_eK@$5ij5Pd|Cjlx(c4E0BJE>Y(SkaCD|@226%s9o>F)p)XPvFY z5!*Y@N0glpCMx7W3PZan0uI=IUv`#*zXjuP$Ug3)jJ0Ds8dpxKc;MB~>py#K^^7ea zRqy?7iU38$cmAjI&_@sb@Zw#48X-AM^p1vuOLzVd?Sp=ZGU~bht%^T-`zFP<89|Ym z$#4Dj@MT;Yy6f?R!VhDe8PAC4at?hxMs+C#uiEpOEAg;8v<;XvHfR!gr;g^vRQDRjT zURZhGG*4ix;G$|lO~;fu$@e8G-cArNSb42YMEwc zWF&TGBe6>glaAdJTpMl{c29UhxeftP|JQ?DHOxUa z%(%CK*({~uvaevZs>fkF46z2-)dsd|hAH6K+T^y02v~X+mmlU}nB!Ma{=pcCnP)^c z&7X8%1AxN)W;`#qv^CIG*G@s)r=6eP<)0g%Ry})QpJ|%b@4p}a`RAkG9r=9!p7-9_ z*7c(Ji(5~9@-e63AvTVG^ZA~ocH@@szw%L243U;ZPGS0?op0wfY4_RzF#gtDN5ou^ z3d?Ho-md27?RfVBa^Nf6=+({M_mMe%)z_SzR94xSB~gXbXZ)vfjH%-4222-J5nlbo zUa5tFOex9A`2w18?c)3Y#vD*wzI5Z2DpTQ#E7shkwny;+YrT>{^Q#}fz6mk|uTS6j zAp=-mWh$z`U3ccW3I^C@V?8C6^BCP$$79>RFcu+pz>v={VCZcqoK$E8d4o%w07HFJ)Lf#R`FIhE5GY zoTkGv>e5OSUA_2%Crz-$qJE$dT&L`k{*~pSj zB#ZOsd<|z4rYlUEi>@%xm*`ZN9M>%agoPzetM@Px{Z-EWXAH#T=cSYTZ^8F~@zM(X zSkQOqG^|+VM7JD4@un$UisHpG7B6xs9<4?#Q@A6-eH&sOBZ8jREegtU*2FSY4gD60 zgt4r-u49L}?6@pPt%;i2Cfd_d19-t>>azlYG8Y1P+rWaU-6Ml=&h#oi!dHo<>>yN1 zxm;@)H*<;xk|c`8x#kp3z1(48+5PLNT9(ibHm#QE8x(}0d)!A)x7;nM0B?%Ke|Epg zL*j<_m3Lw#0@Q35({E)gF>9N*V9nvW2K$$ryS1P94k>x{hNcf@ zf4qBmre78jU)wqVo`?54y(VP0Seh(PV7D9Es?qeG&+IGF4<1LH*flracw_qTMt^Yp zONQyo5-V15mN_AUdy8YjG1G@V!pO)t%QthSxoBKOpJ(L@D>gq+HYpaEA8k7)I$cDk zm0Gi$z$|JT36zo^a3VO8Gg*8%rraFYoxVzQ>#Dwu?{o<`Y;nNC#C=cTZAscaxD@yi7Sg5uvrjLV32JeSeN z?Q-fe%8jUN0U(mujcCOUatR6QH1bLw7iY%%LzJSOUiU4x0!;`9A%~>hb@QM1wAtGh z6rb6UBF$ShzW4FxhaR4_`nAvMKY#gwWm}(`*1K~`K_!=^MzI_dwo&i9w*0hz+KGGK zdTqeBckw!@0W08Qb27c%rhSpW?|^&fsm%Vx?SpL^v9>973vR_T&> z{BF4^rVGovD9lymk{e(o!=JLe%s}w9t2i(+<0U9I7+U$&bLi>zo`9ik96B2)z=X_%m$$h;rsg@}c5C+NuG|du694i^RfI zE|n)74Q9nZ0R>RquW~^|Epp~uAaP>+DRv!XA-3No^;C38IVe#VT#g%UEqMZEmba7f zBWuoK_t>9mGm`g*^yd9grb@2;2JSHj`8niqBU_s(=aZUMJTrK9Nt*@7Dk@P3oNPTU^Cpz%)?L6oM0Mf>(!wm0cH(>i*_{*`>J}BS z<%aX0d?kzF_SLTz%=}{vgVu}_W98K~_DFN=;LPscwYDn{&b;n7B`aS2dzS^CmK+tW zdb0b8LK*wwKdySPNq-Z*E_z5V1b|L zSj};)De|VDNL_L_d~AfYtyd1qByTrUIf-NPRlF0FXlsW=Z?9Li;}p@)Wes*+!tupP zrll$#e2|ZIBXbNAzrP0`QRR=SlBnh(Nou`Snh)g|7wnH+2!|>);VW~Mm?rq)hAqGWtn87WHu9or&Vl2IcvZ3;p#+eq-AC+7K z_3U#>YrL1yLR$Nke5%=vS!<$OH3#ci6~=$l6URTN^OY5Ek-)kXD44`c4FBOq2!}9* zWMPFVrGjaMt*AOH3M}ap5HKyOkH(=-BOj}~hy?;A0%KoNir1F)TnCu`?`hd4B2+@=$kMEJ%WvlAGn`8Q``Hiv7q=#M5ST zzq<5SLoQStuOF*QVUM%%^?ck|Ng!NMEs{- z?3d<^mp#G}ONJ|!+k2B7db*{Vff!ES3Atu%KLjBSezEifT#g~a2m8t_JX(pIkh~6I zV&O2EmFxve-|;j9iMZPLAVA{9RT|;W1l66$^}P144Z&YkMnLsk>6cc5;4}17fiTZ;Kqo^!_ znxPnPBAnfEuc)4(SNqno?pDtd1LTcUz3N-=5YdJs2UJIDbkWJzc}g*Q#20re zfarnGddP_QOv>lKLSA5_b^P>E1t5Gka~O`&`qX|oCC~kQ&ebxM?z;IiZ=O-w=Sv|= zvRQKTDh0HE=;fvg+IokVpm^*?g=z7mVhqx!ADr*a>Qm3F8eDeg!zy0WM}Ja)`SW{T zb9X(^N}-m_@#sx@ca9F>*FR^Arjn5Y~ao|%&K1?^Q zaT0wPzViY_PXJeohk-}^gRA{5v6SQdrYz~q%p3GW%TVVy{e0C+t!j)~6Ef2^gp>b~E>)PK!{V64?DCXV{R8 zi4)nBI0^Zyt~>CxlL1Q7JeG0sfQ=?l1vSMxEq@x|WTT0#8BI}GQ@d|SOV#oC@c8)&Yg$GS zD(sG0suk20Q2H$L70n=DrPX&%QkN1`$&80|l1-e9-htjVCYVlxQ=o!EL z@(PF3t%_69#@e?hbHBm-kSv|xkj5V-IKB8lh85mFuv<1H_eX%1) z(sc28rto22l2Y8(e*GG}^2>?e(gyp)>lc-D#~?ILBG;QbZe0TmN*HmecE?26Zsp{r zCOHDx?&?fwAS%2%c*svF$(=65bj&uq_C^LGh%Q%e(@Qtb)TQRlIbMNUonxUQd^Cw@ zng%$m=BH~>h9)2s!IWNxi}Ee-l_Lb(ZOP%LRb*Dy?y=4bqXcHSZEWM%psDb@fh<1+ zKd8lL6j%WxEy(Cu>;aom^_R1Ic zezNsp5iZP@IHHerJ?teWJ-(xQ;e>00YC#bgg+40CU{O7aeHU7x%(}J-XfL-2uWH29 z(-@hxMeiu?0_JELqN@r;Y*wn??T`__n@46Bk2Q>_4>)k*t=WTKqd${eDpg$DVLCp;H()tqPO{zMz{0c*9J8bCb{+d>2qK{M7r78cSm? zdx_r0<4#}6pl|IVNu?IQD%aNkHTbjJg(hseE)RNJ?f!rIonk@d^xbWG>3mMbX71me zP7PNeT_WibZJ+BY=%VRH>pzK~t zc3r?hFc`j$6!Y>QqE}=L6f#=7i1*YRk;^X1zNyV4vy0DJ$8tj zbpI9eB`gftz{9P3NpIZzRb$FS15j_;o}su6X#eyV;>3hMp!t;FT61JfYPFgNgNHaKiE}@ov8gSd zkSM}MlZF{P6&&T7ZkO$>>%33lD9-M;s__)6JK>=fo;?FjD%e{y9i#KL3lZvR$f*pJe7Pmdh>ktF8&-Z$`HpK(> z1Ar+uRnNQk`@GJ|mUXayR9a>JB;Kt`naf*qe6%fsV=T>LH+c6eLf{Q`P-V|S+ z(8il~1Q(0V*l3(t(I%LwG!j=qjNEABT)poIM|Z<9!|Wf9`OFsSW(&J%#0cY1RoK?d z(r8X;w8_Y*F+v|gO|xzxv#1%Fm5sG#RsPiZmenQTMkbq9mq|hkhQp~&sYvS>7Rwq3 zd`_CFH!sCakuas=E#A^nh&L@E<=EP+)6+(_mf4kiY=j@=^6kIFucOwNHvUNlfSr8X zcZ=b{;hJXOc(a+}g&#k49aZiQV8(TGb9n%$PR1LzSP}?e7oWU=8J!JlUW6&KgX7Hr z!1m&6VSDD7d%icYNVY*MR-;h3!!pK+u}JwueV`fx5FT*yZcgFgoF9gFEzW^p7r`Mn8t8u_Mf5H+CQd3jAQQaqi z<8|G|gQyTHY2&@#NT5!g2MYy=3*pM^w6_2t?qB;r`Pl0&;!l&pWYM4X#O*w7<}8>vs9R1$<*x_t>EZ=h&+`125NV_zE637NYb)9; zO;S-Pv=UNF(H5ymK*M)% zsaQdkL!*?tT~fThP5BbnE|ZybXqLW4qJP!V#W3af3}R=y{y01Igbw-yE%d1221NB& z*+h^6}|zyyt`oU_c>LIWS2rvm7Uqnbgk7%#5Lf)kUAg zXI3Djs!ySP%p0mwi~y&e~D z_@Fm_IMRS062Yu=U3RR_5q9;pLcm#vJu}=R2HJ^{iXl0PQGUX%PPEbZ)jJLo;2p9- z|G|v6tmsfA!gd`*=Q#EQJg|3`n5S_BCW*O<#vp8%c(=D1C&T&O8m}Wv}4&u<6_$ zR`?_Aiqe=kI$gm?d{6E}cAo<;WYO8=eq!h@RkY+n?RYkt%hjS_ zYvfLst-@ZLoTc0~Y~rXDm3D%DO%y%@E?&ve!RL6j39-0DJy+2P5(}QO&NC|74uGu6lgVI z_i$}oLO*ajL_GCkj$gPb=rl$z$YIPKz~-P17dK&=!t~4Gv5g9heX{ zec^Dk+>zlxU}U(d6^0uMS%Fa43g84|z!h2`EGQ7F#8W|mQS&tOsR%3VnB<;idCd z%qt!KuKV?7ly6btgu52b8Z+Rve!S#ttT+t&p`CK!=LFewJ=ap2%EV>@(hgO~>M=gJ zi*0QwXhCO5LO31Nj+4?fY2&&R{R(V z0SkwDTIhk6q}OCc3eu>hvTdys%xSy4Dc zatP)Y;Zi(;xrHIQMVPq-p3(EPz(w<@nK>8FGzu8=C}7TY1O9oXSNZ3O>E>Kf-zs!9 zx6+|mdnd7YX5gj%Vc7w08ZT=muM@b`1YL{^&Y@vBKXqOm4l^ISThz;Vpd;zFuT|c| z)$A6xyH|QY`Io!+*xb8xDK-0+LkO*}=dII2cY(y^iH*2>Gl-10JCEPMiR$Xf_ZY4mTpW<&|Z(l%4m=7{K!D*zWbGhD==a$l7}*59xjb z9pH4>rWMV3=*jg9?$~$J6RAKn?9l!%4yd0y?#WIJV3obvb3= zNvfc>&!9us98s?!Rs%s8VO zrqIw|h!c?gtD4@}a_;5C*&A^6ZU=|p^9L6xmj?2*ep8EVQ~A+D|8M~?d2L|127Isl zDG}(aZxnBJA7w-!n`5r_NP@%wmIBG)J~DYPV691lE#fb5*qt?&?QktD}hU4sSbJ$JP=RMH9a8 zBGC>Zp_I*gq(T#t4|lmcUWK=~Pr0x^?qt>TCK-afT{Wns{H6kK)TB9f1u#K8>S7sE^S!6bP8$W*MP?)fBD?*DSXJez;aqz%&N{H{b@# z95BOg4VdmX2UMA6#dP53n>j@@th~^4e{RuSGdEOt#=pcjV36ov5I;6t_^?8s6a|1l=5ZW8(Ti|Nc6_4OCx9~yQ^&F;4Nml$ zCp;KudL6>`m<%h3fkDC0-#Jbkr7ahD`G-H33qjrDxG^X+%jL^08t%40DTz#Rxy3Lt zWSgV{9KG}cEFz$>;r$NGTlvUiQb&e8L8?Sm zCkTGg=3`tNNF+iO{Yy^5=5w@A^zyeK5g z&#VhTGBr;yWsFU5dA#Cy9F6*za6iOQpEeiO0gQ9!?r>2H13C=AFw9{+BR3&&?(7|S zP?0d4j9gfrIcEz1i~m^?UE3W!V3b?KH5MPr#G(tMNFF*Mx>0(h z6vee4U4!GN`Xa-jkx?B-68W?X>pJs@kH+(dDDZ*G$Wh&i;*!@^K8#hD3gt`+=Y%KOf$&rcjJ(PSjKr$pJq!m1SY`s3i`{fv;M$J7 zA&@%~tHa3YX4?UHM$SN->1J*iuzW_t#j3$iNG=CO_m%M{w5^J0-okZfe#^4^@(=$O z<@E2(4&3THWz!vL!;&A*6<)`OHUWak>x_3&5_(L|sS^S8X1Ejgt>!QnfgHyoBPN^p zAdLYMQuogohpIi6SR_M$5C~^oHY*Z&hOnvmTLl2z+Av3=c&$M^HjVy<<*vKv(CbLX zrW0rurg4`M3!d5c`|lqyM*vJ~Z{39HyT8L7_0Sm$0oL zdh3yyMTspd5ZCjq9ioSs4DA&RDwFzYM+Ua{gmtBAsSNmiGCSWouqjb-KA!`v;y6QdkQ{`VOndruY+pKbuFOdamFqD`^A@H(hnA{V(7$tc!W>@YQixi6W-<=Gr@ zqpBHjbnv9( z$p|^*4>?8?k!F}7-E10415P&qLvyJIE`>WdG$1fvOdM4lDlIJvl?#4Cp}BCYL`dM@ zhX_-Y$1MM4mOrBepZ{7t8CI+R%v4NQ+Nb&Mvh1_iEhJib=Nm%tSy#Ss;hS{aZiD#k z3Mkv-5Yglx&I}4I3aJ=Z&iUYeB;Fl?%qPQ4nwqdDXa;K^PkE9 z>eBLach)@dz}%^mCr_QT^oh69DgC|Y9w@t|*Mb!+J23??LshnDq!gbzp=t^V`my_0X~deNq3GlpKtXT6%1 zUbhW|dqIzA#h88_BGdJTkSP@qk>Vbd2pa&5oIADKTQhyV8gu$&Xo*!k;whij1r9g9S+IH!!oVY;4b#^d}R?2DA!V-Rpj=^ z5JM+h=5|i7OC)p!Z*xk_M*&WPJHw+mQk;*%012KT@D&z0a(Lg*IAD+-&B~V^ zpIQtVEpy1A#}UTnUqiXPJK(%IU%ceP6fxSHsn#rdr>KLMVu<_#2i4N8XpDoyH@5zPh)n zAVqP8H>d@nOAqzN_a}1od2e5ixgC zh@c|XbM>%pb_(mcp|GA4F4ZC3;eo%jl<@fQ_;css{&6vM2v(l=Smt2M!e=@xvx+SL z46@8c;pm+3V+)iN$o~@yrv1*%XMRVwZsOk?!IU>o|Ih&m(afPJ=!S_yE??2}ue#;V z2cCUr=gxOtSbO)N^Dqo8dz9Sw>XAR|zuEch19y(T<(APi?|EkbvG;#@b%^TIb1u4Z z?$aOr^XGTl9-P=WQ}rJ${(9fuV^SgIYkS@FHE$k0dE$e|@4lfUre@~)Pk-TsQ`8@L z?w+yTcnd~t_P@VYrqq_c{|ss(&756#E5Ng#`?#HglGZ90lISE!eGo9*mcSYEOUghA zK1*W*<4{WHp%v9{fBo<8fBx`x1e>?#nK~7fFW>(Cj|WbEylLKqtAlx$^}co8-(O#> za#rGstVhb~TYI0&IV!XIr^`JEL>6T*8xN+qz1^!ezW&eXTs+*REt;}Vh5;ocD0wJz zBCaCRJ2HF8e5fI!;K%8c5rFe>>;r&^RO^ zEJ+|ZV!*OU$z1C2V;(Ro@sJ*$Dq~;Ewd981}xJmsLOkwxAp?tM^z10dz=~QjS~-?7_C~2mhJEu z;9@0k;?QN6iBEp_^|l)lF^FF5tuo_N-A+#b>1*El`TI})+BUy0cjm&`DliU#`K4dJ zdHMd^-b}F6yEdsH5+R);TN7QOvqTqvO{x1418>%K$XVW28w^9&mpM`0ahmJ-dOIhH zy=jz1Sx!Q5lA=@t9Q#?IWopp;l{dG=xD2~0;1@HfNmXEh2n;QcHSrG%DXXb}mgS#W zRY>zQD#BIv04Qk&2*(}}N}~ZGon7E`#|@=9Av~xqTrIOGq`@V2agv#Ic8j3g!-3{G zfU6))6^BAqp(2q7Y04KW3Py5#!}M4Vf4(|L{a;xBqKs&j9YG~UhNhQ75XyD9R|Gla zI1VT<%v6|>MiCP(zLS6xFu;uu;}8ZG@C-^ylmFHQ2fnjCaSFjWp3(=}!Z6E9VS(=m z2$qusi&@Cr-v9dJk7|&UQ5=~q-eZ0?Y?T-{YO-d^QE0U9)%A-e-Z&&OWvH7chpY1< zpTTm4nX&Vk2kyM(`qdTVLeKNm_RQ65m4Ixv*!j}>1-A|AaeZXU@XUS*BH1XTHTvk) z_47-Lx?jKSt_gY9wDFQ|QACfFPv1B5<{rHso_A~BH9Zm;XN4sV)-gv(i7mT6f!=k^KftcpLLhG7Yjf zZpXW$`$N5Nnh|+wdnTAPjG!gbr5UR$hE2YwxaYvqs&!lU${DI-m=d}!BKN~XIFI*8 zpME1}uh{f%b03l!Am6Y|-sXW&VezCTkHtUb{SDg8!^oy>Z4-1;Az4BlI;4TzIAJ^B zc>t7TkO$$>9mn2LNJyZ5EL6qpr*H9Ss3kt{fT3~*!t0O=%HZ-@w}?YLpiW}ZhB1uh zg+hkcAv0Uk9MW9I!;7e>oaZqh2_drbWuiH@kn!kiq+{b47DG|C@cj*IFe7=nvW4aq zsJ`VY@Fr;IM;uBvWM1zJuTI*BHEZYJF{szD#hn;QP6pd@Ep(H!w>@#s)L}iYSuuN5 zzaT!)!cCLmlEWG;K6sT$?W&vS-9F$luucwkWq1>N&&z8irLCx(5bE55YscN>C9mka zs$$%=oiYrRcQuegE=?v;Skk%x9vuPWD94H-4E#^F(xFN3U?RZaQw!47nq1I#MA?1C ztpyKU!?xXrOkOm6=329vk73i74j(>c$;Nn2 z!*Cd>V9vXz_}cL7<*|2LYH^2if@Y^1`maKD6;tR*ELrspbX&#mCgV; zK_ea;K&8Pe2x>$T6jujC2%zqWEQh#Dg)8Au2Skm6+9Dz%tD}MvL>+Gw)VLy|q8r5v zMGcoAApE}XRZqa(-|zqVznQM%)zNjndR1NZ3R6tAtVvf$TT*Ldb+{ZYt>z^0#kZ)& ztE!@L)p1p2#}n1baJ7fy;*gOcial3*BXx45W5@Fxdt}V9ZT~579J|27r=Hn5scgH* zhS!uR;XI0k^Bf0{3gHQN!IM{lPdgeytnoEx7Hjx^QA*)ZV03rLp&P4=0_HDOk*ao* z)v0PVr<7D8SXAP(U>}`+@zMXPn?8RY|2wvo-ig1@ni&QS1+NVghGwYo17myksQLGa zDO7m*l*d0j*zm_kPdxgqV5r5}qvyZ)?QeU&eQExvvt@y4&?^X&EW7yN;~yOS@!$uK z|Kp=gKxAObBF)B<(u=p<_2RaMO}oB*asKGDGpidxj$_B?W-5=x}h;j2iT>|l`){{&od^b z-;L%t%nN^M zSZBP(JD-_uu!|oOT?_qtHg1rMCL7(REv3fv%4HeMk(S2veX-FU`pFn*PsA9!=n-SE zeldFKXGZVnr~KMg#(w?=qtEck@=cE!yDP<^u@M4x$w>wyBr2PmRbS($kVC*5ilCTj z8_!?a#+>H$Y%~{haGois(&*+; zvmV3duKv$&8^7MW;gN~`JAqU)Z81I&tHx*e;(nv=erfxTO@}^zrs}V!Wf@+~_~_i> z+?(!sefy7fzwG|#iK&-kSu1`t_=ojqfRjxF%w=>(es2hoRHGV?H}sfVo-<~_%Rl`3{jT*7jlam@#h@e|-PP z=G2Ui;{NwJO->q9vaZXG0T7lEQ-us`!n^x#6x(20afvj0QnqTe^j#mmq zqgSqGp$J9gB0-cTxeql8gj^KjBTj0tr$#c4r6yo9i4$Cq-lK4nrre-~1D+kisaupe z>e((_6opL=ex`E5sbSPXr23G$4dL)v10Ob0<5UVllDc>~oiVg^aww;h<5-;?hoI-; zaRShTcodJZigg8Dpb*Xsq#SzorcVFa_m3m{zMS&wOGbJ57_(U``&6h(rJl>}Z6uE& zD86agTv5vT8(ZD`?PKQ99-%(+y0~0QwF9CT0JS&4? z+b1{`q?Va(s_A0xFsS6ag!^(Z3|?UZ;`^DYXb6M?=<~)vMPR)h3jTE`3tOQ(BPmSa ziDF&&^A`qN^lT+88V-P(uO!Qkmytd`Ojeu7)sQI$pX3ZjgmE@KrCg|bbRZ3l>aY)274;0NXp$+L}CXVg8F9-75qgsx>kg$v5kz#y#jHIgM5R)2fLsp2R zN=lPzT)1dl1fL~E0Eu0j?MyStjVGe^>n@>LS}oPEUf-Na3b=MA)R~ zS!d07P2i0Typ@6Vjer|24Q$N7`x@U(0hHc(4gu8r(i>npu2RUAdhK>@j;Qqm+>q3? z=Km2ZwVK(94dGMoHuk7&XVfPc-_&h7Q$PC_^gB(_Ex*Fz#;eb;*mK!imc#nAHpX*{ znH|Nh8{TI$EqoHy06obTd{C}L&(N8U*79&S!2%&I3f;)3uYz8Q1w8}lBf_cD$7W#9 zmjvO1$>ApKj%K3uPpynm5AJOY4n9TC6ix!B^lg5#ODC9m;Dp>v3F^lGBX4%FQ;rj+ z(Z4oJTZFY8gdb0S-$a!snh_4zmDl@)a7~r?YWbw^X`?C18N2-VQL`$)W|TFvx&JN%`9*3` zdKX6k*^2^>(K5GeJDFEMbiQ%O*k@hr+gaOj)iIWJ*ZUTez=cAM8RgLWzaRBi1D3{N zD;HG5KiC9dOR7?g##TS`sKul7Y~QBmN?~E^WyoMO&GfU0#R&oHf3CnZTi8QuS%RN6 zccBS6@8;i442K3kDhBoT&I`rhe6UWcoklJ`Y`pBa!K5bm>Ni~kbLKn7ftI}M8@^|~ zo&{G4s6WE~Rc4tB%~(`73%`i0<)c!NNX25TLAI|;&JM8-BDx%!27$$G=iCSQ&&TDt zR(T}~5X97rD}K4w4w*KW?Qdk%@Zfmz2HRu5m@R9bm9u}2gXuIm>tOH8C)(iXFdDCd zg9|4{Qp0k2QG6a-oERQ+v1J3>uf_Y_W)?0yQc)H%b%<{b)^u&AThIk1 zj+5~GdM*3VadOy^;7;c2eKsac0-BQzN0-yhuXCnJveEn;1V_T%60Y-Un2Nz3702}B z{5}x7`nyXWv3)Wo#^NK&Q6S^w~%}X1J^mCar;ITf8mJhBGH_d zxXUCbihjwJY}BJBK8OfOD~3`WRn7TCKBH*PDcZF|_=1N&|4dx$g@3%!r}5WB(m?lc zN&r^M=u>1RcA@p6u|8wn$%ciNb)gj2GnVxOBi4o&apPEPS#PtI?y^E-pr!ZQ8>P!` zx2&bGd-sQ);=&eK`IK+?1BJnG(ORR^VQFaChw_Gx*umpFpzji+(+<|(J_=2yqeVhtPCGbbSTmpRlaApt&d7>$oyf&4OuY42))2>rYV8H9bM_c0mKl zn9-Xq{x64WAx)6eu--TU5T7hU6y6M?(SjQ-hYM$fD-#`BGR3FiQb z6e5$+(#973+W1L>a{1I-NHk8brV6MR?qC3Ez!k%_E%EKmoOLP&8R&(lKM3&@oQ%SU zIU_=fRrWp)=NM|6gRZkhL9!IL3?=AP0Jk-K&VrD-LG)4w8av+$N4)@SsGphvd3V2) z?#KxeWxq*UMwRVT_0sl9HvjQAhm+Yg#Thj7eOHQId}(&#ZzZJh48|kVhUH?YhicO` zuEQVERe6RqJ!gsvj=1v~D!}h$v`_jK-oxa;V;A~ZR2)ObN;aR1;J`vG()>h7f(U}vv>{TPmxLME4K^;)=%pa3pcnt5kPkjo z?pFo-t}xolMXabJoOy%%E|n_Cy3JBOQQKQH#pJJWY6yA;Xub;`edl^ssgqDDG44NG zG*;HgLN1$}52P@%VAQ=Cu`DX5(&`P@IgwErhRm3ye4=t=Y3%o9l*A-#GcP(hFRQf?` zhSIMZ{c5Jim@El=Y;a=UwbS5OO%^now|*ghx(b4;L0@J{{zu01;}abZ=7x?$--*w9 zhYLkOL$&dNhUcT`ASgS7rz0F1t%xg%LC%%9t>{Cy7vQ8|7xNNe zwsIl|4q``*Y5WRGBZ=BHx@>9_Xc++46$ZP zqEg1{JA;?tgFo2*Tj6J z%X(dnSQ8H#(`-$>dK)AEIu8Gh>lIBvJcZL0-+JgbnDTl54cS|2G?OX-SlE+|NbDU`D|;Vy-zx z^>F!)Pe&puZzfr>zk9jbizU3AX~`*`n&t%(o>l3oTRclQr@$@vET;TBI-dRD>< zRASs!7H&G|iLpA@>Vd%plJc6DKP1#h%j%izWmy;DF65{ew0c?5WQ&|EH)_of zrcRBff>!@ze`vKZL{q21Y+XdnlR;~~8%;J3ruvkmf}#EnE9|c&^`GyxY*KEuvw|Hi zfE55q+_qKg98o()1(D?@9mulq>QCLh;5MfRlH9*aak^S%FY-idF}xVXB!f zyi*LyaGdGS-4k!;csmTw|mcT+1l|&wqF4He{tOIxrQszYenCC5U5= z7#(oL5nDa&BmJ@wp;%ZB9vEY5<$A? zJ#r}IRoW?}POZmTdW#YwJh0=bi0(w%{|X}e5P7Q?`WGX07zDVy35YNHQYZ+yHZQ&1oOvvb)>9x>aC+!7v4h6Jh1vE=$@ce%whk`%`6M}4+@H%VAG|IQiJ#`tD z0nko{s@!AC0dF=2tQEAV1Oo3*Q#sGM96U%@4N9i&0ZuzT z!lhKcw1mxog~$VVf>Icf1APJv8n0jNV0|3$Gs*}lllkw3=$&5UDP*@J8`+P$TOmzi ztP60`6Eh_&Hp3!pk(mWVqaiCW4evfx6DRFAOI7hXk1>pa%k^{!@Q*82W57!c49M6hb4tSkYNyJ|Ca=cyI;7Fekqt3y;+ zX*Y%3=|;}cQGCe7a?pVMs{zP$(2Ooexq{AjCUxKssb%cozznG=NIIGY(ShF)4Zh_> z!P~Q)_M`RS00S#@0A?;cF!TW!Ou_M?TNEApHDfp(;vmUJ9`4ZLX2FId!Gn?HiXs=% zzg84MK4{QeAxFfaq$Mi*NF=&{YEs z`cUum5$uXufivzbJ+P==*wG?14KxBUi#a0cP$O^}O$<_0mKR~-8)4TKR~~Up^Elir zjwB448s``W{XMecqW!&f+CJhB{m~g^)#EiT5KBy|PUx(H$V9uqF12%Ucs7PZwb0uZ zh#s)pMrK!5CT$hLrF(XwZ7iab6=4u)8+YeK?B>vOBayZmdtjrHaI1t3vAU5^L32CQ zYBF}E-J;8_Y0$@Ery6u?+?C}tZLqd(8B$Ia7rN2N;PPmsx}5C6<;^2Q%A4aXfZ*&Q znsHI!qV}wEVt-Xbu$ygAISs+Cw%l?oonAKBDo7MqkiFGRn0KJmi{P{!t*(CkXq3ggH)1b?FKnvn1_draaN=9 zUX3DGW>JHE(9>Kwnx7LP`50yFx>87ng?$hXG=5M=Buc$~GFx!%6#SxP+4X#~TPm=e z-@#KWd;)%3#vFQx{X&j3o+wp9qy}686|9V4rGRD$$(5t5f^sNKk4rP~{V~0CHG#%A z!aR`zYP?5ZbQcWbHIL?!VSj%C>!~)O@Nc`x?BKAUwsZdPjGChml;9jQRXv6LjhKD9w~VHs~#^sso(+@DSdXMdj~dI*6Om6GraUG+{3NLm_gti zI*C2p&x(K+6`z-Qk_7OgKmKE%yaf3%&0a>_hDm39GWUt+y>(j-Z0GN}Iym#um%Zid zzdQK(OD77ULtyv^WBb+XU8KsQl;vm@bsX1)o5n*9tAB7|;xcQCEgZ%=JQxIBLH3KK zFwX%@V;3EiBIQJ2bhX98-oY6762Y#^%tmPkFh5J@zLY+-*Eze#$E-tV=1R?a|n#!E+7V=YuuCkYkO;j7GK`$>t=i zycziV{7ZeP&Y(o!a`t77aP6O5jOC3U_J}ay+OuO(~Z(Z#SCTo zP&U~?FrFN`y^JnPN*I}eLeRzXJhnZ+N~x%9VzAzB4vtKs9C(pZ;diCmvdhfBLTO+V z-i;4mG6WZ{W~-0AvFl0CvekO886^0kf7G!JmA-JTw9u+SZ1bi$Ygh-&n!suy#%#dR z6$D8GFmdi;R3$k=cjiR-etnyM4~Z2(3v7hh@)}sZqsYRig*Ikf*st|YbD;>&nEq;l z&%Vl#iPy~}gOP<#BQrkncsgLb5d}T~Jp@O$LJphhYozbul_!8E^)|c^3)&f}LN3EQSzUM-z=Mscp&D;;AF@xm6D)Z4Ki;v!~$Y$SK z>Mkwbc!i&d?lCA6_c4fie%Uk=tEdi_n2E)e*Emf5j%Vzs0E9SVZg2(6{PM;P_h7vH z-q?uD7biIbB(0r%L|LPcF5*K^f>tC(-$&p9v`9grPxM|A4Vyb&hNlb3x34rW(=Y7QfVo(1FzdR(zZ)uSQrOeSB5Wp^Ph5$d9JQPGCcBhaX2y~9j?gQER5J9aqP17?dkQhb51a8JG zhu88*z=j@lBMN5QDu?)29cgFHaCFXiCmNgXc;j`H#ybw4DBX%@Tpsj%#FZbV3HhGg zCW?JTcp{D!$&b!XD60|&T@ndpLv~UOXZ&)fC2Y1^iE3r(3OqP-7>9TqtxBx8ny$)m zwlJ){uo~;6!Vo?&*2$WTud+rC8&J4{u95s6Ij)%x2mVB45BvDL-rn}%nfDJck|FZ) z@(KzHo;b1-+$O_vRy2k^AJQVDiwk)f# z@J@9|Ou>T~D4724wArrPiAJ=0AL4V7faQJbBSHBc4ml1?DABtm>2q7$T6iggBL9Y(lr?qis;w&jTYKdk}=yfRJc83 zS13Nh(PW=#lU7}6#w9;M7BV}8&0MYj3mP9tpt$dQ33`(~d>n#9*K`ORck5 z4@h|hq^|;-59aGP#-m#W95Ixy58&#dG$cU!5fYUg<|PYl;#bq*O|#d-JUFW>I(lr; zPA!tJ5zq1!!l5f>^N}B+*~)#wj02`w1vNXQdm9f}amE^(e)|Mt4VE~hCkK5?HBIMH z-G?;}1=kRLdQ7oVg9AsT5YW`0o{9yp*b471C;OXzK8m$TXP|YUK>;L*_zH^yYwT^r zyNoR9W6?o2phz{>$(|N$_FAx@ZHnLeOx=R*HKxq*Q#RY{T z!JA?nQ#RG{0wbj#1z|)A{O^O*BI2joy#%h-3|<3Dj~K+%Q)`ep!W(4~PN{a4oE*}3 zvty=*f;-*LlQ5bol$m=`dJVq_yVB1ZdFS)>`@R?{Hg=)@=Zw5_)z&{g{dU*-CAXCq zh(B?-vSGb<+rX|0b2d~w-Xr861(+#Zj$?~Ngv2K?8x^7t-+E70lU)tF_WZ_;btMk6 zMzZ!aH0=6q{{dOF;vhJbN&62-;$eG3L&5V6{KJ91fyO1WB9{$R_Xk)D2crQy$FD$M zF)*;X^&RXiqXb>@4x2R`l$zIE%y>E0A!G+aw0(eZUeb&JN{ufYGBW|bS%+FfMT4($A>c-&UQ^bJbKXVPWyYHQex;22Q&N_=h@ zC9M_3&iSJBoUy8wtxw6Fdi64^@1MOSFAL=1u zOt17Yme1I*HT|uR6zxzO){b>Hb57&uDal0mi&u_P?DeFNTBya##-oUsyXx%xmp$`l z!Foc&g3dAi`+fStqAAz)=^_H>dSaU~w%+(L*y7t*u3huMKd+VRl>FgANgUn9)iH;( zi%bOol?tPieq=&&&p-)n(>%_i7m=!CqXQHA63SSM$xZ7ZhFF0H{~7|Zq~SMNJLvnn zo}6!7O?|>#ho_<9xEChg^IX@SLnbUtZ4~+l|B?`NKjufv+lF5G$NevKExBX_-wTD| z6R?RB;+oX69W^lQ$o zFw19L(qac4A8ZVqvw31n8-xyT6a%j!WG>UA($0d@VGC;lL-<`p9~e!CNx86g!IKf! z*EXo7^n9E&{L{%{4+iB0P4h+eup|d`Cw6PBn^LhftdztnS?vVoIQBVydym1l-u16D zOTo9=Nhv6)K)By~@&&{7;x(JmJ%d4wVw57SE57jhsSn=9t3C@rpn+|9Vy~dc<_qZob53{rUVD4ZnoX?`Lt7VB5&avSqAZ@VB#QH;4#X+qV6>BV<8hr% zyP8i%r$b;m@e{#}NR$dd291IQnwDNhGGmDm?ih7&uG%S)(7HSg&J5vDP-$KPF$(o>6fpJSuU_bG;L~aHNx$Kf#%hH$hEEOo&aGH8^iVHRaE;Cl62M zhsU0Q%_mdBDcnE9JIAI#*3T%af(-HM@JO7*9HkW^%sjaMz;QyOJiEH_kCl}y{?81~ z$tu)#aTvneV)YQ;(M?2nY`_W9wK3{SxV-!rM){7St{FJ-7&MH7Pd>R(a0VMJ;|vmW zG8a6+m99~sf@$snfIEC$?`0cSs?GU3d}(95z!s=!YkSy#ygjCY)AgLYkL%)NQ?Kvy?i;p)Q=<@jh7~TGF-ATlo%~wi%lq z=VR^D8=bL~p)##qkgFWyhI5lgVGe22Q4Hz_ln<4z^T&CEn}z@WB4!8$ITw8q;JL=P5q{wUvp<9_wK8Xr(wV1Y?9UY4Le0MYie0FO`Af0D z&2v|0h%5+Ix}3uZ%!3;2JYf9dH-3D}VA4WP#LR=AyL0s3CGYO~&+c#5Exhs39KolB zv4N2#N8lUsRevnoxaX^FyEZ?6=g4!UtNUUf8!iz4C^zgP|8D2!N4BJ@hxas*m^PG- z`^g1g*K~22!X44wcjkyJH>b{XE8!r zk&(SuF6C!)GBLpzC|;=c6{8dVDu()5#(wJ|=hEX)`~`_ye}g-&i`%|#H0p; zx4yZ`7_MGr^!V@4&-;|40DStO(Wl|$*T8=+t2?ms0Kee)_wo$>J`iz+Wnb+6ZY`hU zG*;SqjWKvOwitb3(1~4a|<_oY69>1}P^x2!fOGkeETO2PN^D5`sBeImtngDm4gS z;h{T&A)bja+;t#qv1@?6Wg%z-eET+D}ZJ)l%HnH$Z;8gCAE;{RVD)`&aGo#wFA z5q#1+3b{v(FyUes0~y1s%(os3hB@u>uO(%^6ib$6*&=&p-#ZLpJbwtcJ{^XVzi2`; z3ZjAG2~nHMAsc~2K?B$sOHI=UJSzq+@@W0>Y;H63v8z{sQZRtghFU_A(?&gV!JN%p zApmQ;XwA>KYRbcB4Zl|!yW=pf1do@h9klI^$4k!{K5NDMkeKR9rJqhGK6X|K58Z3Q z;V>1DhF3zl5J$!!wT%d5rkG2XtmK_;tcXEb@ZuONcoA+CoG11j4LTeFeUIlGG2!c_ zoWt}Gh;S^#!Dj+~on|{G_LXQ=h|{;OcaYEAXrp}#FQQ?!L_)(iJo6j+n!xL^!yFSu ziHP9Ug^Aysf$y|(wb`mgT+!WH(j!LeEkm|||b44>U+?52zni@_M)Nox!!*xClPsAL2ka33N87R5~vvP*Fg-nULwwhxX zQnjb5S=4SuDG(FHP9)WAoHtF$-3-`{J2xY=_GGX;V0Tnk+;A=O%j%S3VP=N7q7Vf- zZ@egDkF+Q*dSwbqpV6<4#(NXpZ`!tM=$S;>jzMm`e41Dza+|s@K@&hyxHqB! z5ZIxmj6g8pxNQ(K*agtI?K$iZmQ;7j6BN^G%}Ct-X9RiiS8WL_+&=qq5WfUHc3RnB z!R4PdPz+=p(rj2Z&RC(I@|9-u#sRDfT>hfGXqS04qbA=Yn4h_*3_Scdl*!K-8+IY1 zV)Aw!wg21+``k7jweJdm!CTU35AA{Rq1zkn`P(`9^u~}z%*C(6_``V+Xs}Ljw=NPN zhmLGsIb)V!7EKfk%&%N31qjpj{ETT=NIV685sh+pqNgr+wr&+>eaoxI2;0K6*x@)d z|IC00lasOID~QHUVbfh^#GI@yDhD$zr3!#zy_G5~cO0iYsZjCiO_d<0W6+RQ5V$kRCEGP! zn`H)B8gF;RvqtQ;#(>m}ER1026$ubl=^3WS>yc|06oa*Fq7!^BfmdXy6fIt<|kJY-{@ici3Iz=R0XE;gq>%|=b}GzO4X zr}n?0^8SG-Mj1+N&A)PB#k9q1>P}-6M11L&*uWFhFi&cI(X4hxG^pI@)=b(nKPP27 zo?V!PzoQPZR0N}erz%pzy$#(wOmUg|^C$q>l}>VlS_@=a6?$xkOv`Tg{du-~{W<46 z{K5ADRmn%rg5bX%DKBW*sGS$%b;jA}J^EqA>FrI`ESkclj4fL{6acn&<2qu1z-n#` zsH`1^m|{omK0(m=A2LQ-H{J9&6Jdt?rW@B6hZ=p`8Adl3j&QRSX!nPSx6j9H)p$lf zlQG>jJ7aob7@KhWKo3?mp8YM@l;NhR`F>;V@s-i9dqH%jQfawB2L^&vm`N-ZMjJnx zCBnRGB*HL8rz~-39mqS{07OHxI7(>P?qu!7wh_+dMF z5&7oKsb|$8>?Ao<>N4XQ2i1pBuduor$Mljav(-Z)2&u72Y)9ww(Y*bm_H>7HjO=k= z1~@E^k45o8(-zYLbb}zpAZ;Mk#z8zb{HPT)$w{3;1op z=b_+YOIFLcY&1S*%6xJ?C&F0|h1^P39ptt<8;~_O0G(ULYwRnW=3W%kU7>2FN=+4!l&#rHGL;xbW(+a>#g2C$dmz zL9f9BODT^-E)!*%j0Y`IAW{HvleHyFtdeJ`QFIvTyV8qxlN3oEz0#Qdew$B92ixfJzuHHyG8V=bj zz4FxEq$IFu$Ki5C1tFiyr!%Zde|MJ0MU_fOV;a|IdLWr^PON6sr?z`J5txD0iw z8Ii!LU4gFG`G)bQ_-!8iU&Il&w(BIC^*4%E9TIKGLw*zmCUw<;pu}B=4}N0!xfyhc zv<~E53`zMI>6J@ki;oW?c1`5fuzv@QPDYBf*Mxm1eghI%P3c=DJ{V+@%r&OVq~n;o z_%N}<3jkHpe(|nbrFDLebb@O@yqB>G$mDEmzOg$-dma=+Or7tQ7sA|dogfAv3UW&W zY^iR|YH|n?!|B1@Gom+eE+nIiIaZ|H;|>IarsSK(1kSOiV(gE7 zMbSlyb;^%fqRR0&%3zTb0vd0{!3G>c>Tga^vj~zvZJ^1iEV9cC`(HHUjDVIpE|H76 zrLSSvGD%Vz#npwghTKX?E8?mJ8ve3*!i{h+`BA{vYf?6Fl_KTvEQC`*!%qRPQGguU z4}utMRe(}s3ho&{@PU9$*$=kduxvjDhp_^ve2Ew;j}Gu>ITC|*nO-tZs6EGnH^GMu zPmqOx2Ye;ExcnVh(UStqsrj>xf^czGaaL-0xahJ_QFm3;&Rc*3%L^h!SEXVoQmY6R zmtB;_a$s=0i-k+^fQVm_${hFNBVY3KR5>J$7-vy&sJ6(li_876q15HnOO$XbTag+4PLAmlsdCsR>m|-{7QVuw{r$7Lfk-dt?lWpohcY&M`r8q+kti~d==&-Azb5e~e9tNsU8 z#;}6Wl*a~bKK&18vmn0VjES<8O<%LpnO$vK#%50DQKX>6<5Kpsl9i39V&van%s`mN z8{!oSucx$O`E+ykNWSf;&Kw3H-Nx5XF*;O7bC$8?)2J|3Lm<6>+?W*w%)&7kvI z9O!A7wR)Huz~39dMVcZ4K~y7ggpzsn8ppX$a zu*v_{Z>U$kV0s^;_7OAr+A&8VXr*Z};y%szy4PsW|F4!bxbdb)&G;#o9l+~GK~{7T zZFvCFhW#cnH@P8~33J!1+FUOu>dnK^_YFaud^-Vx4j}trK13!Ik@cAcaR`Ch8?96U zpI5^Vk`(6SII>;a7393XoU;F%Pt4(<93O9HJfE`P0GTKlX&BMM8VczOCfz7P^kb$9 zX&%8bg@-w6+!l!D!r8AGdIb$%YaGPIkzSqF8Ue-643ky#1Vo`gJ&$y>>V?VJAnxE= z$>FcksFIG1$&@XJ+rZTW5eOQ4i$e z2Ph{lWDJy9IEa|sfRpAZIF=IB4OWcVmg;ZtjJpO#E_tZ2R2MauxB4FoA26Zi9BC}< zP?U{q1I!myzJN(XmFA98#gDgsLCMw_{}_=Luho7)&{ITZ1!W5q51`mO3(B)mJc06X zd3$UTg%DnV3r*x4NH`mNZ0jY$wNmOX`*9H`1E&(uSya>YxmB26OyHd;jns*28e{Yd z;}6MdEOwRZYNc|)(G)jVU^j|!7e=_Ywb#KKL)L9vdO22cl$&%hSQFuBaF8D-G`$m0 zmL0O@&Ihy80R@ZV&Q)X4U@g=Grh7=6yYjEU8%#ZppKSE_xH~9k+dV*%i{|=N9vKB{Zlh7 zNtaNNqG9pALB5qWb5VIy)3l-}&Digy0_!>>wQ-B3i2y_ac{UmQ70KnE=kh6j0cMmN z>u^hB4SsAYutlr65N?DTWA@|N;b#D9=jhMKRLThH5M8B4@RZGejOqU-<^%hd;fEKl zP>=W>YlZ|78CW@?PlF|u1bd>&d{nAK!=A=d@9RsTqbh`1VU~|M9`0c&|EYt{kH28F zn2r7xB4a)lvrt@Din>G8J%&u3B9=a)Qt$t@TxwiCHQMLWesnjnWwfq;vTX~oRrfUv znrn!fubB;Zl+NS)pf?L2MsuGPTk0K2k$byo7+VesmYyCb`chxw$e3pg`mp(A?zJ{) zIjrsx{NL0%)v4b92w zmuyb0lJi@oE=vw&F&U8I!wtyrysfOEEM{Xqh1#o72RAv?`}`#I)!ODD3w!~2R;{)6 zXXs|Q<1x?sNgT}0;btU4Wu zP@);J5Q>rjDG@?Zl1ITqI7cy@`%T|4kA;%C^#P27VE+_GIy9_V(bsHm08I1-bHOuE zH|WrMoFx}`zr^Rh;I<3iIVpX=!>3Amfl)Qq;bsj-c$mcN4rO#)L2ez+xJ=C~_S@V! z{O%G_m53BZ$Me$MxSUiZ#{*_^QdyB=$b!hXra0Esl#_%P z3~{vz#1^VmYxHAmaQ^>m!!)rm4tfV!V(>+)o*NU}liN6}dWmq+ZlR1#?uRnT(ke;~ z$6@f$e7c;R*_gOH_|b@VV2<(C3~>cxhSN)}$>MlkyJWa>p_6^WJu)A}cEZ+vPa$!@ z$EpjYuOKc&N~vCW0RR!NK98fsjIQ7#)-g2dGMr>*lTE29n8#@RTrZnjxkg}T0c5c{ z67OkNfK!Ie;0GLLfB>bZgO7>AkcLz=%^2))LZT1_Pg5~f!d?Lkz(C-K5Aa9=a+N^~ zf_?H_g-_0kPaqooWr701{=!IQCNRgz-7CjxV}*J-(PW^|ay;y#hp;OMKFBJLWY!~XzipWAmQwQ(syQAW zWoy;i9auexd3c2=YVFCU2^Arvg%55Z11>R{aa1Q1vn;jR)^ZCuOIk@RZZOxV(^$ZU zl!4w+?qf|4U5A13nj?mlp+shY2OS4j7G!HcZ zfuAUUE1fBN>qxUh*ehRuB*Htt5BEkiD;f}3h46m74M=ZAHvR-%;q(Ds!DY*}WjQEx zmlqiHQPD%TnzjJzA!?d~HH8I9P{Ln2m<`jOC3M7?f>HiPj!(#ekHk<9`tcUTfa{QrS1p>s*v9-Q96=3&Hn6QkP~s3@yZT% z*0K$SzcvCYk08?Dxvz6~XIT6D@4uU&l1?~y@Ze@DvuRiaMdriZ<9dD}W3>8y{P>Ee z4fYT+>WzTe5&Oc^D7~Cb-<=1^pIroJ;j}QVvR%3g%=v3bP;3UoI({y06h5iwPSUWr zABR}nmZi2erH;RdqO&wTOly1%Cc%PKd`!Joz zQqFnZK;pg$U7ms%YIC9A@E#Q80?)MP0VyH+-nW8f@iu`299JAXj&+|P*EMC*gF|U0 zvi29D3(@@z$g~er(uqi(m!uR=v%+qxS;N}S%FXR)RURgl&Ye*{*5;}0awMS}(%gnu z{J<`*%R`|x1O~wTOXzTKo)aMgK!YL|Y*2Xnx_`Zj8{-cj{u#*hx_Zm%3d6s+D9FAY zL~z}w+aAAynr%h?tlPOMQcXKnUBs3C$RFsx8U!Zq-H++owbCGtocIhE2f3qro-GEL z17AMer&W`}OP^fZNdg5bJ$|LetxsFF9D*tR>H5VBQk&lkXj;`9FM~_xVQ5)!I)4p1 z0Q9$}LZL0Pb`=a>UdlGRncJP$@@&iuN?4|FGVn(YO;CCW= zu!Pw^HQgPvy&%=0;{-Z(us(Qh^*;6|(6FUo^d2byOgF~&eDIPdM^$~$J>xo74%)$H z7D80Z({GD(YW@Nc4swRh`5T@U5GVZK^9R1oz z%)>XG$;k!y(gVH-Tu}}X!UEaFxA>uI?eLTYRb@rWt0R$k6(nU>4bGoj{dbx+Ix@R3 zGP@maUd;bHkck+i%{Y^Ef<{&+=zGU8<6(xGwTIg`S-O)Bu`1884>1u1gS0suq0TNpTDNMCxI zAv>jes@Wkr;4#+d##*jPhG0LgCmvy0gXjuoZ2>%@3OswL5})AVAojk?&o}7njmC3P zZZQCiw?&WTHk!c^ORwSM$k}>=!8`iWxxDaz_Md*>I2d@SbU&ZRu1Zt-vC(jme#t`w zH$JOI?-3o~V15RHGsT=2liwJMNk(w;}>ID{c2PnI- z8T}PuE7oVzrFSz%UrT}SAHHgOZIDM#x?(K~Gcp>~urIZFn+vw-Qc532=p_y+CU?dU zmuabah;1fdUg-ocG#q4k_C=2koGEw0{;L8gmhH-o?R>6;%Il%fB)nkR4}f}DP(670 zmK@f=A}N;l5qN@aBJ?eAM==}Iyg?qYg;HmtK*1o zShr(uLpILbRsspwGlx9iiwMXK{_^j)e!Y=ee85S3(6os?7mHWqERa5(vyTJgpCDrw zx3R_@l;+XmvDI8645y49U5S(l1U9^VdAs9Iz5bJf#`S<6Y&r_ne(=Ggi0_tf>o@KH zco;;qkUK^u%jg>9-MVlf91%tIiC7=0#Ohk1`jH2vaQ!kRR8nd9avQlmJ(}KZP}{)1ffZZ{N&iB>i`V(jWFj z>xPY2lW^kbs>`!B?ccwQQL?tjT0{F=|8dw9#hn*2^t<8iKg#5@Jqn)MlY2aSGi&0( zlLkD0-$P=@=oOoP$s%Tj6|c*I3ujn0Djc1yI@l1xjO7}>HKkA`yaoA~c&GlJoDxmh z&AogaJy-3l0QwYM`wTc=_L*o2CCE~GK5xiMNEqq6E4orhOZ<@s3l9Rj*j}siZ2qTWu%Bn#Zc0XlZp+O`A??`XmZ~CFb=&UYo_d$5SO< zu7j1r)Tt`fDXTbF6-Aj9?&MVm=4Kx~4u9l7joZs{KJvpKUXPbF_3y>Sjcg4vp{K;} zTs`uEWL*#WE)7aha|YKM_ysfyhT%sr-!K(LFZVu?=_;e`c|S@Wu^F&d^HhWZy{!{t zX4IT(BD$V!7{rstN{5?0nrl!Ri{XDUMhB}G3w;mvp{WmF$MgpcCT(TEC_X?$ahJo7 zY-W1pE$k)`3F@&=Ifaw0S7! zGRn5h%Mp(x@-V&8429%nyB0+xH)t=&2;dJQCJ`M`{QL&Vk~wHk$oev_D_O)lT}ZFO zJ2-AoM4HnLii=^I#cx}CG7wn}bZ;7nZ@L_zT0E2jMdkgT&BJ@~G6HAe*Bi?lbjpvv z8m-{W!yUmRU`?BS?91M}8`^2J+ff}@S4kEEV?`!=-|xGTs^8#B*F<%I9L#4K4uq(} zG3;H7s^iFgo&z!bmK6ZE8OquN0)l;jhsrqj`+xh5%SRvjZO^Vf`wt-psLgU9!j#~H zfXC4aDhp6Eh$6I(V45D;-%q#V*69syrIFc=IADc2`*;y%Iu#E z{K~0i2sfU158A&n#T5wb*zpT5?E^5wM~~xEhdAgOD0BcxATYznWM4UKq?lvq8q>DM zb1GOTFlSM6r65=__C~8UAO4HLwLZfE8T6ljaoxId(vVPM)}%JovnJXTCsgEAPpDE8 zCsos$yu_&XiILS|9ly$|su-kdMph+mm{6r}s0rU!6R7UsM02o7Ioh^KWi*;)s%KeC4TxKFv2*H&<45w9q=$w|$4$ciInX$j_y?ZVAvfiFPefms=bggQ8&0qOr)%x>pyz%CCt>HA0 z27p7cty@uz)NMmkoK2(&&ww{sRZP|ywCZU@(DWVZh&*~H=H`CYkNuVCiKHSoPz|Q7 zfV&C{`_Yq!dfoI|Q?UmhMdxWdqv3e;qH*MH<;$xbwEsR?uUvW#d|Yq`duutU8XQIW zWbAYWeL z1(Um6u2Q1)mc9vMV5lYfuiW79GiXRc1hc(>ZZ{aw(`NR_@SaP7#hBH?xD8JMYG567 z1)5)oT!R?CQA9W5&1f3Hg923jHhwLtQoP2yVNgm&oR-#t8|E0Stn)mBL z{$-l%!1W3R2b*4wy&tf}Rm=#Hq>0Q_}9*n>+xHR9N*=8Q!bhvJ)VJcr{6_wV0-6ZFv4m8Ufu=M6clwCvhx zI1@&ZF(XDiAa6p0A31Vl2O2+eWo zef$~|{rpPvS{|X6HQ99^zOUwdE-P1K1A1}l+E!w>KAocoq z*T$FsgL(x7Do<1N<>D4)XuSC9va}?`<060$m{dy z$76aLhRkC$CR69?{>2tuqKUAPF#JLGl@74n0H!fSrlHtP-L|qqAlaoA)Eg zzJsBJ=rP<$B0hUion3~bp_nkiUw}Q|6QIruuPD0bqVp{z?i=$VU*0?;L?e2^$D(~- z?fm8@a$y7%{|)=|kUos|`LqL~d~t9n9|=`-JEV^79|(fBF|T*2!+s_9rvGF^DZLlk z#EdH^nijRMc(6TFEM@H+@BcGcI zr2XT%0c|gsP=9dWZ(Gc~FuQKkrg`iiuOG8z%a#IqW6KtKU-{-OThLNKx2q3)c*-Me zVy3~7TKVQ{X~}`C2Gio-I{$w0KzoL8&M*`|+le(SYV>#;DpzCrx|eYtmNbmi<37J; z+jL<8XaXxQe9ku+t!@Jv5n%a>@BYbtj(pJaT4)*%CK)`3C_H1QA2{4}H)EIfdj_hO zHy^OnKE~0Ai&4Uu%jqMasG#Ob4I0@&=(-~+lt#n>ARWSsFbmF~LTT~R^`5Iqifux=S?S+^uC z>*`z01;;P#I|&PAnZH*sXGy`F7|yHrTd))hWBdnViJTb@2b};;o`rD$F6!Xm7)xX^ zC%-Vr>L^B6!-$hgr)AIw+jNu}2Q$qQSrFPkaIA?1x(Fw(F{z!2@Q5G|j+4LBX-#Mk zXq{j^SFc`eTCK};8Soy2kPF7*7W2|yZ($@u0R_=90r;F;PEqo4f%YFX0nN}Leye*b_g*=L9bG?ZVb zC>?|4k=w%5|L(gAN!=#dvM6}qo)HTY!%%qd!2RAa=#V|ej2Uy|1F(7I7GCtsv*5Bo zyWTmTR(-o}&BzF~XEwZq+ykpndZ3EJ+Q$^8_xPd=2vN%|9e7vWS&#}7rVjVwRV_N< z)^WHs)$4N-@IksC_OBYUJX6tG=c04FW2mJI%KRn+DnxoaMELyxGGvkPAY2NY{gg|Q zoDW0=l^;*9BfxnNF%-!DWF11yUI*7A&@E`J|H>gzqEPGbcn}eS*_i9|BlSDe2%epuoF`Ee}z?3qS7g0L|h8&lW}>2yJB2aIE>@4^zMK z{o(kCrn=J+c!gA!6{N4 zsB7@M6|xCfD|fTaZw-RSpeL4I+$RP-#$m=@zTBox7!fFC(eO*J8`R@*m_Nny86Kn% zp;D-hao0_`Y#LcpbAc`7{l!=!O+I%Hyil|dvQ~cjXb94lP47fV=W@REB^c=IaVtWz z`%(xaWrrxUPQ3KtAsDTRi~8oOa|=72a{dilU~bm0-%u`{=(+v+aeJl}SN3n0)$T9< zIPW#KQ3xxz9Y__Nb@R)x@q8K+xD_J@TrjYr8aNc)#_rCFE zO~gXaCKBU#`Nwe_a^i6X1877X^9wu5^9oDzpodw6#)>pfxnj;9Xgr?kIzppFYLsq^ zxqb79?Fl(;4|)HH5J?#EUbUFNWPIG9jNa>BH@K2|3@FLs`E zW6pQ2atNnVledRclTy{e+p1GF-h}GB@cO~CMo+4$S~F|%s99AbM@;y8)eRG7O}cl~ z;7QA>C*C`E+23c~FuS6nVpheZQL`$RSG}VgS>WlWc9^rRb*d(1+u7hz@jvzitNO1K ztbP3xtV`qs>jjTCpN%uE=w2u(kWj8G=8%{|+71-{;crN2^s1*nN8dG&)AZ9%x1Mjz z*Zurq1|5G2BPq95lrJFN;+E~KGt`Wu08ft?meEz2p;>J%rBlCfPyeKiaU$FcHh+q5 zrhQ=={RkyCb^s!M-C>S&=bk;AUp~N$GCoE9`!66z@7C4T@v$)bAbQgE9p2otY0WQZ zz!h!aw6}LOY<}K9??f-Y`|e&kn0#4BzZShW@*u|rx_8p*HrMny0d@%WoJ1`tRvozVp+cV=ERPs z)P+meQM768jT+ZB3H#~hV11GE1whxazbw$&W+FxS(A4!IH1*G@Jixm>OK{0rHiaFL zql8mVfb_@!)sM!_1^Q79c|!H)KmEy1dySZJUZ+v$v!D#iNWPbVv~Un)ss?#n$oCx# zMH)u!WAIV^Yj(j9^q&k2l$fJsh|l>b$Qu+neSWBv6h{*I3Lsod^MT_@j*SLz)l3hN zngRWW5EFfXdl^F5R2R_c04~)5c}m|ZI~nUZCfwJ2KWNo?h`{1hq!9oF6J>!2i?YZC zak?A6W*%(ViU!8DaXfvz@8>4e9bO@r80%h#emb~ys-#gP!c>4m|1*pV+Ph)Tn-qKl z3;^E-(2Q>CAD}Dtx8?%Trv|9neP2^D+X$m^42fEyo3#O!DpYo{`qh7TAEi)`tr-qm>4@u zE%0&>2}3PlbvP6FF?^0qE~|XV6^tlbPnS+6WWc^0HVMX}vYwRc!`a8W|i3KT|5%uPyGuzf$Td9XtXu))Hp zo8k-eQhlgHD0NvpnG>?xhwSX-l3$Q(bjGZI{0*`Tw-_C2&zz-{0qX zW?*JmWWXgAmj~QfTyVi9H3wY6Y;o7razsU3DoO?1@&Oe_G#!@&w;VNhttqttE62<& zvuv>}H7PSSM98w#O#a_Ld^3SO zR{kR1S;Ys*iZ414gm=b$X4dgX&U83omSq$Kavti>+78jMYd3|K9SrO9CvbLM(s56@X22uZtw3flQk@zFqWH)O(jcqP zM{ZrmbjF{zniLAsFwa{912f3%h5(BYWJ0g&@afq5US5mT1fTHLI?3JBg6 z`-lT4gD3`mdEwEIi+&GfTX4zwW;6?e%psuwvI!pW7^q(GxM}J{U|jL&L^JU|=tu&~ zxH^#^YENd~$$(T(fX@GBUF|~F5oFOE%rx0leH1(2;0}qzWmquVdXwE?GK(jFF(LHIm4|sw@%hon~-xMi3Q$?*fMa(U@Rl>^7FU{ z+SA*!CX(h*Mqo@qJ7j}Gm+{l_*W6Es+04S?c7C>#N?m3&Z;JBN_z z4~zieTH*&gSAap2f-*jsm|-!!#BmV7u(CU9&h2khVJj8eZbZLeqP+W_DIQDYRl7V zOS%sBYZV0Tb=9Y~f*chH7lnaZ!_SE;{>-h1eg+bFlt*i=#fe0acs zwW}o@<$UY%)uj3`!i*g8fsNQB;-bQ?`*&OgNvfj;*h5E6iCGEJ=5yS#)vzNt+QWlL zNXNs)Bb6sQmVpi_Z4$OmR`IsLWWvkY(SkPT;(%s%qsZ}0u|r~v7xF zRcYaZOLq^5di>9{?(tM3W=|AFZ+0TCX#Sg#wfp0Jg z_{|Su*N~MP0#3B%Qs}t3A$1;0Y!8rSg$dOot3-p7zmjB~Ndn(rj|huD_7Pj7vPS(8 z*iQuK3f3M}G3CO-P_+`%niF4Ow8?#5$BJ46zoubKsl9d-JmWgaGkaTGaWNG80*}Kj4=5DcB%J+5CIlEpvEWh zTkTfCKPkitAkqUSGUkp4D-fe`k&@)AWX?){_B{}3SA?yek(DKc5TM^L=WHZ1&iDyA znueTlzV#}`K8!>JR^&im3Ci@e!aUd9vq1xh$hyiFjWElC;uSaqc7_{LQ*;NKg#+tO zAX_+EdqB~yLJ<$JI8tDMj4Rp%-AYb`g$EB8;7I|xn>B4E6df7ESGdpc5YySIOk|Re zEjj+-6Py4V!JzZ(G3NMokX{o13V{tW$s28r&*iZps#WSEA!R_$Y|Mto2V*QcVxR~i zS%;VjYa}1y40KP%l*rsfvQ;=4X07YWcdy*%?xWBm|3a@Ky}32`HDpfcBtNHt2BRU=WeR zD(U`O+XM@!7KXrTso!Mn6Ejz}29p+z-11;S%0`>(KCXDM`Lm0I)7Q``b0-TlFrnbVC`s*>B%-tP0RSm#qy;jQCzcLxIzrtvI-R0@r?w z6W3>O5w$bnP^pC=-JgwQbINS4kLP&v0yIrKR?*F|)?q+HF-~HfHyaZckD`7USCMF= zc>KU$Y$;{LC1ov`M7weeEBKj80!z=|uYOu){Hd)X;ih!DB}!T;n$od?9|v(nz~@Kd z`{2@s7$*NBAUmfZYmAeye{9l?YFNT}JV{)?lAEXGV`>{QI5pmY1O^mCjuJ$azlAe( z4wPw91GIaYaJ%O(tYAUx3M?WE#|2y@Z2kDkCU;uU;~rRq$1H(QCJP(8p>8}UE2_hc zUO?Y)5zRDcGBhiH%J(@;dVA5goUpE_4{u&YIO5>}*d?_xF@>EL-9YCaLB|7enVAi8 za$77kT{IXR#Z)>nSJZ{T`fytz;^8AhI2yi>!g7@Wu)2b3e_UATtZnbUtV&q&cD^v{73LH8$#(J>~;gNtyNK7dPYmG;^{I`QH2O;)YmQPh8++izRritTrKDjjNKUF`SG08*@b_Z zC_8ry(K-GfQ@rP&>ae>!%F!v}Y770B%hEcdr_lGlD>Tw`r0#4f%0Z=Kq~j&AuEKq; z`kT=8&#no*M7_d#(~snsx}tu^)_N(=utvX0U-IPu=aepuR6xYp+jzm_?9f^OnBnmvu=NdiW zcV=ei(({cB?q}xfw3Xzb=4Ahoy$VteV#2)v;U4VMFZ6`#UPWGy2`H(HsYWz{m5rJ= zhSY_DLinR}4LWLM4QnuNW31O9y#xHnbHS;}%*hW}lOe7{xqr|f`jFA|0;TSigOBKd zbnP#@@eYd-NbD1G%nwcIq5?iN4X#!M!o45nuO09n9aKaVTxmvFHr~dsL`H#=-*8Zb z@);v}hrsCUx2%~6QCQpf@D6lG0mh3`cnrQa#tBjfU>}ooVWOQ_Uu)8cF86^nX;f-_ zfUs_nZXfvry(^{(*bs(~d%4pGf_qx5M1%Ph5`IYzwif(xn7XZp5HRMW!UfF|)%meTkK;6$UK( zbaL>Hkkri@@rY+mWS_q%9~)&8CE5pH0UtiZaetY^eq^S;e4ANJ zFU)P=@7}~$&waA)&P16&p-Tv_f3Za7Pe4-E@TkUxAkaAE#u)j#kzum|mEBwlLt9s0n5F~Yhunjo0~!U!-*NsWwE>&L2*QMSk^sH%6_15bpod85!FSSpE7 zlPDhJ(hlzYaJRLd%^C>-oWh|1l1^gk*)&ZJ*D2brE3s-_B^KzS0KDA+^%ZNRu6Tzl z-nzB|5mnJe?})N3Toz5kAcvn(0e2b?6AZ{JaF(i|kZ4Ie`UCn2Mmz8i#=A~YWW6VZ z4?rlv&Yjg36IbBvdp)=!ebtJn&-;zJCLtUIHCP>HsbUmi5^uoYZyZ*(ld|k|2|>NWEpV>~6ySazGy(>Z z+`4qp1lDeL!b%T44-ctHxWeo$C56!A_3V2v`QNM?+))E(Wl@6G$gOq&v9r{$}M?Eg> zsw%1dT@Yk=Q{)!}oQAU6}xu01`z7_^pTEyWq*Rha;Nf5`_nD&95v+Mc`?B7(CuZhrtWKDnsW1?|V^K(Sk0 z{WWrJqr5B^4Nc3A%1n3JGxM`4KQmu*WlnG|z%PHRGkfTGXJ8+fGjM3OGjpLet=MUE zS&H=RfoV<)CVT&+4_yj7Ddj0&ViO2E@tXu;P$j&iFcdP{2fH|_vQ$umZ$=VnU!c=r zi=z{Dx@K;WvOR)AQU>AfTQ~>XP_eP+X$=NPKtS6^zd&#-V(pn~NtY>)*9IX|2LxAl zJG*|DKvl&F9n?nav}YOzH|ih1HUO0{>aPa0Ka`wPsiQ@Ij>iU`Rdpl`>Gj^cz8s6* zZHCGlr)V^@27A$jniPXw<#&2hiWieP6Vx#u?0KZ|#;z^jYe28OglTx8IXjfPw5G?- z)XQkPjPz$!C(h*(J92XLMpmPX6WJ? zOo%wn&T@Qje!`jCH!Muf5qgJ2s_`v_^_&|<>!0qAFm{?<<2)y9cWn!GoD1zdYErY< zYax6P1MrA~Q|*I=!b&))InO3*4EBNAfuSYh+)QJB5|<0LGv(c{d{|}lJUVXWONT&S zCSMosN=fg0I3_|)BW19jxO3;q;XCq67~Hr%2uD3olL*E_@l);74DSjjIc55bAXL zJ}e)4fkoiG*2!o4kn*7jBFg0qekel7x40j08Y_?|pzTI#`JksVZ~ZlbOc3VJ(4G~K ziBEY2=ubmS#IMN(IdHmHOcx$?gi519GH)F$%~0DOgOXefrT3AT&LItZ2kq-4JuzFM z8Y@W?Ux0BXX6|Khn|I}fpW1$fnG28HKo#_=4*qo^Ovjj!q~>6Ms|KnSGh207)m+66 zVS0c-7oJStN}0jV@tG~gWM*1YXQbn?ZCmbytiawAf-=*G2IfuZ9hl~vXC3OC9yq=* z+nF~LV|K#p{R1JOj3$#Z$*oNEC{7lZeIAlf{8XRN)zS*cSrt^Dw72=s9KrjJ327kAt9o{D*>aNz2??RUM-L1gbG(NB643N=jmFvd1>BgwOnVk6j^TXu zqlqB;j`yu+TF}2^?S`Fw*a;o{i?u<_5pUdU78;v|@7O8LHrQ%`DQ59Aj?<|r=^ZyG zC!gdj=o%=G0;57GZz0Z9f>qu4; zkND{rKBt3nwOT)&=Ryl`BcZ8}9$7wPPd~#&2Xi(XovPLb+Q;4_m!jlAL|DpMzkbE6 zlcG{!gNq1ju7U*{6+=GYUd`_I`G%qY6fTy4cB_`mZvaym#uUI!zlgDbz>Mj9W?HI{ zRy3xVZ#iefMgNNH+<>b2$s13bsTMRO_X*BOGaJPfw5FU{96$tfs1Jp=y5<8Av^6y3 zfrg`m~i{hG{Qv)3h-t|3)a+8@&IhncZ6~}REIP~@IY?6-VA7E$aR`FZDs%+nWeN{Ro<+?r zr>43x(-a<%c`e4|da4@E+^5BJc(uv$Xr-gOrpca2aJbQ zkT?_Std<6WuK>oAmY(GurjEve-`rFN&B zP_iXD$T5ODMcp?|iJhClFmCGf{v9wD?V}tgFDciX5AS?DlE4?MzZe2v>eZZ_ZF@g4 z??iF(UafiiCC|bwGd7OW(f^0hlmLfVdR+%As4u5S&fo-&qhB<^HyNWjKA7O;``kpB zoeMfrVm@x>pjjzoUbel~q~?*+!yOUm7W*{X6+}1c)7xRS^mFJ;1=LBU?X@_yo6k2g zZyKG{plMIvCJkI_@xd`u_kY$F%fM`&>+IAbC}RHWfyo`APb&mBGd=IZkaYL4e%NU_C>Tez5xBh+ZbUBquOpsB-Jf4 z7^o-N%}MUShho?^t_Pq4b%HhzNu*<(KYQ{(Pcko;vjyWtmOSGNX1Fv55qUT|47=l_Zut_rrLlgY}Vyi}E_~ zJLYdB@6gCB=W;k2OG;kRAu(Rr-N&*FdevOsO2EenYbh(cQAth-<&AWu z;R@))g(HhHV{zMa=&E#F0nN-$o6yldv?xy>;nFAN=T36vj&NlZ=6T8b_&n4 zyTj9ad4{J!@+d!zCHG<6i%d#E4DY4cblyt~Px0nFST>b5@Z$qNfzI;M9zr%sX=rZl zR7F{q3jmVN`#oad6R}%X4rKM45tH}q|2B1T+oN^I!mx(f0)G7!b7VB1V59FSt2S@k z$h4BWapT6;&;vJa+}wcF1353uc^y`EIk4cRtc|$!uxb^M#fz&}tr9F5?cGb3aOZ2f zWXX~<9BlF8#U~@4dUMjyv{C5Jr0#0JY0BVHd0kI9g3Z{Jp_-V|#f{@{;LUw9XjliR z^15~ohCx_t07a~^v)2iri|!`?SCK_rS#)5~dQ`{Ds_udB_qC6|cA$4p_^eGMWkJh6 zy~3&14h{);8$GkrJAj6;@S?cYUogZG$Om#DL|pB+2hPrXek-tJCe{j@uxtam>&#JE zs&^}4;H6Hhd5DPk@z_sNF2UJ0{8Q;kTS7=#!{)|(y%M8_QH2rHZ<+U3P$Tg?g;OO* z*8mw)3!tEiDIM1)LcAQIx`DR@=)n}Q4YcM3^ur)7Dk+Rm0GczJG`=ltn6&t3Lf0HCojKJGOBf94|v%T3CM^+55g5R)PqWckfUVPCuS9F zXP_m2<(G0GfC4olKr+d)(MkxWVQ1?lq!23#1vl*DsI=~D4&Vr-rsx6XaBRg)`s^rj z#T{gnarU=p?2LC{eWhhkqtkBz5G_QteV^v^6h9G-?mL``G z)WzPFy}_45-`k(Oze&BC!KMUXb-;~L1J6dqwyNloH*#q2gR5r@X+OQN1Ydk@+zMRH z-P5`9#!sD>Ojz^QwS&U|Jh2-0n_tt`?E#?p;H0sj+AHKdUdXH-f~#ft&7i1$vu9 znU&{FvrZO@ZE~hJZGvZ*%aiW%WV<}u`A?bUaz{^xoGiZGzjv&Jdu9@Rx7v1cBESFW(l89^4_RF4vbo?LWc&GnFkQt2rdWm0M>JVn3= zcOvaBpuIV=Wah7&uR5Gz8A&Md3}A%~Iz|$pn0ACxUOOzB)?-8rY{Q)lGyZEPa(>s8 zZhhQ=`tE{5H{R9mnCPfXhTbK^*1s3@G?k*hqrNM?0=}m7Z{-#U$dg2jg{7YXzhI*T zJ1Pu7UaulwIqx9+v+j?3^o0fk>1gH5XmYd7XU$Cwb%egEYTvqlda)3ZC< zOi5qtLP?|kEdTZEeys$(vf!}n`XB{%au7%H{%f=5F8MJztK7(l*l+c?XEZxFea~*D zS5af2C4v)=UNHvBiO-F%&Du0xAS(n(EWRZNn%GDt-vKXUYXE;|*c%6=k!vvQ)>o}K zNf04w&Z-ib<|DV>dz^#F19sLw_FR@Z_2A={6CRlpK1u2Mhw0`Mz}c9puuWP%s;ns+ z&_t+QRMM7%?|J2tm-BY}`&q}gvK5qY`%^D`0<%UA*gKsVeRLJbDHySVa`JJerv%^` zKm}{8WF)8~q-NIOBc#ttC;j1d8NP{)5`?$HS$e2u4$ny-0W8xX4alx!({ep~p=Ext zWo+IP>3K7X@@7P*7lHexbLaH*VbK$|MUTyoo!DH1ER&e+d043C#iY3MViruy8kmr; z4b*bAfeL>14D&#D%Cqj6XWg06%XDk~G1f>NXNu?QENVfFt*=!B-4%^-@A9VN)KbCe z2p@4+aLy8?c@-R_Js9CQXFF+ZmO`i$uh2B5p8G+F($G-mgR$HdwURLHG}n+0X`a;q zkop5hcm%XO6{DYpG}^yNe+FWTZsyX^#H$^<)@{06#LA!i?weI7T3Ibl5CPn`Z@<`r z$tP;H0xKraDHfb=&Te_J?$eDP3CV@OGIdwCCz{&25Bv0KkpQ`HLnCiCxyyM!izm6z zb8q&67q(|{9vQi>LEY%KmU>5y)|bH-LVzFyDG7^jSqf!b(v|P@(^%v!dNTRxZkEOY z&+#Q5x4ATQnx8G^Npr^d)X zSy`m3iNC|Z7De4U^Rt>U`Wq%n0JX7Zq|6j0Y%|%<@lbOv2Zv(}M?q+S%94+YE@0>Y zxp+Zl%oonMuGNZKo|&|N#2$rNBF*$MKBvGB2akFU5%f(tG!ze}PB)0RQ)zp@f1Vr~ z0woU_VpB+MfYZ&JszCvwsp`RN(|H42_%=f_2I}CTzGiYDugC12~tnJlSWw6VTDwe>kTek2?|7zc^2fnW z#9OiR`%dJ~%7vb8WP!f*$-HnOsRhk)l|03Y_S>^a>)gAHL)lZP|6| zPd5zIk3%v*g7^CziF2!}$8R|ECA%p#AGtQ9J?93nV*jZxalF6~`N)+S1B8i0J;C_$ z>Ywf!>FMGW5&fzD#q|qCC?I`Ad}>Vj#6fZffa|=~ms`YQCK1z)<;dY8e<0 zV|CqwCm)m_b!*nq6uMdCz}PqgQU;746P#JB=jGe-`er#7WZUZG=UEmM=gv$^eRghC4 zh->L(O@%NOwZMbx240sJoHi^1#J$K`A9#DAHPS_qcw)LUEgI~#26Tnom6M>Y>M5Eh z-0ls>?FQRA!S{Y;##Oty`hCIe&iw5$1x*cZcjkGQ&JIWn!1AwdNZ`CfPA6&8xTCiGG>g?K~MXZ>^p?IzP($_4IM4bEX>AG@U%$e>_HQ~8b z!4mvLjpx>l>;F8~z(;cRFNaMU3=|xLB@o9oU{L`^C_|q`D(Y6B7zy)AMhcEZkZc;o zP*ZjSDDR|Ws_hfA1?1ZBKyo+~tS>t_cFzjM*a|0~RzuMV%<${ro*YKI&qtX)MX!UT zSxN!E3)?UvPoIQ>)+Fz%qtHh~l(Od}9Kli4AQIb1>DPw2_RSS+-ar7HwreUZnDOb9(DFB0`o1~3dAKC=Xv{&T`L4PAvD$wa=( za+=?J?<1U1!=)zLO&&nuhAnWsckxTZC;{T{z@h}qhj|gqNBG$L$SqGb(UehV*xEj* zx&*4(kO@)HV_JRYFuy830O9YGF!3ITNtQ*iO7Mk~if@Dni2?QnmoM55 zLRVKEUvV$oJwuX+od#}?YSm(hX5yU?MmEQhZaey<0X~lJMLPzly2z3JNF!?KYZJ=3 zgQ#l9d+QL8bda+c|@wC6V#vIWup{LjpG0Gvi`E))yJ`!~}@jo5Hq z))X;K0=Tit|Px}WAc3XcV}bM1EK67 zjT9YkPMNE!y|81*2rzE-s8-yfsFr(tSX7GjrkN2PP+eRGmZaABe2N!#5^s`P<07nh z3IW^+v5+mQ;lVzzfEfj)wMM~j%#m(-GC$d93)=s$Gwp zFlBJxZnn01-HS)h%oP!Mv66$U8BhLv=9^5<^<%TzqD!j*eP93SD_6z`)yGC7LcQ@j zeqyWxiXma{O17rf_TK z@%)8$j9X}pcd7UvnW9V{@3jQtAU77fyh>zSJpUC7`LNuSU|+GNaMrocgAO|mk;6dl z3_PCD6M44S|0Jygk<8G$l2zN76E&28KIo^aV#)!VdOPLWVNvxPtTQ(mqYkXIG)m`@ zgWC^BK~(}Qp-^V+=<`GaQ;X5jKLlaAQcvQ9u-j$p@aTGh%T2>xR&o!eMLQA9+C7Z! zNx}=Zw;S&RIsi=U{jQQ=K7C=$Kv*8_LT%g8yjMd4FhywQu3?T5K_V2<^k(a5lDG9! zmuph~t8ffOQaK%3d|h-q-GKIsq zJ}9(t>jt}@UD2HRM1DbHAa(&|wGWL5_11l#TvnwDLxt~qtUA?%kso!cpYzz1e`j-r zV?Gn$EX(a(oBBrsy7tS+|_opD!L~d4VY0qE2;gTLmWlSSpo!!XI^4B zv9irbz-Pa^zb|%ciy-h3O@^h!_Ny748`Q{OtDqh*D_~i$)j}yC%{4p@gtlouN`JZr z4Z?o#ae-~YcIh)wP-lGIcK~-m6tcIWvaY>8TGI#dhv`7ASmh_o4h{Uo{@x?)-?=+#Hqi{|NCZ`XmlL5ydwe7{~ zC>a)2rS8NI%6C6hCQ--_$!hdKJqky{pj!ZT>syef<2}hUw@E*Yl?thPsq9I^D?HEV zzWCVVb&R(RG)|i8xcN-2>e{V)iJZWy50903y}$nB8PSeY95iRnx9!=T;qdrbs;#K! ztgS!N+Wy-mE^4s+mw#`}kH&Wx66!uFok1NOat1h>)!8)|&SfpP)!#EN7J11#TI=Ky zhg}h&=@V1Ufv2)dhaP_gw;BqcI{rz)_u1XgpZsQHQ*Lu-%R%X57xF!C9sl^D!gz`k zgHq>-PJ_>0F(0!h>6Z@QisF=2a6X;-zC}Ma@u04KY;*abr6!m$=}^G>F%<{@;;tlm z4D*jp-rz1y7qO~*NtQ;s$?`>5A!!C=fDubw-EK{3~_pM9vLw8;(AT9>3!Jh(&w0L^SI=F#8Q%kuwes zci$@ai9XO6^Q|Zw-%V5;yaBAT9$RK<|lbWjkL#Qu*7* z>oWWAY++Smg9XH&xgm^&E8&IPU;7~?#RH%!Jgg$Bi$#%utwP)6lBN+ z|1cQvBBWk2AFiJXclGM5tPW(}Gk1fSt9AEZ2S-09J58fWxEhlCEcP^bLtUdz^WDF- zxDF%r8Cx)AN@du`JQXzjco?ox9lThXH3qiCnxVDH)g8ez$or;}eAb8X6&>IMaPEWh zE%meY$DY-g0P1`qVuT*`Gw_3R3FOS}9^P`*cU+_Pg+r3*iDxPJs&}hT`LsSc3(z*r z+INhimP-?-KgSh3&AB2Z;X9@0x=;6-=?Tash!SNPh9>2P` z!7**ALBk`JR@Bv&4U94B@9TxTUc>T@hunsGbanuS*)YA0sdG4CGbJOo@^VApXGlcP zs!!Pe$R?X7U3tC>sPe+f#4u_jx@lEtSv|&pM+<5=&#CArbrxXl^E@ssa^=voYT05F zt=d#y-#TGFq{h^sdmYJ9#G~01P9ALQ>ND$Q)akg495W^i3TzH!%RMJ5h!Pst8`G^d z<8os({_^x&b<@f5CGag>Sf)&w`}xnQiJdd^bDmP0}1ICas<(dVkO7w!vv{)~g`Cf=)rG z5@QmdvoQ3TEYXRP6I~!9jXZ1=X68AxQ_mVnijU_XDtiVBKqGPhfoi)s1TRm}nI^G7cE`a!7Y+3S~19iEGfJ_hQY zGB0EQ0y`jBy{>Of*#Tq+xe@olFaLI=+V~FOG)k7i-R^rlN)kSwhEBpjy8q9wBkR)( zWg^6P>6EyYL*&G*rm^1=>aE#D%n@Hhf|{s9npcaKV)>;~to}P4_%dM|YT^_wDz!rw zoW*GbPQ<14D=?fbrt4Wt-ocsns<8{rO>vcQ1^YYDaQD*Y4SQ)<7F*Gw>C17nym8U9 zW|}km^TgnM%uJ0|!lUyet!Y5fl=;5JdqqH+7eUj=5qX2Vck4Qj zj*k>tVEbA7`>J%eUe=|*uMJiUqqdw_C&+KtQ<3vNJoWYh1EX5s8P2nh^Ru$W7{4Lz zeR=3#oVGP%XX(0E(A*3fGARETQ~7BRV+iXULAvX_*TdIZ9n^pOJcpv#YQPuKNq=*L z#6)Z%1!T5;w@kIe0tIRF=yZL`ABF^)9-9?+a9}UaSd_V{7E0cJf8W5KV#fSt-p6Kj znREAmGl!PVUt&s%uyUNI{^)SZ+P8nQb>sXxfgG+8=mmTSH)ZTN%u&4HIUOyIR~PZR z$m*RZG`l;WaEOW*BSscVz`w|ve~&Ebk8bE2%dTE&*f8xFpfMh&-|_FQOXSO!&W-*>D9mzfe=-~O@Ca93)M8J{(;nWzkB z)~Ofweqk07IGp!xz}!O<|Kfru=LW7CKV#L7@7nMVdU8^kDZ3XLqyRQUH%BvlU~^5N#*No0*yoZ()M6H0}5#) zki|i2Jgy&tCH0)5K^g~cWZ?Z|u%${iK$zlm0{a;=BwM`$APqp!b^L0U5VHWMu4WkJ z-+8=J%5c*U{#Pgd{L2rw{czV0_wL;-lGVs6GEHVtZ4}I!uCRhSk#&fL8IL4Xq$2B5 zY+h1Q5TYrxpJS6M5($w<39?E_QI;Ae-1bp3NOzQp+EjEAMfVfJ4z!tW$I24diBOpt zV)!5f0h$Phf%X$}Ww&yBYMac2=SqPXp2h`bKs{mxJH|)YnR6vW&^K9Y)QZmc#~2&39AOY)!T0Z>KoSs7W-YE`O50!BV@_x#NJ&NvBeKJB&L ztaL^&y|MMoW$DW48|{a&BFY1?_jN0ye?%W%6b-8@>f4UGzYx zL@5Nth;idjmOh`%WF^C?DdQZ%ppiBPvm?*{vP~y&=`rNY`yHVMcnnI`5V+G}j!;v$!qzlwj<1INDX{2W}kt6u_ zweHPj?R1U|@E3a5_&zxz4h%+qcTH%(dlw*j+7udn`iK9jMv+^!vPn%|m+Hkqm9(k< zc?F0uhH>`)shDV^HBUv|FiT_kRrx_Gzai^_Dsq;JT#>%~Pm>Jv$3J0F{#;i_{An@J zeO<|3fgVG@gDqdiDi1JBm&^T5*BXM@kZ`KfANf*+kvNr?8|hJne}|;Hyd76pb-7)Y z$yQfF44X)zx}wsb72GnAdiO*fY3jda zur$;^yj>d9A1cHQhK9ale}6NYEAJq$Lm=(ZSt7lfeW3S~ra8`knV`X8S@ooe z!Z7XHj}OLs*|`*mgk*@S8c`?=)Njp7oS+FXZ~gm8zv&NSI0e0Xzn65YQTnry%PQv2 z6S%-y(%8>9FW(MtB#W(-Sk*3-DoF9l9iUn zm~fpyI6!%iMfWkh2Vx*cRv8sVVA^Ar3A^1=@(@>%c%ZqpESc&}iT2{AT#SZyCvknw z11gER%L7@#+iJL8=Q;Urb+SktncyovelBsd28U8e_!6G`9FiZnT5Qw|98s3yV$8E*pAjpW;?mL}Duyaf2sk() z@#5wVPDIwmex<3=rMLpo<@cH`{N{}SQ8S{(3gvoA)DNzObNvLwfqO&shJ!a-aB@J8 zph(;d3Mb?G5Dd6FWaCHWaerI&<_KPD$#ty@uIuFrE+tZ3x2o6lx|VrTa1RML!#h8y zL!REy<_-;QY)1Y-V=FpZQr9&srckx0SOub1H1gr<3)=BMMGk2Ci@BIHq$Kz!oj*yGn#+7Z8PP z78Q;d;idouM?&m$>fHCy&#Y?7(>g=-nR6G;1)@@sDYjoacj3a<7&iACe;*Wz%05r^ zHSvv^vz zluQ7zgn_0Mif`k}P%6K)0(kE8lXbf{8X{rC4D<_ciKBGCH^3vHrEkbQU&ww^KpFCBU( zc&B;idGqPy{b~bkDSr0F2YtrReh+~g5)p^wp>t(eUnNTTlc`zX;n7%8FlhGM<{r9F ztGfuUhaR{lSOa@h=OkDCjnY(G$U*j)ka=@OoZHp&k`r#1%g9&FbLSW-+1-cAD#7(b zj^(@WCnY8INa{rEoe6vmBdG_VdsiQNsCeJeig)92!S*4>M!sPC&{e$u^;}M~RAnY4 zBv@kYuRt2yBF!_6qH#}GwZvy}C0Sw})<>sFnOq=H0wBm9~dn95Vm$g31v)VSp+#J3Q(WO=QiK8iz5os{StcX+O~K9&&H ze%82oc%27zS(DKGs4CU$&|pAyVu5AXFq?370=HWkFX1o72{({wx)>F>Jy?OVn5d$n z!bvJ?6&3J&Qy8EiJ}Cj^pjXw3a?q+=XE~J>zoVG`I>Xip9Pol4ghnVx;SPMz5p{eN z0&MX_9};1>J>-@WP*GWV4Q6yN)dC+jGbrK95wH>;AWvS9u1AztzI)fM%G;mGb;pEEhIcLBl;Heij1i|Hp(u_-JTm{NKRuqemT@Oo4$woM9 zRF`YegPw)W4LQ?5kfA7tH3Q472t1k1_bcSWn1vcbE))df&0_h4K%gih|1nv59xJmg zoUQ?OoZOEV>`Luq~BblYjp{e>r|JUPat_vgrh!^#ymCWKwqNN0(= z;cApRK$r6$tDY#AH;08a=2E1YH{Y=L0%!kX1!itRgP1{biYf+-Vy?Wwi3bV7Tq(Ew zg)FO;*V)Zfag#L!6&gG$Rp9|AxTq9(t4fqPJVX@OD)Ofv5}{rG%nF-=kIhd)EHDhP zgULT6*`koJnK?tMxsshsOIYRFl;H2L%UIHvXvBy1J4Rx3%5px~`alyil^}LKm=03H zcng=@Gr3=CccALNyJ#nBlM}!m29)yR-w)wS+puxSQdUEGDB>t&n~gPwt2>^N2C2^2 z%t_<`3I7$8!h+rk3*_!%LM5ppLw>nq8~5v5E@$Z{IrW+NAEK~(WpQZI3scnv>)Y)^hsjqBJ)=W&BaPVY-BIy&%0;IlL?*xYL&NUtJsZxt zSn~5?3f)M$=(Z#xZ-91pCDoXO0n6yHe=|iDZ>K5O4v>|4pwAu)G_G#3>Am07*JI#nIvxh+80T zfw%?Y7KmFQZh^Q3;ueToAZ~&GNedW=0I5+?@mzzE2i$*W;4_`)kaP7t*WM^5oeYmyUyb2Khi;V0E z!b_Z6`&6HnAi{*44F25~9az?R*cSGPbvNrVo*-D_9EH#%5}uHk(2m9>cHn@q5HcP| zh`7M#T^`Csy9z-B!vl~7nLY5!uK}LBxrA904)|kp@U6HV5Ly_$hI+_=n@OcZBf3Bv z8Uv2i0ZO_d6Hjpq@D_k~0?Zy2V~|e9%DQbTH53AmQh^)2Ft7-f3Z;nnB7};O;-I(% z;ueToAZ~%U1>zQnTOe+MxCP=C_@A~wyOFmh6|?k|7Y|;W$Hb^522)-;l5%aF*YIZ= z))=e~c-*64je$0LjSRt}lPrtM;S~YT6%*fC|In|GW!LcPKtE(d*cPAzB9YPSKm@Yo zBX%~*cl9!CnlkXdTVLM}I+$+&B|nChlHZ6*;?S)xwR@HZiYTQVWIyEqS=-?M{U!st z8kTe#wtM1x&=_@s{~0dU-MD?(Y-iJvKfQ4PkfN6;{MLU04aO;O*iV%>Mwk5hZo6SS zSg-L%=*>S}-uyAo9KH2oX_m!a)t;one7?7(m0Q_z)14-xw=MK$V}f8)zkm5`_m(@# zJJWXNio++@v^Y9s!@GY5SV?=0Apo^%0F#4s9>`{F28W~XA4%_a8%)Du*(z9^`<;SC z2)uL_PjL&xEfBXr+yZe6#4QlFK->ax3&brDw?J$yKvHyWjx8&;N>odF{5RfenJ-Qp zw?NziaSOyP5Vt_w0{>zQ4F73SLM)qqVsG7B-pCr>n~J^?#@=)3R^+5j)`l&PUs>rT zRz+C{^Qh9dSN;BcrzdvI?c5c9V0kf$Cb;~w7SvSJqlY{E)MpU9Se!76^_ahiqWn%z z_qMd^o4(N7d1IF)tULK6TW*EdlKEtSUz%z-O*l2X^_IDv8hgFOweTxZ23f=MB%E@&w)^>W35D=^!cM&ms)5&W?B?eP z#LlhX>Y9Z0d&k`AWXlub+aBXYc;}`D@0`TlA1ybLu8p@F3~a|fbJxybUF0g{WG-jm z!=L`uXpat3b7~Um;CSHo0qY*T_ry2ToqJg?u?6UINgc%1L}|2oX^#y=aqney&Bm7O znil8Fk+BZEIkYQ~uMI3w9tf^FZK%h zzLRyeCCM{({uNJMRar502%=o4m#~bpKre@?Cqsxac)#3Bjc+#41YS^em1!QJdJq~ zT0=_JA6|2vJYa=);E1IV%vsREmz)etg=|7e za-}rhW|N?V%wEyy&&_Qg`!UHiAb0O!w^9iA!tA%VP`5P z02&9xEfBXr*cN~{n(^8kJbX}2mL>z~eF%XS;jS@giwNZX?}jKBAsF|>PrZ!3dMnO3&brTS|Hr#`zow4 zXo?V(h-cy!h+9C_0+D;Z4}S?kRiQXAnien^45;H%`-^G#eG=pv@d`f9ATfgrF-fK@ zSg>-DupcL6&{!c86J4}~a;n*A*m{+#H5Pz@$8?l<10o$RMf$X((RrcQVwsHKEtwwoz}tbKW$PBsyo5ftU{ET zM9%LWr0d|_?>(Ax)RdUGg*{__VCB6JY}!DPYD;f5tB>97hJZv`te%hV{djpA>7sXf zJ%(ZL-`=o`qh=+mb2oaMgFe?9jNbHyH?-(n3@hi(vVM;~^835(_o!3UAQZNfM#w_K zF9;=M@u$t&J@VU5csaavoBqH497(C9#&5#-mlDPA(zB6;xYtsz{I{Rp%RCM`Yu~EiAqvCBj z8=NyY-@bN5<`T!aHZNa39*-(Ta%lYo3E^7xCrGVlMw_)<<5$jXhOsmC3UbR{d-AEH z^V4>H)OzuX`|qp4w`UbiloqY!@b)h9UUOe**4^)~SW8LkqMbK-ZvbcM6H^_c-R8Kw zFfCvU_b12*{>mAqs&VwV1^#swfcGHPpR}UzE9bw?*zpEmsuobb8F#7b8P_Xrfw%?Y z7KmFQZh=e10ymEM0ZxZUD?YpOMB~`$#4bb>`_r`z>orX6a@X-GA9orCi|4UO1c@9L zV*mQ4b?<%e^Y>wy^Oaa=!OJ8j*fRfl6gHcG^)ypw(vmKNF2+a~8#19^Y{)%Ylj1;-#``#nJTXDm8 z(|Y5z`Pd`}%0-!H7WSUJW&i4&yZ1k|{NteC#@48k7Gh1mH&*2>X!+!$@AZDPqDEx} zV{Y~oreAw}T>n+?e|+cbpI!)tTBCJFZ>A}+^7c=MY=5Z5?F~ECs4R|od@-ZuZ5)>n zw?NziaSOyP@c+jGPaT?8K6^50N36_nYS-oF z=hMgTiqigg%4QmOt-7jP##P7fhZh}ZDl4NO_=)a^O2<7xlMiwMav_03a#L5C?t z2-VIBi5+=_F3)~5zv(xPZ!V)9X-wqVhvYcHJ=?tbxz1NQUF34ui6AChY{6>7ZQjcK zK_o(5LMmJX2xpI_WY4PoY$amRUvv%d55bbyzu9q91IORFu;Q}XkL%ES+0l;@nVhlR zkdnHS|C)DG#X0kE82vGMcz!$k!JVt`eE`BRfj4xFb~4XzHRgu?W4{{=dw9ZLVo^^P zTRO^^F!-0%CrdB;0XE0^JL($1t3mVAXU^?EZF>nqiRIzRuhJrfam(E9>6YY59QJE# z-gVS>pav4n&4~1>4t@UA$o`WyEMz^2YryYL>m`g3p0)9H35$x(PyN7cNx&P9yCcL` zFPp|){@C`@z#Ybu3~FStmYY>xxYo<6N_br-_H5L&$G#a$XtF^e91$*Wzt*dv_zgB= zcpKnoXgA^XtZtiTem#Ri#V5*G+9{W>K@>so*}qb9pRfIFQXWf{Q4!>;lNE97=k#55 z)6X=A#p5+GkZ;`bQpddJHy#-Zhn$bnsdFUJVD|mgQV99Y!JU49V>Y6_hHh<~doaD^ z@YL0AlrR=TezbptYusf};(y*rJ;c(e5$qggj+Fa@A>nyWrKc2cpLUt>t>xWbdtb~h zF1#-XhJO)KAc!nTCu3!6WRqRk4dhzbt^8vK_A@RNxFhj39m4%$6=0MCoCw06Jxys#EY?B^nL#MDFE?O0EnKBpPVEcBwh--$#jsps5I8gurvzz z$NC$ec+5o$^_r&-_)!E@|5g2PxBL&<@ zBXn8qhV?FokM8DCipuB}T{MFnwR+FJ@a9Y3P3k>2dV&Kak;r~#ce|%|)g8HMNff?M z36L)U(rwm@1K;R2Y{m6CKg}0_QNLQ?TUA6f{ORrM{(3EQ__tKl6_5#>4IloUt4STi zHDsAsD9HC|?-i(n96>_vP+vo?=0R=n19G4a5~=ZQ+yeg(EfD@&Pl6G`ek}^#Ie=rN z9Rhv)8n-~)0#&wv_|_ApaqhQ>pcE1Px1N+5#LICD{A(;gYCKMAzLFH#x1Rnrb`H_d zDt$2!f@i!sZh^Q3B4dHJ!?wNNV%vk!{w7AIJS-u+_gdTc*M2tinhV^5tgJR#1qzXr zYiG)9pFMoxvj$VhXo*#klS4ZHI&}Nmb+`ZiZo5*I`sjchD97PE*tW&J`>r4P-I9Xn z2vwDGvn}o7d(Sr{k zb8?}pYVSvJz@b09?z!XjwvW@XvjynE-gf=qJO3PZ=jB}- zvC@NpS|IW34oqna; zk?@HlMM01hJIA3W1;lim|Pr6uo5*bCgf+*MtCv8rxiPyOhHINp7 zQNDuP&8qYozJ)pKWHQRKK^$@HxAezS8=Sco@+DOC<#Q5k)(iFuPr_JOl-Ax944$+Q zY#F4YqDAYRCly>Zkn!-hO&Zs~^7P+EFR_tFpfrBgQX`KH%!Ul+a%711whV&u=9Rivubv48Kw=RY}*72AXe4`sBfe zg@aiyv7HK#AC=>=qicMlj4f7Ip8^sx#9Z9$E`) zM1l^c4(Fd+*dq=6^22|P(kA|jBA2K6mZWmtZD;Bhp)~J@a`Kul;jfyyO&aLJUnz3C z9&{BlwYiM=nVI~u`xT9ohk=<>i#(4XXl zet0~0iCe&6X4$EJ9SLM*{3R}@0J)6i4nKws#cIjTuR}an;}&ob9;FO9$jZ22ok~as zWT%BZLL|$&Nq;e(3*Q1R$q)^nTBI1=|f|FhRWsaKhN z;^wSxKcT}RmEpP)G7e?t*S>kogPsX#Fw?4Ha?;Mb=I-rtm#5@8Z?AI_!-U7r>bCy! z(D}QWVfQTk>g`70Z>lV%`gk= zGncye?)Iu+!2Tr9~Dl;qe1WC(;mGS*icZYh%`g zm&1kFhg-FWUCMo!ZH|pO*YYAi0YY9$8SJBtapwXokD_sExCa zw=9|8@x`)2@5F-rOkYY>W-wsFsf@f&w>;TAGk@ZOve&w}bawq9A$K8_1CCd@mb>FM+r-r0IM3uT^vW!IQDyF7PXE)m;+l7rMs_M)vv zulb}IMH!!ca{OyM=}%~NJRQ(+f8KJSL$;yAOS=aipENgQcZX^;k@4FXi+@<~sLR0W zZHx{jc@%k|u)3Z3Qc|IoH=;7gNHN8HsKyA0dhuYzE#i$9bO?S8L$3bD1Q!`@t$w+B z8xtCj5at%VaMoBZUEUFE`KIL8zyn?68y&9ML*a)uqR2sSV`O-E7T6*8fRh&4Kt?{N z3o+x9-6xc^Qm*lQl`S9yPo9A?c*<*G6+}AO<7amYSd|s2GNIMoy%J`XM4ri7LLt_j z$xed@n{Lx8WJ6wLZ@6G{<{QV`S{7}bY-w4zvBHvBG<8x+%cA@#7CQ|$)^fWSGjPIv z%vnsC(ijr6H%zf`XLR)$*z4)^z5{!X$6TJZImf`SJM#!pT;yylb2q|?B~e0u8==Ts0bO<{~vDOt7iT5=WhH!Mdo z?t_L_Fd5t?^A%B`+M4QMpw$=^Qb}V}YW*O+cz%jV<@7d2FCOR>tHxlt@*txOL7dS5 zMcSBPF-M9QDTC|eqt}Rf@%~@UD)l!yw6%i<8J=8%f;}Q%{?Mr^A`H)oWMfoy(E#d~ ztG6+F@iY_)3C6R)_M?Ur(>!K%W_(K8kh0?lp+l=O+{!7GGrs>(ixP|5vjoeGo-(82 zNOm4f&PGp}UcqFqvv8BH^W}Cr%!b*E@?9ow@@OyG=)u!S_)|1(0!#`G7p*r$zS%u--myjAj`I%}?jwyAibeTA0gCycB=GXM?F%=Z zP9I%2@f568ewBhb+n7KuDv*%D&d%KQPsZrF=Ga$<7V< zfWn6nt;Q(kZ#E|z?t*Hh`bw`y=*0uQv>F{Lm9_Zjm#eoidhxV03JWfe8_LFL1sNqC zCo*601kC}X+89a8!;q`m7zryOs~1nekK9#nW0ZJ!7I@QqaZ5rb;sD;)l8kbn!=d0R>^BcEc zxVvaQT}38RvNWdm7{|Rxd2oz~aG@o(_r!m6lw`_MX!&Si3Jjo3_iH zNr%D@S}({=IFr74a36QGTNa;reFr1|#mvV*U!$e7Pk3hBe$SkvnGY`hXYr0BKp%+& z9=pBg^N(lVHE(RjRvANhMb54L4ud4-;6~@megZIfESrJpH8OsKyr zg6Wql^k68qdhtT>sY1Qp#^}YX`jaQ#GU$fkhgAJrdUBGik!Du)lYhW*GI&QI`*`+9 z%12NC{_4*%vS=3U?gE?rBKUu;J<${!D2sS-;nkz}!#;Lv1M9+qXng|?-#u&W&gAg# z&SbGxHh!<~gX31*e_tgPS}aQ;7fpu73lGEWxx4vI40&>Iogb+1C?kzc<$SZ``+jc~ zRYaPxZhl5@okWYL!<34tWwdUh8bCC#Z9g5n{nX(*e(yxq0DGM*iF|{rDXchsQmB)l zX3*FLuU&W8bV;dk&ql| zsFyfWcs3&c`T-?_N0q#h8!47m<3m7S7}KjpE2XN!u4ORG;f|=Z5J?LgkNpdF`~F_a zk(&mopJ7SW$gX8p4`aM&ERgFE-I68QvwXuW=e8xC2c3+rWSO)?i?w`27eNZ@EIC!NB4@4{9XAD}d^O(5~2&j6i7_r)E)YXL=|*nayz1 z;u!2&*7&4_N~-8VT@G5;9yq#RjTl7-Q)3>iAXfb8AhL7Ggg@-L#zoEpI>>4Sg>Yr% z8rryL8@HyNA*SE75B+0_;Bsb|Dt}Pa3XP{7sPwY4=&{rb$cc*SFj+(dJ=tSG?e|+z zy|Xh=R#Imw!(fjH->NXNfG%~$O{eXMEll3n#%-oELw=XylcV*-syN#f-Xc%3SgN91 zoF;C8xCP=Ch+80Tfw%?Y7KmFQZh^Q3{x>af`StJr*z%6o#>Py0s`z)as(M01L%|rD+XqrsByHo1>zQnTi}1f0wzlea>Phes3^-uSx@}fx46S&JKzk*oM_92mmXcLyS5E}4vu;V@d&&On^+L-cEc(6E6+1^Ka93q zPM!hHHhT;)@NkG`9s`AvS-#n}sJDJPd{9sxb*Uf_-WZ$3dJG~~j@M*n|YE(VR7a9gfmZ>H?%Q#dkL?BhX$yHW#Ehrho zcgiJXxG;cc+%dNC()CKPkt_G>VlCxMou*$$^{|Fac1DOq9lZ znDAy65!MN|q?56#iRe7O*R*%OeRC6uPzDH#Nirr*sl!g`KE8})oJBFw0A>S-stbol zw_{=?-zT;XjZk|>40Rw7I}?_+CxagGBv`;zZ4m-x{2I4F+ybf=(DiAdoKow{L9KT|wMEGJ#!GPv#4Vt(K=_{)!e_lGtS<+( zMJRNM7vmO)g$0D;hG`Ivzck{OM$QPC$bNgIaw71heLO1RB7Ntia>t`yqUQobXNe~Q z!w-QpJT?9ZFr2yMl6Yt}e>|$>wA2`@ijP_CI^j~eQbQ%-Vd`xY%8t0@2#EIK7tgs=VPXM5^Y6 z2^~5V1ex^OTP8nuBL8bjQ4#j>oZf{0rVJR}<%c5*Mtsp3c%-H(9srHyZSHy4_~4Oy zaGF?O8`peYl>r9_ z(`=Birq<2qeA%yCo`Mg|Ddcdn0ntRn+~fZ9<`dp7kYj?`MMsH*E;sI;1F+$wn#VL2al_-I;a%4rNuA(S6B(~nh=*0^u zFnk2PjnRwuzcF0$5qD;wU@om7!#7$b$OwcrcGc?{+QC?PlA_1 z#(YF<#^BJ&EPG&(mryFaEuk-^W>}<%@)GL6KfOpX)bfW(z_j2;?Bq(W-dO2?>ohJ8 ze{yk2Ut&lpp1kBYofJncufPPeN|L-JR1n<3>9!mwr(NWL=7z{@AQA-x_>eBp;-5$mO2t2r7_eg+#U~#{cgT$_FDVR(Icg*Yrn-^U zXcd?;BWIuFOr9|u2=O3EP40wQa3cq25-mySiN42OG%ny|hWVT<5nckyG9Q=KL&44{ zAm(y;e<`ZvMR;!XJ>$#azLEp+K|-jszzU;VMIDR$t_9u2!rRvZm9A?DO#*qq8US(x zf;AxFVQVmn6xhrJ4YCR`;b8+YUNoTKsyo>!g#9e?Nm#zn%~_u&Haje7^1&UJ3H^R% zQ;r6+S3VkK*t}9Ca#6a~izp(jI)yP(!@~weJs|=|MB5USS-6^n7Yf3|4ytJNE4eVE zjGD;mv8qK)Bne7z47p4?EYGsELk5C^nF%Lw06~oVU{*FSq-mrYb%>h!OZLW%H4`qv||Tu|M(s zNh9xW>(IdEz+nU>!edN(?%p2Ek8ATPOp8~)?WKb;)?MeMdJnp+-{&{xKKRXs&dnIt za=nxxf)dflIQ^sfQ|6z|?7MYBlfUHHj*vpMf|5P=v4;B#-y)v>)xG$Iv$T?r4c>lJ z!gQ1q+$r5kejECDGt==`W;L?XE(sG}6ASDJu7eRqkpv}F>!&8o`j7kNSoU*mT<*Rl zDjdviYq>)O?guF&8*;imhGq9RCeE7otaZ6>+vZ1=<_yRTMh_=h-E}4$G_^@7Upn-e zPP@GLU3|5~`8cjHGh7Hugw0(?-f^6zq3J;+k_eK~qqt+sr^gsYbvkWQ&W|}v$saS^ zA@~W64@fHR2VA$15{eL62%x&Bbm8EJ5L=yUh=U62YI6$*7q~#b%B-u+xaMlY(*ch{ zQl~P2>+eDcA#_kWyW2@iIm37e;`TvV$t(Rj$l)5LaJo<@uCocLA*>4EQG*7xuWi|T zL$jy1+6xOf2H|v}9M@j>;<(i3cJ^p}Zv`{|&ZM>-nr&m|ciwPkmQS&8x*K9io$}|t z`?Xu^`%vE9Q@;Cj+_mdVS0-<0lT(FO@+66eT4Z)~e7*GK#;)UAZ!G=HIQ0sIib*9u zn{*UeX;Lf1qDG4ref`VA>vy`06~BDHzvraJGuCKEBNKvq2za|B1$1{`V(fMI)!uTC zr{1x9I!~#+{HwGw33@=5#6!mu#I2q5oYGG1lXCJ5m3Lm#w&RN94^v7e0Xo3LC9OTQ z>aBWu&yDwt`Q@whecLX+dr$I8npJWX%M#a6X}o|&l;4)RF30}#mV3VRZ2!26?TAnj z@=CxvlYxzg4@X4DrC!d-B-k(-9Tdy55Ja>RsI0QE92smaE|Gz2J?p7+ z4-Xr7;KArn=M`mW*qWG`aK>eu_Fd0JsFXFn=0-&Q%5m^i)X*(Bpd zQ}Q1CDyN;!%Fdxk{8&s7EX)0f0yt>oq-lsXZm zJ~P!`t}?;9RI*m`d;@|K+(I-&0zm*azzldUrlXY({OMyj2@I>b!~uN7Fr#ZpNyNq) z1T(Clp&17y-AhV7DiYrG2-Ql3vkCHoUM&uoaXJ}w9ET4qI-8l%8I(9@W7V;_Ta{0w z+}U7q5#%n1aMEx#U!CCJSj!;J#$hY|$em5Di4MbxAxyy*yh+8`1Z>f;0aSVb!JJKQ z*r413gHkYOgK6i~pzM^D(!k-s&w-2*gD;0YiluBMQBw-$^5Q6#Y63MRas=Qi0R{`) zrTktcspi76Y!r(jELbaf4svIcaD}Xvvsu9Rky*Gm$Z5oq;B2y>DP}Vo&K9hdqQ=k-+L29T6br2cCVoxKNl9qbEhkvo&sr^KrLzAftl#v@_OA&z z1Ai@4xd*~ld}{nuVa0_FQ_IOrw>kHgzt5O;+`?431O-|dgr^|$SJjdp3BXEWzk>QCd+R zraqcKrxjDw`jXP@+}xI{IyKtUsAp+voxkR<9{+=r$nN&eTWMqLt=ByNT4|r}k2gQy z0*RDNs;|HS)r12DDCVSPAsZ9f(8SqPmoKMGFRXm3)mukO`%W0%`s*9({Fw%IPGCpE z)t`WehHW7_P?$nZIMAD@Ztqi9%q=9^Q`}|-07|%+0Dlz;f+ni~`HCh)csb#LTKNeG zWPi6X^$Qw}UxZrt!3b`=*z}Mf+-n3RxYxn+2FBr%NFe~ESJL`dkvaQA)MO3H05~P7 zgSj8v3e}Pg-qO<&u7(etWA@O-FHMAZJE}_wD0aqLu5>!S4koj3D--biS6{I70|Ih} zUc{jW=sd+$@Gj~LeKJ%M-jY(C4scj%6au-=>>21Rg?&oHOTdsSE-LT{det^N#FBC$ z^nw2d8=Rs9KZOW_ITo-9mIyGt2?j{G&tpEG+L1vNYJuqy%%`=+pok~UV68A!FtdyJ zqKm>90b7NUU~mFOsAn`7gX%6CT2j!+;3p4sQO#iKC*pGc$e^TASut43P^2y@tB{6QWJyX&YCH5$F)g65UTVni zI|pZu6acZk)&j*6VRm^5B&rrX%qA|q1T=viTsOU+^2qsP-JTwNoVo-Qa%7)s$(}Rp zz7%wUfZQBR5_?_17%+MFY?|}dp~E{;vPn=XmdRl+&p^rE440RhtUeo!x^~Segc{{W zU2LF0Qv!P>oVeIIC$OppArItCqPiwj1=c3bTCusXXXF8T=TlAl{^dlGkcz#t4 zN|&>4w6iH??9{vfjqdd#s(s`|(q6c7wz1^VwfN)r4}NJ|+WOHf^y*nsPE}(iis$Tm z_#zs2&p~aM_JQX%*TL<7-r3Jxf`%uM@RgMXZf2{>(d2ex3zK$yU1RD5WeoG&1PY=~ zIXtM|4y&F?*N}C%vy1;*Q3 z1IlrTOqByZG~Ty~MmbKX4nG;-smVzrhYBTy9Qqy^OkH69h2Q$AyR1K2p5a9HFGytd zuDqA{=uZY^1+09i$l^d<&{5P8846(bgqMbV36>x9Jv!n>C?3M?V68a}LP_^7NB7q% zMAnO$w0VE`EVSrZ%Ew;gl8(+J6Wy3JJJ9_Imd~(>EV5+^4MOfYLh~_OAgTus+&8(i zP{!9V0>%L}W!KHdcZ)to_Z8;|II^pOvfEIRaXmVKlDVybWVNaV7*mo}6;%}c*zB_| zcpW@KBO3)HF{=nYa^rpI>Q%*s*6rgSA=;b-s3P5LJc82Ec$ASt+zgB1W<|B6NTQGr z@F<9NVS?QVxTPQxT{q)H^vf}D*w-%aXZm0QCxe8w8oan;MpHYsRmSA0#DkaEBHm1>V;~sPf zC4n&_84km+u0$)!xD9&AMO{H5MgrEzh8GSsm;?bEP#PLX+Yy-PTgYlAQaNDNFm+ou zkk~&0j5(>a@oRKBnuG{OiKuJ9>J5iV=D18L;tRYlK7sfZJ&jD1wGalHFZtCKASze= z;V^(02M@J{p%kJ5)uuWzsU%tsE47gbhZY1s1;`m= ztv-M8iQAlRWlf~ZAKrX(v!$!nKd}FgHy6L*p(D*+%Flpob-s2u?iw;=csFn9g*mgn zmDWZwZ-Z$(s*BJY-Mim(XxzB@Wo7ds-#F4?#XrXR5|L%AQ`EfafWbqCuFvau)v^!v z2sPEY60lytGAbM{BZ!O*8xFR+|9<)2+M7HUwquS-;MWbQ(snIw5jyUiu`lP4a3^_2^% z6dJ&QE|K7wV9`nCS;>}^Rlol1;g1i!aA9qyFIW3VhJ-Y%tn`%Mun4P;Y~Hb`dn-rF z)5ELIA_A0|?yXk0b#VkZRhfmu?*4!+hqcLACL3C0EjL$fltfPNLRcV+sU2WzpG7E* zjkO&XfgNJ1!zJO92+amN#OSw2pHtfq3FCtCm11@20CqSKE8#x5aI-caiLnlGJMbM4 z?+e}z@S-IPU!39<sGjC#8`(Cke?o79pZN2Isks$zEt`GUjHz9 z@lQFiNY+Il^@dnT_RF*mhy3mSax8Mz9FW>C7LxrkW3GdhLDu26*!M-u{lV^rzOeZG z0ak-xODlgjMJH!v-S!+f6<+R>_>^BVFQ=FWCChn7WeVczc3slEZp0F-5UD zoUihKv5ShX8%$Mv%5S$xe~7Qfxn#)p4bme1uEn2R9-F^_a}f08iI9U_#&U-rL3WI# z!OVuazk{Dq^mFSY_Y9Eu(KQj$zsX#7D*kM{`f1e0_hw(L-@p2d5KDaY#P&~C=F5Kt zaNtiP#bjk%5_3vUzy%Bv&aNPN%;bmrUjOCaJt@9aoxiQ0HH~g@;DHEmav27_vorHu zx1&1h@U({C>aR<@Vzz96RgG)%o}2gHRHd^=mGaU`9{6PWv8$EfDwS7!Thw;Rpb9B1 z9DAgF0xsdPGL#>i#2*H8iMa5>fS*TopT)qtf7MHbr@^vouRls**5Hs-b+|Q~T{-*I z)_r7kc<{8D1K*ooQTGv%k)1+{Fe_Gt$`B7f{)o%ob7CmG(DlZr)7O2wZ<1UQc})1oh>iLxAeK+UbXWex~Ir z4P3k{lkP-Zn8%FoGWUpX0!>81Yk1(r&#$005yGgjZXItLe;_+23=$1gGjsR)d0YS1 zs0|PNLfg-;THSQAo`LLev6_iGLoqHsG%W3&BlQ@4=y;UM9w294<`q7h@c7%-7UFCB?u3 zPlUomE)a`{=o{Kv)YXNDC0&2G(2rT&258Xh4i{mCwXlY^hGsIbJlenp39NFFL4t0C z6T0vnQ<30-a)SZ{Mg8GIKW3m56^Qvz{T0tOKT)UkpxG+5Y9Os}vA6DRzqYz_lVRM= za}L-CO{rKo`j;gx-QmLQ$4oeDSuyp>zv^e+TsYwE&s#sZ)>-?zUH`Nz@hc+^SU*T3 zmw33rsK>|NaMgm(dp_0Xp;xc@W$iH2Bcuj$2(aPo_wAc+C_QIz&s+A^yySDo@Bhl3 zT^O85Yrql{4>VdhXQWl@GI}io{z%<)!;BLfI!`||YDu$~ zDqhO;%_y{_t1=UUfN4}g5BFbdXPmiy=hOywe7Bkkz_zlC-bbrS!<$pbH2R@z>XZ%j z8nj%NXVG2*asgg+!)FRfH1 zt?YD6XBQe4tddq%CfrGF_!OWoo{(yLiUH1|l{X4r$YyoCctLSgPR1!em`lz}yA)h2 z4LX$_>ohP0aQGAgWes2PvaH^)!>bZL1*Y@DPZKrVlq#^Vl2$GdM1>S)gI_m<_Yy)< z*xmNfQ4~&h;i70r(fSd>4}IZNfHqYMOZr<;Ma(K`6@KUopQ4KG36~6qKI=tPZ*NlV zW7DTz$nPR{hyZgG_g7t+rXW!c@^NV`rjV_$IY?3!|z1NtX-F?Z9!X3Nl zu_uqTJ$>f#3QwXp&s6NB${1vWor_B0*ZA@}mE~60fdwbSq_+)|iWqN)T$iI&pK*=T zey?>VW8%b=b?av2?VGy9HE>vQ(aPN=cb)WDU^E9N_{m=_`2+G*D;BVZvNylY&TO*d z)q=yj=;!wzXm`d-xTpZS@I#-Eg7DBRQ5?unkU%1vkgch!otB^8rXKq6RuftYC_)CZ z0NOwbFbNxgCI?h0$jDjLK&eVvRVm2Sfbw#dRnjW_K&GF`8NFcX29JhoG=elEJfq3kSRHHma#Gxk%q5RL6Z{+r7Aj-KwhKQD(Y4hN2&@< zjkNuN_nj(ytKzWQG{oNDsXiZdRgnWAg(ImWg@VElHglfd!l1N@-2Y6+(GAuJ<;m(X zD-KjsLvxlJ%Ow*gUN?c8^`Ba;w=L~FgE%!e&&e)#uF|q`i`Cyc=9X9Elxz^ZLmG34TD{a3UbD;1f&sZM% zw>51?!fa$Ib9>X4JDQG5S>2afjk^T;&P7#s7c!XyC=@#uz1n+fY1=*BrVg0TAZP|` zR_UUZXu--Mo!mItu<_CBD*Eqw;-dK?X6-;`H^EAL5utl>>Ok{Y#?HE1;->>RtfPz> zU2){fii@?q>!&R#rgj{2xAl~#Kp}f&&dHOdca>jE^^$-J$?8Q(|IGRB>BBz`zV9=4 z+9?d9ke!kJG>U=9d^R7NXLBAgrTrH_=@5BzmUEHCzRzU-CqLgq^;^;*J#D+md^&&X zWG1ILy?$m9+zyDY3Gg(MT~?Y?$D6U}%=~sM>0vZb%etg00|R*YoiE`6_02@8z}0Vb z=_iBEdj`3v#ta~U_Xg_{iUgcc3ErWgr8rJQzSTHqcAsW%C!C!`LJf3@iua1T5mMry zd904l$-QP{=ND$)mGQ~!w4aWHskoZJF5%e$jmZ*v{3vA`-6;v~PM4dW?o^VU`C$lC zVGY0qU|9^-tXwoVXfY&ovKVP*VDg^a?iVW=byJBvz*N5Rg9iq73Z!q)g2Nh497l#k`9ntr=&olgRAt&@^COM;1{f>>A<&elh36;t#sRZzj9RQUA+NLjk=J)3g z&P{9XDW;u_w921%2(Ie~;kRw<6ao{fCFAL&W!+|C5<_lDod(CA)zOAC?eRJFA zZLY}S*4Z_W9Mw9QPmM9*G4$BFr@ig{&9|-GknI3x1KX%>tca;8f`vd}QDdZp8P8Zg zzM+7b|Hi~|x7+OH4p<-7b~Yf!fk9U?x}UH2RTgnXOm!K66q+W@Uin0$0*+P5H&&NfDk4{k>>*pE3f8D+;GzxJ8A?NM#1YoYAX$K{#vk7bjZ0RTt=4`5gBbc*E z4NK~GHeSDA&V~a=h!7?;lD|rWQZQ%Z4J*Z#^9f#x=qMIMnHlWGde3sj*d)@OAZk)h zhEl~>oBOCD1^|N{sjEGTgU+jX8FFXigOVU!>TCi`ah1$w6g9z|E&Nd|VXon9oF-bM zSVAsl1H()(XM?+xNhA%HF+!A{hno9l;RXshg!voJL8@h?&>slRfSX)7DFR_}1?P!$KZMwr#^6#jL~QgyTlu7!a(n>zdbS*8 zQ@DiHNaY89G11+f3&1+?A0eDWJS*`3EDYk3fqPyvPPfo}**G2fSfMLA@F4gLtoVF0 z5tPxF=JTc;Xg;43e#ciZfzKMaB(w1ku6$A-62mBYT3|U^P(Amv!=kNwOZ0R={3eY@keji_6 zF>>Om_1A<{1BgRPZ8)%9c$#B+WM7?pB@M=gpCzlHB z9RDZ+-}E)!IL8+xu=!Q|Mu?zk9tW0(6CXiaAsBaf7;?$iBB?d_2wKdkfrlm}a9^_p zx5%FqEmG?o?1#MHh=#@Vz5%&GI!Fb>N(QOKZ$zmc3bLHv_y#H5)~12?$^C}2hHpUn zd=smY(L_*#^8+C+h|D3wtHIqiI57BuPYs_T0BZ5w1UNN#Vj?{enbE7Lg_Gt1ToCkS zp~1w~rMHn{5kjQ2;Dej2j0D5xbUAbOF7M(4BL!AG-$=eP9 z*_DB$Z^fdOzsecOKi^K41A&;%#GP5Y9=C1n6Lrc+O4_Q>(o%A3FgO;#1_ zv>)(NULvlSH36Q9NDRzph2eusIN_S4zn=K)D#**40A(u;hf&#O7R4mKwD+9}Uo&=E zC<<5Jyh2HGwbBNJuGzC@M^j3KnsEE^2k6Kq!sMAelbxv?@yqv4AckafAZUWbymD;5 z#F@9{oGUWhL7Ddu4UOR0b+*eR2eNMhdX^shS27d=FY=r0TA z<+#<}leYYJf1Qx?$i|RGVqOm-UEWP!mR!`b6vu+VKVFE5^N^OS{_~5}`S@DwmV5$_ zm!y1_#WkjlK={`hxOjIxsH9**5`5PGg_^XJC ztkhxU8gN1+-V0@m4Y{NYpka(p7w*D2tcH1chgE36QF;6>hRXrw6#^uOtoBE4UNh5O z8X%n+&tsv`O00_x`JH>9bgs zMgg>0j1ZEwlFw9(di{b8=*TNytUK4c=V$5&0#vW51bqKhu)!ZWwpwG>u5(vZJoMOm zu6}+5F(=!vU4)7mrK!XU$S7@KGCXQBz5e^}zy5VgpW#wO+$C!JwF@}my9h^cWkV zMz~hPEd>@KdXB<)8rwaP4yX%#uLqUC3qC{zPm9*K<>y zdopSwB76U-`w!|ewa`j;P>2d6peQI|7Omng@bkXOL#A$!Ay>nDgN$I)WZFs=7?1m& zbXC>ysZn*e#0*?CVQ*zz(}iPJ{Ykw~C&>j5O(>FxuxA;!UV|#3S5y+9K?-M5FJ;2p zAb#wr)F$!#|0@d&9ZIx#e+gaVZK_o(Gd=yrwU1{nt!Yv7H{{>EdFIRS+}h{Inikch z#8G#_+OJ+HC>TlBtSp-?ll8unJNloc&C?5NN)$gyFd2*pz5PfAVb8wrlM#>0manQW zIdFQHm{Yi{T0f}SI#92DC_!pD^@=K;)1hE)GC*?=2JN-wU&*1 z^3fqfx)fHKw@4UtK4q$FAeQZZWA%sS=Wh@xs*v$_Ng%N>pE4ox*3IWCf9v7g5s0J* zqU;hWjeMFR?Y^jS@`D z8vgfOQj&RkHALxpKr>U+<6mt4@u^$;9IJ{bHB(9WIB;HAH)xtmUf*%S%l9pMwUWYc z-6@!Q% zhpq0swd>{<%s)+wD7rG*T>`&Ms~*kU++ule#E9P?Ywp$S6h1{_aoXw1n{$e?9@s5T zCS!2wiY0f;%6oP#?$8REJ?Xmoq{fiAK-jIu*z00!rT~r9g<4}s=E=aiM~$%$%@|&8 zF;g)6BRSM?zuO4SUl#_PusIo-VrvK1Z5?Sz_S~vV;hI4*=anwWxq0E5UU1*w-rK1e zF+sVS);|2eiI4Ap_)EZQ6%@0^D~&i-wyAM3+5+JH8RJq@Y;}N84R$cZ)(+>7SeIIA zUyNm67&5Lpx&7?16#Uh6ElRu3qEJI@1tA$x5Tqpn-tZwd!F1^gN(-ar;qW=Bag)Ko z##D!6-#7XprQw+dv9v=|@{bF-Ou^9@>LB#?pF$>9SC&P8#@Y@QXA=L2Rr>|Gb8HjJ zg?#)Pw?NziaSH@k0Csk_kaVEdWm- z0Xx8RR&l_yyZS!x4uqU;HOs+kfGkZ0)@X?c#Ee7Y7WkK1z(vE}k&eK|6R=3G_I|Am z)(^yu?qy^)*bhW?mKv@I8P|9rZh?QZ1)L1r^~9Mft*KY zWf50U1;paRKO(00NeYZ`Elr^$M<8lVK#0j(*BFwYv=|N96Nq01@NXcvSn^4la{Uzr z9Zciy&N6oFSdg%(L8-AM`>5E)L4VAD+G_1El}4F=6sn))Lc%M+j7T76Oz@K!d^$sk z05M~NzEyk}monrKX`Ha2FnQTQ+RWS|pZ(@AHfM<~J&lAP|8qqy5zFA8?iznt0{rZB z`hqSFE7MZ-?v6LBZNQMoR1h)I93%5Z0y&?2o0?GnD|ADnPQjUX^XBX#=^dZB;qCGc z!BO?rTHPizbyqKRwEbynM3%<8UU0Ii$gk-jYvW9-WiIrEffS)t&SY*)AufBafS?zY zq0&d52*QVuM|(SsY85z;VNDPc9aXM=tqfP6Xk{EY=_BInxAeYNhmV(YP9Er_ISFbj z42nU~_Gog)L$ww9;292;n~hc z)ylxTl$by*Ne@}$>y6&r?T#IV)+463w)XCvpYikHqvINUrAAvtn9-Y4MkCt*bl2Q% zUlnfpb=At>y(WLq55^Tr(A@m?)^{MemrGLEox##$jD=7 zm4jSeu5I`1JN_7!_Ve!b8+YZ1C@Hn_Ctg%rBH}TG&}Y#x1mKDF{bf5>wm9!JNFoDl z5M)WDCRNa)NR^B{EU6&)K~z2qo7*p8{1AlF>l5eCQ7*+%2vUsd=hGVyXV1E*kc@(c zJ~=fhFLLcTd(K6%w6Y>n<@AOIV+E@4Ks(6LJYfl6U+9yuYSolDc~S-mUN8j>OF#=N zm@ed`DanniRa4{1lltO!aN+)13bIa05;C%y8V{e8pJl>hTLPwnA{UL&C)Ud{jtmn@ zBr49~jZ_q22@WRlM;;lGSO5xAEXXsYzoK6+W7*zNtmM=}iKBATLQ(sQW$g1+hadX1 zDH4;+Xy|<-*iA>CyY;s@$=@;S9r@Z?1k@-@s;Ey~y8PDNwzYPbsas|~TO8!Zs2~hc zG`d@y>T(FxR$ZeDO6v@&gOzs3sX+kDuv~ns(7dbjG(4`HUb_U_M!_VJgg@AwH4+va z+`<}!!NJf>65l_u4rCxI93m751c)^ZWmIsDZrI=|$|-nv%au{Zb;xnV5=({G-6CPA z3>5-FHj4;P2aWFb36u*CjvGS9$W^PjktXwSaDjf*h<@Jj=&}^P=yl- z0%a*6MmJtC3YPqWC0N=zB*3Hhf)A(g^E{ryAt6Na1Hz>>Atr@VRApXsHZ&P#hVVP8 z5?{4!Wc1DuF=h=+CRMdk1=@A4@xjk%K`@=P;>i#eW=K_edVyw|B(LQBlRO4Xn8N}O{JKd;^ou+@-|=8Ny{TJ}rDZt}{g?<5Fls3|c@#^5(n)RU zaH4|F;~N3fq|HZJ6AaZ1&QyiOhYKx2X^0MvDd%9+9awm}zZT@k8C0u;rJGFO|5Koc zOR#Vur*C*?uQo#$$zjzh8UG}9ckguL6j~=2J>kN5+lUrLYuJ@rl#xzTh7ee^5GKrZ zy}SRW^;_YTT7<$Guv!dZ2kpo5@2G!SaghCq)oHb*r)H!l|4{nSZjvV?yRf>LLr z59+vL*?X^iQX)Y%7$k+jtes^*{d(~{OK1NT(tm0 ziuW|FS@^d3yjwx!(0I}a@MlbC zfi)TuuG|{&!M8{L`o$mEj6^25GmBoMphB7`b^ny2 z`11~Um}Y%N;#8B`E#7H7QxapkAp?#CI(}pnZdE2^W+VfFc;&Hof8I51*B-w}k_his zMIn#0MI9!Ov@cn3;0PshRJx5=p_Val%WSe18GHQm(JRj$EfmouZX>?|i5SuhiSqCR zkqk5O!3Nd`Mp*4xK140Ik@6Aq*lpwkfvYlG=gpb%k`GJ8ZDhepWQH?-x8cElOv!v# z!)?^Vh16{nqImxVARN#^$?Ad?6mBD+g7hY$e+m&ys8n9%;X>gyeq=Bmcs$o~T@g$e z+BSSFE#>Fq&uN&>kD!(#;RZ;!)o`i7_e#tJWS~} ze$22&oSzK&-9{l@GU6oYzO3rplCM*^jRZ5S5hsxu%xxG6o_bam1b7q;w*k!Q(y*q^ z&jzo0~rLBO$Xx3Asn@wKgnUhUjPe9jv+wRtj24s-6D}zHZ}zQ+SCGuk6!Q< zk)0EkvZWe}6=FuhLB(deH0F6}P)qoWGUXW#LxUP)hUTDcak(Uo66q?YNqL>T-Jk)b zK=7lh`3exol1l4*^(r-T*?4!n01}a>!-eNsQR>8qwNGdhjt>#IDzaW7EJ%yt0>ZkT;pdvv)4dp9<7eH!t)d_Nh~IggiNa44 zNpMm!cgVKGhM$<_N%EM*RoLL9R5^yU1o0o9kmWSCfM2C;R_#|8C>P5}IpVy-?pA;t z_LJpuDGCH!gvU8SRno~Q5I|GdmeY@+s17JnE?NOo(^P>X6k)b(AOg1i zt$xM3@+m)JHUNlHM^}Mb0)FU%t^^KpRG8($Bv_d!4j^_V&61C*i`xEXKb~~q zqN%d`dgdfGdQJO1joc|y>HeBv+vn@)uDR-)XF1B3BFqP zhKJWpZQQ-D?Sq28TU#Ee|4*$+|Hs~!z)4YD4_8mm?hNd*Gb#!yGK2TcD#)?d)E*oj z(M?Pu$!BCqO!S}2H7ej<(|d4ulf*=f2G{66(OjTLQTCW-7X(i*8pU`pL?b9Fiz4i@ zM}M!nySl5Zt7m38lJEQdEVS#@yX)1f>SKGlnni3WaZ)_NyOxsw|7jw58nFzWl2D14 zm>jIiXy^#6A8kY`NOTRQf&io$P9cu5mUZw*v&Drs-qR0>u>r;j)e2)2kku%!tv z!{L%YH% zJ~}0h7S>e=C>i_h&<`H##UbtBU`$cJEjpE=HN)!$skB9*Qna=R>^M=RQyM%_`23`t zMs4}1H=WXe64-Rgvn`Rd2II)`xMAvnNLmGUN_R`L7%yp-#|w_D2>xRV$xqG3r1pSV zc5obY%xuflCx?EDgxc=4$>C``rn+_5(xysJX+l5Z_X)M3th21cU$>}s=3t9Un(@Ze z=ct08E3vhMjf2gF-D}qM#g&^!PJQv2kqN9CTwq&En=7_Lxp2vp;0-M|e{t87;kcxv z^xT%>Az%Ls*KM49{XZ67Y*5@WvL&z$Iv~05t~1&ee+GX-_2MAx?r1Se;xC@-Yk%w; zDxgFjU>(iy3&5O-3H8ZJLf|!|w9IJAJM)WWMT;vUEK@;?ZX%r01gUd)NK-9FjlG61 zr8M=xZQ0Z=c!cV-OEkEo={TjdP4&2?&69~KO?{;N&f4=*iLvdiA)RffExtviE&Ys5 zC$Gi9ps6%L>2yzNM#u!Qd$iL?vmUUPb%iiq=`NK_AA#L1ZH|}e8>uh@J3()0-NT`- zI=M|=qd0}kR!qfl)k)Es+L~*c+RNI@Ynt2Sf~LAod0>%yfiZ-LZE3K1O?N?et-PkY zO|EGws_$-J-_bnSJg~lfQPKJiKfh=o&DRFp`{DnZW{`@j?ThNO>ed%6sw4arafS1U z@z2-x+JmJX9WB5;1O8u6>Y*lvAq^<+A8H)zR?={Jlg$Ub-VA9d2QOdmSJT)KlZ|XG zFRb?~VulQ`IM!WTw@%G~Q^W!}o#<-os5nr0c*rxLmBrFN2jFd&_y!@n2v*L>`&eml z7G&0{V>$7mY8M0fKhv=yU@F;~#&inOXRztb3zq|(K0+|yv-34n0Rg*v6=obbw zsk^4Dox7~Htflk6xT32crZct#Yyjf`Tn79aovOivHPP-rO+e3WoT{!(n&= zl&-{)5BUT9qv@de_6ng2OEhYref$eDes~a?vVU>ABM@vquDRShBfHw{ah(LoG9 z_1DNp2CRp+iv7ll$;(m+%H2VO5yJz$)Wzre^(lOK1fIl2Vd{#CK?RDUAijfyKhP77 zB?;hN;0&h-sdP|;M9>VUEJ!3#A%hZ13Ukn44@b5pu|qOyw48(D;4m7VStnVq5k=s} z*(G_zN0a_`P36n0=Kk$+i%L7Fu+L&+3cNzJ|E#C4ne*w|j`EskZ@OdWki7Y_oiU)1E!WL?_rZ?lf(!x*3>voHk%(ncaiD-<9<%C@xu#=5zkick2zu z{A=RJQJcvF%KKeYKPHc~51`~VH5U>w!PH2r&G0!bn}8wEGyY#Ewkd?dPb&ep6BW1~ zy<4SW18GeSCy!SvP&(nasuwX0KJeA^>RW9dm>N;Ll?DoYwX|MI!y_U~Z%=JqnLNM0 ziY3CN6BemwOzXr0;_y&X#dk#SJT^mYDS)=trM zYdd=a%7#xX--}L*clprBug!s%DUA|@g`O=6+c3W#5N#L=`X77ux*kl3lI%POhhqvs zn+03GO1)D1uey&+0wEfQH`DFngJiW!FycHHoVNJYB=U!x%RFWk@E(NJqh5;@9_2j2 z2VqIJ3+I7Okj~zBtj$3>*2^?`4|Gz4|C7Qa_MC39MdKC-KqKb?p@ku}Lp!cjH=2xo z$8-tD9IK@6ff=3P_th;_fhkKWUqDMTO$aTtXmsTHN4>rdSRCPn*Kr3}38x1@nRU)~ zNIARk;U!(LsRlIls)rkbHj)dk(*w#D`0)Nor_u%=UfKmpr?s@^hI37r53gf@DL<*D z`=(6Lb5#^xi_y>c56IXZZ~eSd99~O=^8leyuijg2XS~km3@_&a;mCbQ-Q~REY-GdB z4=@pl;(@C^8KDr66p;3qa|RubdU&sT$=aUP5nc&?!r+{9M9%2kRqtGAT$QGuKZthj zw8!dxIys~H!*ouba&4uNy_VBlc#ZL4!|U7o!>^~EKghYD_-bxEezR)XU_8)qFnI106fIyDUm&SPuYa46q>Mk4Q9`GENU5wB`WE>aDr)>12_UmL z7gZf;El)?h`e4NGnm^>?*eX)5ej#ixPtw4OE5qFKuq3JP|VKPAHbaaya&I#irBJ^on?c_g>4D8|YTw5V+ z*BfjW6SMuhg|h+hK>K;iy6%Q|U(dC%E{L!xF06qtUaf(zhI`P-3w-o~JFxw<)4o

5xj-C#kQ8Yhd zmij#BJtWN3JeceyKBYR>d(L2{GY65q#)sTe#emv56z3nR%d=)iPQY^z>B4GbMvg_B z@IlOAM%zIRiQ}znMkxUy239$2L&C=3unh@a#UNDBBhRa}DCw@p3H{f0zduflB*wX$Up4S$jQi{hvHuph7*<)tS`r7&Dy>O6f(urdnMZThe!D>~FtLc~zh;Cd7lN+?) z`>*|d>@3RVwYp;y9;M&DVHNY1o{1&3}jd0pa4sP$Xl zS~&gDKlCu>4+AJkBg&f}x+{`LAD|ru8`j-GVDIbij=%Kc{fEN_JU5_s19(DD_4T3m z-dl0#?YM4IN4(h>zoWG^jt`wJdr`G@hJ!7YU(4UQ^N>xZ6gOzT`AGYhV%t6>P;f-n zm$yZSB$8h$`7#1uM&Qc`d>Mf+Bk*MezKp<^5%@9!|9?jyF!qj%tNsgJkXx=Py>QI* zyYPR(HX8HZg%f+GS1bQv+tRbAlwH#NWMil&I63*BwEfm8Xv#Mp_g~W!Jf`|TX_wf_ z*?I4-kI#H3jL`V}|D^5fZ>yRy{^JV^wV_;v*o|l^ss1BH8r*%tkpGn2=!> z5RC? z{ZZaD)R5KLn)lF{OT4)bo+1xO=w(GqX6Fla`M&+&G<4dkqtMSDSgr~=4k1$Bw>Rq* zsWJBbD)<2cOlDf(j|I8gmSEp+Cd`rRCqE5nOi64F(~)f&Fvu!yYH{&vMz z$Aw|>c-Z2fRdp&lq4Q#8M^$Hv^{omG_@5`R zR3C1YvhKefy<2%MiaZEgoF3}phxy`Js(Y5aeZ=lv!IS0w#+=!I_O}L39J3hpg$A&v z$YCQ?R|g}zKMTm?VmCy`_1}7q8fe``PG$eCB9Cosmp9&8rR{(5rT>->#P0O(j(=#x zQA?-2v_H`BPlcUMnRpdB6z3#jNNPL2Z^T98s-D}_e%QkG@xzzw_Uk5IrsEJu$>01Y%2|Ga^6K2J+9%Bek~HY_CGPxb)*-N%LdC2XKKNB{cJujgy&JmL z#TQfEU#@U5@rVHbOJSZt&>nYv=*0UQ-yIJhIgRg$4uAJBkJTH__8vco7WhC=@7%Ir z$<;5va+207Ygc>lq(SrF&Pc;#st4XXr4L^dT9UWD~c zLkoFw1oFo!VJnsB+08#PvTLIolnRN|AA>aUlVbw!T}C!bl8r4*ez zVeQvoLxFq#G|*A{?~itcuIQ_&ec>NNSFIe-FZq?V{ewM`9g4ToX!Ob%Pmdj4*D-h7 zN7}?o`)ZmuygGDM=e|A<89%~nwLbBLHLXOnC_PpxrdrrQuOBO^`tJN3j&#TG zpZMM16-(R$kX2>g;Go`jg99ckyIfSTa?%T1c5T{l*KHeFq!7mBER|`={mzWv9=*jc zCsPaXI^D|!BcJ!zeSQ5m-u^>{!MQ

b~MhaV3Dz_dZ&cd_;*x-KtECA5J&J={tzc z@712zusTfUsaU$&dtn*BV(ZrJAN~*f%@(RG2wO_s{>OSvr3kjh zY$FAX;Q}aA-V|zVk~}~z5z-Y%PROwIB~w1QWP=MNoL=D}=NN|FOgrP`^WzwF?T{A7 zV#sQS&45(QB0S*CJME9B7-?BkJq#Qmnp{bm#Ti7371fz`2KX|4K9{NS+`B8`=h+&z z-8z`rh_9Y~XYSI?`<}Sb(kGaW;2RX9(o`_K(+JFzL@dDd#FC`t-tEBSvT$ zdD$DARzGvov}^BIP~g66FvWYD@+U_ozZd^y}m|-5WUkFaVWN-eDsTf@Doc#G;ZnT`Icv0={bZ$7b{CMB| zxVFybe^i<_Y4sZ~WHeW8DFb?C#)6;Skho7}@!_!}D=)uvJB#v!9YOTZ+_-eh6F=&4 zap(Me)3>#TA}zwsqDF^7$XPW!rB)=od~(X(wN2UT2MG4qAmN%0g59xIXe#ytg{g99py_RiddWneEVx{(-Is0prr_6MfL34H%-~#p@2jCgSkguUh)QA@Hi@I z)@LnR^~lQfpCY58BbSBZQ|rCPZRF<5H~i_D3l>uD^tDu~J}6E8bQMC@G|)jJdGcgP zf#ZIorqB``5kzoX-MG;~}NNN9>r}UNAN0V*YeuUt7?t*ZWoFi! z?+iy`DE)`3F^Yn2y~w?jCk`EVjQqj>C?;lD73pDABoR|1Ns58GBy2)J@aErW`^f>P3SrGbM{e!ZYzo+ zJ^GdferkHut?I^^YF)==n_^dJ_dG1`+@0>xLDdJHFTH@3ZTaz6o>5NAIh|_KYYd;b zBcXDX$g!o>pFc2H>q(hCmdKW5LyL@p*-E_od*O5cx-sy^7K;Y?6K9he&Ix4D5gHLd zwe1fibvVsI+WD!r`OTrb1fNU(19?FBE|3RFF2=3ZbtF7 z^I~0P;gUO78RD?|q)JgGLnHFs=KZ5bN#l=&-|V8;%x6*EAP#L!g0!D}7zmytiAzf3 ztVm$x-{kV|@0~v{3$QSyAgfyIe!L2H_aD;^=o#5+A zw49{Yf8~K5@@z?5;OF-+W*nj!=0C?eKk0?N%c|zbKKLoKxTYr za@42?UU{=_EF~%pcU+?MeTxi*Fx}y|{rvo(q8{~_Hy?QD@g4coteGbJv=KqP<}EA~ zz$lUw5YaZTS(dmW4-`z1Ki!4)_lr}M47$R}xSH_Yxluzu+VSz4 zVBScw=5tqla_^-TS#NGlm7<)4%L07h{U_R%s@b^pSNwHqZ6!_K+p2qbDmtoqC8>4D>!?R z$ytK7e1QTA*-LUUIytTqDU8XH6s#XvfFekZ$w4aVZfkBUZ<~H$|6ZBdII%cQTk zxxc($?w4yT3gv!Z-Jrj&Wnj^?Tc+k>P8!4jRaQ*&%l+kTgulG4b8UOYz>K09MJ)qa za9n5X?47;3Zz&e+>{r_H#)QtYfkj0X1B4%riBjSwU$XY?pUvYSx3%X#n|et}nVZ66 ze8pYum_tIp9PEwhkmfb4~o zfl6mRQ#$zcbC{yj)czWSlS;#rz3NGu(+II-J^O5FU;BpnPhFz29BE-&8txc+zPMDh zt6WLr+D@ES6Vh_dpM05!?Si})&Py*pa?-S02Akb{MeClK{rvf=NU;N#yXppE-b~4T zc3MTDgVkyn^My16^M`yj!bdDqZr^x80(Es*LJojx$ldisrQ)C%jl#vT(yW_{X2|^l zqJ{DotgG4gykfO&;DVPf`k9J~W;E{?Q3)UMAh-I)P+WC$#FY7?F8Y~*>$3cgs|+G% z7inn3t+6W7pc2P@TcM*yWr1XE}?x$YtY_fzHVDihiU9ZM8wP;a_{wH3{e?mBN z#m#G$Y}Bvvin1y^{U32-!L0Tcg|SM(3oadbzN%qp7MGVctnH+DXPujWV@P$P^j1v0 z>C&9@D8*|8vw9mF`ht{;RWtoU-#|-&VxkHc`-LZ)++CknVphz8FUY&?JjH4iX0M-) ztri93qM57r-cDO=UE+IY#B-BW1BEdvcy`8gg=Ns;-26Q+$Azo=lu@(yNDV6C($5iv z*^317IoD6#q$_eAEbq-TOwqc=4txJc%I@o^Z&GQ4sS5(CHsfYMFz2V*8SCh?3Wie* zUU2F2yRkzWp~67cbTK!NEm^zzwh^BzjBUbEIJ2Tp7H*sl7!ILrn=ZH2wb=IUlZnY8 zl%}>x&29v0OPd@iq^SpT8=d+tX?G-8!v?H14h%M{0K249*fvCo2e653MO?+mSKR4S z47737rKyiBfraB*cU${91yI>j_Xjc5dchbjkyRZ`rGa=%w?Sp()A~&mWBsfF(XSm? z2QbQB{mfDYYIeGjrg3sHAp2;?N;bEF&FZaL0;iwMY=hsl(9x4R<+lEr3A5LfYShjc z*tQ8(t~KHZJy_7U5gP)oqpiUyjpAMXgMLNl&z`y_Xh3>Xc;KXwZ=lnsVO8hNUhvFD zMJPp(!~U_OP4q#@F}OuuobXN-#+%R=GY>cJv5y z_JW%?ss!K)Hgm;_qLrU=ZP9w0naM#LnQ~sy(Kpyh4#iZJk%e>wqjrIY0&Am%aQZJb zv=r=LtYVAINdunCPj>AgQqCN3+BT%YI3}E=`08V&4UMV}8wZjD36n!4O?aUX*x+`0 zPjGe)dr2FTrhH+CUttkQZi`FB&2aLFs}8)g-_!%QeeeKo2^C?kI2DqX+*Y)R$J#wG zDORqqoQ4NN`{pb(edNZ=)hSl^?7fZuou{oUCOz>Be*wtBvlnck zjY&#U(LhaqPPJ%GplDrlt>Q$bU7T3oDZ$(KUO}X85QA=~1!lH*I4foD|7(G)SJW$wEFss%J zrlY6RaXd|~xV6~*PrWGn`c=0uV9e}?1SCVKCm@E3E7IVy?Fxq@&6v{B-PI4tqss~y zbd;}X^Y+KTt_aA*6=ii;(;U(eMa;5hf3fMzo-sA>(A!;L^!lCBgvZw4?|t!;#=%{4 z>h;?gw>&#~PO7hX)}C9=?`nVI7YlCp9+aZ>axE(-m9!L8^kpqv*IQ9D?Y60-7Cx`3 zm}TJL;)<+&d!UDMrG+A)yTKdWFvZJ=+XCU|1SiKEbmfiLP{1DvwuS`5vF?`03%4noL2A`;z-BDqR0=H{nOs66K-1$$vqyiU7q+v^g zT=o}z7h$N$1AlJ(VWOokS^L&Ny2C$kz){jnG|DSAly(+Z5ck>1U@Q=YGuQQD%II0s zduPI6PAAj$+248j!}Ze?Pf3$7lqh9*pw}@?)k!5y=N+@`nzUOLXmRu25rnv9v2bxm z#gdzJWtkiVF`NEeHotu;r>R8t73(Nd+Khq8RrsvGh7;hL#oc9fD)HRxwG~Ae)$}EX zhsw*^S!pU`WP}upNfaZhi)NM=tc}^W4W5`v?|r0>&L%}9XyOV3ixz5@=akXTMt@)7 zEQ%G?1s7)ReErykvk{YD$)3MxnyQaB44Qy7mHnbcPhEP)kHvmd z8_v@;#gaN!nm;G>$l4NYAw?d@OkP=xI&j`G&;2ZGpDpdAnyq+On<^ws{&{KVMfnOx z<2~@zm#pP^h{y$(4LqB7CMFs!kD9dt&TOhrDJob~Rt z9auAPx(5zvB3~~UOj`$YXF!mgbb9&XEVyp(7PYT0Pa_}x{Vs_7ALYhy1OFti(3K6DlcdU*En6xEIx}@5riO=j-pTX(sch zpu0x(l!9kM*^hr?V102G9s>R~z`X|k)pQqT`u1?a#uB-?RxVguVp zupII09x3`!d@wkGhNEHzdBEpOvB@_G!P~vS0dK3uwIr$C#V-%Jkvzr+w_oOU1W$rL zcLd;44Zm6l`{B1M=EzxiNFY-waCrXl+GG!h z_OYLE*Re2NXgb(RJt2E2eM5_Dx{KpgosW&>hZ0`%wB}5 z7`->BDHx4^5GShKI!w&D-|q0On0Faox$KI~XMJ$=o{>Z3w?}*#|6qRNS{_N-#x#ZD zkBX=c-1x@)O+pbkhB@hUaU^Mv7-NbZG==c4W6veJemZJ4#4$sEC>S=80oxydob zM2nJ#u#Rq!8hp}RPO|PzeaHyL6N{UP| z;i6PhdeywS`KwzmfAO3hM}I!DKT8@a?QGpT=bArHef`*-d0(_xD=$nj_@2LeQTqG; zDsbbv3Pe^-eeL+2qrOKyI7h))TIV(0hIx*1 zEU@euh;<|iKX=fqPEm?=j-rSZH9M5l3gn!%umxag{S7)jrjn3DaW1rS&nXWh6eR>w z__fda0-YjC`7z1M5~Wch2e%6_AZ+pLKNbZHhSMe#xz{=!ktUSuVx_I5mdFoj|_9*C&O zBnE?X=PFwpZH90}OWG|0MV(FBG9o$DS*DtpV2Hn>2uG5Xo7OQ!f-x~7`R=>s!*FX* zC>MNIy2pYUz>Z|+P!|RQRb}(|VIGT$GNk!T{JTR<7>1W06h2j`795924w*T)Li1tQ zpzx_eC8IG-r7+a*4s|8?LE%!x;D+=L(NT~F^r?}UE5(>m0zN)fbWuSh;gV}@NvEtB zs`cM+I#R{NyCVrR1tzRiKiFc*@`DlvG8Azn$<&a*uFk>_3Qd(%k?hobiue502SwLS z4^|{YNwC}OJ%8=wH$@$hlwu`5OO?Uk?jW@pBf&g&NRbgt^nE+Im!2e?arvsy%WZ-j<$EXfzHyPBR(M52;s%gcx zC6xn?v85Yj$huFqD~3;&rzT)|Dkc*HPL~;1i0@|qi7WnG(sXhkivBxYc6C?}j}=C< z|HKs|9##j9Fglg_Rg0Wq|KX|7h?`IW{Pi$OI_p*Rt29l!?05fa&-daYZEgLt?p@>6N`}k*K3dS=xX`hn0a>xqY zwmC0UbbF&=opmj#^CeqU(RsBm3-n-hJ2Ek~$$9g=jkru}mnj~;g0 z6)J}Yk}kg(cV*r3-d*hu!tfe6>IxO3z)Y9M#?q7TN<6iZp$rGKvR5p`&R{$vuI+tT z%$FLQa!BAJjagJdQSAL$Q)4AlF8Cz1Ql6TANEA*aT8XDm`h53F4xOr*(U1rfjF>HJ ze-+?Gs9_8VFCHEt{r17P#v798yTH{Tq=CN zBjbtts$F;w#4v`$sX;pUpj>BulQ~T)C$D;XXy}Dy-&)DM?Pp-pD2-ci3>Pp{Wd#oq$>GH#CN~nFxb6EcV z@k$>2eSB9)5E!3qQWAFL=?@<;&wLIxp#>{$~(b`)dmt zd|lVX)oA^H!-szy`}XKx-*Ch^@L=eDXn~<@6oBVx9;lo2Tn%D-CYX3@Yj>WQ0DfWdR;O!u=y~k z&P3&LCy$dpKj$<5bv?mR2YOYXiJeqD($LzRM0DkEAO@jTsTdC@EuOq2o{aSUx5i=W zjQw}K|JxI_STbzM!GJ z>0dIlT0zm0s)JW|KpG?@odw?*{|FzD^uY`L<1Cc(>mSA=aH;l* zdhqHFNW+MaKe}{aFW_77t25$F&*X!@?f-1) zAgpxGQs-=1t=CLQh1%;f#Q2+b1`IH_O?8OqvkLRV38qMu0$h!`_Q5qrQFQ@TtcWs5 zJz~0CDaNP4N-ydRRyZtnBgQf^LxL@W378TpW#H*{gk6+Wh*5Jrs25N?EQrEF8PWw% zDC~)pP`(E(Oz}yrxeI+3x#>rS-9Ovla()EuA{aaq!A{vZuyv zN6uJV>fnY3*$Oq*Nq)lORmfYUQ895}@6e``ZV!GSpvK-y`O*A5C=3@+q)s)00mhcl9NhQySCW_4!i7Q)h?s3VRNrDj;+>?EPnDVYn5p zK%AcL$4GOm|FYEHq zw12i0oE1eVmZC?n>xrjit9mR|D_<^B7HcYsfRt5aw({-VUcxtB(x8tWs>tCHi zG)b`CUOv`!W`1C;o2N8g`2CNoNiFs7>&c(3=qUJumUy`^shL*N;NG5SCDoa!R0`{G z#sfdY(OI@mYq_vDnfEYcak}VI-@YFFRR>z1+V&>%;1euMe367}s=*}1XM}PuEYSBR*;xBH4PMIy#|BNFGqiOh z{Ds!}mqoYxa>;#6lu=uHPegA;BPRf+NCe}w} zd3{thbluX^{&Mf&AY3exg7u?$m^o#1>%x2%GYq(OVPO4xZ*04f7s*~A%N;HN#leaenBe_ti97$)Ulv|tFu-tJGVMrd=@U=Yrw{;AOxxD^ z^839{Ri_g&b?<$Ynp?i?=2y_hSu?(ynvGI@e;+%Ph`jkP!O4)a zkovPPnlHQgiN~MAEHWjWAC09OEF%f_JXKl6pI1_9PD`A;`u=cEaN(8Fw4B3W2u+-G z!PNT-&c1sXGHF>!61@0&=!Fg0ekCpQFc{>^7CaVyC6G057&7TuNfOj{KfYGLnOCxM zNv0|16xWmB>OD_a36miAE3v#6vZMoebU1l+Z74Ul?%PqPIkS~_=e~}ixf8}ZA8-g+ zS!5_OchA1gqO)&jk#HTu&Ytsb8dq>u7=D74X@Or~=Dhg&=3IZ4pZpq^Jv9Uw9fd^= zpMH*(|1`?rQiAe=`;yg_fvgjCE^+~eUHouCtRl4G_3Grql~LBRi_=EPZ7=PPZpJo3 z2nB&**XR8kUinN?6d?zXlsCPZ+}z6WS|>oB6WZat(58(7yZ|M?@Mh2ER+fMzw&j4Q zof(-pS9PGg8uErWdwL#GY%*+y3@1Qi;9~$;MS@3h(MW88YGG0V@~dz4sLL_OLaral z08L=L8Jgb+gms7baCdmQf}19z@|*~ ze*Q)8kpUQE29W~%d~e-k4}U?*nR3JNclX}9B5%E*+DZ*rFGeoe@WvtW z*Xk(Q?)>AuaqYvuCg#q)E|u=Uwt2Jm{4Vp;80^PuEJMFT_e?-P-1>vgO*yTK+G>xv zedFn)N>OQvC{qHLpjqtd>wut`Rm(?RiGH#dk%w@Zl%5@d+u)Q%ZtY!RBnuY2KEp0o2t@-Q zE|iZiLN%FaCp@8K5A+ss1QdG?>P>o( z&WnZ?R9m0+fF44MYO04<rM|z^D0AE%mF?@iZasBzj%{Ua*KNGZxp$eAvi!1UM`A2kE|`tpLor+ z!W6Y6VE&VwK3X|{s*=^2M5`Q%oeh#EcE;(n$hNQIiF2>(C8uz4fzMvS+=R&)Uykyh zPyn?5U7RqicGUlu`k*7;LiqIQd&~Xao8b*zg)Tl&a zn2MMwN}IHPTDs-0Xlg(5DHM}&O_n}strDAvbrCZ~dGWFGp1(Bos>BkT^%NCJh0%zmzJ1eBi-^eJ;I(g+X~H~br$=i- zORVCSK9f6&j#!lqLmhb-6|oA0Fvbj8hP|ZCW*k;RvX4ltbZYK%pTQ&*Xvm}oQ*)kOUDm<-WVFRw{yKT>t+m}A zq_Gn^8pmwe-X1yNg;9s=~6n z;GCX|uWyI(8^|hJM5GcDcJX|P_%Z^+k3dZA@avg412U_nyY?5?;514~eC;#p%R!1{ z;*6rKwS(qdfQ=U8>f*gBCm$Eq)$|sfCMM`V<{qvi9-62cB#x^bO#+N*M(kKTwiU z7tW;eQ*K}KMI*gxGl=lbGu2!*oKkknmbX`Hld4PGoMNbUam6!=PMW^-+0~db!5KFO zwvA$}Ta?7sIM#3NhZR$;erIGE>kDFWm(~aoBTD(pfc#%g+N>h;=eVt-byfomL_g)OCH-09H;ba0k ziP%<5V`k;tHred~k%rsv6!P91S7Uwnh#}Ety#x>M&M$0xHYO=LOb)gOhGm_gl@4|^ zV;yuIv>DTDW+@EX(7;HOojfXHo~JYCw#$YhRytMh33v9?nxJZknMatCm_;}@sb?t! z=10(J8yH}l_qufo0OI0nUt0o8m`OCG?I^<62eod;M%ivHwgP6 z;hLqi0ktV@#K5^tIK#z&j?K#%+eAaUU3`cddjzO$TX`n(u$|%3I7vFjwxJ=W%C#MS z#b|gQef6b-byXaKQ$K_`<+by_f6?1o+@%EPu$9PD%3ybJD3ylfMVoV;#Z?}KG7m9(`{w!u9t z&xuG(iu-&;s!N)BklHrcX1uGPY$zidZSR;H=E%|9W?bMmC?$H+6n&d;fi&?O{=;d+ z$u{Y=^j5^?P1kFiQ;i~7xJfLU!9 z*e1Q5q`RR(ByIER5I0h`_gJtpP+7|ZPTHP3{H+6(q_wWWB@OcWHtCu8ja_RQ(QU6i zmXFjVTZ3gyn2SY=%GanSr>6#i-&DCM`zXz(sAIdNp*$%qgmqThO9`wy*9A1K0nmss65On~I4(%SE>=q+QGkBJ5FFp}^FA z{I$Yad4I}&QgQP;F;)-^G}+;F;`03?azB3VGkBiO@CrR(qzPRbR8T|N{5vl^Ur8u{ za2A&aM%v)Cuk2|&oEjW0mvl@mFD$N5fyqmn@xUcbq`aSsP|KzUp)_1u!nG}?tlqVU zG+f&kroXci^^W$}*q@f^CGUftD=LSj(&XME<+ZixH>)2ri<7;-D$8LRV|q>Lwq?R4 z7uHuKRM03Yj5H-XrJd2E7EE53H1pt)s^jjg7!59I(HIc>!rg&Aq#6CD2he}NYI$SM zc@JsA#%ti^hkN0{jXqe`RD>CaNCS-8#o&%Y`AQHM;5l4ERew|SRosAJ4c^tAhxNG6vx$ie}(t8 zo+}}v0i-n;XQ}>~=k=XO*V{C^-<}?rv5u){Kaf!Lokwf2t^kf8l%|-{ESF)^woGX* z4={FIco=jqX`;oUL1dYF?7j@3PSASNt519#(*~>tCd{I2ssM6q)oi(9JLvAdQwei* zV@cZ_>7?P_q@t@I69G<|c#vM&+$NnH0Ktd?w!SeQ(^NymI${_AKth$ez_oFT zZN5n#Qp>wH+J-=YSf-^otGvL;?UYXEL2*S>n}Tg?PH7wO1r;rl+iaRhI<*HR`{HJl zMAAfes&Lrx^OCl`b3?`5r45ZG7%W(SyK^(8z!=tXhPmQaMC3E@8?swmthobZ&p31Q ziq*u-T^7y4DWqZcfd4kLz=M?1yu7OqX{13OvIpA^IH9!CB~5X)bWm|fS$#+=P47z` z;?U4FWFQUR)0K}`k&!g9UtXiSWHEffL&YrG(Xa?Hsc!wco2{k+2rSe5Br9#*qIR;O zy}l`UL)Xh^jLdh)Thdk(&9&^DcZ$~6&yWV1*TBZ!=BD6X?G9O3_Trjqp$rYw_Vk=` z*mhjiutj$P4rzl*rCq!_!uAoG0%`CDizO-6(jbsdZJTP)YfouQKU~ivA8dDidR!$E z@Rtl^ypS}KG_5jGCE!e}Jq@EQc)c71dND;xT_{0dX)HOOP7YAUB+2n41XM!;TJX~s zcqhLKHaQ6E6408fz`-CG5=biTsaB9r{~NdOW1Wn^4Y|c|V`RucnQo{cEj17-6kX{F z(+Ef$9~ZUV02FltULt_vui6Gj!;S4h>P8@FI*w3I4`Dk(frB9dhUR0^A{jdE0R@hv zKnGDUn?QyHfuNYthym3sdEt`?1a9PoDNc_DbwfwONDL?d>Cu2dKmk=pG(!ie8w7Wf z;f5JMz55VY7^rbk>IUh)9&v>J)j(v7qW1{g2&fdZP45x363~6Kgz^kPcf$bb_!2iD z5A<>pfuKhuek2f3$oP^TQ8!2n+>pA)Bw?UP(YtTB0R(fuTTO+J3+QD1fNfBR&ib`!)1yq(QRM^vo9uSR^vOU?|z*iaCr3d;>og%~d)ssg`IvtIl*gnh<_ zf{Z31NMu2pE$*!uSt@KrNt?P|+R$sZgARhL5mT|lD&Iku0HS-il1&iR2E}5ClFQY! zg=|GhHb9M(+9y7b{W1l*j9o2lS@$lF{ML+y5*Q@1t0N44L@KLS04dnvmHckVU6?_M zG?vyk!iOCB4UFkto1C!qc5MLmaE+8U)oYGOLz076`s%_7I(McFB_nGa>p2$2D+h(| zel#YP4mvn2uLzb&e%G$F5p8H~*`Q(zM`NnpAQ0WQl>CHVpzW&Mo!P{tCQGf^1ykJ{@jufEt=7HJBab2gOVkZ2Tt7uuqQGEDOz>P_Gh<&?`b9 zg>GOBXKak(bU;Im6NJS`I%z*Z1H%` zUp|s3RNB~Ki2>mB%BHwyN}y-rJgGymSa`2Mmrm0Mg(Sbl39D-Bs?MaywdfM1D`4u!8$LPvM#Uq z(}X2y0rHMm4@$trw5C`%05G4xED>P+ydXcVtc++$C&@K`on+dQ)unfGE1`S-lKpPy zqe8MG_(4#=7$g9cOW^-8stJI<07`Yx^b?$ZGXQ9#7ob9ocw%h9MO+dUp9pE<;EQSig|xlLoj9R31s{;o zZNbHz7Z_#Co>#%Y$=a(7i37o%7(aO!7WiG4^MuFH9GmEcgKe7jcZNC)P(iVz7dsaN z*c0>S_3J-dc9-y4!b=I@VMv^73&SglK{)6id)(Q5eK)EMV@fbes2L3jr^fF;SJQ4( zIIOstI#eK&;xi3?-#MDLM-?k(u)_@%BN`>JJah&^H>m{lE(g5dAbGKnrb`r&4_A7Z zjI-CJ!Wo0(fAW+4@4ugDb+HkI?T(wwWiQ9T$G|5C1KFJ=CHWj*>ZT}r83|zwWO->k zUd)~O7$FzPEGRA-AUtEBEyXBfJB|g#WyU!qPO>Bg_N1P{zz`E% ztTOJT4HfXFUsSVHG=3SQl>-kVROb1|jDa8QZ03~H=J2wYN|D+fgDix`t1OENA4v9^ zqQV%Y>W{#@j`GZ2po(K4p{C>hnY{4ASze)YnM#?xsm{T&EX6LH^*5J75h7jodY%Ev z42BEakhV&s)`=4?C?<<=NGx$92C;3B|Eh%)1H&H@s>g}}hI^cS?^85i8xPCG0fFp= zlJR9xv^u1T?Hp97hzm*Lh*@y-I!)U_TXua2LS-u>-Bzo^W5oSo`7HQ=XRvqNEZr91 zUQjF&0aC|695;soy^%@wx}^xlh(R?9q>r0;L7{4`?4<~%rZI@7+!>w-9Pp5k8yk8B z8Pk@b;5f#<2N)cYJ;Hk5`a~U`)FI(M?xgwt-;aC#7FcOMQx1v7 z;VF?44~K`8E34tjT^tWj^w4nQ(okI6TCE-$ZNt_A?+o9mV!Rsjm(H!4n2npQLoJwx))@pr}=-~SxuR6Ik8L(3NLyep>8yMCBTm0|LyV2|FZ z#`((}k<3Hke#8v5Mn0RWd{qhvk(%+mc!X4R-DHlP5{u-4T>;cG&@ zEuqZaLu?PX49iz9<+wBl)}72BXnE;5P<%|gO+nEE>i>b7as0C#N9Gt%+RD|@GL^84^WQq#0 znIR)ffv-0+-owC!5&Wt?vQIq>sxwg;(}tEJJYk`LR}tV218PE0*njJ<${CqFAYVs6 zdWH0-U+-81chh~mg5j;Ja2`B>^FVf?Kd|3F&@bDGYrl2b+XdKDISD#Nt z)wk8hRE|Bv=5ZJqFTjcUt-mX#`j%w&GD|pSL2lrG+^Ebk@L(`M1$Wq@$rj`W;8iHI zj3u?C+6YR3+I*WYV54m9HjYBo6o5xzP#~Uy{YYe;24v}BIOrviG-}atRMf&|G&mM_ z<^aSKRi$8$TC{8nasxz*S*j=?sG|ZLgOXHem{DiYI!NNlYz+)8XR&vkleQ|96);t9J{_s(y7}ts9;H$kNv;bEGe>(3e?(OMzbqEuT3u~zm zz3^AmBe?`ZDwu#9GWctYC94Axd5aN>O9DQK*|WPsfY#Z`pBEgHC5Czog8NvwYg@%^ zH##%SfFD!F*gj3e`KxG?uVLZx^wz-}lg9TK%`0`_Xal}D3}pZw?`Bf++NRAb6Ecc+ zWF(9-oTl6*uT8Do?03YIV{#3g9Tt`|)|`lpuD~qnz+m~Wo5ohA)WatLwc{9rSa(`I z5R(iBu6%2?_OX8yHF7ZZ_J@{?jiyl`NB~*2W%QAH@1YH2!`Mb10Hh~df3kSLRNj(C z+?4h*FHV_7x{Pz+nE3_N_yA!6JmX49sMKe8w)C$5H>(hz{=5bR=aQ$K@ zK-A;1BJbw9>mTC=1$%g-0sfWq#l*atx_RfN50SnCfZy`2SHyZkTKS63XjQ_soVwY` zz{w@=x(NnGzVotZHFhnhUhJ-gn!aT&Z_3xS^~`TOyfruzfb!X^&wI-P#Ycrptt^szj_^8_=`e4!4ZD6_%B27dk@rp>-Bhxl0qdVNf*|vNSME%qBxIw z+)R2)E=&<)INw%!qwCLX-C+)U3R5Z9^?C5aHw`3#w2?ohc5B__nw<3m8RDb`t(e_Hf zPzb2#`CyRcuIBeXHq>)me3SHjH1R~?DbS89`gN&6k(-+TouSUb6~}q|$=LCimz8h- zB*|brLETU2TwFn3d2hYLz?2+_}S^k>5;YQp$7 ziS8z+p8_r&HJrXTceS|Hz}1wG!!E3Bug`4UGasWlMa2zK&JJycwkB_`jlJ;Z;Lrn~ z?Y;Sy(cNWbl*HELw@fg*>=F#pU-;G4mv{eTsHAiE7x5)42kK+nc4~8%Eb+HYJN?87 z*XNBKaqlTu@rpdT{!ysXtWWPMs*k_7Q;W>?Nf@6!VcXY2IYH&?_S@Ics$Cl-Un%R!xlwhm)^ zJ?ZzT;2iWGs3*aSLhULsE#4Zf%D3=GN}t0}%+Fmg)(IDp9-=Eo_m0@5y2liicFxS_OX`Z_Vz{M=drXH8iSHdTw^$^ z`;IVDx~m?i_^Q>=LM8Bsi{ycj#=rab{J*~Y2ugY|IMF34TGJ{wqdc|~P8O?0&T0y2 zWvwmxoe>bt`vuhUyRu#!x#z~)q(cU1cEyu-bRYZk*B$6fc*G4IqS1+q%u`FJpO1M?P z5rzGLGr6Z1i1Mft8dRy$Wdp+t73Mv>L%?$hcX08I6B<^Le!!ovtO9RasIjEgJa`}y zYYna3h_0P%sPbGzElXs?!&WO(ATMz(dgKz|*7qFNl(ce=ifn)FT*3vMEGjVOcC@enFxFzpQXMdv z{|C=0ArcUS(7Qr!7U*MFy+q(nsdMP+MHlFMgszqJ=!> zjq3*Po^3VVfRUnZynaemM0gi2aM>d$8b;&IPeO4^pLdF~D?X!gmLbxT6u zbKF9+0HGKw1No-u5_F2D8f1)FHVSh{yE*#dz>_NFLZUCX5hC;(GfX55X0dmS0|fja zPo)IU8Z*-HXMS_XIzTF^q=Al!N~bi)FJAYIL4Z&4>j2^MD!eqpHP6D1d2!WCfWgBb z8+bnG@6g&IUw5sd^HCoeAqK?cffHsCMt?KeyVc!qJJuwE39Rah#U6lBAdq6wnUR|G z^GUBX$U2$jpa@sQ$+v*rc$YNz4W=srEGu}`sca6>Lc1fqOZD=7LxTL57U7Gcg64)AMFlO5pY}rnwA+>Ewmj_@- z3-nvI4M7115DtZHMn7Z;QR-Y747xr%I+?cJwsj2WH{~X^-3M%rHbXV!Qb^mpF3YCJ zG1azUbF&bkyQPg4+~7;6A+}Puo}AA{I#JHnj}k9V(65=XiJ$sT@|ZorzO16McS|peu3UP>#xrFt zZ?EDMbF{8`=4Jb`4Jh*2_btCDw>MBL50<{G?wS+-bK&cueGX5Fd|%eWMMX_3z{={S zm6QItWqein z>yeL$4*jVl{CphlJDkODJcy{AP(Ii=v`VW2$F_{wozqLWqOKc>XhhQ~td1bLnXwDp z)4~G|v*1{gf)+;8SPp^~yn+C?lkh+QF4Wz(0!my<5-GO0JvFI%fLKoC1Eq9&8diX; zK$MdXHmecP_?tn$?EJHRJa>%rOu>-TSV-ZvM~^La@`lTF#rja zI_OLTzN`Qm6b7S%f(w6h67rmmp^PMr^^+{8Zlx|!6CoZ-SAJOH)QXP{0_fA!%49f;2!;~oI!wZp#meMi%}Q2fIJ9rT(+V9*GR0US2kh=XLlz7rLp4FO z`$8++%w{JcqloD~#{qp4JgIX{50c3$qmTjJ>jy)$L0CZeq|^UmgM|yMC1*`BcmfL= zEW8Qm-G^VF#a#Kpa}qOxMak?we;FbVWSzf!P*PYet|2CCwWAOY#lpz1jXz?KJY*v* ziUn91Zehq-g^82=;;F}TZ_7Cc(8_Q|1DpUP^wUvwO)DM@RmGRhiG1|iI~cM7T0<-h zi*ee$@;7j%j-!=J7d&$D%Om!?bm#~ig|9~T@-#CmH1cK`=81fp&LajNg*AXt>b1TZ z62j~hC}s3>ppl)N72)Kx);CUrJ(M zXtSA+F_hL6)3K?B(o#UWuNQtO2~s?iPAhQQHP=~oDCs2h9!kogn=#i}ekjTG5L^uF z{yGj;CNt+9i-Eb;vv4e+PeM`mEhg~Gyp*v}Zh-B+lS)?Yxa!1-S~#|~CoCe67*o7^ z5>g7yWFZGPl+H;=HCRJQ5u{ekhWAjK45V^qGQDdt&o9EEOjx#n6(YFSfR+(#{^dohHqR3_wpvgn_B?5oiG^PPr&9gq=V)jDTHA32qBzLT6*{CJ}e)q zItC8g=&m$IG<3AN*$G601_z;YHyRS)#x!hzW{<3`S_n2lsGy3tVeYB|ZU97gebGg& z9vK6rcYiKOpuGEaIypf1jeS^{9zirp2o2|(!2wvlg$9<%0~%PMpH2jzj2Vc+Q6FF0 zRd56Hzz8heKt^}N2<(_-lH}ptbf+QsCcFYF;M5F7G4+~VkqOQS2v)V~vZ^y$DdvOq+%w%24R0HUvhHzEG4Za?~Zj zjmJqtADnf~Ndm(bL?ryiJ~~F{bu=6dH->;}mTE;$UvR??zd4JHgg5!*KM;U8ZV!pS zE($o)4MGOz_4G9y*Sjx`B`-;Dvvvz2lk_Crjlhx2FMrILU_}EaY1o{Q6MHeyO^VSY z8b1TqfQ>pqa}(THYfMt&hB*z5^q}+3kbu0=r?3$rz5BWw*TbGBJq8OoGCd+q(tSOB zeHxjrMj02?CnTMB29gGrS)-&rLec}P=izd623!*81%||@1C;=o1$;8|#vUBG#nMaJ z5Q^&SA?W1?r6B)rkI1_wHQnQGvm z`=Y$jdjuT%X~W>8E9jV{wMNpsG3K4Yrmvt`y~b$+>o}N@5RwFSgW!xFnG@2;O~VbH zpi7b5KJP|H<6X<<%BoorkgWJjIk4XZK7UNe7W{<{2 zvrz!uht5c-ly23&?5X2IRtvI>fe~tSv}j`>=dtq7X#@E~NehOnl9W0Q6g5;? z58WD>BtX7^#wsvHe1RTRU2zAcgky?HAQg5$2q9Ra|EEPOqEJDGg95t*Bbs8j3<%B# z5fyIJ*H#z(Hc3#r0s@dC$?j)%@MTMbIO@40HbxXvfq{Z!RPf6KJ36CoJ$jX;z>sM~ zj^BlblAsZSZbxU-t-o58D!h{13qNZOGSQKe4>XQ?%}}o&`S2QDa70IX?7L4g5`1{g zE?{^_f+?FM-+R&&g|-kMUZx8cPqaSpoNqpD^5JE=07Ihc%cMV~?K%4@)r>hGUPl** z+6EMjB5-nY{tpzx3@A_y$T&I%SVtMr2|PG`{5LlPBCIQ5M1mwkoCTA=uqY_|_ak3} zp>c&5!H`&`QHQMPWY2i{@2>2%GlVc3^M^(!qU@R2y!4-z0#mUgyru_C^1b-xy-kuc zdmZ8B2bgHx_-0?inx6#W3@`5iFdYB$qmu0^VWq_pUfKmp)azq^J5^uhDAE~TrVEq- z$;q}OC(X8+FfIQtdtU+{MUg#RJv~WJAY_Kigex;33MxVrgbH(Sg(X2* z#nqK~tOQ6hod5!&5|0&65)@bVzq&+a3QAN|KqM0m0XY)x$u<4G>ie9T0Lu4$ zzpWpnU%jhdRlRyueN2_Ac_H>_J6PZ&`;UB>Iws24;bvJ8htUNdqDl@CCj0;6kc8GY zuN}ELiGr`P1&?L+BhL$zkj<;MU{YaMK_YbXsw{{b+0^xI1`CC3UX2B@ce``3M9AjV zSU~t+pk805svH(KG_S@2L_AA=1ic}}L|($syqfWYPkhQnsa_UT13N*(9}GYm3;*L; z5dbWJ@;O2xRwf%xJkmfl{Th zi(eClNK6PC?NNx{icfQMADgNm6o_Y&HegrK3IrAy7YHI59)#fsc1ohNO1L_494lA@ zn?k~n2mg_NmVjYZ5`NEbc)#T3N|7T$#TOQYXU~2sCd&=-VdLszS0p>q3z}QhR2pN< z+fBnFWg^+HU*gqW@$>gmgd%7+c<~d^3Cme!zkV5|0{AxYR+PYsgI1xi;(%ERFju$4d$lT5ZQ$j~Syw`)magO!a}=Be!3Z{ZuSA2Zx(&GC*(ZFnpUfGc zqKhDduUFJ1q9Xp%nA5MOr(K#~)D$aNczI#+GEf#UX#=hiq3*QI5f%vBRG`7M4QA1Z zND8Z7V2HqbEj2YFZw+lpRS_%$3eRKGRI)7J2Du3JkTF(3*r00KRJ=B+4WJwtAE8p% zF7f)U#N{^NxJxYE!)3(>3w!wh5Hp_0M=+FwEX$G2F)ITe!C1^cz^!frc>m5dHUu&X zi_Fwqg)WiQ+R2|6E}Y%18!DAB5nP4NjG#+Yx`=964AUw^I5138fw*=_AnGIbAr!-e zQ!dd}u~$OL*HG>P{qr?}7BVt(C_o*1^?YI6h2$E)CWzMI%Zc23^%lUg$QdAOPkbE} z6|O#Ex+LuU>N%8`zb1o&=quvY6LsE(a6@hYS$wShj{VUsEg?ew zNKKEmMN`MgA|?tykfS41{tK|h*=(rSaeX$c#fQqlk|h-R@@UX6C!t^FMlOw)aFKJvV-&o=37>3Hr=UITYeA!nlnJ%ZbdNCqy6@5lc(@#p8)crHaY)QHY8zgV;|`L3{jC{Sv9^IG$oV z=CHD9*9M@1Q9r}4h@zNlQ@m6v*%V71B$^B6U28jb-f*TSHejb{pC5T7>#VQV4}Yy? zBB9t`e4$%l;#>7F_q%JD2qYqkNwS`nPK`cz_0kQ$_|B%r(Y z7xO~u2B+*(N!D{p9j&;8osmJM9G+A1cLnvh*L2mAm0Uc>&flk1X z9k6caNM8`^)oUQE+(V-ywT5eo>WSs`$;)fiV~zPcRxKP&+h`!@EdW#mDfOq+quU42 zW}G$y(sZ>waLs3Q20w{uE$|xp3;!(MSW^1=C(;II^e}W|b|l(KxG3mQTIX&kx$N zV>Hck%Nn$@IA^2bk0J3|lXn!g8)@o`27W!lM)Q~fD;ADr_-8;NEYQoU@@c0B@4B*> zGHA=zL^2Uu<5lU69P|eJ|Lgh8LSac}^nZj*M}S&?+UY^to*T=kQEJ#sC#I|P4*sIj z9-TykuUDFk7PT@JtkDa;lwRb5HZYDr5HTBrDop@-1N}dF;XX!-lyl}zzvLzHj}g*0cOl5I2&z(M8Awyvyy&299{;@E-Ub|e36U71aP-hoBeX4pzn^zA6Pg96))w^tZ~;BM108}xP;RcF+n9<# z{|}%4T}E_x4Q56TJi^-YPh6~c{&$QSS%WjP;MQ+g;2FFp{nTga1`Apcg9fad+jZ2< zw5B{{m@P#Mw0hyoAT)PH(1#5XlCZp$a=PnFF9yf|UV{tu+61;rQ1RlWhDVjR0LlNX zL^s|j6F~yM5;3>Rg}?D@l6i4-lwKL&wzMRe{JIPzOcL?+*C;;Tguw5L?qf_SEk=>melEtrkegVwDC~;fgkaCEmL?7a z7!x6mIQ}!Y@HYy@fvE8B34Ow@z!7cg0;;*=$C!n`VTD4}E^1|!Vvd_LY^;NumsU8$ zy^f$BzGpvYU=Ldla_e3h>Lpad)}VTH!ta@Qe`}n{G^C*6Q+$5ds^^zWPkzEgkEBiY ztF1C+Fhk-Y1vj{w^u!b1ty}#bmmaTSy28w?SxkC5vi0zo>M^kjx(~Cuts57&yFT;{ z9*oE*87yd))Dkdmz!!u7fGe)oB+H9oN!h$<@ZhXN4`jL=mJ)&}cnS7tRl=G=Cw?$9 zz>}Vc5|@+`wst)*dMPsV zE*F2oN{A3{ty&w9-Q&guzufNFDZQwWn(vlBB)Y_)_DxZVL<_Ztn}Q3NN&bcw&5c+TscQ*34Z99%Zwe37Vle8hEVkT*I#OEuuU8ykQ8Z@;6h0~P;3-&c5imm zf}ISplu=Pd@aMA?W1{KoY#Q=J@=ofZH#w*XJm9$TwT!+4N_<(xZs+*u^h0NQuGNGA z`#W{*hq^q+-g5qMguBql#oF^9-?wxH(dTVHkuB;**ra|ao57pR&OZQY z+`fuiYs8TU%GC}E4@@z}t(Y~IVihn22_8Uu{Wz~mw6Ke`%z;VJ#@hYi1%8-epw)e> zkdyz0*#uDL69H8NDHx8G+4xdm@ERsp-38@u^9Bs0A^q9=~vZegJsK=m3jbLED*Kh9$5;~_- zOXy(eTvb6;g2SEjo5=XX%EI6Gw#55FD2C5Hn9zt%Ik*ynqDh*2Y4bIPi z8~RnO7NAmneK0iZN<%3Kd$M-3B7TVRrAZBBQM7@X346qFz3hkF4s_4KVnYKB+O_0_v=R)p)iY z4I&XH8lFe#SdG$_7KqYnx=|XdS98e4MO3!Ttp_yHf`0R|1woAkG!4&oS&c@MQ`t0h zRUI@X3rsXm)(Mzr&7{FNm40(MfrE*pzbqA4mMutV+Q_H`q@mGhH9_CvAvB0(t2y-2 zYRx)0jO7ZqT0cC|6bKq=fg^&l1t}9>fgue_tdeMfC~eYT*!O5FX^aOAj8RUI5K0!y zFjrz~v0B<(OvLhbo9kEC+MGi`8`EjAAfq%UXf&wG z7KP+qHkX#T_^YWj$3j%0aWPg&w7^takkZ&_t|o-Exqu2w`&hkwG~X}^4>i!Sl7+07 zeY}Itaru(97SfQ_Bwy9>^>b*9@}1w6KbygjkRq}>=93}M450)8vY}}NEefl9Te84V zHkA3rsVCZd*%|YIj|$(qHy@?lg^3qmF0IUA^dkpBa>ICx_&I{^7p#B3Za%VG@m>Mm#U?g3WSaH0X}LU7yhSnVO4a&WC$E$O=je#XPwBf7 z-;in7*E%KA2}oK`)-i#9cyuUZR;j*`X7aSkgejEMiCt+n2vwT&)mM>1JkC>b37%|G zie_YM2ce=ux!Y86Fg)Glah}=m2ywR=x+_H^jGl~35pT9lo&gDB>LF&l%{5$vA!gQGv0wr6N=(ZYZX5e%<3M&+%EaXAVzWTIuyD1_0!N;Zd@% zECr!caS(TdH3fD3xKgh_0l%QcH&Ci{$KE3^)2Le^=upjb*y+6fX)D;<5jyGYn>Bkw zL8a60$lA8^hF5;OuZtv4jOs+fE4ELPpIE>mN6PI5{Q{UoEE!#3!(T2D?taA&oR7vj z82mJ%7#}r4OVZ+%)*&FP!vI3TK=&#cpcwB`3~MkG9)uHUh5H~LRQ`ZITFC)qkQOa4 z;DME2w`wdj;o_N?NbEMVkGIW4tGd>_1tvUVy?_GM24j_^mKLx^jL2=G*T8GSqnB1% z$y{2YD^_YDfx}!{%W0hpB^}NJzISMiZe6`YX)u;-+1r{Tgp4FDm!KM7D!UkRn|ZopB~CI9>5w5QW8tZvSlW{#WFz{g_4aD zT6=|Hs2eRJT~)TVhO1+_PQOXy(sA%6s>|l-AO}fhH_vAJhZcKVMHJ*LrPUUwOXf9T ziz`?z10zSyEFYG(iBZXl(!QegMS0D$^XFi7 zQk$#UZ0lb*zWr?T@WSS_O@DoJ-By+-(%>tbXFT_zF1NKn!wk;e z3+h-j!~uttP_PVm1)uc9V1hC6ix67Dn|*|)P}Ptf*5M)wv_uS(4ea1g;QG)^AnTR=V! zi*x+#Km^I<_G`F29$5Y`ie?Q_&Jaz=;ea)$I=S4ixZ@M1swsL{=oAHO-YE(lAdirP zrVOaxsOvigPSk*9t|o~U6k3w1Q)#m76j5HPpsrI0_+&{_0-sSdLsqIF4Ywv@a{3`c zSau3e#hi-QNrJ)^#vP-@=yT?-0wcv;VKuC<3OqD)6%;g!onkgLl%iD?RFL{kQ7AG5 zodjM+$a(NOoHQdvH#B$#xM35$PNIrlXXq4pXozx3r$`9N>nuA(V$^j?5Z^TNp}@3L zG|sQ@6io$X{icAfQ=oznMM4fM$@%r2B2-YKnid8~?7$wFCDh5Zf@%g42D9{G1TT-o zq!}6#QoQ1kB0E3e;5%EV6hGp_^9CQ0cOL9=zI7@S|zDP9Ji6qtkp zlOplN$!PFlypCw@_8lUZFJg0$NvCkySwp4RmiMQ7umAk#9&jFK-iU_nDhhxeRi|hv zLJl8v2L~=zuxcW$%6JIPY^6{rzcQ9}2PbJyUQ$HVkbEi>k2F7*O5}$F+jz)$aFRq( zD;8k_AHHUGRG0}bMxCbx0^Yw;PV!-8yfzsv0>lP>;&F3DDvJ83J@HsJ6*WC(e=*nv z!$V0XM};XwsKTWiMjjIo1&A-uL+cJJ!$-OC6oIXfD@u&%pDZdN;qbMP&5Iq&%S3;B z(|4+3_2*JaB&-c!8jpboa~>OZRw5uciOnZ2O%$96xfwfuw3f5z{=zPdOqQEh#;5_& zLt@)?dTyKt*##dEnzu=s($)L`dwiHeRDJQ8d%kpXncTb@hm`~+4My%Lj<2>@2`Fv} zts{q}-FqgD`N3?>E2v z4%eT(eyYNuwm`vZ2H;cP`?pU}Xu7l0L;^Ig=$3$RZXEyNm~5pgQU|M)z(gVpKyXWp zc67O~%lt{DGKav<85tG<2#zq-NT~_YGdjLu|Cx~xN(U=$ zscBv~N|ajKXX1OVa!EvEfq>Lj4!pOi%_D8s_RHp}S_=ThyQQXb`~cjNu>Z5&fFjx_?n+TZ}nzxbL;jH zO|@af<-uCy1l3X{y93=e^f9`eP=#%pA+_rXU9~uF>B`#+Q~{GWQKb|@oYSw$;I1VO!$?dJ6<0{b`LVAH?BN zV}T$c#5}1;<+Mf~qql%Z>ch6ac>}UC0EDohi+QGABT3<~0E-SM$4D4_HS-#ojxEqt zj-S+I28Rm{8<2AVXe|&ZA)D8Q)Gj8fiu}Atscdb`(>Jfi0!buv^J+R+6wy{LbO(!J zJ8bhx)=1N;F6Kqh!C{s?AuYhA8F=z^+HNa*pSu>ha9IuoUNS&*dR zKfkRf1|VRtFc;>D09?kxe||o|fPrj}1@2)i5HPue!c7B?$kQkTK8-SAmXXWi0aaO- zLAEQV2qyD?u_?lFq?J;=ghIh~NE(|A6>JgNaQ-T89;Nl-`TAR(pc=^mvl?aCcx~(1 z)n(zT>ByCGPYKNflU7X3ER@EvSQpeCCX*0}2|+{g$)nSgLorCSPzRX7YQgFoIuM7C zrMwky(XNC#Ntfj{+!0XL-=jbq1iAdCrXP#_dss{Tn1IdEuD>+w@a=AvA3*05~bVwd`X1J)2SMch=f7LsdTMG)<184(_c zw5b$LPB?^;t1q|NUW~Y$NW51mK!&IuiCu4;F~iQtm{353-fN;Fnnftn1m7}Lm=Wye zc9pMmwY#rnO7+|jP&n-pU&8@y&@_KOGr7j^(coK_kphaj6z`T^uA$f!GtrzQBCIN) znD7g3$#E6jKEJEeWp$1)|5yf92A3+0;GCMtLwl_%d(^~|(5MKSN>RXZf1&9SISL{| zD3Yd0IzOEE%%UYr3NNd3OieN!!6f%4La(>1GN#%E zq(xArX0>m7Fg4W`Lb3%UtDx%SE7NCw!g>S+j$!mqSKa$=(RjS`&hWq67eR@R!qXCL_# zTiz{-ao0t%UWcco5vPb>ga5Wa%6t7EOd8AI#73+?B1Wrs(-y)hG}z#`ApRO?FY8_l z_f$i#r48_8t0CK6`%cxbQDKWRytHp799eaL_;!$%wHAu50G%>>$dID)Cnhs+E1Ml! zkNAu{G?SWYxd>1*0o*6S;r-tZv0YhsYO-2KyW%X<5)GczOy{LDds-rLm3iHMS8Ar4 zyS6s4XHpsW;=<)lE+9Mwy~>r0KwbN>toz14tf&R2OIR6_AB+r1zfz(ju{XuloeZUh z3fK*UgRvBV@7XMA76eD42fn;(=~C8fZ371{)?1XErQ=Q=>*tTLQ$c$R7ZfcjClqtc zPxC7#bgfvm2$ly^LNWgGv}hb7UiszAu`Gc))*)`cv%cviYmTeAgfcCK%FG(qXTsML zhXb{}DaTb^N}84sHJhL=q|cr`ljP7yQ;uH;31_xakCMTlc$Ud;K@|g3))He08 zxE-v{)YFMv^{g#_nb%+0@5l~_ay^)SIyr@ckAP7V&W{Wan!G4!+c=d#`m974$ZBCE6E*UWM#e>cMr9V4~D^9vYvt_`vG znHijm_4zaC@cNzp@H{Q%_(yi@h7U7SPJ>{p>=sMQ0{FnwM~d?+SQ!%$5c`z{TYnL+ zdl5*`0zNgCHD%yG4FHN0$qXW&aEa37t~^7CDv~Tz$+zxd87vgE5ZwlzH)0W@&LGHJ zKxT?lApW*Mn}n71Zg}^q#ml+NS$a$DgIHOO7NDgpD*?JIt;(I9SwdH4bD}tk*2j5|Whw-Idnmwb@vf;QL>T1%DVP9ny=2 zYiQDdAlaP~Uc(>_I5=ZgW{?>9tIVLmhRYCE&Lp>WI0n+92)TtV=;r7)v~fW2oUAk; z+(tp(L~tmfNF#$_qRe1aU{wdKk{_-{5#@nlIEHWyXrkL1)rK~SC_Fnv6d`FD95HtE zh}f9?8?fnO74=(i>GsPR{GeXw#%|h5mLK&W58GO+OWAEq>Otvr?BcFFN-26o!a$& zc|O~@x69>V1Ql)dnpAl&2VeV{h8-0X2GjGn3+L)8Z$pfZV3O@&f-^@EXAW}}710aE znv=Cy{G&xiLQw&dm0k(CxA^t*# z4$&JvMmzIhHBxtIR6>UQW(|v+v2BZCSp-saNYq5r$+b@Pz1H}|URV)scr+2;C6NxLDYA{8n=ey6H@P9R#VQ5o2 zkHkaHE|Yz%3|axixf4!b;=I-U{d1E9`yi%XRWj(d3ks7>CF{KfArU6*vL7OB3NySppO|YBN45D66EAq3DjPR z=72MCmHipJXMjgb(d6WU6s)R7jDnRQXKnq()plF=ckaHt#-)rp;ZPq2k<~j61gFJy zcugwk>Re&kJ$K^R)zPKWf*+YH;~nBJ%z#%tg_s*mkIT4X1JpG#Q5Hd1Gm(mQ9NYQ# zY=yva^D4Hdrlqafak5+Dlb3Q8O5{-V=0iL7!b~nD_%vm7vfXxg^c~l4{QOTpo$L1i z!;|oVlSLmD_`ZFmzsmz;GV=g9V>y$UF~nxXVCY(rYp~?B@`=1-ch<`aK>Esf^O9&jhhWZ(;nK=qb)OFh6met>=S?PTqF)Tb+v{YV91 z5+*n|cF#%hKPR>@L721%mg3^slSJ)pT&LJ z;N_r#k51p?FLvaNB&$zwC(m*{7iZ95es$BSvY2@RX+^#X^#)O|x1=$z&i3MRTt-z< z)Gj}Fq?T*U_nyAs{^wQ?Eia25(J1vsn^3{dUHg4SS2>^zp>Y2RuDBHqWe~`nlf~_N z%@DgfPt5wdW)C(Oj$OkGNQC}4)$6>Iom``u!0)<^tT`EM%4dL_m|jH^$CSyvzL7gO zT?&6$i%0EA&IT@Z>Y88eXLDTh?-P+%`Rdvh`B(^pGEMCKl6)?#o`;>Vv+Yl!UD7cRzFI@`S2tqjF zj4YzlC)GmPVv37FAkol7rv}|z7QVW}PRE)%n@uav9EsKbEH#R4^2`B^hM{E6lW>(f>uKGk%?iecn-%Z~g z@#W<+x30{3cxzH7-5YqY1ozRu&-S<#GLYnYkcUnZ=K`EF#^iWaV(I2!G>JMs&#)j; znr+L-arI81d@CA{)}mu)xP3>1yJOjR@YfVlA}(G$^TQz_*R6a}umsnG{NZ@;L7Idx zDzPb7aBB3?m-!6CwBAPFUs>hKOmZ$-$F17c{NC2!xl6v~Z)6M~TB`6v*Mf%-e!qaa zZmoBhZv}3v(S2-l?y;*GHE@t)em?7cDz|#WBrjL-eIR$VeeCV_i7B=l;tNd;L$D<& zUii`)HCH(kZp@k%=VLCzD(qMuQ*~Cz7E922P0Hg5J>`n>Zoew@MYK%BN9QO$8e!pL zSc;Gv*us`@HI{K&|47n9Z&;r+dqTrNlwTA~#i3jth{RA3cTvS9`y4b)cGE^KU?Z9hW$b6P{nBU6wiTAE{tY}{DbLI26=o~ ztaUKu55raKRl~_>?`$T&kWCNL1g-}ZFp=D<=D;vMEY`3yx6Fj=2N~>*jEt0kWMUue zv%L__NMT8zA2cB<6XlF2w+jrSKYEb6&e+3}4qQJ8MkXHo1o6o>bmKxMw}B>0c}8kH zdg?xHJ#Z1S5*>i+r$cG7Q^@sk#6>^4t`(HQ;xe2bRNR|7+>`uSWMq(!_4+kZs*HKz z8V~l94;S;lC~u8&8yH+hTq&w6qvi{0z(+ZLlX$lu_lG{wA%5&D@|IE04IqD5i=dlP zD`v)8mZbSI{AEdLC{!G0RtP_%7sc9YL(GN z<;h!^q#DD_cd6$N*TUroL6UHdM~rYsLy45#dR$kW#mXWF6jOVFc;1qmeyO9~}#0znjA z<%#xI@&5iOxq1q*7KVZEe1rGnfdApS`cYu+qUK>qDGiStLu7y;<2^l)xuFN!_KeKR z4?#H>F@Vi7L(ME1mi6Fdzy>}!e`u50Hg~6e&PWMkeZ^LJc(Ma8tO=_L5?GXG$3ezjed*-siEVN> zBUkv}VIeHx2{U=&B82W>0EC3Y67~Cc8@QR#A50KXA^+mwZasg5iVS39>pax1mF*5C z$e5MHyC?@%nJ$z=O|L=c0mD3Je@PrK-%5dtA17ge;wNOho}kfUPjbOzn>R`j4RKcLY7*R*&B!ShMr zYk$L?XIYf|`Ry;aUshNESU&Z?C+T-I582kgJmr~-Tc@wtzHV+))H9ceW>z@w4)!Ky zfy2Ib)?600EdA!skG?r0X71pjO<0V2I{9;aCDOl{^6|@S?xX?k|KfAg(l%~f5Ay3$ zw+Pb6H@l%Jibd3g$M3s!-<%oE$InA7#6IAUF^4Rd!IW^q%!~!&E-QGl)8ELEC$8&Up)qEzXNvPTIm|Qf<|eZ zit;cQ)L<@LP*idw1wF#9+A)76&O?2MG;L;4%C-@?EOMM%HI_kMV#WfFxc5O4G^UZK zUT1&}E!u-n%DfS|m0qHtz{aA~5yKj=j5^^H6Mbw(^TvF9a$(B0xw#Ba zj>=uT4J@XcMr>YFl$JgeAG?!2w6OVqe-!8lwe3A;CX_!bZwvBn89Zbp=1|KE^_ktk zaOp)E8(7e4%WK}0J4TYXr_62ydH%#iq4khhTVNIK7iyorF>TnK6b4b}8^9(hrSUx| zw7+0&+9oeCY!(X)M7d3<&%6<<;m8Q?!nkMa80gV&C)1KRN}FA@9&qJ!uz6F|;91eE z+Iq1*Q2q=rcz6GHa|6`h7-VZwtGx~HLzV^0doFKuIu|6_BF?V9XT{$;1SB_16(XkBiAwP zEtYLm(N&;j*(x4u5TrSED*~swY(db%0_ZR7E2>sBmBu}VGh%)@ zK^dk=Cyn(R?3V4r8Z;K@q%qwnZR{1y2gy_#8iMAj!XFbJ(E<~SP`1iK6CSJ>%iyD!g>er zs#MW$>We~_2)?rI2^ED034g*n;vJOjh0X(~VB`m0A2WSqe)lu|29N4W1!3;^_NVsg zg&rVLWkX)cyklSM08G}zTlbd&3PSSvm|Oge0FHBKKZS^s&XGI*_(lbNe=xDGsv-{; zOEo^5I|Jq+c(nM3AnMU$dn+0p& z>`#tMSU&v&OrvJy6=lN}&LnMrvCGNYSO%qR+F0(v7!KZGOUqV42hm_(a|iyo55Fp) z{2kg~1W{DLuv{sA^nuX0&kl475)2wbZd~F~4jIH;M3x6?n~aFE)c9JA;M2Gr`}w=4 znR7Nav%-zaz1#M5!!)qtfR_WMI6*fsFe@(U%i^0EVr~EHe}cS_RxJ7;1w*+}bJA!5 z_4`fVxe>}Ey#%p@6;SxEk3qDoqw0o)Q;bC0HoX*j?IM*|G$##tc=`Wn$RF3OH zmZjcwP(s>%y!cNt5EUeLdtR+&@a?Y)O<#4aZCz`GW<(9!kS#Ha*NS%%U-0iiq|X;dZv-#vgnG; zc}%u%nb3s70CKJ#uqqGV4vpz$E5APN>lbcw(pVh77x!|03?i~F=Ht`@o$BXQ&}?;P z1v9Vo(4?4pOzFQM-$pwF)633{#aT@$<5GCNEK2KD@%AW;jN2Ma7gS6*~Vxo-*;`fv^?#h6UTgl=f zCgh9{ZI$W^4Wvk!L&k%WSjYrUr{geIA09dJH{s@r zXs=&K5ItFkwa8-k@UiemU5Zup;n|GcpvSB$`4z5#(+Z*MN+`u)@MN8=&iT!{vQY>@ zIzdfV<^18gQX^-YF*=>tm8w!q1$iVMVp3NUxAgq&X9BD0r8yqjz3wSrs`D3(O8zB~ zfcgEl@A!F6Lx)Mr#?h4pIMPW8FGJ3vDDS3xMnn-`_xRro413@|@yz}cnW|T8@VP^) z{WawSFdZlIWc_0&5wXAP$EkDj2af8#8uz_F0fH4Vbm!OakZHj}dmBe}Ht6PB2O7ZRY?JO;5_i*7 zM7(rgUF(<5d6|?$hu>m)9UHW-W8Lw#qK;`5RXq{9;Ygc{Gor{13`%^_QIU(a2cqry zuoOvo>z((HIk(^X@!)}b@me*p(#>Qpeg%u1ZX@d0HLVl^GZ{SFN<=JrpG`ub;_e=WM{rD_Ng5g4xcL)tUDgeRvT>@O_S#rP?CaY0Y@u)4 z+KELKP*=3@wFzqlGJNcIPc2@B)x`?4JyF0_^~1%M_m2^$e?nNN?Mdu#2IkozZ`)`| zR(mB*I~@Urffiem$OeE$l+nHYn5VnX>36P%^%2(LW61pS!X%gw6W?x!shTwJ^)4rH zB(@bKR>60GP2|qrT~0vWdvHlQdkk2KJ?;+pCceT=Gl;6X1=mNH;%muf25>fBgxs9e z{%mw3=8=`|@7DedVhU(~HSmqwajbnq6sAM#|7lV2@5_b>3pczxhPq%oR;KnkC~$~M zclK^`!O`U78aCU?fe?Rw-IJwbFh{FP%bGHMWrOx@uRYKOBM9&0@-OSQgO220hcN;h zCa#ZeIfSR3fz;vkv==IO^qzkE&$fCtkTPfbPu{xgVwj32u0MhWhOf`BSpi+G*ugD5 zu%xaE+#&s@8qAS8bf5i`Q{dY0QN*`#7cR_C+3ZH^Hr?5)_q%g2o$+l+O)JDhPu5{e zpSjry>m%eonf49lL8$YxTPN2S_#ax?_Cg>!ckZQ;>#uL$6yRte@zbuCUI%Z4sDLAD zZMWli+D#jpFv5?wHmu0$zZYx3DelF@!dpszTi4prKsAi6pVCB6D+9C`s|SvH6LcVF zeV5*L>FXflyo2@V;w)Zcz}iQf1f`6yY}b;^a%Qnw`0Ch-r;@VO$hSke>VUeN7iFvbWH}Vg? zuoAf|4YYb}=j#WV{HTjF%yGD2m(E>=9XriG|17YuIgsDr?!rR72iSKhFqr z*MY9>&M=Y`ynB*A{=+!y?rEI;@tAdo#uaCe+uqOr_MGTOQAdPT;$!ueQ5s_u50@jL8KdmsGEuzS^|(-E+XacS zT2+czePdUe$61(&q=NL=ngJWv13pdtBqahaknJSB3MBBbw#MF zU)!G1QH@!n(}~rGfsCT|^@*+zp#w)zv#yk;UFukTS$3x*tP9!@d$#Hu|;iurnu9Tma;RoZqHsz1dl^hX>sUYspPwKZF zNY<9(kqrx1yCTnH;1^8i<7%&)TTiBXc#^fHv|%xfo1NnVtv8Q4+o9dDqGpEIU}Y&! z6#qyI75-j;V-_*X`&D&0AbmJ$oKCRQb@}Z1E#H1LTw zw-h|NF>%7bYn_b1Cj7+#wkc}8ed!WOdE#9hacNR}Q){(&nEiOhk`&g{T zCg{Sa?ZNZ^#Oa9(1Ozy?D$$H8BZ2tjFXzVJRi-Sxgi#iU8H(Ke>9qJTHX_jFx{|!d zvN2wB)}8+3FQAikaQ-z^o_OM&fgP7)iIAu5VFuV&j^C0bX|((0#~J8n1V;%fNq-z1 zrr>ky;26;8QNdPoS(tR4evB29%Y&4=CXB;tGKX6ThdZkuc}>SKNL zG9>3$1R#?5l$?0ppQdth2TN?4=9OGbQ#q-FB?(RQ@-8M(ap`8CK z4>NRxK8n-iv`RzsYL4QhwH*wj9y+*1&C6(AOlZ51&8zKTu~Q=-KcYaGvsN`0NFtGs zA4x(N(n5|O9N!>mJ)TmJAE9{#Hl7s+NSHpnjDpj9yom)ikP!}Aq(QC&mK6Xy3lSey zMo0wkd&GFa@xKmMyd+qR@O8w&dR0;RTKV#DPl8f>rYpz zddjMT8k3;}ToG9fJQaz_L+M@?kfw<-#mY$hT-Z9@hUOVs4xc2vH_)es)s!kBZpU7~G+$j#Bz0x~UVgI}ymLbZWpFL*DctJ5Wt zPQx{<_hJ+L1~Cz#?Gh_46q+rJyMWeEZ6FnT)x1|yC|&tp5HnX0P$F-Guw7yq(pu04 zk((p6gLI9&4O(E1q+GgQej7*{#OqZX!KC%pa4iTZI_u1WD)F423CCVY7W14iRfh^F z#x`(rLnQ4_x~GE4g#5%)$JZ>AfU(q_NSTk#T3KROpp-iG{o11JTH=dCx3c zGHG&>RF2j43Pvz$bD&l9n1TW)Yo%PmM;=fl3Hj=6nBTkC>-7zW@11213AoHI;l-b7 zZru<97F;tqZyQ zrjpzx>U1kdNiIKXtU1{Ba+PZ2Xb)r!Z{awuF}-|dPR`>iHO!Zpk^0h1 z{Qe&AJhSFW~9dLqrEj}&W^oORhII2J>~Oq-)Fiyr&=;y zDM4H${MqB*OnqNfcBSe}z!6+?UPZOYqA4}GsdQ5ezLMdz}z9tgW~45Q`0bTd^yz-s2K3KY zQSarba{4pV6R1;lIRrvQ%;jjfIx;a^bNapg!b6u<5FFF_8rp$tLpHfuFF;MrZ!kVT zx`qSU_V|)EC#SsDe4IhAY(T(& zcwl~Q(y|BU&HC>B^6CD5M;lc^y{Om&)ibL@HS|nTDzZSpsYZVrv9I_AGJBVQBGdRj~KensFHTE`Qcr#Xu}N-1|B8QhD;ZV-X$jC>SuqbE-H|#HpG|_9 zUc$2ypu5sio|ORImDc4W$hM&o#hUX0Nl4)s2<_A+Ya=sd&!(6j6Kw@=PK^yvPk`+QPsjGic>P<@iS69l(T3asamp4WsUm@L?Gw#;{j{g;tQ- zM#Bva!nY-^q!|Nl6p6v4@p!)k-7zsy=;4?0TX;7T``MGvIo{RSqs|ZwjYW+a6}#{$ z+wM9Ct){`~h0nl4B=Y{k8M7d?5l2qW>Na6JY=D$Aw#RLy&-N-R$|wZ6Xgf3Vc6Zjo zDia-4lZ~&l2? z^8Rglfh+e;85rf%5_frG0b?}d<$u26ad~y}6 z^N1gR3iRmY7(F0F!u0^Y-=rD@MwlP2wAc&S|z|8u+!->LUZa1~FnWt2WM{3RP9_#h~6sb+-s zlunw2?e=l(xl5^?+v|I(m2k5{H6goX3R=&vtYA{ponS}6Z3q+ zwAyoiYZfgfC$nF#2Us3AV@P6~XGdzu>Rd{1=bQhjy>woU3k6f7W}W)<_cYJaA?}$D zFIYD=`PGjX3r>_`R*HqtiGh%b+Vm~A=mUWzdUytDqb=FVQ)j6a+v-HRb`u;^q093( z6Des|9aSl~55puT3|%DK!?O8UJxc_Y*Na%?X)6Mcw+^Nu#UdkG2GhvIuPK*lJOPW` zX14}c$5N{zRoV&bpM0vgE(?Uo#d_Fdjb_30vPv=vnFdoM+y9*y_0I`4W8HGc$cXFV z2)7C+qvs@#_GHhGP@`;~_`}%fIR|?rXoa25(RU1LYOd#Ytns@NhS}e-+ZmPi`)+Ud zc<$_t_eYbnxBb|>WZs%Q?b8NkY+qPg#eKfa9k12fW>Wl9)BnV4a+W;QJftK#1|<#m zxIE;>$zyLVPQ9_)l11zD%Nv@Q)Sm4gXV$JSUHR3Mmy|OFRzBu|HU6A~Kb_P@=D11Q zp4p-V8ZUS8?Cc*tJ4jT07k7l+d+Eu4`c>A$P35TjTTxW}v?>WZF~wWw_)Fa%emn}d zcedM#|Gwxwo)kB!V&sW#6#9R=W*`=1({@TI3ukR_b6{10*Bxji#%|9n%~r|q zAhZK4(d&mm$C8Ubsre|GR&x+Z*E_QoRF{+ia>w4B-ERTan_hbUjIZRA1J$g3_|TYW zlrz$mjm2QSD!UxqK(r!~>ekLG$9g}i`G{sfg`C_yJ1=^2u^*Ii{vVI#7C#7X#3%CH zY3DRgdVWc(*g3ridR+Zjksx2PNA}$nU*XLLy)S=MQ>oR9x^-;5=tDnzRgqfKnnFFh zN_u&5imRY6oaL%_`LpNGCTr|b{KHmAsS*I$3Bi8=u`22&%k z2Enn%jK7rTaj>VM_wgzo+5G%&z0?@P50{eOzqtNMdMXTDD@?d#Yj(=HFX$}p;=%%)w zs;p^?PdzbVcg^(1hE@|!KmXaL8@sBWUyyIEspG}W|NL~aY3f~q4YlVQUTxkoyM6?` zbyd>ApBq-RdH0>y!LO{axD6PiW_K zj~~>~Es7*2Hr(I6m{Cg!a9rXF)`%4_ZDQ=h6^8*Is694tP1=!@jVDf{y5)|!2RpUX zhF2>^Ey8=*?k(cKTFm%PLQ1p*T-~d>2ouecdzz3;P-1*Aq3Cv;uzDn_NJd;RqsSN` zuV04?U$9QVBI~ih{}-*a0DPlPD>(xc=CP)MKSBh2CLhXoeyJOUa5o7NTHhiz_;nB} z{5Xj3k{$154|}OM%xtzpzpej!MYE#EBY}2%wBzB&Z3&$vgw<7+kQq8HVp+(z>tYYU zuG4T(N?+UX-Zq82-&@T;qlVu#H*_j{^?FY%GiNT4yD%g?7{pT zOo-tk8a4GnliyoR04;uI#jX|GG1LBMqI-A3C`f7PS(h5!RUWBh$KKS`1g<9LR8fg9lrxj;AK!J);_5W`GpLWF$~=!z~+Bb@=6ZYu=v(Q(AB4l?dP-jxnf#H zO$qCwS&9(^d!s`qQJj$JMiVuel(+*QoU;-{=!=?y-LamcrKMmj<(w8g@jn4U4*o_D zWAB_>l>`L9!P2Nt7xeR4tSKeWTU6sKL-c5Y_hxkX1d0RLOQU>0QAkt0A!A222dZyG zyg5c8~t&7&H3G#649x%C}tr=&s^{G^SY>Uivtq} zny(kp_+;|i!?-gbrR4bz)>OGqmzUJWo$tMMX9x1)8Pp0BSn{1xT!xr#L_eKEZCXFW zuFb1pVX08Dhg!U5{3Jthu}`sN69AS&R{kmiz&g^UMkerZ(FxNQslWB!>QylOP#ym|NNPcq!*W>A#-pxZ9Pc|*`Wu5rv z*RA^q8GSwUWt8*^tj-bTOfL5&M}5z?Rc@nq(?(ngF%^%GTC*Eha26t=K{B$PBEdzR z?}|>JMzPc!IxnB|<9c`l%coe?!fvN~5;doAJ@WEd8{q{BoaW!>pl>EBzQcs4a?`hL zax)${E1pkdi5>e&Qkvj>SZ=LpFFY zEy>eraCW;Bch}c9@nu|WwmgKAxX#w8 z12MgcO-U_bf$ZkEU5D)nt%LAh8&aHud3I!Je!BM%QPuIm2Wo#fhOvd><@dHg7M^rY{eiRl{(d^O!OUD;*v zmm0Z>y}_Fku+g!xb_-WNczdtZ@_@jq+>>=Gie(+SqBOhfA%_xOi370FDtUk&?T===p=l!s7 zQ&A~Mz#^O5wT>G#y?NR@O+C97&U*&UN-GI+#ZkjDdmTL*{QfrqV1jgOm%^e#h6iEf zRDZYlrsj*o|3D_Z2`I0Z?%J*2>Dob;;PHFU&Ng_vb)AcdBC$#8M0{Tp+( zB}h9WXR{d@e#H7w(r6{JHXZoE$Ix{6LSV=HP)vB&`UWPi$rlWI+*z;_%=dwW-u4dH zp;m2(s3<0ss7#{*5ZOZ~D29by?qduLlonJrKM$pRi}YQ8!#Br59N4SHyHcL zcGQtTzWH>wwpG-W-ozQ$yeSP4eaeWQdO;HTD7ITh8TDu%;uIX4GLt2m8Fch8{xXjL zSk=Vrk?0{7p*A#XT#v5HL{rjf^kot%*@K)8SQfMBuP#Crf#LL#WyIE0^p~|PaXJ$% zu0rkXs9+17*tU+k=P>&75flU3LZnRUNGkEtR@7k)V@s0xhT!MTRI!{V>!M%lO`NSn z*KJ@>Era^+L~Yxl87%8&i>WVM#M|po0yGW2D==?t(i45+6?Ee<nEr)I0a-Do=0~07lXQHJ@7(Kd`?PkVyE1GtZCRdaqH=GRh zC2G5q?UExaNJYeCL@kFrv;p{zNWd#gT7*ShY%wulKJj%&lq(TLx^%e-J^mk-EQJPv zYK}!yiWs*`qM>9&##Zre37x|*ZofdJ$OD|7{`Lh2?KgaZ_nh3XAW-=WB3$lloW@$2Tu6WE?~g_ku_qV@4MY~5>0h& zK(;g%v2|3Lh!na+#evAzP|x;gMh5H70rX~KEew%dWUFJSotueu7myvoMpX{=O)Q!K z!9gvlf}w*)phQ}53Z+X_9B}v$qP{@w+OS=42X`Rqhoh+wN8Jn>l!``th1#YwXdz3y zvYYq_1_q-R6=CRwlYBW|sHy{dx!ff>?j5LIkU_I^qPM$3+gZqbpbRPAUMGlRXl3&2^7`FL)fPhdtGQIN275bG;SHXz9X99q&rXc zP_L<(C0Oo-3)~zV>y^-9uFf1ZXg=x#M_$qpqW#p$Y+`>Lnnjz7X?mey*ObEFJd@)} z>I%%*Das&Hos0mVQQ=3q@PG#-51yZQ&_-nJ_pz2gl6mI&M)0?}Zb+*BIj;|Dq4 z`X#P5k4@>iKPNgd4#FTqQ+^k5$%YCPZE6|5J=6#<3t^8(XqrmR0!_vED!kl^tX&72 zXQB?z7sicV#JX;&Losl8D)d6BhdOL08c9^Z((W=3b<#mJ5U2pmVewcX_y-PvR)*4R$)L?{O=$mAzI#luJl0S#Xq_Re@Kd(J&m9f|M(JF1|O&yS~^sAfd?STxO# z;sbE*DIzKuG_DR$&Lq~x;9hMFu|{~dgk02lyYO=punib^3*f3s-sA}@J}7V`1Qyym zy+Er7pi1fzHI|3XyF_La=N~yQ0?yig3lG*lXzQ@B^2_Z>F8pxW&?SnDb(aV{`k<;v zVu)-GHhle{ci_IdIYYA@wC?i)nO7UuWwZ>a+W^|#DtTwV#V!2tgM@!fit5*y)^q8? zx}L0W14d*p{a_59GHt6q)TMwsn>*AMk2WcxtffB+3OF@a?4VLtK>xI_mKKv?%UdC)LDwYV!Ro_$!rNt0= zsbt$Fc0avsS?0?RpGUFLvW5^%1tgiRB|xws_(^@gxBtD|b8JpjVp|%A9m}Q?+yx(P z`~tcO0)&!;!a*8Rza#&<-PA*#Sa4#TpJy@rf#EMKFF>mYsEZe`?;nKTf3W>0F{Ntm z*y;D(Cg-Wf?nfChXFSkx#i9>VF!3qJ=6sP!xIjsq$HT+#&w27+$J#PHtUJ)r?MEI* z$-J^uOnjP}n@aOsxb)1Y_P*0l-_L7(1p42G^5xaYQaU-LXhXo-2(}P{p z{aySJ*Wxa8KT4~4cHVeqA$$?i^|9_*U?=g?y?5+UlL3}aA|<@;sZU{A>%f8lg8?b+ z>uS%ZQO~>&zgmt+fBf$26OT(I{)&&UWO#@e7oVN0p$X_FvMzSPj12Gf-O4k@BeHbN zE!7$yl1?JU+(edp{8R8ChFEh4%z?Z-k8UFG{JbzYJB>rT*xrczApMpanF#N#G&_?M zj+OSVxC50xzr{mGPHm%UI&a_JjBIe}h>bzfpA zp6slPFP@e=Yj>|k2cr@-fY5OgPAtVnG?DqUgGJs9XBnKS8-KughQEF1*(lx7wSZKe zRGY=DD4V{ge_f}G@V3*|l(E5IjiwcY3{r-NRGN5o`xmjJj$MDgeIO=i5861yR^((A z_uAe5Sj-u+iBY(P;gL6oMJ1)<>IT)t2ii16fk0se;qm^T_MQa3iDLcPByG}Cns65c zk|IYDeXS=TM;7EMUXS{GT!OS%1biM8MatD?+d_-VQBef6KtKe%@p_;l-2y&%PlT!* z1%-x++|VM0UP<=OYwLbruUz?r%=DYTrnQxAro!K&i*8ZZt98)W6%*!64 z2b;nx3$u$DVy^0uwHP{UFlsI9a8U_++{i4sIi-V5+A=qEo|DkMs)vh!^&c(>#CSa& zE`xl5*Fd;@xCg%K;lSYf50;4UBx8o5*<_Ea>riS}0;ha9fc%=;6WK$lUG#3{!vPl7 zl+?i+EhsMTf#WV53BsE48AylgiaSeu3bT9pfOJkt_Q2-JN*+x$UWLxZMp{<7J0;u= zc@_08v3D3aydbIrVZwo8lfYNw(DkXIJM?sN$|lt=<8TZBFN%G&_R1)m6;8FARY2;0 zJ$~zcY#F9KW0n8Voni{-%r7=D)-M&c`kl95?dyv$Lqr_w(&h>U(~{Q2B1DX8oMgBjEwldc_Pn9L4tjmx zRDC%@W49mCT61BfU_#Rlly}BdeJW}TzMOBad~RKW_F}S0Zq=QEH7@hMIcY?w?}_RC zX20F-x2%GE%fL#5?%fe!pqK`A&GkYO)2>~*Z6Ln3x$^wWAH{2p6Q8R(+u@*Yz@fkV zf(*OOkFPVK=tutXb#7gBoNc8C6fE7Atj(Atx-nZu;Yh`_l@H;oMK-QW+|2ydB)-88 z5PqyfcXUlOP&3~hleO`$|Eh_haFR|lqNH4hW7m}n=8r_g>!Z==_o#NGn3Nce8=;bHTGtgc#cV(4?TwVss?#B1qAl43{kFCHWV~s4p3{V zZz%73zH3x%+o!*pZ$$CaGY@wsF6{3Byi*)13!p0jxI$&ah!5b^1tUMDF4$hsQH zwWb8sgaa$Q7mzaoh9P z;8s>ocgtB~+6HsNEz54}bc$kVOv{FZ^EI(F>LUAO$!4WFuHck_vQyts@+v7^?l*`%rb9u^%W zA3mua-tXO&pV(rBg_uCS-se2ST-Mmpp%Jg`{{629+su(2SlACSJ15k=_4TPm z#jh6Ovx#kf?oXwXG%f)h*0Drp>!`S7jV!!$9)ED~kVOkFvjtoLjz;t7rYzayJWs&cVC4 z&pNUU$oF+P7h9BkQG4aTnCCMOpj7X{^cWH`bC%PWx%iE$aTd(G{ln`CU_NEYgl#kD~0%2CdVcif?

e(7s?_7;sf#+8_Z zHZ`@wbvrBkd*f9_b_D{$IkK`nn=HhbL;+~1vE&0As*+L99#Ycph8m(I zwO4J8qBf%%!@F+T(8t~@z;~*QA8Qt@tp>L=M`qgd&iUA(X^>?t8g>)`li54V4s3*H z-0{?dPnn4mso$NB%b`k4EiS9RUJtYu!jLk)R~2T1%a^t;&Yuy<{C27D$cf$cQqzoAPf~LX$RXb7-a0w@Zj;<4CZWXL*DGlgMH7)tn`$F zLr<3IpsdtHty}sKv`Tv0^8&9aKV4m5MDnus1=%CuWVIIZ!N!ckHPGpi?{6QSO-?Io ze|_${I%uX}89Je+V+{peu4gOM1~e6+s|#}1{sZO+fC~<3XGb5cSpbBf#5XV#hV|Nv z9~)FHYacy`Jo5JG7Bt4Jolj^f1SN7E)A5UL5QC1T;^YV@0FFQ~+VVC~a*+J8DlexR z5!b*3sTqfAqSCQm1=%UuS_?s9>~=HLG?}+tSTcLj53|h#jY0ApC?+^!3wm0I-+4fo z_>;NRu=!?N-0_T9uMy;!4%ci3X;s%fyY^t)OR5h}K0p2tg%NMLpfl_lIL=0c&|tD$ zX^yt&Fnn2UYT7}7r=5ND*tIAQ|Or-N)O%g~9=cGH%z-9h#E$oy>Zo==~F# z_IIH@o7VRswRL?iH^@#J9(wH?8!uRw!Z4Y`j<(hv*r2_Vt6#bo>ZzXYRgC+qhJ(E? zIRp$WEnIk)C3lEn%!3sOoF!o)zRv4V-a0xvk?G#D0g*5X<{0iUXbt&OWjsvn=_^_F zU^r&CBMY!y6|Mr{gggD06Jm>K9JfHinB0@_oeo!hj{qw}oaKv?#1>y2*oXsmM8O^= zFQ{+ru?ZoEJcA0?I*;Io_8dycG|rtp>WB6&TCgqcy0gdp(4IredJ_%^!2^4vu6>XN zfKz<=Oz1}e;4i+c)_xQKf#S<@@}mHFYL6#_=Ac2~SUdx%^Mv4gl^2lxj-k^T%9a+r zYP6}7Eyml?1{djGskr^fq;Vurs>|+tJ{3=P2hq`vU)OK#Jar8_60zyS{)mt69<;qG zBeb#KKO@wAWtrKpJeoegh{S12i#E;mu2fP-cXsZ4qZ#QPZOlXBxQl0rr(IKuL|X3( z%`-yuZ~hY+Pqs^Ae^RNMve-OXDN$1g2dM*f9!(t_q-LdO6;nU27&u|bM}&!;f@7Zv zu2Ku#XQ|ueM}{Y(I*+#!?7R(!B}9ca?IQbO$x3%*iBs8jO&?C4J8a(_pDZbII;^QZNUN2Cn&P9~=b^!Rn&oG|>fCSL z?i?S~y1`sG*s2l()I5_AYEZ}8vrO%MCigsM)8GnTipNPg5pG%B%O^CZVz~)iL7fnoN`N^F=Hwg;;#|QXffg4I z7ZgN8&%zZa zy@(B*ctK`~$AF!~*4+jo>iBXf(rpl)EO8b~m`eMXaFzfXJbe&Z0ye#BaYDrU{fPESWP z4*1hXn)xIEXAk%4g)v)V#cxh|;}VvoiY$3lT{XbSR9KMhR8PJ(mp~ zx-2J0LyGVbO!n2GMZ{LC_4qp7sV%0;s8U)f$izJX4}x-F6AaoN>~^PfyK}#>kBAuN zQbYfmbMJ2QnmKXD_Hid@JfN`^AgKBW>gy(+-TdC<$>|y&oP?7^Tnvl`;5E^T#@eSA zFYfK5JPQlJDdH`Xp0QS|uRYNzVfEa(g}#{yrAl!z7E*t0seEGMTm z<(~kIEn0dp68JKbyt6K;>`p=mcO+&K3!334E+L zJ}hhkc*9oxwK4r}JzifoETke_EfQqO zImC=ym#BhTaw>#ESl>$lPsoyKOyd&^7W7C<8)^!n5T~htr=Dr7DtV{!rlfb~&3lOq zoe*c~#Vi?S>cD}WR;^mL`lKl^H!q?Xn93|?@^M?k*woHS<@+5F@1*N}M-mmw%`i8LqsVR-9v0>j6-C{N!DaXrL!Waq*=dPJBL>ER0 z5!Mk4L@W@ofd3YNJ#Aw!RmUgSoK8{K$uj+CVWZ%Cxf=+pp2u?gA09hxON!ho8yx@x zSg}X>LB{iH&PA=J7Irb-c(P#*yg^XwvSAB@ni2w(3d`EsCovzrA)%F zE>n2Xngl3&aYKh)87t;YrxEIv-XtQ0g;ckTO51(6bpxfYQ-!hzDlCyd<^7Q;=it+H zyG(tVEEU$}+1G6G3aqHtaqp$X16N%(driBoPpq2z46QfC zq)U?MiIXB*e_C*m*RiD^jNE_3ld0!NnsRzy%jz!?3`s^KGUON$mcJP znm;dnBaM{x{s`2Y{J|E-{b+1~<;OlJ+NFLxs}-=84HPG_W9XzJO2_NHk}!zQA^MKU0F{h!VA@3i$cUW05-D41?uWYH_aV6GzU$| z)H3xR&9sfBkA89hB85pdrbk&up4T7BOfLMg2>NuX?x=+K*B9;xp}b2*m)5Pku@z~cqrEZ-&Y`it(F<0TD7~D3Z@8>y zP8}_p*>hd0M%B7AtoebyX0rqnsWVi)+<(@_j!QR@P+-FzDL5VWcY8cAbZi+=HKb7W{;;C`aN7=RyLO=AlEphFsnna9f#3~k1KUhU$K;a~i{0E^PK<7&kb%2qLgGMzlhN z%%gDi^NzFjaXAYQ9Fo1K*@(xL^tku9rET|4F&?STddecO_F+AW6M-dU5yT8=s|T-J z7`uFyTJ&oF_#X0~mk#ou#4K?3(m3J~aQ>4RCYb*?(VxIRE^7}jS{6af9$tjL0fe(C zd*HE8=9MuHJT@a)mqXj&PAFat#|cg@ZcrYCmZVsP5cMdc2pqAZQNrm9JW-v#gLM`{ z2x1oYq_f}CESO{?wrfk0EdaCw?$ATH9yE3&(k?E0 zc;Yuhrw&vA1JItDIPOUnPO}k@E9tnm8IWcZStz%)!f7=AfnYPMJ?i#r&jxTCO4<$X zykR|xC<0sJoRp4;)54w!KJIY=;Vl3FXA6MSS$Oclw6@?2AqrlRPI=w$@qvx6tc#}T z_g6Y5ZFb0UUStk$rlu&81;QF_HY#?F&zSZ}yV)Hx55+61Uy34GbPDBoZ6SuZB3MNC ztD9h=(R}E+!aF+N^WmFuTPC4|I33b@e;n@4fd7t1u-w}9m-|ii3%ASR_9VQ zKWSjlLEl&I-rr}{K=~(~AKI9%N)Rd*0dbfAvL^1~Uk9w)Goa(Xcc07*h=(xIfUl44 zKHTT$)v1W$Acbg}N&fX9(*2878ApgSO0qEn#l}K$HPB8jlr!dvaJ@|k1-pOF%qQ5T z!XG&1UWZ`OWmt3_?E3tepMNt6c5_Vpngs%NH)2hbPlRL%60*~5h zbOVxZcT#S^N%^k-By8D2iM>tM3Ge42paggo-*47+u-`N7TQ%E2%eQJ?Rs%2P;Q%}o zghR*VCf};zkGcG{8Vr<#zo*A4s9;YdUq>w2)C|CC6nre{@rZ7>_|Ty`->TWZ+4jq7 zosK?5$KKzYQp85P33J?lF-RPp=UX+~(A~FcJXUM>bvZTQzb8?`=833(=0T*HHOwIv zAt(-7_kO%H$rT}OQQxZhu-c$Y*U*#^tI*-EhoVQ$>CkC;CZXt-8GI#@Y&G~;=I8Uo zhZ~FKj}6?6hILDoFOMVPU?>WTnCmf}F>pPil}OZT+>y7fCm?4k&PLb%OvatRN_V+5 zgN}RXW=i+blSq?kw{IYo$f+_XqT|H9>llf;UYkwd`RR{T#^}{3Vb%jEZ`~p^zjZ1i zjp=TboSV?N9u7k|E)G@PFq2q3d4_!Zi4;OPKVAOnf72CRUL|Ewu9=h2xE_vNkEonU zo)ul&NXA`$(b(nORiqktBkH{5QIu4@1c@BCVe5^Fc|{L#g16fGxpd!>V`$0?AEGnk ziqRuix}#k?`w{nbCsFNGUD$>EVFge=HIux8-6xj-6Md`Z!PH-$WuQG6z^%`Lyf?ZF zBAt*=Eau~{`jkt!^cX1})d!{@Wdgdf%@y3cu@amnOx#h;rR-q%0@@GRdZx|~?2w|$ zO}_U)45_d{nHjE9s+gFVS!kevhGDHVHjkyEIFNU;7;j9hq;(7sOx(ugppu4 zgfDPdn3&@rZFnMMLpzT9FC{kg#pqyNrRbm%BVu)hd>>914A_N;R?HAvmrog;*n9#e zf{r~*yNap{n3NRjVwN5O7b)L^I7|S;U!TV~JDEzdt`nORfg0d@;EZwc;MpKLr-U(e z60M7I-2y44>4XZ4k3)AOVx~zW@E#APL6tA#Yv4E^gNhwmSJ7gr)Mlz(<#ZTr4 zepOf-O$e>aDgoCJS6Kh7%h4mW6;`@+IV#|2=qs#5cj%!8{L*ml%QbhbUBL(U9&4CG&e9A*(_ghJ=eW2DQs+QB#;D+sG%8L7z zAX#hyh4Lm&MR5<^f@CoZjOf;+EhxI9+M%JezEq&rg>%huQw~9S^*XXt%|zPOAFY(v ze(@8^nfX<^`b0%4!PL|tLb20?5*Gd)$>UXc?(0k@Ogz;BSvO33Gp?{W66U@Xd<1kA zb_^QuxK3F$_akKaJ?ry9Yjb9HT6yi?(BkN=N>cvzSp||UBq=lm8lmU}&`D;5pNF8e zkA-Vu3jFj0b?4LIJXgYjws_v(b1Dx&eg&0$^M#m(tXVOspFTV41L!nBP|=EvOAW}9 z#ZcM<}w(a*AUXN|77F;VQogv1Q-9dC}%mvLTsdNN$dSH}p$DLo6Y+Ez-vjR#o zp;Q4<@#JmpIv6!ead*X}_dXnB>%vQ=^dHZb>HK8+x|n|%`&56|r*S<$Lb#z?6O3{c zGGgn)pXVly8at)DDZgptj3!{u76WwhEK8;vW4he-T}kwoa6dUOf3O8{5&YiMw|`K5 z|KcTS*Kg)U3p)_L(T^`jDk5$56j`f>mN9g7*hTUQz-_>^9l53f-`STO z)+8{ZvUk+qZ+)hwkiwXmrHjsiQJRcy@1`^SDG!?M{g-Pr)RFHFFGZn{0 z^62OSQ9h4cvtUQF?vZz+FgCrB~1f?lPd;8+7UQ;D~QI?;fFJ z3qeqwAhLv0HGP}PrhuRZxa%*;1`^S1&+Go<%LXDfn8PSErs6Q8aGtB1-`6b&ZuWK1neg=EZ6R3eH z*+Hq9{0rwQ7A=B>HMqT;k7PO+HzeANJyM9Lq-9%SXj6)i^j`}RdDVY z?tl?P(hK{hbQdFwq(>|evB00u0Q=^dYWtj6!$z$bH z(@!a)GkW!EHFBh_^s%gjuf9iK@smhLWliAGL3rNTB5;{o9I7#xcSy-BWPS z5(y2^7&D0m9AK{hT6^uYf`S-C_DD^+b>g}69o%_{-U*nhOS;;6fU#`9^81WQr}9lG zMyH8SojrSjC_i2Vr>XD&YfvUqwv{uxESAeB*G)`8F?T2w!_@Ih-x`;m&Py#3m`&-; z)kt{hz`%`kJieW0}4IJ=P z|C>72*F6-Z6rTQ+o`F>;X;oFr78b4=rckyI{+ua~PJ*pxe!)<(O*Oo~rfplp?AZz` z$j1P3XD0#T_9A45x1 zNMIeEG-5F<1~G>z65YQA77U7gZ)yMDLyRMvt~aqUDc3bq7CcoxdBVyvZC&g_Sa^Sv z6e4LjI%#MwtX`{1A760pz&BKz>u3fXD7@U2)a0?)_QOZ)d!dCPW9ZT1XcE3dDALx! z+tg+W@?4b5?C5psRQbk=$296Qzh}bB;#-?V6}!-0tH|R|HYBxMq&!y*8zL~P;lyS8 z0zY5tOT(iu>{P5gWA4<(kp5Qo3zWX%B@@)!Oaw<;N(>}^NzU-EnU%=G4Q=g{`BkIN z8nRNRt+=ueQXY3p6}*An#7t1k{HyDN+#CtL6OgsSIIG>a1y5b7gdLN5QhHOjbTO*% zrOUEcwkv&PQ~l4o2hxAlsOyNc;BunEC(kEmd^EOMZrHZGzT~5WP-xzO&k7flK1srw z7~4KIZP@QezJKb*(+gXYdS*|xx=uAbtRe-O%cE|)-F&=L)^96%O<1K%D^GkK_7Ry5 z-}2>BERFb1EG%XyuDugZo?y20J<%iQ=?h=T7_^Q?s5|^akX+Mk*dh9n<-XDGn3bT$ zmrq_(fg8^|!1eVG5%WOU^g5b5;Kc_@EC+A@<}z-?*CB&RmL3SF;A!w#*^OoQw)=48 z`-Y0V#uGcsG^lj~7^r*jyaVF!{zzo_MltcRLl11dX6{?^`<>9c(9*D7k1il~K@(vE zRqY$rYE$|;;t7>QHKXvmbY1c#gv;a+pD1PU=lj6d)A)r$!*E^GT3pztQRtbc4>*NY|;LxV4W@@YRqw=@ve<@yo3?;h}yr71y zIb?T387m-#pDGL~winc;H~w0qr_;x)t2KBl66W9ISI?$~=#6hu;{BNQ__S-`a z8wumpF=X>CL22@x9+Fg@C>OIRM(_xV;92}ZIm~mGD@ud(es?hZdn$EqH6na>rZl#q zClBWY8&N&jrnZLir9FXk$bfx+7=u=BLNaFtgi%m4u=aQ?E!6}%h{~|-w;OInJf56i zoSq*XQYei1{(Qls)@qx={m++RMndY^TsyCMYRZkNPEr%2BHf61ZhW|Kxk+rmjrqwC z$-~wSB;vx|aCvVAD`{Rh2T1yECODV@N0l_Ux$~~IYC2%n{Dku~hWt7Y^OF3iE-1lO z(ghZZ-9K*U*w==N3k1?h}6Swy@H zIaHy7QVEV2(s?jY!lvrJgGc4|i7d(Xam^HEI?txKOXf| z#6S)81pU9b=WS86VnsI`rt$Wzyi)WgUT}DUI?3#tcGa*$E!%KNvWkBKMDVwsPoY5e z0>m^$U(YTaUKcwmiiEx&Nx+$uA$O^sR&cwAPcC3OP(sr|*$a?1oC9h%&G7M(_VG3H z=&F05{poJK$}RhZO=x>oBwi~bH8kx@0Ebo*iY~_dBR5x4?fXZ^^XlJn6jm}O&Rr8l zeJR9ILG9Ow8xHRE@hB>!<8by76(4~+iA9<#-wHXoz~ROoLq<-LcT_@iR`#jw^gM-h4s2n`xH*H)|Mw+zdT_1Sq z<=mnU=KJo7S$(fb8f~ChninXPxIc#~{bc>5&6|~VSH7R~VrI3Io9}YYq8Vs5CVC~9 zg-Fq9tx5ga9dZAL@6VXq2u3zO?X6!riINH+kOFWA_#Os~T7U@a3%hT>bo}}MY%nF) zExOp|?l{UfT`;MdYbCV)5JSi}tj{psK`=2z`u(@9Uw&75Q7~MEX=Pov)+rmdAkF8W zU3`ro4JU8P%`rcoR($GSVVb~G8Yx8VWKjMztLm=(xeIlq+T0^mUHi&2w?Fd|88}-Z zO1Z^^P~WvC%``ogYbyKn@3#^RwM<_3!<~6qKS(Q|LabeId;i=(8IBQ1x!5XGc4%ad zUutm+tZagVll9fDP|4cv$=_dY=^_bHObaEYgqmRb;gfO=c*Pm;s^FqZc3UU&;q{?< zI6+O@#%(xHm+aJj{I{Fh3nGOp0e_P_>5XRmCRfiub!>TJ`i3^IhU*zc`5Nj0W#pIt zE*f(Gvd1q^^JDHcWL-ZQQDRldj!#*odui1DS=)XQ)!75Vziy!Si#naQTFqGdWNv7; zedoZO-U%BkCT4nw2*<6C%ap2TX!5S!hi8%7UmJIE0ZAhSe@z>RZ5A0bVROgCq?!F! z{HL>ka04Zf-~x+dxNtc!Z6?yRJU!jm>Zsg;Ho*ozr2#T>@RXn_M>;GXVi5HLLdKb6 z52yCdT{f{AYJ*q+5=somk**bc%?1N^H|G|TZt3B@Q8K&C z&`Y+e7F^mq+L$!C_li$aLMM?DVFfh`SKLl%wBFi!aZ$n(vf75?gd_x;3pQ1N1xv(N z3}JYDYRB9ssz>Z?ee0a*=bN;DqBT#eR<&%%E^aNCS#xh3({s&wx~ZNX;m6+#XLz67 zG`4hbnU-mYc{$_qJxwMa1nCCBmwWN8vsbErWf*47%F}6zvZj;|OT7^D|Z^1=fy=kCQF({jfl{Q1tJ!RSRi77hy?<+0A6{|Pu$|_ z+_wlOH(hCtL1f|(hG_si4aHa~gj%Qb(*z4`d-LnRF(}W76dm*oL(7PG;F!!%_vAje zoFQ}08|6$Kyh~Ql2=#>JiqHd&H89pxCG&eaLTw;nbe%#@l7tGDPeP)=lS$N4w)~lR z!@3?O3y}qbvl~c+j=???(C4`eQTKdimUS^A*qZ|xwy4dW8CqM12sI@32kyiJZ{3f{p>ac{W5n zlS3k8p^(lDJm=F|2e zFbsor^JzoH0?J@eG9|@@ +#include +#include + +/* FNV-1a of the decoded pixels of other/YAPF.YAPF (the YAPF logo). */ +#define SAMPLE_HASH 0x8891c2b4u + +static int tests, fails; + +static uint32_t rng = 12345; +static uint32_t rnd(void) { rng = rng * 1103515245u + 12345u; return rng >> 8; } + +#define CHECK(cond, ...) do { tests++; if (!(cond)) { fails++; \ + printf("FAIL: "); printf(__VA_ARGS__); printf("\n"); } } while (0) + +/* Five kinds of content: noise, smooth gradient, hard-edged blocks, + * noisy curves, and an image with an opaque alpha channel. */ +static void fill(uint8_t *p, int w, int h, int ch, int kind) { + for (int y = 0; y < h; y++) + for (int x = 0; x < w; x++) + for (int c = 0; c < ch; c++) { + uint8_t *d = p + ((size_t)y * w + x) * ch + c; + switch (kind) { + case 0: *d = (uint8_t)rnd(); break; + case 1: *d = (uint8_t)(x * 3 + y * 5 + c * 40); break; + case 2: *d = ((x / 7 + y / 5 + c) & 1) ? 200 : 30; break; + case 3: *d = (uint8_t)((x * x + y * y) / 50 + rnd() % 5); break; + default: *d = (c == ch - 1) ? 255 : (uint8_t)(x ^ y); break; + } + } +} + +static uint8_t *encode_mem(const yapf_image_t *img, size_t *n) { + const char *tmp = "test_yapf_tmp.yapf"; + if (yapf_save(tmp, img) != YAPF_OK) return NULL; + FILE *f = fopen(tmp, "rb"); + if (!f) return NULL; + fseek(f, 0, SEEK_END); + *n = (size_t)ftell(f); + rewind(f); + uint8_t *b = (uint8_t *)malloc(*n); + if (b && fread(b, 1, *n, f) != *n) { free(b); b = NULL; } + fclose(f); + remove(tmp); + return b; +} + +static void roundtrip(int w, int h, int ch, int kind) { + size_t n = (size_t)w * h * ch, len = 0; + uint8_t *p = (uint8_t *)malloc(n); + fill(p, w, h, ch, kind); + + yapf_image_t img; + memset(&img, 0, sizeof(img)); + img.width = (uint32_t)w; img.height = (uint32_t)h; + img.channels = (uint8_t)ch; img.mip_levels = 1; img.pixels = p; + + uint8_t *file = encode_mem(&img, &len); + CHECK(file != NULL, "encode %dx%d ch%d kind%d", w, h, ch, kind); + if (file) { + yapf_image_t *a = yapf_load_memory(file, len); + yapf_image_t *b = yapf_load_memory_mt(file, len, 0); + CHECK(a && memcmp(a->pixels, p, n) == 0, + "roundtrip %dx%d ch%d kind%d", w, h, ch, kind); + CHECK(b && memcmp(b->pixels, p, n) == 0, + "roundtrip (threads) %dx%d ch%d kind%d", w, h, ch, kind); + yapf_free(a); + yapf_free(b); + free(file); + } + free(p); +} + +static void mips(void) { + const int w = 300, h = 200, levels = 9; + uint8_t *mp[9]; + yapf_image_t img; + memset(&img, 0, sizeof(img)); + img.width = w; img.height = h; img.channels = 4; img.mip_levels = levels; + for (int l = 0; l < levels; l++) { + int mw = w >> l ? w >> l : 1, mh = h >> l ? h >> l : 1; + mp[l] = (uint8_t *)malloc((size_t)mw * mh * 4); + fill(mp[l], mw, mh, 4, l % 5); + } + img.pixels = mp[0]; + img.mips = mp; + + size_t len = 0; + uint8_t *file = encode_mem(&img, &len); + yapf_image_t *d = file ? yapf_load_memory_mt(file, len, 3) : NULL; + int ok = d && d->mip_levels == levels; + for (int l = 0; ok && l < levels; l++) { + int mw = w >> l ? w >> l : 1, mh = h >> l ? h >> l : 1; + ok = memcmp(d->mips[l], mp[l], (size_t)mw * mh * 4) == 0; + } + CHECK(ok, "mip chain of %d levels", levels); + yapf_free(d); + free(file); + for (int l = 0; l < levels; l++) free(mp[l]); +} + +/* Flip bits, truncate, scribble: the decoder must never crash, and a + * truncated file must be rejected. */ +static void corruption(void) { + const int w = 130, h = 70, ch = 4; + uint8_t *p = (uint8_t *)malloc((size_t)w * h * ch); + fill(p, w, h, ch, 3); + yapf_image_t img; + memset(&img, 0, sizeof(img)); + img.width = w; img.height = h; img.channels = ch; img.mip_levels = 1; img.pixels = p; + size_t len = 0; + uint8_t *file = encode_mem(&img, &len); + uint8_t *bad = (uint8_t *)malloc(len); + int truncated_ok = 1; + for (int it = 0; it < 3000 && file; it++) { + memcpy(bad, file, len); + size_t n = len; + switch (it % 3) { + case 0: for (int k = 0; k < 1 + it % 7; k++) bad[rnd() % len] ^= (uint8_t)(1u << (rnd() % 8)); break; + case 1: n = rnd() % len; break; + default: bad[rnd() % len] = (uint8_t)rnd(); break; + } + yapf_image_t *d = (it & 1) ? yapf_load_memory_mt(bad, n, 2) : yapf_load_memory(bad, n); + if (it % 3 == 1 && d) truncated_ok = 0; + yapf_free(d); + } + CHECK(file != NULL, "corruption test setup"); + CHECK(truncated_ok, "truncated files are rejected"); + free(bad); + free(file); + free(p); +} + +static uint32_t fnv(const uint8_t *p, size_t n) { + uint32_t h = 2166136261u; + for (size_t i = 0; i < n; i++) h = (h ^ p[i]) * 16777619u; + return h; +} + +int main(int argc, char **argv) { + static const int sizes[][2] = { + {1, 1}, {1, 7}, {7, 1}, {63, 63}, {64, 64}, {65, 65}, {127, 1}, + {1, 129}, {97, 61}, {200, 130}, {256, 256}, {513, 77}, + {1000, 700} /* 176 tiles: big enough for the threaded decoder */ + }; + for (size_t s = 0; s < sizeof(sizes) / sizeof(sizes[0]); s++) + for (int ch = 1; ch <= 4; ch++) + for (int kind = 0; kind < 5; kind++) + roundtrip(sizes[s][0], sizes[s][1], ch, kind); + mips(); + corruption(); + + const char *sample = argc > 1 ? argv[1] : "other/YAPF.YAPF"; + yapf_image_t *img = yapf_load(sample); + if (img) { + uint32_t h = fnv(img->pixels, (size_t)img->width * img->height * img->channels); + CHECK(img->width == 512 && img->height == 512 && img->channels == 4 && + h == SAMPLE_HASH, "sample %s decodes to the reference pixels (hash %08x)", + sample, (unsigned)h); + yapf_free(img); + } else { + printf("note: sample %s not found, skipped\n", sample); + } + + printf("%d tests, %d failures\n", tests, fails); + return fails != 0; +} diff --git a/third_party/stb_image.h b/third_party/stb_image.h new file mode 100644 index 0000000..9eedabe --- /dev/null +++ b/third_party/stb_image.h @@ -0,0 +1,7988 @@ +/* stb_image - v2.30 - public domain image loader - http://nothings.org/stb + no warranty implied; use at your own risk + + Do this: + #define STB_IMAGE_IMPLEMENTATION + before you include this file in *one* C or C++ file to create the implementation. + + // i.e. it should look like this: + #include ... + #include ... + #include ... + #define STB_IMAGE_IMPLEMENTATION + #include "stb_image.h" + + You can #define STBI_ASSERT(x) before the #include to avoid using assert.h. + And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free + + + QUICK NOTES: + Primarily of interest to game developers and other people who can + avoid problematic images and only need the trivial interface + + JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib) + PNG 1/2/4/8/16-bit-per-channel + + TGA (not sure what subset, if a subset) + BMP non-1bpp, non-RLE + PSD (composited view only, no extra channels, 8/16 bit-per-channel) + + GIF (*comp always reports as 4-channel) + HDR (radiance rgbE format) + PIC (Softimage PIC) + PNM (PPM and PGM binary only) + + Animated GIF still needs a proper API, but here's one way to do it: + http://gist.github.com/urraka/685d9a6340b26b830d49 + + - decode from memory or through FILE (define STBI_NO_STDIO to remove code) + - decode from arbitrary I/O callbacks + - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON) + + Full documentation under "DOCUMENTATION" below. + + +LICENSE + + See end of file for license information. + +RECENT REVISION HISTORY: + + 2.30 (2024-05-31) avoid erroneous gcc warning + 2.29 (2023-05-xx) optimizations + 2.28 (2023-01-29) many error fixes, security errors, just tons of stuff + 2.27 (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes + 2.26 (2020-07-13) many minor fixes + 2.25 (2020-02-02) fix warnings + 2.24 (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically + 2.23 (2019-08-11) fix clang static analysis warning + 2.22 (2019-03-04) gif fixes, fix warnings + 2.21 (2019-02-25) fix typo in comment + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings + 2.16 (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes + 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64 + RGB-format JPEG; remove white matting in PSD; + allocate large structures on the stack; + correct channel count for PNG & BMP + 2.10 (2016-01-22) avoid warning introduced in 2.09 + 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED + + See end of file for full revision history. + + + ============================ Contributors ========================= + + Image formats Extensions, features + Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info) + Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info) + Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG) + Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks) + Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG) + Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip) + Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD) + github:urraka (animated gif) Junggon Kim (PNM comments) + Christopher Forseth (animated gif) Daniel Gibson (16-bit TGA) + socks-the-fox (16-bit PNG) + Jeremy Sawicki (handle all ImageNet JPGs) + Optimizations & bugfixes Mikhail Morozov (1-bit BMP) + Fabian "ryg" Giesen Anael Seghezzi (is-16-bit query) + Arseny Kapoulkine Simon Breuss (16-bit PNM) + John-Mark Allen + Carmelo J Fdez-Aguera + + Bug & warning fixes + Marc LeBlanc David Woo Guillaume George Martins Mozeiko + Christpher Lloyd Jerry Jansson Joseph Thomson Blazej Dariusz Roszkowski + Phil Jordan Dave Moore Roy Eltham + Hayaki Saito Nathan Reed Won Chun + Luke Graham Johan Duparc Nick Verigakis the Horde3D community + Thomas Ruf Ronny Chevalier github:rlyeh + Janez Zemva John Bartholomew Michal Cichon github:romigrou + Jonathan Blow Ken Hamada Tero Hanninen github:svdijk + Eugene Golushkov Laurent Gomila Cort Stratton github:snagar + Aruelien Pocheville Sergio Gonzalez Thibault Reuille github:Zelex + Cass Everitt Ryamond Barbiero github:grim210 + Paul Du Bois Engin Manap Aldo Culquicondor github:sammyhw + Philipp Wiesemann Dale Weiler Oriol Ferrer Mesia github:phprus + Josh Tobin Neil Bickford Matthew Gregan github:poppolopoppo + Julian Raschke Gregory Mullen Christian Floisand github:darealshinji + Baldur Karlsson Kevin Schmidt JR Smith github:Michaelangel007 + Brad Weinberger Matvey Cherevko github:mosra + Luca Sas Alexander Veselov Zack Middleton [reserved] + Ryan C. Gordon [reserved] [reserved] + DO NOT ADD YOUR NAME HERE + + Jacko Dirks + + To add your name to the credits, pick a random blank space in the middle and fill it. + 80% of merge conflicts on stb PRs are due to people adding their name at the end + of the credits. +*/ + +#ifndef STBI_INCLUDE_STB_IMAGE_H +#define STBI_INCLUDE_STB_IMAGE_H + +// DOCUMENTATION +// +// Limitations: +// - no 12-bit-per-channel JPEG +// - no JPEGs with arithmetic coding +// - GIF always returns *comp=4 +// +// Basic usage (see HDR discussion below for HDR usage): +// int x,y,n; +// unsigned char *data = stbi_load(filename, &x, &y, &n, 0); +// // ... process data if not NULL ... +// // ... x = width, y = height, n = # 8-bit components per pixel ... +// // ... replace '0' with '1'..'4' to force that many components per pixel +// // ... but 'n' will always be the number that it would have been if you said 0 +// stbi_image_free(data); +// +// Standard parameters: +// int *x -- outputs image width in pixels +// int *y -- outputs image height in pixels +// int *channels_in_file -- outputs # of image components in image file +// int desired_channels -- if non-zero, # of image components requested in result +// +// The return value from an image loader is an 'unsigned char *' which points +// to the pixel data, or NULL on an allocation failure or if the image is +// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels, +// with each pixel consisting of N interleaved 8-bit components; the first +// pixel pointed to is top-left-most in the image. There is no padding between +// image scanlines or between pixels, regardless of format. The number of +// components N is 'desired_channels' if desired_channels is non-zero, or +// *channels_in_file otherwise. If desired_channels is non-zero, +// *channels_in_file has the number of components that _would_ have been +// output otherwise. E.g. if you set desired_channels to 4, you will always +// get RGBA output, but you can check *channels_in_file to see if it's trivially +// opaque because e.g. there were only 3 channels in the source image. +// +// An output image with N components has the following components interleaved +// in this order in each pixel: +// +// N=#comp components +// 1 grey +// 2 grey, alpha +// 3 red, green, blue +// 4 red, green, blue, alpha +// +// If image loading fails for any reason, the return value will be NULL, +// and *x, *y, *channels_in_file will be unchanged. The function +// stbi_failure_reason() can be queried for an extremely brief, end-user +// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS +// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly +// more user-friendly ones. +// +// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized. +// +// To query the width, height and component count of an image without having to +// decode the full file, you can use the stbi_info family of functions: +// +// int x,y,n,ok; +// ok = stbi_info(filename, &x, &y, &n); +// // returns ok=1 and sets x, y, n if image is a supported format, +// // 0 otherwise. +// +// Note that stb_image pervasively uses ints in its public API for sizes, +// including sizes of memory buffers. This is now part of the API and thus +// hard to change without causing breakage. As a result, the various image +// loaders all have certain limits on image size; these differ somewhat +// by format but generally boil down to either just under 2GB or just under +// 1GB. When the decoded image would be larger than this, stb_image decoding +// will fail. +// +// Additionally, stb_image will reject image files that have any of their +// dimensions set to a larger value than the configurable STBI_MAX_DIMENSIONS, +// which defaults to 2**24 = 16777216 pixels. Due to the above memory limit, +// the only way to have an image with such dimensions load correctly +// is for it to have a rather extreme aspect ratio. Either way, the +// assumption here is that such larger images are likely to be malformed +// or malicious. If you do need to load an image with individual dimensions +// larger than that, and it still fits in the overall size limit, you can +// #define STBI_MAX_DIMENSIONS on your own to be something larger. +// +// =========================================================================== +// +// UNICODE: +// +// If compiling for Windows and you wish to use Unicode filenames, compile +// with +// #define STBI_WINDOWS_UTF8 +// and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert +// Windows wchar_t filenames to utf8. +// +// =========================================================================== +// +// Philosophy +// +// stb libraries are designed with the following priorities: +// +// 1. easy to use +// 2. easy to maintain +// 3. good performance +// +// Sometimes I let "good performance" creep up in priority over "easy to maintain", +// and for best performance I may provide less-easy-to-use APIs that give higher +// performance, in addition to the easy-to-use ones. Nevertheless, it's important +// to keep in mind that from the standpoint of you, a client of this library, +// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all. +// +// Some secondary priorities arise directly from the first two, some of which +// provide more explicit reasons why performance can't be emphasized. +// +// - Portable ("ease of use") +// - Small source code footprint ("easy to maintain") +// - No dependencies ("ease of use") +// +// =========================================================================== +// +// I/O callbacks +// +// I/O callbacks allow you to read from arbitrary sources, like packaged +// files or some other source. Data read from callbacks are processed +// through a small internal buffer (currently 128 bytes) to try to reduce +// overhead. +// +// The three functions you must define are "read" (reads some bytes of data), +// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end). +// +// =========================================================================== +// +// SIMD support +// +// The JPEG decoder will try to automatically use SIMD kernels on x86 when +// supported by the compiler. For ARM Neon support, you must explicitly +// request it. +// +// (The old do-it-yourself SIMD API is no longer supported in the current +// code.) +// +// On x86, SSE2 will automatically be used when available based on a run-time +// test; if not, the generic C versions are used as a fall-back. On ARM targets, +// the typical path is to have separate builds for NEON and non-NEON devices +// (at least this is true for iOS and Android). Therefore, the NEON support is +// toggled by a build flag: define STBI_NEON to get NEON loops. +// +// If for some reason you do not want to use any of SIMD code, or if +// you have issues compiling it, you can disable it entirely by +// defining STBI_NO_SIMD. +// +// =========================================================================== +// +// HDR image support (disable by defining STBI_NO_HDR) +// +// stb_image supports loading HDR images in general, and currently the Radiance +// .HDR file format specifically. You can still load any file through the existing +// interface; if you attempt to load an HDR file, it will be automatically remapped +// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1; +// both of these constants can be reconfigured through this interface: +// +// stbi_hdr_to_ldr_gamma(2.2f); +// stbi_hdr_to_ldr_scale(1.0f); +// +// (note, do not use _inverse_ constants; stbi_image will invert them +// appropriately). +// +// Additionally, there is a new, parallel interface for loading files as +// (linear) floats to preserve the full dynamic range: +// +// float *data = stbi_loadf(filename, &x, &y, &n, 0); +// +// If you load LDR images through this interface, those images will +// be promoted to floating point values, run through the inverse of +// constants corresponding to the above: +// +// stbi_ldr_to_hdr_scale(1.0f); +// stbi_ldr_to_hdr_gamma(2.2f); +// +// Finally, given a filename (or an open file or memory block--see header +// file for details) containing image data, you can query for the "most +// appropriate" interface to use (that is, whether the image is HDR or +// not), using: +// +// stbi_is_hdr(char *filename); +// +// =========================================================================== +// +// iPhone PNG support: +// +// We optionally support converting iPhone-formatted PNGs (which store +// premultiplied BGRA) back to RGB, even though they're internally encoded +// differently. To enable this conversion, call +// stbi_convert_iphone_png_to_rgb(1). +// +// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per +// pixel to remove any premultiplied alpha *only* if the image file explicitly +// says there's premultiplied data (currently only happens in iPhone images, +// and only if iPhone convert-to-rgb processing is on). +// +// =========================================================================== +// +// ADDITIONAL CONFIGURATION +// +// - You can suppress implementation of any of the decoders to reduce +// your code footprint by #defining one or more of the following +// symbols before creating the implementation. +// +// STBI_NO_JPEG +// STBI_NO_PNG +// STBI_NO_BMP +// STBI_NO_PSD +// STBI_NO_TGA +// STBI_NO_GIF +// STBI_NO_HDR +// STBI_NO_PIC +// STBI_NO_PNM (.ppm and .pgm) +// +// - You can request *only* certain decoders and suppress all other ones +// (this will be more forward-compatible, as addition of new decoders +// doesn't require you to disable them explicitly): +// +// STBI_ONLY_JPEG +// STBI_ONLY_PNG +// STBI_ONLY_BMP +// STBI_ONLY_PSD +// STBI_ONLY_TGA +// STBI_ONLY_GIF +// STBI_ONLY_HDR +// STBI_ONLY_PIC +// STBI_ONLY_PNM (.ppm and .pgm) +// +// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still +// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB +// +// - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater +// than that size (in either width or height) without further processing. +// This is to let programs in the wild set an upper bound to prevent +// denial-of-service attacks on untrusted data, as one could generate a +// valid image of gigantic dimensions and force stb_image to allocate a +// huge block of memory and spend disproportionate time decoding it. By +// default this is set to (1 << 24), which is 16777216, but that's still +// very big. + +#ifndef STBI_NO_STDIO +#include +#endif // STBI_NO_STDIO + +#define STBI_VERSION 1 + +enum +{ + STBI_default = 0, // only used for desired_channels + + STBI_grey = 1, + STBI_grey_alpha = 2, + STBI_rgb = 3, + STBI_rgb_alpha = 4 +}; + +#include +typedef unsigned char stbi_uc; +typedef unsigned short stbi_us; + +#ifdef __cplusplus +extern "C" { +#endif + +#ifndef STBIDEF +#ifdef STB_IMAGE_STATIC +#define STBIDEF static +#else +#define STBIDEF extern +#endif +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// PRIMARY API - works on images of any type +// + +// +// load image by filename, open file, or memory buffer +// + +typedef struct +{ + int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read + void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative + int (*eof) (void *user); // returns nonzero if we are at end of file/data +} stbi_io_callbacks; + +//////////////////////////////////// +// +// 8-bits-per-channel interface +// + +STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +// for stbi_load_from_file, file pointer is left pointing immediately after image +#endif + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +#endif + +#ifdef STBI_WINDOWS_UTF8 +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input); +#endif + +//////////////////////////////////// +// +// 16-bits-per-channel interface +// + +STBIDEF stbi_us *stbi_load_16_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_us *stbi_load_16 (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +#endif + +//////////////////////////////////// +// +// float-per-channel interface +// +#ifndef STBI_NO_LINEAR + STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + + #ifndef STBI_NO_STDIO + STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); + #endif +#endif + +#ifndef STBI_NO_HDR + STBIDEF void stbi_hdr_to_ldr_gamma(float gamma); + STBIDEF void stbi_hdr_to_ldr_scale(float scale); +#endif // STBI_NO_HDR + +#ifndef STBI_NO_LINEAR + STBIDEF void stbi_ldr_to_hdr_gamma(float gamma); + STBIDEF void stbi_ldr_to_hdr_scale(float scale); +#endif // STBI_NO_LINEAR + +// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user); +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len); +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename); +STBIDEF int stbi_is_hdr_from_file(FILE *f); +#endif // STBI_NO_STDIO + + +// get a VERY brief reason for failure +// on most compilers (and ALL modern mainstream compilers) this is threadsafe +STBIDEF const char *stbi_failure_reason (void); + +// free the loaded image -- this is just free() +STBIDEF void stbi_image_free (void *retval_from_stbi_load); + +// get image dimensions & components without fully decoding +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len); +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *clbk, void *user); + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit (char const *filename); +STBIDEF int stbi_is_16_bit_from_file(FILE *f); +#endif + + + +// for image formats that explicitly notate that they have premultiplied alpha, +// we just return the colors as stored in the file. set this flag to force +// unpremultiplication. results are undefined if the unpremultiply overflow. +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply); + +// indicate whether we should process iphone images back to canonical format, +// or just pass them through "as-is" +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert); + +// flip the image vertically, so the first pixel in the output array is the bottom left +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip); + +// as above, but only applies to images loaded on the thread that calls the function +// this function is only available if your compiler supports thread-local variables; +// calling it will fail to link if your compiler doesn't +STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply); +STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert); +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip); + +// ZLIB client - used by PNG, available for other purposes + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen); +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header); +STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + +STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + + +#ifdef __cplusplus +} +#endif + +// +// +//// end header file ///////////////////////////////////////////////////// +#endif // STBI_INCLUDE_STB_IMAGE_H + +#ifdef STB_IMAGE_IMPLEMENTATION + +#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \ + || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \ + || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \ + || defined(STBI_ONLY_ZLIB) + #ifndef STBI_ONLY_JPEG + #define STBI_NO_JPEG + #endif + #ifndef STBI_ONLY_PNG + #define STBI_NO_PNG + #endif + #ifndef STBI_ONLY_BMP + #define STBI_NO_BMP + #endif + #ifndef STBI_ONLY_PSD + #define STBI_NO_PSD + #endif + #ifndef STBI_ONLY_TGA + #define STBI_NO_TGA + #endif + #ifndef STBI_ONLY_GIF + #define STBI_NO_GIF + #endif + #ifndef STBI_ONLY_HDR + #define STBI_NO_HDR + #endif + #ifndef STBI_ONLY_PIC + #define STBI_NO_PIC + #endif + #ifndef STBI_ONLY_PNM + #define STBI_NO_PNM + #endif +#endif + +#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB) +#define STBI_NO_ZLIB +#endif + + +#include +#include // ptrdiff_t on osx +#include +#include +#include + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) +#include // ldexp, pow +#endif + +#ifndef STBI_NO_STDIO +#include +#endif + +#ifndef STBI_ASSERT +#include +#define STBI_ASSERT(x) assert(x) +#endif + +#ifdef __cplusplus +#define STBI_EXTERN extern "C" +#else +#define STBI_EXTERN extern +#endif + + +#ifndef _MSC_VER + #ifdef __cplusplus + #define stbi_inline inline + #else + #define stbi_inline + #endif +#else + #define stbi_inline __forceinline +#endif + +#ifndef STBI_NO_THREAD_LOCALS + #if defined(__cplusplus) && __cplusplus >= 201103L + #define STBI_THREAD_LOCAL thread_local + #elif defined(__GNUC__) && __GNUC__ < 5 + #define STBI_THREAD_LOCAL __thread + #elif defined(_MSC_VER) + #define STBI_THREAD_LOCAL __declspec(thread) + #elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__) + #define STBI_THREAD_LOCAL _Thread_local + #endif + + #ifndef STBI_THREAD_LOCAL + #if defined(__GNUC__) + #define STBI_THREAD_LOCAL __thread + #endif + #endif +#endif + +#if defined(_MSC_VER) || defined(__SYMBIAN32__) +typedef unsigned short stbi__uint16; +typedef signed short stbi__int16; +typedef unsigned int stbi__uint32; +typedef signed int stbi__int32; +#else +#include +typedef uint16_t stbi__uint16; +typedef int16_t stbi__int16; +typedef uint32_t stbi__uint32; +typedef int32_t stbi__int32; +#endif + +// should produce compiler error if size is wrong +typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1]; + +#ifdef _MSC_VER +#define STBI_NOTUSED(v) (void)(v) +#else +#define STBI_NOTUSED(v) (void)sizeof(v) +#endif + +#ifdef _MSC_VER +#define STBI_HAS_LROTL +#endif + +#ifdef STBI_HAS_LROTL + #define stbi_lrot(x,y) _lrotl(x,y) +#else + #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (-(y) & 31))) +#endif + +#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED)) +// ok +#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED) +// ok +#else +#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)." +#endif + +#ifndef STBI_MALLOC +#define STBI_MALLOC(sz) malloc(sz) +#define STBI_REALLOC(p,newsz) realloc(p,newsz) +#define STBI_FREE(p) free(p) +#endif + +#ifndef STBI_REALLOC_SIZED +#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz) +#endif + +// x86/x64 detection +#if defined(__x86_64__) || defined(_M_X64) +#define STBI__X64_TARGET +#elif defined(__i386) || defined(_M_IX86) +#define STBI__X86_TARGET +#endif + +#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD) +// gcc doesn't support sse2 intrinsics unless you compile with -msse2, +// which in turn means it gets to use SSE2 everywhere. This is unfortunate, +// but previous attempts to provide the SSE2 functions with runtime +// detection caused numerous issues. The way architecture extensions are +// exposed in GCC/Clang is, sadly, not really suited for one-file libs. +// New behavior: if compiled with -msse2, we use SSE2 without any +// detection; if not, we don't use it at all. +#define STBI_NO_SIMD +#endif + +#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD) +// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET +// +// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the +// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant. +// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not +// simultaneously enabling "-mstackrealign". +// +// See https://github.com/nothings/stb/issues/81 for more information. +// +// So default to no SSE2 on 32-bit MinGW. If you've read this far and added +// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2. +#define STBI_NO_SIMD +#endif + +#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET)) +#define STBI_SSE2 +#include + +#ifdef _MSC_VER + +#if _MSC_VER >= 1400 // not VC6 +#include // __cpuid +static int stbi__cpuid3(void) +{ + int info[4]; + __cpuid(info,1); + return info[3]; +} +#else +static int stbi__cpuid3(void) +{ + int res; + __asm { + mov eax,1 + cpuid + mov res,edx + } + return res; +} +#endif + +#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + int info3 = stbi__cpuid3(); + return ((info3 >> 26) & 1) != 0; +} +#endif + +#else // assume GCC-style if not VC++ +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + // If we're even attempting to compile this on GCC/Clang, that means + // -msse2 is on, which means the compiler is allowed to use SSE2 + // instructions at will, and so are we. + return 1; +} +#endif + +#endif +#endif + +// ARM NEON +#if defined(STBI_NO_SIMD) && defined(STBI_NEON) +#undef STBI_NEON +#endif + +#ifdef STBI_NEON +#include +#ifdef _MSC_VER +#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name +#else +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) +#endif +#endif + +#ifndef STBI_SIMD_ALIGN +#define STBI_SIMD_ALIGN(type, name) type name +#endif + +#ifndef STBI_MAX_DIMENSIONS +#define STBI_MAX_DIMENSIONS (1 << 24) +#endif + +/////////////////////////////////////////////// +// +// stbi__context struct and start_xxx functions + +// stbi__context structure is our basic context used by all images, so it +// contains all the IO context, plus some basic image information +typedef struct +{ + stbi__uint32 img_x, img_y; + int img_n, img_out_n; + + stbi_io_callbacks io; + void *io_user_data; + + int read_from_callbacks; + int buflen; + stbi_uc buffer_start[128]; + int callback_already_read; + + stbi_uc *img_buffer, *img_buffer_end; + stbi_uc *img_buffer_original, *img_buffer_original_end; +} stbi__context; + + +static void stbi__refill_buffer(stbi__context *s); + +// initialize a memory-decode context +static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len) +{ + s->io.read = NULL; + s->read_from_callbacks = 0; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer; + s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len; +} + +// initialize a callback-based context +static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user) +{ + s->io = *c; + s->io_user_data = user; + s->buflen = sizeof(s->buffer_start); + s->read_from_callbacks = 1; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = s->buffer_start; + stbi__refill_buffer(s); + s->img_buffer_original_end = s->img_buffer_end; +} + +#ifndef STBI_NO_STDIO + +static int stbi__stdio_read(void *user, char *data, int size) +{ + return (int) fread(data,1,size,(FILE*) user); +} + +static void stbi__stdio_skip(void *user, int n) +{ + int ch; + fseek((FILE*) user, n, SEEK_CUR); + ch = fgetc((FILE*) user); /* have to read a byte to reset feof()'s flag */ + if (ch != EOF) { + ungetc(ch, (FILE *) user); /* push byte back onto stream if valid. */ + } +} + +static int stbi__stdio_eof(void *user) +{ + return feof((FILE*) user) || ferror((FILE *) user); +} + +static stbi_io_callbacks stbi__stdio_callbacks = +{ + stbi__stdio_read, + stbi__stdio_skip, + stbi__stdio_eof, +}; + +static void stbi__start_file(stbi__context *s, FILE *f) +{ + stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f); +} + +//static void stop_file(stbi__context *s) { } + +#endif // !STBI_NO_STDIO + +static void stbi__rewind(stbi__context *s) +{ + // conceptually rewind SHOULD rewind to the beginning of the stream, + // but we just rewind to the beginning of the initial buffer, because + // we only use it after doing 'test', which only ever looks at at most 92 bytes + s->img_buffer = s->img_buffer_original; + s->img_buffer_end = s->img_buffer_original_end; +} + +enum +{ + STBI_ORDER_RGB, + STBI_ORDER_BGR +}; + +typedef struct +{ + int bits_per_channel; + int num_channels; + int channel_order; +} stbi__result_info; + +#ifndef STBI_NO_JPEG +static int stbi__jpeg_test(stbi__context *s); +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNG +static int stbi__png_test(stbi__context *s); +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__png_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_BMP +static int stbi__bmp_test(stbi__context *s); +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_TGA +static int stbi__tga_test(stbi__context *s); +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s); +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc); +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__psd_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_HDR +static int stbi__hdr_test(stbi__context *s); +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_test(stbi__context *s); +static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_GIF +static int stbi__gif_test(stbi__context *s); +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNM +static int stbi__pnm_test(stbi__context *s); +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__pnm_is16(stbi__context *s); +#endif + +static +#ifdef STBI_THREAD_LOCAL +STBI_THREAD_LOCAL +#endif +const char *stbi__g_failure_reason; + +STBIDEF const char *stbi_failure_reason(void) +{ + return stbi__g_failure_reason; +} + +#ifndef STBI_NO_FAILURE_STRINGS +static int stbi__err(const char *str) +{ + stbi__g_failure_reason = str; + return 0; +} +#endif + +static void *stbi__malloc(size_t size) +{ + return STBI_MALLOC(size); +} + +// stb_image uses ints pervasively, including for offset calculations. +// therefore the largest decoded image size we can support with the +// current code, even on 64-bit targets, is INT_MAX. this is not a +// significant limitation for the intended use case. +// +// we do, however, need to make sure our size calculations don't +// overflow. hence a few helper functions for size calculations that +// multiply integers together, making sure that they're non-negative +// and no overflow occurs. + +// return 1 if the sum is valid, 0 on overflow. +// negative terms are considered invalid. +static int stbi__addsizes_valid(int a, int b) +{ + if (b < 0) return 0; + // now 0 <= b <= INT_MAX, hence also + // 0 <= INT_MAX - b <= INTMAX. + // And "a + b <= INT_MAX" (which might overflow) is the + // same as a <= INT_MAX - b (no overflow) + return a <= INT_MAX - b; +} + +// returns 1 if the product is valid, 0 on overflow. +// negative factors are considered invalid. +static int stbi__mul2sizes_valid(int a, int b) +{ + if (a < 0 || b < 0) return 0; + if (b == 0) return 1; // mul-by-0 is always safe + // portable way to check for no overflows in a*b + return a <= INT_MAX/b; +} + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow +static int stbi__mad2sizes_valid(int a, int b, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add); +} +#endif + +// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow +static int stbi__mad3sizes_valid(int a, int b, int c, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__addsizes_valid(a*b*c, add); +} + +// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM) +static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add); +} +#endif + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// mallocs with size overflow checking +static void *stbi__malloc_mad2(int a, int b, int add) +{ + if (!stbi__mad2sizes_valid(a, b, add)) return NULL; + return stbi__malloc(a*b + add); +} +#endif + +static void *stbi__malloc_mad3(int a, int b, int c, int add) +{ + if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL; + return stbi__malloc(a*b*c + add); +} + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM) +static void *stbi__malloc_mad4(int a, int b, int c, int d, int add) +{ + if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL; + return stbi__malloc(a*b*c*d + add); +} +#endif + +// returns 1 if the sum of two signed ints is valid (between -2^31 and 2^31-1 inclusive), 0 on overflow. +static int stbi__addints_valid(int a, int b) +{ + if ((a >= 0) != (b >= 0)) return 1; // a and b have different signs, so no overflow + if (a < 0 && b < 0) return a >= INT_MIN - b; // same as a + b >= INT_MIN; INT_MIN - b cannot overflow since b < 0. + return a <= INT_MAX - b; +} + +// returns 1 if the product of two ints fits in a signed short, 0 on overflow. +static int stbi__mul2shorts_valid(int a, int b) +{ + if (b == 0 || b == -1) return 1; // multiplication by 0 is always 0; check for -1 so SHRT_MIN/b doesn't overflow + if ((a >= 0) == (b >= 0)) return a <= SHRT_MAX/b; // product is positive, so similar to mul2sizes_valid + if (b < 0) return a <= SHRT_MIN / b; // same as a * b >= SHRT_MIN + return a >= SHRT_MIN / b; +} + +// stbi__err - error +// stbi__errpf - error returning pointer to float +// stbi__errpuc - error returning pointer to unsigned char + +#ifdef STBI_NO_FAILURE_STRINGS + #define stbi__err(x,y) 0 +#elif defined(STBI_FAILURE_USERMSG) + #define stbi__err(x,y) stbi__err(y) +#else + #define stbi__err(x,y) stbi__err(x) +#endif + +#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL)) +#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL)) + +STBIDEF void stbi_image_free(void *retval_from_stbi_load) +{ + STBI_FREE(retval_from_stbi_load); +} + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp); +#endif + +#ifndef STBI_NO_HDR +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp); +#endif + +static int stbi__vertically_flip_on_load_global = 0; + +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_global = flag_true_if_should_flip; +} + +#ifndef STBI_THREAD_LOCAL +#define stbi__vertically_flip_on_load stbi__vertically_flip_on_load_global +#else +static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set; + +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_local = flag_true_if_should_flip; + stbi__vertically_flip_on_load_set = 1; +} + +#define stbi__vertically_flip_on_load (stbi__vertically_flip_on_load_set \ + ? stbi__vertically_flip_on_load_local \ + : stbi__vertically_flip_on_load_global) +#endif // STBI_THREAD_LOCAL + +static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields + ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed + ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order + ri->num_channels = 0; + + // test the formats with a very explicit header first (at least a FOURCC + // or distinctive magic number first) + #ifndef STBI_NO_PNG + if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_BMP + if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_GIF + if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PSD + if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc); + #else + STBI_NOTUSED(bpc); + #endif + #ifndef STBI_NO_PIC + if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri); + #endif + + // then the formats that can end up attempting to load with just 1 or 2 + // bytes matching expectations; these are prone to false positives, so + // try them later + #ifndef STBI_NO_JPEG + if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PNM + if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri); + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri); + return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp); + } + #endif + + #ifndef STBI_NO_TGA + // test tga last because it's a crappy test! + if (stbi__tga_test(s)) + return stbi__tga_load(s,x,y,comp,req_comp, ri); + #endif + + return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt"); +} + +static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi_uc *reduced; + + reduced = (stbi_uc *) stbi__malloc(img_len); + if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling + + STBI_FREE(orig); + return reduced; +} + +static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi__uint16 *enlarged; + + enlarged = (stbi__uint16 *) stbi__malloc(img_len*2); + if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff + + STBI_FREE(orig); + return enlarged; +} + +static void stbi__vertical_flip(void *image, int w, int h, int bytes_per_pixel) +{ + int row; + size_t bytes_per_row = (size_t)w * bytes_per_pixel; + stbi_uc temp[2048]; + stbi_uc *bytes = (stbi_uc *)image; + + for (row = 0; row < (h>>1); row++) { + stbi_uc *row0 = bytes + row*bytes_per_row; + stbi_uc *row1 = bytes + (h - row - 1)*bytes_per_row; + // swap row0 with row1 + size_t bytes_left = bytes_per_row; + while (bytes_left) { + size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp); + memcpy(temp, row0, bytes_copy); + memcpy(row0, row1, bytes_copy); + memcpy(row1, temp, bytes_copy); + row0 += bytes_copy; + row1 += bytes_copy; + bytes_left -= bytes_copy; + } + } +} + +#ifndef STBI_NO_GIF +static void stbi__vertical_flip_slices(void *image, int w, int h, int z, int bytes_per_pixel) +{ + int slice; + int slice_size = w * h * bytes_per_pixel; + + stbi_uc *bytes = (stbi_uc *)image; + for (slice = 0; slice < z; ++slice) { + stbi__vertical_flip(bytes, w, h, bytes_per_pixel); + bytes += slice_size; + } +} +#endif + +static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 8) { + result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 8; + } + + // @TODO: move stbi__convert_format to here + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc)); + } + + return (unsigned char *) result; +} + +static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 16) { + result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 16; + } + + // @TODO: move stbi__convert_format16 to here + // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16)); + } + + return (stbi__uint16 *) result; +} + +#if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR) +static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp) +{ + if (stbi__vertically_flip_on_load && result != NULL) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(float)); + } +} +#endif + +#ifndef STBI_NO_STDIO + +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) +STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide); +STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default); +#endif + +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input) +{ + return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL); +} +#endif + +static FILE *stbi__fopen(char const *filename, char const *mode) +{ + FILE *f; +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) + wchar_t wMode[64]; + wchar_t wFilename[1024]; + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename))) + return 0; + + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode))) + return 0; + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != _wfopen_s(&f, wFilename, wMode)) + f = 0; +#else + f = _wfopen(wFilename, wMode); +#endif + +#elif defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != fopen_s(&f, filename, mode)) + f=0; +#else + f = fopen(filename, mode); +#endif + return f; +} + + +STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + unsigned char *result; + if (!f) return stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__uint16 *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + stbi__uint16 *result; + if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file_16(f,x,y,comp,req_comp); + fclose(f); + return result; +} + + +#endif //!STBI_NO_STDIO + +STBIDEF stbi_us *stbi_load_16_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *)clbk, user); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_mem(&s,buffer,len); + + result = (unsigned char*) stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp); + if (stbi__vertically_flip_on_load) { + stbi__vertical_flip_slices( result, *x, *y, *z, *comp ); + } + + return result; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *data; + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + stbi__result_info ri; + float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri); + if (hdr_data) + stbi__float_postprocess(hdr_data,x,y,comp,req_comp); + return hdr_data; + } + #endif + data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp); + if (data) + return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp); + return stbi__errpf("unknown image type", "Image not of any known type, or corrupt"); +} + +STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_STDIO +STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + float *result; + FILE *f = stbi__fopen(filename, "rb"); + if (!f) return stbi__errpf("can't fopen", "Unable to open file"); + result = stbi_loadf_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_file(&s,f); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} +#endif // !STBI_NO_STDIO + +#endif // !STBI_NO_LINEAR + +// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is +// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always +// reports false! + +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(buffer); + STBI_NOTUSED(len); + return 0; + #endif +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result=0; + if (f) { + result = stbi_is_hdr_from_file(f); + fclose(f); + } + return result; +} + +STBIDEF int stbi_is_hdr_from_file(FILE *f) +{ + #ifndef STBI_NO_HDR + long pos = ftell(f); + int res; + stbi__context s; + stbi__start_file(&s,f); + res = stbi__hdr_test(&s); + fseek(f, pos, SEEK_SET); + return res; + #else + STBI_NOTUSED(f); + return 0; + #endif +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(clbk); + STBI_NOTUSED(user); + return 0; + #endif +} + +#ifndef STBI_NO_LINEAR +static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f; + +STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; } +STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; } +#endif + +static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f; + +STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; } +STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; } + + +////////////////////////////////////////////////////////////////////////////// +// +// Common code used by all image loaders +// + +enum +{ + STBI__SCAN_load=0, + STBI__SCAN_type, + STBI__SCAN_header +}; + +static void stbi__refill_buffer(stbi__context *s) +{ + int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen); + s->callback_already_read += (int) (s->img_buffer - s->img_buffer_original); + if (n == 0) { + // at end of file, treat same as if from memory, but need to handle case + // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file + s->read_from_callbacks = 0; + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start+1; + *s->img_buffer = 0; + } else { + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start + n; + } +} + +stbi_inline static stbi_uc stbi__get8(stbi__context *s) +{ + if (s->img_buffer < s->img_buffer_end) + return *s->img_buffer++; + if (s->read_from_callbacks) { + stbi__refill_buffer(s); + return *s->img_buffer++; + } + return 0; +} + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +stbi_inline static int stbi__at_eof(stbi__context *s) +{ + if (s->io.read) { + if (!(s->io.eof)(s->io_user_data)) return 0; + // if feof() is true, check if buffer = end + // special case: we've only got the special 0 character at the end + if (s->read_from_callbacks == 0) return 1; + } + + return s->img_buffer >= s->img_buffer_end; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) +// nothing +#else +static void stbi__skip(stbi__context *s, int n) +{ + if (n == 0) return; // already there! + if (n < 0) { + s->img_buffer = s->img_buffer_end; + return; + } + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + s->img_buffer = s->img_buffer_end; + (s->io.skip)(s->io_user_data, n - blen); + return; + } + } + s->img_buffer += n; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM) +// nothing +#else +static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n) +{ + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + int res, count; + + memcpy(buffer, s->img_buffer, blen); + + count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen); + res = (count == (n-blen)); + s->img_buffer = s->img_buffer_end; + return res; + } + } + + if (s->img_buffer+n <= s->img_buffer_end) { + memcpy(buffer, s->img_buffer, n); + s->img_buffer += n; + return 1; + } else + return 0; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static int stbi__get16be(stbi__context *s) +{ + int z = stbi__get8(s); + return (z << 8) + stbi__get8(s); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static stbi__uint32 stbi__get32be(stbi__context *s) +{ + stbi__uint32 z = stbi__get16be(s); + return (z << 16) + stbi__get16be(s); +} +#endif + +#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) +// nothing +#else +static int stbi__get16le(stbi__context *s) +{ + int z = stbi__get8(s); + return z + (stbi__get8(s) << 8); +} +#endif + +#ifndef STBI_NO_BMP +static stbi__uint32 stbi__get32le(stbi__context *s) +{ + stbi__uint32 z = stbi__get16le(s); + z += (stbi__uint32)stbi__get16le(s) << 16; + return z; +} +#endif + +#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +////////////////////////////////////////////////////////////////////////////// +// +// generic converter from built-in img_n to req_comp +// individual types do this automatically as much as possible (e.g. jpeg +// does all cases internally since it needs to colorspace convert anyway, +// and it never has alpha, so very few cases ). png can automatically +// interleave an alpha=255 channel, but falls back to this for other cases +// +// assume data buffer is malloced, so malloc a new one and free that one +// only failure mode is malloc failing + +static stbi_uc stbi__compute_y(int r, int g, int b) +{ + return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + unsigned char *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0); + if (good == NULL) { + STBI_FREE(data); + return stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + unsigned char *src = data + j * x * img_n ; + unsigned char *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=255; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=255; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=255; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = 255; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 stbi__compute_y_16(int r, int g, int b) +{ + return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + stbi__uint16 *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2); + if (good == NULL) { + STBI_FREE(data); + return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + stbi__uint16 *src = data + j * x * img_n ; + stbi__uint16 *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=0xffff; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=0xffff; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=0xffff; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = 0xffff; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return (stbi__uint16*) stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp) +{ + int i,k,n; + float *output; + if (!data) return NULL; + output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale); + } + } + if (n < comp) { + for (i=0; i < x*y; ++i) { + output[i*comp + n] = data[i*comp + n]/255.0f; + } + } + STBI_FREE(data); + return output; +} +#endif + +#ifndef STBI_NO_HDR +#define stbi__float2int(x) ((int) (x)) +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp) +{ + int i,k,n; + stbi_uc *output; + if (!data) return NULL; + output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + if (k < comp) { + float z = data[i*comp+k] * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + } + STBI_FREE(data); + return output; +} +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// "baseline" JPEG/JFIF decoder +// +// simple implementation +// - doesn't support delayed output of y-dimension +// - simple interface (only one output format: 8-bit interleaved RGB) +// - doesn't try to recover corrupt jpegs +// - doesn't allow partial loading, loading multiple at once +// - still fast on x86 (copying globals into locals doesn't help x86) +// - allocates lots of intermediate memory (full size of all components) +// - non-interleaved case requires this anyway +// - allows good upsampling (see next) +// high-quality +// - upsampled channels are bilinearly interpolated, even across blocks +// - quality integer IDCT derived from IJG's 'slow' +// performance +// - fast huffman; reasonable integer IDCT +// - some SIMD kernels for common paths on targets with SSE2/NEON +// - uses a lot of intermediate memory, could cache poorly + +#ifndef STBI_NO_JPEG + +// huffman decoding acceleration +#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache + +typedef struct +{ + stbi_uc fast[1 << FAST_BITS]; + // weirdly, repacking this into AoS is a 10% speed loss, instead of a win + stbi__uint16 code[256]; + stbi_uc values[256]; + stbi_uc size[257]; + unsigned int maxcode[18]; + int delta[17]; // old 'firstsymbol' - old 'firstcode' +} stbi__huffman; + +typedef struct +{ + stbi__context *s; + stbi__huffman huff_dc[4]; + stbi__huffman huff_ac[4]; + stbi__uint16 dequant[4][64]; + stbi__int16 fast_ac[4][1 << FAST_BITS]; + +// sizes for components, interleaved MCUs + int img_h_max, img_v_max; + int img_mcu_x, img_mcu_y; + int img_mcu_w, img_mcu_h; + +// definition of jpeg image component + struct + { + int id; + int h,v; + int tq; + int hd,ha; + int dc_pred; + + int x,y,w2,h2; + stbi_uc *data; + void *raw_data, *raw_coeff; + stbi_uc *linebuf; + short *coeff; // progressive only + int coeff_w, coeff_h; // number of 8x8 coefficient blocks + } img_comp[4]; + + stbi__uint32 code_buffer; // jpeg entropy-coded buffer + int code_bits; // number of valid bits + unsigned char marker; // marker seen while filling entropy buffer + int nomore; // flag if we saw a marker so must stop + + int progressive; + int spec_start; + int spec_end; + int succ_high; + int succ_low; + int eob_run; + int jfif; + int app14_color_transform; // Adobe APP14 tag + int rgb; + + int scan_n, order[4]; + int restart_interval, todo; + +// kernels + void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]); + void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step); + stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs); +} stbi__jpeg; + +static int stbi__build_huffman(stbi__huffman *h, int *count) +{ + int i,j,k=0; + unsigned int code; + // build size list for each symbol (from JPEG spec) + for (i=0; i < 16; ++i) { + for (j=0; j < count[i]; ++j) { + h->size[k++] = (stbi_uc) (i+1); + if(k >= 257) return stbi__err("bad size list","Corrupt JPEG"); + } + } + h->size[k] = 0; + + // compute actual symbols (from jpeg spec) + code = 0; + k = 0; + for(j=1; j <= 16; ++j) { + // compute delta to add to code to compute symbol id + h->delta[j] = k - code; + if (h->size[k] == j) { + while (h->size[k] == j) + h->code[k++] = (stbi__uint16) (code++); + if (code-1 >= (1u << j)) return stbi__err("bad code lengths","Corrupt JPEG"); + } + // compute largest code + 1 for this size, preshifted as needed later + h->maxcode[j] = code << (16-j); + code <<= 1; + } + h->maxcode[j] = 0xffffffff; + + // build non-spec acceleration table; 255 is flag for not-accelerated + memset(h->fast, 255, 1 << FAST_BITS); + for (i=0; i < k; ++i) { + int s = h->size[i]; + if (s <= FAST_BITS) { + int c = h->code[i] << (FAST_BITS-s); + int m = 1 << (FAST_BITS-s); + for (j=0; j < m; ++j) { + h->fast[c+j] = (stbi_uc) i; + } + } + } + return 1; +} + +// build a table that decodes both magnitude and value of small ACs in +// one go. +static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h) +{ + int i; + for (i=0; i < (1 << FAST_BITS); ++i) { + stbi_uc fast = h->fast[i]; + fast_ac[i] = 0; + if (fast < 255) { + int rs = h->values[fast]; + int run = (rs >> 4) & 15; + int magbits = rs & 15; + int len = h->size[fast]; + + if (magbits && len + magbits <= FAST_BITS) { + // magnitude code followed by receive_extend code + int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits); + int m = 1 << (magbits - 1); + if (k < m) k += (~0U << magbits) + 1; + // if the result is small enough, we can fit it in fast_ac table + if (k >= -128 && k <= 127) + fast_ac[i] = (stbi__int16) ((k * 256) + (run * 16) + (len + magbits)); + } + } + } +} + +static void stbi__grow_buffer_unsafe(stbi__jpeg *j) +{ + do { + unsigned int b = j->nomore ? 0 : stbi__get8(j->s); + if (b == 0xff) { + int c = stbi__get8(j->s); + while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes + if (c != 0) { + j->marker = (unsigned char) c; + j->nomore = 1; + return; + } + } + j->code_buffer |= b << (24 - j->code_bits); + j->code_bits += 8; + } while (j->code_bits <= 24); +} + +// (1 << n) - 1 +static const stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535}; + +// decode a jpeg huffman value from the bitstream +stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h) +{ + unsigned int temp; + int c,k; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + // look at the top FAST_BITS and determine what symbol ID it is, + // if the code is <= FAST_BITS + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + k = h->fast[c]; + if (k < 255) { + int s = h->size[k]; + if (s > j->code_bits) + return -1; + j->code_buffer <<= s; + j->code_bits -= s; + return h->values[k]; + } + + // naive test is to shift the code_buffer down so k bits are + // valid, then test against maxcode. To speed this up, we've + // preshifted maxcode left so that it has (16-k) 0s at the + // end; in other words, regardless of the number of bits, it + // wants to be compared against something shifted to have 16; + // that way we don't need to shift inside the loop. + temp = j->code_buffer >> 16; + for (k=FAST_BITS+1 ; ; ++k) + if (temp < h->maxcode[k]) + break; + if (k == 17) { + // error! code not found + j->code_bits -= 16; + return -1; + } + + if (k > j->code_bits) + return -1; + + // convert the huffman code to the symbol id + c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k]; + if(c < 0 || c >= 256) // symbol id out of bounds! + return -1; + STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]); + + // convert the id to a symbol + j->code_bits -= k; + j->code_buffer <<= k; + return h->values[c]; +} + +// bias[n] = (-1<code_bits < n) stbi__grow_buffer_unsafe(j); + if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing + + sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative) + k = stbi_lrot(j->code_buffer, n); + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k + (stbi__jbias[n] & (sgn - 1)); +} + +// get some unsigned bits +stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n) +{ + unsigned int k; + if (j->code_bits < n) stbi__grow_buffer_unsafe(j); + if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing + k = stbi_lrot(j->code_buffer, n); + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k; +} + +stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j) +{ + unsigned int k; + if (j->code_bits < 1) stbi__grow_buffer_unsafe(j); + if (j->code_bits < 1) return 0; // ran out of bits from stream, return 0s intead of continuing + k = j->code_buffer; + j->code_buffer <<= 1; + --j->code_bits; + return k & 0x80000000; +} + +// given a value that's at position X in the zigzag stream, +// where does it appear in the 8x8 matrix coded as row-major? +static const stbi_uc stbi__jpeg_dezigzag[64+15] = +{ + 0, 1, 8, 16, 9, 2, 3, 10, + 17, 24, 32, 25, 18, 11, 4, 5, + 12, 19, 26, 33, 40, 48, 41, 34, + 27, 20, 13, 6, 7, 14, 21, 28, + 35, 42, 49, 56, 57, 50, 43, 36, + 29, 22, 15, 23, 30, 37, 44, 51, + 58, 59, 52, 45, 38, 31, 39, 46, + 53, 60, 61, 54, 47, 55, 62, 63, + // let corrupt input sample past end + 63, 63, 63, 63, 63, 63, 63, 63, + 63, 63, 63, 63, 63, 63, 63 +}; + +// decode one 64-entry block-- +static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant) +{ + int diff,dc,k; + int t; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + t = stbi__jpeg_huff_decode(j, hdc); + if (t < 0 || t > 15) return stbi__err("bad huffman code","Corrupt JPEG"); + + // 0 all the ac values now so we can do it 32-bits at a time + memset(data,0,64*sizeof(data[0])); + + diff = t ? stbi__extend_receive(j, t) : 0; + if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta","Corrupt JPEG"); + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + if (!stbi__mul2shorts_valid(dc, dequant[0])) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + data[0] = (short) (dc * dequant[0]); + + // decode AC components, see JPEG spec + k = 1; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available"); + j->code_buffer <<= s; + j->code_bits -= s; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) * dequant[zig]); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (rs != 0xf0) break; // end block + k += 16; + } else { + k += r; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]); + } + } + } while (k < 64); + return 1; +} + +static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b) +{ + int diff,dc; + int t; + if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + if (j->succ_high == 0) { + // first scan for DC coefficient, must be first + memset(data,0,64*sizeof(data[0])); // 0 all the ac values now + t = stbi__jpeg_huff_decode(j, hdc); + if (t < 0 || t > 15) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + diff = t ? stbi__extend_receive(j, t) : 0; + + if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta", "Corrupt JPEG"); + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + if (!stbi__mul2shorts_valid(dc, 1 << j->succ_low)) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + data[0] = (short) (dc * (1 << j->succ_low)); + } else { + // refinement scan for DC coefficient + if (stbi__jpeg_get_bit(j)) + data[0] += (short) (1 << j->succ_low); + } + return 1; +} + +// @OPTIMIZE: store non-zigzagged during the decode passes, +// and only de-zigzag when dequantizing +static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac) +{ + int k; + if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->succ_high == 0) { + int shift = j->succ_low; + + if (j->eob_run) { + --j->eob_run; + return 1; + } + + k = j->spec_start; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available"); + j->code_buffer <<= s; + j->code_bits -= s; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) * (1 << shift)); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r); + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + --j->eob_run; + break; + } + k += 16; + } else { + k += r; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) * (1 << shift)); + } + } + } while (k <= j->spec_end); + } else { + // refinement scan for these AC coefficients + + short bit = (short) (1 << j->succ_low); + + if (j->eob_run) { + --j->eob_run; + for (k = j->spec_start; k <= j->spec_end; ++k) { + short *p = &data[stbi__jpeg_dezigzag[k]]; + if (*p != 0) + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } + } else { + k = j->spec_start; + do { + int r,s; + int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r) - 1; + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + r = 64; // force end of block + } else { + // r=15 s=0 should write 16 0s, so we just do + // a run of 15 0s and then write s (which is 0), + // so we don't have to do anything special here + } + } else { + if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG"); + // sign bit + if (stbi__jpeg_get_bit(j)) + s = bit; + else + s = -bit; + } + + // advance by r + while (k <= j->spec_end) { + short *p = &data[stbi__jpeg_dezigzag[k++]]; + if (*p != 0) { + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } else { + if (r == 0) { + *p = (short) s; + break; + } + --r; + } + } + } while (k <= j->spec_end); + } + } + return 1; +} + +// take a -128..127 value and stbi__clamp it and convert to 0..255 +stbi_inline static stbi_uc stbi__clamp(int x) +{ + // trick to use a single test to catch both cases + if ((unsigned int) x > 255) { + if (x < 0) return 0; + if (x > 255) return 255; + } + return (stbi_uc) x; +} + +#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5))) +#define stbi__fsh(x) ((x) * 4096) + +// derived from jidctint -- DCT_ISLOW +#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \ + int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \ + p2 = s2; \ + p3 = s6; \ + p1 = (p2+p3) * stbi__f2f(0.5411961f); \ + t2 = p1 + p3*stbi__f2f(-1.847759065f); \ + t3 = p1 + p2*stbi__f2f( 0.765366865f); \ + p2 = s0; \ + p3 = s4; \ + t0 = stbi__fsh(p2+p3); \ + t1 = stbi__fsh(p2-p3); \ + x0 = t0+t3; \ + x3 = t0-t3; \ + x1 = t1+t2; \ + x2 = t1-t2; \ + t0 = s7; \ + t1 = s5; \ + t2 = s3; \ + t3 = s1; \ + p3 = t0+t2; \ + p4 = t1+t3; \ + p1 = t0+t3; \ + p2 = t1+t2; \ + p5 = (p3+p4)*stbi__f2f( 1.175875602f); \ + t0 = t0*stbi__f2f( 0.298631336f); \ + t1 = t1*stbi__f2f( 2.053119869f); \ + t2 = t2*stbi__f2f( 3.072711026f); \ + t3 = t3*stbi__f2f( 1.501321110f); \ + p1 = p5 + p1*stbi__f2f(-0.899976223f); \ + p2 = p5 + p2*stbi__f2f(-2.562915447f); \ + p3 = p3*stbi__f2f(-1.961570560f); \ + p4 = p4*stbi__f2f(-0.390180644f); \ + t3 += p1+p4; \ + t2 += p2+p3; \ + t1 += p2+p4; \ + t0 += p1+p3; + +static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64]) +{ + int i,val[64],*v=val; + stbi_uc *o; + short *d = data; + + // columns + for (i=0; i < 8; ++i,++d, ++v) { + // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing + if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0 + && d[40]==0 && d[48]==0 && d[56]==0) { + // no shortcut 0 seconds + // (1|2|3|4|5|6|7)==0 0 seconds + // all separate -0.047 seconds + // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds + int dcterm = d[0]*4; + v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm; + } else { + STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56]) + // constants scaled things up by 1<<12; let's bring them back + // down, but keep 2 extra bits of precision + x0 += 512; x1 += 512; x2 += 512; x3 += 512; + v[ 0] = (x0+t3) >> 10; + v[56] = (x0-t3) >> 10; + v[ 8] = (x1+t2) >> 10; + v[48] = (x1-t2) >> 10; + v[16] = (x2+t1) >> 10; + v[40] = (x2-t1) >> 10; + v[24] = (x3+t0) >> 10; + v[32] = (x3-t0) >> 10; + } + } + + for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) { + // no fast case since the first 1D IDCT spread components out + STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7]) + // constants scaled things up by 1<<12, plus we had 1<<2 from first + // loop, plus horizontal and vertical each scale by sqrt(8) so together + // we've got an extra 1<<3, so 1<<17 total we need to remove. + // so we want to round that, which means adding 0.5 * 1<<17, + // aka 65536. Also, we'll end up with -128 to 127 that we want + // to encode as 0..255 by adding 128, so we'll add that before the shift + x0 += 65536 + (128<<17); + x1 += 65536 + (128<<17); + x2 += 65536 + (128<<17); + x3 += 65536 + (128<<17); + // tried computing the shifts into temps, or'ing the temps to see + // if any were out of range, but that was slower + o[0] = stbi__clamp((x0+t3) >> 17); + o[7] = stbi__clamp((x0-t3) >> 17); + o[1] = stbi__clamp((x1+t2) >> 17); + o[6] = stbi__clamp((x1-t2) >> 17); + o[2] = stbi__clamp((x2+t1) >> 17); + o[5] = stbi__clamp((x2-t1) >> 17); + o[3] = stbi__clamp((x3+t0) >> 17); + o[4] = stbi__clamp((x3-t0) >> 17); + } +} + +#ifdef STBI_SSE2 +// sse2 integer IDCT. not the fastest possible implementation but it +// produces bit-identical results to the generic C version so it's +// fully "transparent". +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + // This is constructed to match our regular (generic) integer IDCT exactly. + __m128i row0, row1, row2, row3, row4, row5, row6, row7; + __m128i tmp; + + // dot product constant: even elems=x, odd elems=y + #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y)) + + // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit) + // out(1) = c1[even]*x + c1[odd]*y + #define dct_rot(out0,out1, x,y,c0,c1) \ + __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \ + __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \ + __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \ + __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \ + __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \ + __m128i out1##_h = _mm_madd_epi16(c0##hi, c1) + + // out = in << 12 (in 16-bit, out 32-bit) + #define dct_widen(out, in) \ + __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \ + __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4) + + // wide add + #define dct_wadd(out, a, b) \ + __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_add_epi32(a##_h, b##_h) + + // wide sub + #define dct_wsub(out, a, b) \ + __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_sub_epi32(a##_h, b##_h) + + // butterfly a/b, add bias, then shift by "s" and pack + #define dct_bfly32o(out0, out1, a,b,bias,s) \ + { \ + __m128i abiased_l = _mm_add_epi32(a##_l, bias); \ + __m128i abiased_h = _mm_add_epi32(a##_h, bias); \ + dct_wadd(sum, abiased, b); \ + dct_wsub(dif, abiased, b); \ + out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \ + out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \ + } + + // 8-bit interleave step (for transposes) + #define dct_interleave8(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi8(a, b); \ + b = _mm_unpackhi_epi8(tmp, b) + + // 16-bit interleave step (for transposes) + #define dct_interleave16(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi16(a, b); \ + b = _mm_unpackhi_epi16(tmp, b) + + #define dct_pass(bias,shift) \ + { \ + /* even part */ \ + dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \ + __m128i sum04 = _mm_add_epi16(row0, row4); \ + __m128i dif04 = _mm_sub_epi16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \ + dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \ + __m128i sum17 = _mm_add_epi16(row1, row7); \ + __m128i sum35 = _mm_add_epi16(row3, row5); \ + dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \ + dct_wadd(x4, y0o, y4o); \ + dct_wadd(x5, y1o, y5o); \ + dct_wadd(x6, y2o, y5o); \ + dct_wadd(x7, y3o, y4o); \ + dct_bfly32o(row0,row7, x0,x7,bias,shift); \ + dct_bfly32o(row1,row6, x1,x6,bias,shift); \ + dct_bfly32o(row2,row5, x2,x5,bias,shift); \ + dct_bfly32o(row3,row4, x3,x4,bias,shift); \ + } + + __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f)); + __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f)); + __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f)); + __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f)); + __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f)); + __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f)); + __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f)); + __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f)); + + // rounding biases in column/row passes, see stbi__idct_block for explanation. + __m128i bias_0 = _mm_set1_epi32(512); + __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17)); + + // load + row0 = _mm_load_si128((const __m128i *) (data + 0*8)); + row1 = _mm_load_si128((const __m128i *) (data + 1*8)); + row2 = _mm_load_si128((const __m128i *) (data + 2*8)); + row3 = _mm_load_si128((const __m128i *) (data + 3*8)); + row4 = _mm_load_si128((const __m128i *) (data + 4*8)); + row5 = _mm_load_si128((const __m128i *) (data + 5*8)); + row6 = _mm_load_si128((const __m128i *) (data + 6*8)); + row7 = _mm_load_si128((const __m128i *) (data + 7*8)); + + // column pass + dct_pass(bias_0, 10); + + { + // 16bit 8x8 transpose pass 1 + dct_interleave16(row0, row4); + dct_interleave16(row1, row5); + dct_interleave16(row2, row6); + dct_interleave16(row3, row7); + + // transpose pass 2 + dct_interleave16(row0, row2); + dct_interleave16(row1, row3); + dct_interleave16(row4, row6); + dct_interleave16(row5, row7); + + // transpose pass 3 + dct_interleave16(row0, row1); + dct_interleave16(row2, row3); + dct_interleave16(row4, row5); + dct_interleave16(row6, row7); + } + + // row pass + dct_pass(bias_1, 17); + + { + // pack + __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7 + __m128i p1 = _mm_packus_epi16(row2, row3); + __m128i p2 = _mm_packus_epi16(row4, row5); + __m128i p3 = _mm_packus_epi16(row6, row7); + + // 8bit 8x8 transpose pass 1 + dct_interleave8(p0, p2); // a0e0a1e1... + dct_interleave8(p1, p3); // c0g0c1g1... + + // transpose pass 2 + dct_interleave8(p0, p1); // a0c0e0g0... + dct_interleave8(p2, p3); // b0d0f0h0... + + // transpose pass 3 + dct_interleave8(p0, p2); // a0b0c0d0... + dct_interleave8(p1, p3); // a4b4c4d4... + + // store + _mm_storel_epi64((__m128i *) out, p0); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p2); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p1); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p3); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e)); + } + +#undef dct_const +#undef dct_rot +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_interleave8 +#undef dct_interleave16 +#undef dct_pass +} + +#endif // STBI_SSE2 + +#ifdef STBI_NEON + +// NEON integer IDCT. should produce bit-identical +// results to the generic C version. +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + int16x8_t row0, row1, row2, row3, row4, row5, row6, row7; + + int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f)); + int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f)); + int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f)); + int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f)); + int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f)); + int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f)); + int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f)); + int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f)); + int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f)); + int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f)); + int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f)); + int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f)); + +#define dct_long_mul(out, inq, coeff) \ + int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff) + +#define dct_long_mac(out, acc, inq, coeff) \ + int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff) + +#define dct_widen(out, inq) \ + int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \ + int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12) + +// wide add +#define dct_wadd(out, a, b) \ + int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vaddq_s32(a##_h, b##_h) + +// wide sub +#define dct_wsub(out, a, b) \ + int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vsubq_s32(a##_h, b##_h) + +// butterfly a/b, then shift using "shiftop" by "s" and pack +#define dct_bfly32o(out0,out1, a,b,shiftop,s) \ + { \ + dct_wadd(sum, a, b); \ + dct_wsub(dif, a, b); \ + out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \ + out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \ + } + +#define dct_pass(shiftop, shift) \ + { \ + /* even part */ \ + int16x8_t sum26 = vaddq_s16(row2, row6); \ + dct_long_mul(p1e, sum26, rot0_0); \ + dct_long_mac(t2e, p1e, row6, rot0_1); \ + dct_long_mac(t3e, p1e, row2, rot0_2); \ + int16x8_t sum04 = vaddq_s16(row0, row4); \ + int16x8_t dif04 = vsubq_s16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + int16x8_t sum15 = vaddq_s16(row1, row5); \ + int16x8_t sum17 = vaddq_s16(row1, row7); \ + int16x8_t sum35 = vaddq_s16(row3, row5); \ + int16x8_t sum37 = vaddq_s16(row3, row7); \ + int16x8_t sumodd = vaddq_s16(sum17, sum35); \ + dct_long_mul(p5o, sumodd, rot1_0); \ + dct_long_mac(p1o, p5o, sum17, rot1_1); \ + dct_long_mac(p2o, p5o, sum35, rot1_2); \ + dct_long_mul(p3o, sum37, rot2_0); \ + dct_long_mul(p4o, sum15, rot2_1); \ + dct_wadd(sump13o, p1o, p3o); \ + dct_wadd(sump24o, p2o, p4o); \ + dct_wadd(sump23o, p2o, p3o); \ + dct_wadd(sump14o, p1o, p4o); \ + dct_long_mac(x4, sump13o, row7, rot3_0); \ + dct_long_mac(x5, sump24o, row5, rot3_1); \ + dct_long_mac(x6, sump23o, row3, rot3_2); \ + dct_long_mac(x7, sump14o, row1, rot3_3); \ + dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \ + dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \ + dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \ + dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \ + } + + // load + row0 = vld1q_s16(data + 0*8); + row1 = vld1q_s16(data + 1*8); + row2 = vld1q_s16(data + 2*8); + row3 = vld1q_s16(data + 3*8); + row4 = vld1q_s16(data + 4*8); + row5 = vld1q_s16(data + 5*8); + row6 = vld1q_s16(data + 6*8); + row7 = vld1q_s16(data + 7*8); + + // add DC bias + row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0)); + + // column pass + dct_pass(vrshrn_n_s32, 10); + + // 16bit 8x8 transpose + { +// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively. +// whether compilers actually get this is another story, sadly. +#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); } +#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); } + + // pass 1 + dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6 + dct_trn16(row2, row3); + dct_trn16(row4, row5); + dct_trn16(row6, row7); + + // pass 2 + dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4 + dct_trn32(row1, row3); + dct_trn32(row4, row6); + dct_trn32(row5, row7); + + // pass 3 + dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0 + dct_trn64(row1, row5); + dct_trn64(row2, row6); + dct_trn64(row3, row7); + +#undef dct_trn16 +#undef dct_trn32 +#undef dct_trn64 + } + + // row pass + // vrshrn_n_s32 only supports shifts up to 16, we need + // 17. so do a non-rounding shift of 16 first then follow + // up with a rounding shift by 1. + dct_pass(vshrn_n_s32, 16); + + { + // pack and round + uint8x8_t p0 = vqrshrun_n_s16(row0, 1); + uint8x8_t p1 = vqrshrun_n_s16(row1, 1); + uint8x8_t p2 = vqrshrun_n_s16(row2, 1); + uint8x8_t p3 = vqrshrun_n_s16(row3, 1); + uint8x8_t p4 = vqrshrun_n_s16(row4, 1); + uint8x8_t p5 = vqrshrun_n_s16(row5, 1); + uint8x8_t p6 = vqrshrun_n_s16(row6, 1); + uint8x8_t p7 = vqrshrun_n_s16(row7, 1); + + // again, these can translate into one instruction, but often don't. +#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); } +#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); } + + // sadly can't use interleaved stores here since we only write + // 8 bytes to each scan line! + + // 8x8 8-bit transpose pass 1 + dct_trn8_8(p0, p1); + dct_trn8_8(p2, p3); + dct_trn8_8(p4, p5); + dct_trn8_8(p6, p7); + + // pass 2 + dct_trn8_16(p0, p2); + dct_trn8_16(p1, p3); + dct_trn8_16(p4, p6); + dct_trn8_16(p5, p7); + + // pass 3 + dct_trn8_32(p0, p4); + dct_trn8_32(p1, p5); + dct_trn8_32(p2, p6); + dct_trn8_32(p3, p7); + + // store + vst1_u8(out, p0); out += out_stride; + vst1_u8(out, p1); out += out_stride; + vst1_u8(out, p2); out += out_stride; + vst1_u8(out, p3); out += out_stride; + vst1_u8(out, p4); out += out_stride; + vst1_u8(out, p5); out += out_stride; + vst1_u8(out, p6); out += out_stride; + vst1_u8(out, p7); + +#undef dct_trn8_8 +#undef dct_trn8_16 +#undef dct_trn8_32 + } + +#undef dct_long_mul +#undef dct_long_mac +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_pass +} + +#endif // STBI_NEON + +#define STBI__MARKER_none 0xff +// if there's a pending marker from the entropy stream, return that +// otherwise, fetch from the stream and get a marker. if there's no +// marker, return 0xff, which is never a valid marker value +static stbi_uc stbi__get_marker(stbi__jpeg *j) +{ + stbi_uc x; + if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; } + x = stbi__get8(j->s); + if (x != 0xff) return STBI__MARKER_none; + while (x == 0xff) + x = stbi__get8(j->s); // consume repeated 0xff fill bytes + return x; +} + +// in each scan, we'll have scan_n components, and the order +// of the components is specified by order[] +#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7) + +// after a restart interval, stbi__jpeg_reset the entropy decoder and +// the dc prediction +static void stbi__jpeg_reset(stbi__jpeg *j) +{ + j->code_bits = 0; + j->code_buffer = 0; + j->nomore = 0; + j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0; + j->marker = STBI__MARKER_none; + j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff; + j->eob_run = 0; + // no more than 1<<31 MCUs if no restart_interal? that's plenty safe, + // since we don't even allow 1<<30 pixels +} + +static int stbi__parse_entropy_coded_data(stbi__jpeg *z) +{ + stbi__jpeg_reset(z); + if (!z->progressive) { + if (z->scan_n == 1) { + int i,j; + STBI_SIMD_ALIGN(short, data[64]); + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + // if it's NOT a restart, then just bail, so we get corrupt data + // rather than no data + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + STBI_SIMD_ALIGN(short, data[64]); + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x)*8; + int y2 = (j*z->img_comp[n].v + y)*8; + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data); + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } else { + if (z->scan_n == 1) { + int i,j; + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + if (z->spec_start == 0) { + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } else { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha])) + return 0; + } + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x); + int y2 = (j*z->img_comp[n].v + y); + short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w); + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } +} + +static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant) +{ + int i; + for (i=0; i < 64; ++i) + data[i] *= dequant[i]; +} + +static void stbi__jpeg_finish(stbi__jpeg *z) +{ + if (z->progressive) { + // dequantize and idct the data + int i,j,n; + for (n=0; n < z->s->img_n; ++n) { + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]); + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + } + } + } + } +} + +static int stbi__process_marker(stbi__jpeg *z, int m) +{ + int L; + switch (m) { + case STBI__MARKER_none: // no marker found + return stbi__err("expected marker","Corrupt JPEG"); + + case 0xDD: // DRI - specify restart interval + if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG"); + z->restart_interval = stbi__get16be(z->s); + return 1; + + case 0xDB: // DQT - define quantization table + L = stbi__get16be(z->s)-2; + while (L > 0) { + int q = stbi__get8(z->s); + int p = q >> 4, sixteen = (p != 0); + int t = q & 15,i; + if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG"); + if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG"); + + for (i=0; i < 64; ++i) + z->dequant[t][stbi__jpeg_dezigzag[i]] = (stbi__uint16)(sixteen ? stbi__get16be(z->s) : stbi__get8(z->s)); + L -= (sixteen ? 129 : 65); + } + return L==0; + + case 0xC4: // DHT - define huffman table + L = stbi__get16be(z->s)-2; + while (L > 0) { + stbi_uc *v; + int sizes[16],i,n=0; + int q = stbi__get8(z->s); + int tc = q >> 4; + int th = q & 15; + if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG"); + for (i=0; i < 16; ++i) { + sizes[i] = stbi__get8(z->s); + n += sizes[i]; + } + if(n > 256) return stbi__err("bad DHT header","Corrupt JPEG"); // Loop over i < n would write past end of values! + L -= 17; + if (tc == 0) { + if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0; + v = z->huff_dc[th].values; + } else { + if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0; + v = z->huff_ac[th].values; + } + for (i=0; i < n; ++i) + v[i] = stbi__get8(z->s); + if (tc != 0) + stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th); + L -= n; + } + return L==0; + } + + // check for comment block or APP blocks + if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) { + L = stbi__get16be(z->s); + if (L < 2) { + if (m == 0xFE) + return stbi__err("bad COM len","Corrupt JPEG"); + else + return stbi__err("bad APP len","Corrupt JPEG"); + } + L -= 2; + + if (m == 0xE0 && L >= 5) { // JFIF APP0 segment + static const unsigned char tag[5] = {'J','F','I','F','\0'}; + int ok = 1; + int i; + for (i=0; i < 5; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 5; + if (ok) + z->jfif = 1; + } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment + static const unsigned char tag[6] = {'A','d','o','b','e','\0'}; + int ok = 1; + int i; + for (i=0; i < 6; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 6; + if (ok) { + stbi__get8(z->s); // version + stbi__get16be(z->s); // flags0 + stbi__get16be(z->s); // flags1 + z->app14_color_transform = stbi__get8(z->s); // color transform + L -= 6; + } + } + + stbi__skip(z->s, L); + return 1; + } + + return stbi__err("unknown marker","Corrupt JPEG"); +} + +// after we see SOS +static int stbi__process_scan_header(stbi__jpeg *z) +{ + int i; + int Ls = stbi__get16be(z->s); + z->scan_n = stbi__get8(z->s); + if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG"); + if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG"); + for (i=0; i < z->scan_n; ++i) { + int id = stbi__get8(z->s), which; + int q = stbi__get8(z->s); + for (which = 0; which < z->s->img_n; ++which) + if (z->img_comp[which].id == id) + break; + if (which == z->s->img_n) return 0; // no match + z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG"); + z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG"); + z->order[i] = which; + } + + { + int aa; + z->spec_start = stbi__get8(z->s); + z->spec_end = stbi__get8(z->s); // should be 63, but might be 0 + aa = stbi__get8(z->s); + z->succ_high = (aa >> 4); + z->succ_low = (aa & 15); + if (z->progressive) { + if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13) + return stbi__err("bad SOS", "Corrupt JPEG"); + } else { + if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG"); + if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG"); + z->spec_end = 63; + } + } + + return 1; +} + +static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why) +{ + int i; + for (i=0; i < ncomp; ++i) { + if (z->img_comp[i].raw_data) { + STBI_FREE(z->img_comp[i].raw_data); + z->img_comp[i].raw_data = NULL; + z->img_comp[i].data = NULL; + } + if (z->img_comp[i].raw_coeff) { + STBI_FREE(z->img_comp[i].raw_coeff); + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].coeff = 0; + } + if (z->img_comp[i].linebuf) { + STBI_FREE(z->img_comp[i].linebuf); + z->img_comp[i].linebuf = NULL; + } + } + return why; +} + +static int stbi__process_frame_header(stbi__jpeg *z, int scan) +{ + stbi__context *s = z->s; + int Lf,p,i,q, h_max=1,v_max=1,c; + Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG + p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline + s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG + s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + c = stbi__get8(s); + if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG"); + s->img_n = c; + for (i=0; i < c; ++i) { + z->img_comp[i].data = NULL; + z->img_comp[i].linebuf = NULL; + } + + if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG"); + + z->rgb = 0; + for (i=0; i < s->img_n; ++i) { + static const unsigned char rgb[3] = { 'R', 'G', 'B' }; + z->img_comp[i].id = stbi__get8(s); + if (s->img_n == 3 && z->img_comp[i].id == rgb[i]) + ++z->rgb; + q = stbi__get8(s); + z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG"); + z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG"); + z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG"); + } + + if (scan != STBI__SCAN_load) return 1; + + if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode"); + + for (i=0; i < s->img_n; ++i) { + if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h; + if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v; + } + + // check that plane subsampling factors are integer ratios; our resamplers can't deal with fractional ratios + // and I've never seen a non-corrupted JPEG file actually use them + for (i=0; i < s->img_n; ++i) { + if (h_max % z->img_comp[i].h != 0) return stbi__err("bad H","Corrupt JPEG"); + if (v_max % z->img_comp[i].v != 0) return stbi__err("bad V","Corrupt JPEG"); + } + + // compute interleaved mcu info + z->img_h_max = h_max; + z->img_v_max = v_max; + z->img_mcu_w = h_max * 8; + z->img_mcu_h = v_max * 8; + // these sizes can't be more than 17 bits + z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w; + z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h; + + for (i=0; i < s->img_n; ++i) { + // number of effective pixels (e.g. for non-interleaved MCU) + z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max; + z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max; + // to simplify generation, we'll allocate enough memory to decode + // the bogus oversized data from using interleaved MCUs and their + // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't + // discard the extra data until colorspace conversion + // + // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier) + // so these muls can't overflow with 32-bit ints (which we require) + z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8; + z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8; + z->img_comp[i].coeff = 0; + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].linebuf = NULL; + z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15); + if (z->img_comp[i].raw_data == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + // align blocks for idct using mmx/sse + z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15); + if (z->progressive) { + // w2, h2 are multiples of 8 (see above) + z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8; + z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8; + z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15); + if (z->img_comp[i].raw_coeff == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15); + } + } + + return 1; +} + +// use comparisons since in some cases we handle more than one case (e.g. SOF) +#define stbi__DNL(x) ((x) == 0xdc) +#define stbi__SOI(x) ((x) == 0xd8) +#define stbi__EOI(x) ((x) == 0xd9) +#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2) +#define stbi__SOS(x) ((x) == 0xda) + +#define stbi__SOF_progressive(x) ((x) == 0xc2) + +static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan) +{ + int m; + z->jfif = 0; + z->app14_color_transform = -1; // valid values are 0,1,2 + z->marker = STBI__MARKER_none; // initialize cached marker to empty + m = stbi__get_marker(z); + if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG"); + if (scan == STBI__SCAN_type) return 1; + m = stbi__get_marker(z); + while (!stbi__SOF(m)) { + if (!stbi__process_marker(z,m)) return 0; + m = stbi__get_marker(z); + while (m == STBI__MARKER_none) { + // some files have extra padding after their blocks, so ok, we'll scan + if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG"); + m = stbi__get_marker(z); + } + } + z->progressive = stbi__SOF_progressive(m); + if (!stbi__process_frame_header(z, scan)) return 0; + return 1; +} + +static stbi_uc stbi__skip_jpeg_junk_at_end(stbi__jpeg *j) +{ + // some JPEGs have junk at end, skip over it but if we find what looks + // like a valid marker, resume there + while (!stbi__at_eof(j->s)) { + stbi_uc x = stbi__get8(j->s); + while (x == 0xff) { // might be a marker + if (stbi__at_eof(j->s)) return STBI__MARKER_none; + x = stbi__get8(j->s); + if (x != 0x00 && x != 0xff) { + // not a stuffed zero or lead-in to another marker, looks + // like an actual marker, return it + return x; + } + // stuffed zero has x=0 now which ends the loop, meaning we go + // back to regular scan loop. + // repeated 0xff keeps trying to read the next byte of the marker. + } + } + return STBI__MARKER_none; +} + +// decode image to YCbCr format +static int stbi__decode_jpeg_image(stbi__jpeg *j) +{ + int m; + for (m = 0; m < 4; m++) { + j->img_comp[m].raw_data = NULL; + j->img_comp[m].raw_coeff = NULL; + } + j->restart_interval = 0; + if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0; + m = stbi__get_marker(j); + while (!stbi__EOI(m)) { + if (stbi__SOS(m)) { + if (!stbi__process_scan_header(j)) return 0; + if (!stbi__parse_entropy_coded_data(j)) return 0; + if (j->marker == STBI__MARKER_none ) { + j->marker = stbi__skip_jpeg_junk_at_end(j); + // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0 + } + m = stbi__get_marker(j); + if (STBI__RESTART(m)) + m = stbi__get_marker(j); + } else if (stbi__DNL(m)) { + int Ld = stbi__get16be(j->s); + stbi__uint32 NL = stbi__get16be(j->s); + if (Ld != 4) return stbi__err("bad DNL len", "Corrupt JPEG"); + if (NL != j->s->img_y) return stbi__err("bad DNL height", "Corrupt JPEG"); + m = stbi__get_marker(j); + } else { + if (!stbi__process_marker(j, m)) return 1; + m = stbi__get_marker(j); + } + } + if (j->progressive) + stbi__jpeg_finish(j); + return 1; +} + +// static jfif-centered resampling (across block boundaries) + +typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1, + int w, int hs); + +#define stbi__div4(x) ((stbi_uc) ((x) >> 2)) + +static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + STBI_NOTUSED(out); + STBI_NOTUSED(in_far); + STBI_NOTUSED(w); + STBI_NOTUSED(hs); + return in_near; +} + +static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples vertically for every one in input + int i; + STBI_NOTUSED(hs); + for (i=0; i < w; ++i) + out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2); + return out; +} + +static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples horizontally for every one in input + int i; + stbi_uc *input = in_near; + + if (w == 1) { + // if only one sample, can't do any interpolation + out[0] = out[1] = input[0]; + return out; + } + + out[0] = input[0]; + out[1] = stbi__div4(input[0]*3 + input[1] + 2); + for (i=1; i < w-1; ++i) { + int n = 3*input[i]+2; + out[i*2+0] = stbi__div4(n+input[i-1]); + out[i*2+1] = stbi__div4(n+input[i+1]); + } + out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2); + out[i*2+1] = input[w-1]; + + STBI_NOTUSED(in_far); + STBI_NOTUSED(hs); + + return out; +} + +#define stbi__div16(x) ((stbi_uc) ((x) >> 4)) + +static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i,t0,t1; + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + out[0] = stbi__div4(t1+2); + for (i=1; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i=0,t0,t1; + + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + // process groups of 8 pixels for as long as we can. + // note we can't handle the last pixel in a row in this loop + // because we need to handle the filter boundary conditions. + for (; i < ((w-1) & ~7); i += 8) { +#if defined(STBI_SSE2) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + __m128i zero = _mm_setzero_si128(); + __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i)); + __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i)); + __m128i farw = _mm_unpacklo_epi8(farb, zero); + __m128i nearw = _mm_unpacklo_epi8(nearb, zero); + __m128i diff = _mm_sub_epi16(farw, nearw); + __m128i nears = _mm_slli_epi16(nearw, 2); + __m128i curr = _mm_add_epi16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + __m128i prv0 = _mm_slli_si128(curr, 2); + __m128i nxt0 = _mm_srli_si128(curr, 2); + __m128i prev = _mm_insert_epi16(prv0, t1, 0); + __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + __m128i bias = _mm_set1_epi16(8); + __m128i curs = _mm_slli_epi16(curr, 2); + __m128i prvd = _mm_sub_epi16(prev, curr); + __m128i nxtd = _mm_sub_epi16(next, curr); + __m128i curb = _mm_add_epi16(curs, bias); + __m128i even = _mm_add_epi16(prvd, curb); + __m128i odd = _mm_add_epi16(nxtd, curb); + + // interleave even and odd pixels, then undo scaling. + __m128i int0 = _mm_unpacklo_epi16(even, odd); + __m128i int1 = _mm_unpackhi_epi16(even, odd); + __m128i de0 = _mm_srli_epi16(int0, 4); + __m128i de1 = _mm_srli_epi16(int1, 4); + + // pack and write output + __m128i outv = _mm_packus_epi16(de0, de1); + _mm_storeu_si128((__m128i *) (out + i*2), outv); +#elif defined(STBI_NEON) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + uint8x8_t farb = vld1_u8(in_far + i); + uint8x8_t nearb = vld1_u8(in_near + i); + int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb)); + int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2)); + int16x8_t curr = vaddq_s16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + int16x8_t prv0 = vextq_s16(curr, curr, 7); + int16x8_t nxt0 = vextq_s16(curr, curr, 1); + int16x8_t prev = vsetq_lane_s16(t1, prv0, 0); + int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + int16x8_t curs = vshlq_n_s16(curr, 2); + int16x8_t prvd = vsubq_s16(prev, curr); + int16x8_t nxtd = vsubq_s16(next, curr); + int16x8_t even = vaddq_s16(curs, prvd); + int16x8_t odd = vaddq_s16(curs, nxtd); + + // undo scaling and round, then store with even/odd phases interleaved + uint8x8x2_t o; + o.val[0] = vqrshrun_n_s16(even, 4); + o.val[1] = vqrshrun_n_s16(odd, 4); + vst2_u8(out + i*2, o); +#endif + + // "previous" value for next iter + t1 = 3*in_near[i+7] + in_far[i+7]; + } + + t0 = t1; + t1 = 3*in_near[i] + in_far[i]; + out[i*2] = stbi__div16(3*t1 + t0 + 8); + + for (++i; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} +#endif + +static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // resample with nearest-neighbor + int i,j; + STBI_NOTUSED(in_far); + for (i=0; i < w; ++i) + for (j=0; j < hs; ++j) + out[i*hs+j] = in_near[i]; + return out; +} + +// this is a reduced-precision calculation of YCbCr-to-RGB introduced +// to make sure the code produces the same results in both SIMD and scalar +#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8) +static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step) +{ + int i; + for (i=0; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step) +{ + int i = 0; + +#ifdef STBI_SSE2 + // step == 3 is pretty ugly on the final interleave, and i'm not convinced + // it's useful in practice (you wouldn't use it for textures, for example). + // so just accelerate step == 4 case. + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + __m128i signflip = _mm_set1_epi8(-0x80); + __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f)); + __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f)); + __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f)); + __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f)); + __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128); + __m128i xw = _mm_set1_epi16(255); // alpha channel + + for (; i+7 < count; i += 8) { + // load + __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i)); + __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i)); + __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i)); + __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128 + __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128 + + // unpack to short (and left-shift cr, cb by 8) + __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes); + __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased); + __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased); + + // color transform + __m128i yws = _mm_srli_epi16(yw, 4); + __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw); + __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw); + __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1); + __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1); + __m128i rws = _mm_add_epi16(cr0, yws); + __m128i gwt = _mm_add_epi16(cb0, yws); + __m128i bws = _mm_add_epi16(yws, cb1); + __m128i gws = _mm_add_epi16(gwt, cr1); + + // descale + __m128i rw = _mm_srai_epi16(rws, 4); + __m128i bw = _mm_srai_epi16(bws, 4); + __m128i gw = _mm_srai_epi16(gws, 4); + + // back to byte, set up for transpose + __m128i brb = _mm_packus_epi16(rw, bw); + __m128i gxb = _mm_packus_epi16(gw, xw); + + // transpose to interleave channels + __m128i t0 = _mm_unpacklo_epi8(brb, gxb); + __m128i t1 = _mm_unpackhi_epi8(brb, gxb); + __m128i o0 = _mm_unpacklo_epi16(t0, t1); + __m128i o1 = _mm_unpackhi_epi16(t0, t1); + + // store + _mm_storeu_si128((__m128i *) (out + 0), o0); + _mm_storeu_si128((__m128i *) (out + 16), o1); + out += 32; + } + } +#endif + +#ifdef STBI_NEON + // in this version, step=3 support would be easy to add. but is there demand? + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + uint8x8_t signflip = vdup_n_u8(0x80); + int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f)); + int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f)); + int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f)); + int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f)); + + for (; i+7 < count; i += 8) { + // load + uint8x8_t y_bytes = vld1_u8(y + i); + uint8x8_t cr_bytes = vld1_u8(pcr + i); + uint8x8_t cb_bytes = vld1_u8(pcb + i); + int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip)); + int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip)); + + // expand to s16 + int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4)); + int16x8_t crw = vshll_n_s8(cr_biased, 7); + int16x8_t cbw = vshll_n_s8(cb_biased, 7); + + // color transform + int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0); + int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0); + int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1); + int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1); + int16x8_t rws = vaddq_s16(yws, cr0); + int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1); + int16x8_t bws = vaddq_s16(yws, cb1); + + // undo scaling, round, convert to byte + uint8x8x4_t o; + o.val[0] = vqrshrun_n_s16(rws, 4); + o.val[1] = vqrshrun_n_s16(gws, 4); + o.val[2] = vqrshrun_n_s16(bws, 4); + o.val[3] = vdup_n_u8(255); + + // store, interleaving r/g/b/a + vst4_u8(out, o); + out += 8*4; + } + } +#endif + + for (; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} +#endif + +// set up the kernels +static void stbi__setup_jpeg(stbi__jpeg *j) +{ + j->idct_block_kernel = stbi__idct_block; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2; + +#ifdef STBI_SSE2 + if (stbi__sse2_available()) { + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; + } +#endif + +#ifdef STBI_NEON + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; +#endif +} + +// clean up the temporary component buffers +static void stbi__cleanup_jpeg(stbi__jpeg *j) +{ + stbi__free_jpeg_components(j, j->s->img_n, 0); +} + +typedef struct +{ + resample_row_func resample; + stbi_uc *line0,*line1; + int hs,vs; // expansion factor in each axis + int w_lores; // horizontal pixels pre-expansion + int ystep; // how far through vertical expansion we are + int ypos; // which pre-expansion row we're on +} stbi__resample; + +// fast 0..255 * 0..255 => 0..255 rounded multiplication +static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y) +{ + unsigned int t = x*y + 128; + return (stbi_uc) ((t + (t >>8)) >> 8); +} + +static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp) +{ + int n, decode_n, is_rgb; + z->s->img_n = 0; // make stbi__cleanup_jpeg safe + + // validate req_comp + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + + // load a jpeg image from whichever source, but leave in YCbCr format + if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; } + + // determine actual number of components to generate + n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1; + + is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif)); + + if (z->s->img_n == 3 && n < 3 && !is_rgb) + decode_n = 1; + else + decode_n = z->s->img_n; + + // nothing to do if no components requested; check this now to avoid + // accessing uninitialized coutput[0] later + if (decode_n <= 0) { stbi__cleanup_jpeg(z); return NULL; } + + // resample and color-convert + { + int k; + unsigned int i,j; + stbi_uc *output; + stbi_uc *coutput[4] = { NULL, NULL, NULL, NULL }; + + stbi__resample res_comp[4]; + + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + + // allocate line buffer big enough for upsampling off the edges + // with upsample factor of 4 + z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3); + if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + r->hs = z->img_h_max / z->img_comp[k].h; + r->vs = z->img_v_max / z->img_comp[k].v; + r->ystep = r->vs >> 1; + r->w_lores = (z->s->img_x + r->hs-1) / r->hs; + r->ypos = 0; + r->line0 = r->line1 = z->img_comp[k].data; + + if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1; + else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2; + else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2; + else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel; + else r->resample = stbi__resample_row_generic; + } + + // can't error after this so, this is safe + output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1); + if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + // now go ahead and resample + for (j=0; j < z->s->img_y; ++j) { + stbi_uc *out = output + n * z->s->img_x * j; + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + int y_bot = r->ystep >= (r->vs >> 1); + coutput[k] = r->resample(z->img_comp[k].linebuf, + y_bot ? r->line1 : r->line0, + y_bot ? r->line0 : r->line1, + r->w_lores, r->hs); + if (++r->ystep >= r->vs) { + r->ystep = 0; + r->line0 = r->line1; + if (++r->ypos < z->img_comp[k].y) + r->line1 += z->img_comp[k].w2; + } + } + if (n >= 3) { + stbi_uc *y = coutput[0]; + if (z->s->img_n == 3) { + if (is_rgb) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = y[i]; + out[1] = coutput[1][i]; + out[2] = coutput[2][i]; + out[3] = 255; + out += n; + } + } else { + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else if (z->s->img_n == 4) { + if (z->app14_color_transform == 0) { // CMYK + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(coutput[0][i], m); + out[1] = stbi__blinn_8x8(coutput[1][i], m); + out[2] = stbi__blinn_8x8(coutput[2][i], m); + out[3] = 255; + out += n; + } + } else if (z->app14_color_transform == 2) { // YCCK + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(255 - out[0], m); + out[1] = stbi__blinn_8x8(255 - out[1], m); + out[2] = stbi__blinn_8x8(255 - out[2], m); + out += n; + } + } else { // YCbCr + alpha? Ignore the fourth channel for now + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else + for (i=0; i < z->s->img_x; ++i) { + out[0] = out[1] = out[2] = y[i]; + out[3] = 255; // not used if n==3 + out += n; + } + } else { + if (is_rgb) { + if (n == 1) + for (i=0; i < z->s->img_x; ++i) + *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + else { + for (i=0; i < z->s->img_x; ++i, out += 2) { + out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + out[1] = 255; + } + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 0) { + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + stbi_uc r = stbi__blinn_8x8(coutput[0][i], m); + stbi_uc g = stbi__blinn_8x8(coutput[1][i], m); + stbi_uc b = stbi__blinn_8x8(coutput[2][i], m); + out[0] = stbi__compute_y(r, g, b); + out[1] = 255; + out += n; + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 2) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]); + out[1] = 255; + out += n; + } + } else { + stbi_uc *y = coutput[0]; + if (n == 1) + for (i=0; i < z->s->img_x; ++i) out[i] = y[i]; + else + for (i=0; i < z->s->img_x; ++i) { *out++ = y[i]; *out++ = 255; } + } + } + } + stbi__cleanup_jpeg(z); + *out_x = z->s->img_x; + *out_y = z->s->img_y; + if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output + return output; + } +} + +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + unsigned char* result; + stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg)); + if (!j) return stbi__errpuc("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + STBI_NOTUSED(ri); + j->s = s; + stbi__setup_jpeg(j); + result = load_jpeg_image(j, x,y,comp,req_comp); + STBI_FREE(j); + return result; +} + +static int stbi__jpeg_test(stbi__context *s) +{ + int r; + stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg)); + if (!j) return stbi__err("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + j->s = s; + stbi__setup_jpeg(j); + r = stbi__decode_jpeg_header(j, STBI__SCAN_type); + stbi__rewind(s); + STBI_FREE(j); + return r; +} + +static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp) +{ + if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) { + stbi__rewind( j->s ); + return 0; + } + if (x) *x = j->s->img_x; + if (y) *y = j->s->img_y; + if (comp) *comp = j->s->img_n >= 3 ? 3 : 1; + return 1; +} + +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp) +{ + int result; + stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg))); + if (!j) return stbi__err("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + j->s = s; + result = stbi__jpeg_info_raw(j, x, y, comp); + STBI_FREE(j); + return result; +} +#endif + +// public domain zlib decode v0.2 Sean Barrett 2006-11-18 +// simple implementation +// - all input must be provided in an upfront buffer +// - all output is written to a single output buffer (can malloc/realloc) +// performance +// - fast huffman + +#ifndef STBI_NO_ZLIB + +// fast-way is faster to check than jpeg huffman, but slow way is slower +#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables +#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1) +#define STBI__ZNSYMS 288 // number of symbols in literal/length alphabet + +// zlib-style huffman encoding +// (jpegs packs from left, zlib from right, so can't share code) +typedef struct +{ + stbi__uint16 fast[1 << STBI__ZFAST_BITS]; + stbi__uint16 firstcode[16]; + int maxcode[17]; + stbi__uint16 firstsymbol[16]; + stbi_uc size[STBI__ZNSYMS]; + stbi__uint16 value[STBI__ZNSYMS]; +} stbi__zhuffman; + +stbi_inline static int stbi__bitreverse16(int n) +{ + n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1); + n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2); + n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4); + n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8); + return n; +} + +stbi_inline static int stbi__bit_reverse(int v, int bits) +{ + STBI_ASSERT(bits <= 16); + // to bit reverse n bits, reverse 16 and shift + // e.g. 11 bits, bit reverse and shift away 5 + return stbi__bitreverse16(v) >> (16-bits); +} + +static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num) +{ + int i,k=0; + int code, next_code[16], sizes[17]; + + // DEFLATE spec for generating codes + memset(sizes, 0, sizeof(sizes)); + memset(z->fast, 0, sizeof(z->fast)); + for (i=0; i < num; ++i) + ++sizes[sizelist[i]]; + sizes[0] = 0; + for (i=1; i < 16; ++i) + if (sizes[i] > (1 << i)) + return stbi__err("bad sizes", "Corrupt PNG"); + code = 0; + for (i=1; i < 16; ++i) { + next_code[i] = code; + z->firstcode[i] = (stbi__uint16) code; + z->firstsymbol[i] = (stbi__uint16) k; + code = (code + sizes[i]); + if (sizes[i]) + if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG"); + z->maxcode[i] = code << (16-i); // preshift for inner loop + code <<= 1; + k += sizes[i]; + } + z->maxcode[16] = 0x10000; // sentinel + for (i=0; i < num; ++i) { + int s = sizelist[i]; + if (s) { + int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s]; + stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i); + z->size [c] = (stbi_uc ) s; + z->value[c] = (stbi__uint16) i; + if (s <= STBI__ZFAST_BITS) { + int j = stbi__bit_reverse(next_code[s],s); + while (j < (1 << STBI__ZFAST_BITS)) { + z->fast[j] = fastv; + j += (1 << s); + } + } + ++next_code[s]; + } + } + return 1; +} + +// zlib-from-memory implementation for PNG reading +// because PNG allows splitting the zlib stream arbitrarily, +// and it's annoying structurally to have PNG call ZLIB call PNG, +// we require PNG read all the IDATs and combine them into a single +// memory buffer + +typedef struct +{ + stbi_uc *zbuffer, *zbuffer_end; + int num_bits; + int hit_zeof_once; + stbi__uint32 code_buffer; + + char *zout; + char *zout_start; + char *zout_end; + int z_expandable; + + stbi__zhuffman z_length, z_distance; +} stbi__zbuf; + +stbi_inline static int stbi__zeof(stbi__zbuf *z) +{ + return (z->zbuffer >= z->zbuffer_end); +} + +stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z) +{ + return stbi__zeof(z) ? 0 : *z->zbuffer++; +} + +static void stbi__fill_bits(stbi__zbuf *z) +{ + do { + if (z->code_buffer >= (1U << z->num_bits)) { + z->zbuffer = z->zbuffer_end; /* treat this as EOF so we fail. */ + return; + } + z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits; + z->num_bits += 8; + } while (z->num_bits <= 24); +} + +stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n) +{ + unsigned int k; + if (z->num_bits < n) stbi__fill_bits(z); + k = z->code_buffer & ((1 << n) - 1); + z->code_buffer >>= n; + z->num_bits -= n; + return k; +} + +static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s,k; + // not resolved by fast table, so compute it the slow way + // use jpeg approach, which requires MSbits at top + k = stbi__bit_reverse(a->code_buffer, 16); + for (s=STBI__ZFAST_BITS+1; ; ++s) + if (k < z->maxcode[s]) + break; + if (s >= 16) return -1; // invalid code! + // code size is s, so: + b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s]; + if (b >= STBI__ZNSYMS) return -1; // some data was corrupt somewhere! + if (z->size[b] != s) return -1; // was originally an assert, but report failure instead. + a->code_buffer >>= s; + a->num_bits -= s; + return z->value[b]; +} + +stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s; + if (a->num_bits < 16) { + if (stbi__zeof(a)) { + if (!a->hit_zeof_once) { + // This is the first time we hit eof, insert 16 extra padding btis + // to allow us to keep going; if we actually consume any of them + // though, that is invalid data. This is caught later. + a->hit_zeof_once = 1; + a->num_bits += 16; // add 16 implicit zero bits + } else { + // We already inserted our extra 16 padding bits and are again + // out, this stream is actually prematurely terminated. + return -1; + } + } else { + stbi__fill_bits(a); + } + } + b = z->fast[a->code_buffer & STBI__ZFAST_MASK]; + if (b) { + s = b >> 9; + a->code_buffer >>= s; + a->num_bits -= s; + return b & 511; + } + return stbi__zhuffman_decode_slowpath(a, z); +} + +static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes +{ + char *q; + unsigned int cur, limit, old_limit; + z->zout = zout; + if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG"); + cur = (unsigned int) (z->zout - z->zout_start); + limit = old_limit = (unsigned) (z->zout_end - z->zout_start); + if (UINT_MAX - cur < (unsigned) n) return stbi__err("outofmem", "Out of memory"); + while (cur + n > limit) { + if(limit > UINT_MAX / 2) return stbi__err("outofmem", "Out of memory"); + limit *= 2; + } + q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit); + STBI_NOTUSED(old_limit); + if (q == NULL) return stbi__err("outofmem", "Out of memory"); + z->zout_start = q; + z->zout = q + cur; + z->zout_end = q + limit; + return 1; +} + +static const int stbi__zlength_base[31] = { + 3,4,5,6,7,8,9,10,11,13, + 15,17,19,23,27,31,35,43,51,59, + 67,83,99,115,131,163,195,227,258,0,0 }; + +static const int stbi__zlength_extra[31]= +{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 }; + +static const int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193, +257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0}; + +static const int stbi__zdist_extra[32] = +{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13}; + +static int stbi__parse_huffman_block(stbi__zbuf *a) +{ + char *zout = a->zout; + for(;;) { + int z = stbi__zhuffman_decode(a, &a->z_length); + if (z < 256) { + if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes + if (zout >= a->zout_end) { + if (!stbi__zexpand(a, zout, 1)) return 0; + zout = a->zout; + } + *zout++ = (char) z; + } else { + stbi_uc *p; + int len,dist; + if (z == 256) { + a->zout = zout; + if (a->hit_zeof_once && a->num_bits < 16) { + // The first time we hit zeof, we inserted 16 extra zero bits into our bit + // buffer so the decoder can just do its speculative decoding. But if we + // actually consumed any of those bits (which is the case when num_bits < 16), + // the stream actually read past the end so it is malformed. + return stbi__err("unexpected end","Corrupt PNG"); + } + return 1; + } + if (z >= 286) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, length codes 286 and 287 must not appear in compressed data + z -= 257; + len = stbi__zlength_base[z]; + if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]); + z = stbi__zhuffman_decode(a, &a->z_distance); + if (z < 0 || z >= 30) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, distance codes 30 and 31 must not appear in compressed data + dist = stbi__zdist_base[z]; + if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]); + if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG"); + if (len > a->zout_end - zout) { + if (!stbi__zexpand(a, zout, len)) return 0; + zout = a->zout; + } + p = (stbi_uc *) (zout - dist); + if (dist == 1) { // run of one byte; common in images. + stbi_uc v = *p; + if (len) { do *zout++ = v; while (--len); } + } else { + if (len) { do *zout++ = *p++; while (--len); } + } + } + } +} + +static int stbi__compute_huffman_codes(stbi__zbuf *a) +{ + static const stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 }; + stbi__zhuffman z_codelength; + stbi_uc lencodes[286+32+137];//padding for maximum single op + stbi_uc codelength_sizes[19]; + int i,n; + + int hlit = stbi__zreceive(a,5) + 257; + int hdist = stbi__zreceive(a,5) + 1; + int hclen = stbi__zreceive(a,4) + 4; + int ntot = hlit + hdist; + + memset(codelength_sizes, 0, sizeof(codelength_sizes)); + for (i=0; i < hclen; ++i) { + int s = stbi__zreceive(a,3); + codelength_sizes[length_dezigzag[i]] = (stbi_uc) s; + } + if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0; + + n = 0; + while (n < ntot) { + int c = stbi__zhuffman_decode(a, &z_codelength); + if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG"); + if (c < 16) + lencodes[n++] = (stbi_uc) c; + else { + stbi_uc fill = 0; + if (c == 16) { + c = stbi__zreceive(a,2)+3; + if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG"); + fill = lencodes[n-1]; + } else if (c == 17) { + c = stbi__zreceive(a,3)+3; + } else if (c == 18) { + c = stbi__zreceive(a,7)+11; + } else { + return stbi__err("bad codelengths", "Corrupt PNG"); + } + if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG"); + memset(lencodes+n, fill, c); + n += c; + } + } + if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG"); + if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0; + return 1; +} + +static int stbi__parse_uncompressed_block(stbi__zbuf *a) +{ + stbi_uc header[4]; + int len,nlen,k; + if (a->num_bits & 7) + stbi__zreceive(a, a->num_bits & 7); // discard + // drain the bit-packed data into header + k = 0; + while (a->num_bits > 0) { + header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check + a->code_buffer >>= 8; + a->num_bits -= 8; + } + if (a->num_bits < 0) return stbi__err("zlib corrupt","Corrupt PNG"); + // now fill header the normal way + while (k < 4) + header[k++] = stbi__zget8(a); + len = header[1] * 256 + header[0]; + nlen = header[3] * 256 + header[2]; + if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG"); + if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG"); + if (a->zout + len > a->zout_end) + if (!stbi__zexpand(a, a->zout, len)) return 0; + memcpy(a->zout, a->zbuffer, len); + a->zbuffer += len; + a->zout += len; + return 1; +} + +static int stbi__parse_zlib_header(stbi__zbuf *a) +{ + int cmf = stbi__zget8(a); + int cm = cmf & 15; + /* int cinfo = cmf >> 4; */ + int flg = stbi__zget8(a); + if (stbi__zeof(a)) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png + if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png + // window = 1 << (8 + cinfo)... but who cares, we fully buffer output + return 1; +} + +static const stbi_uc stbi__zdefault_length[STBI__ZNSYMS] = +{ + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8 +}; +static const stbi_uc stbi__zdefault_distance[32] = +{ + 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5 +}; +/* +Init algorithm: +{ + int i; // use <= to match clearly with spec + for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8; + for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9; + for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7; + for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8; + + for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5; +} +*/ + +static int stbi__parse_zlib(stbi__zbuf *a, int parse_header) +{ + int final, type; + if (parse_header) + if (!stbi__parse_zlib_header(a)) return 0; + a->num_bits = 0; + a->code_buffer = 0; + a->hit_zeof_once = 0; + do { + final = stbi__zreceive(a,1); + type = stbi__zreceive(a,2); + if (type == 0) { + if (!stbi__parse_uncompressed_block(a)) return 0; + } else if (type == 3) { + return 0; + } else { + if (type == 1) { + // use fixed code lengths + if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , STBI__ZNSYMS)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0; + } else { + if (!stbi__compute_huffman_codes(a)) return 0; + } + if (!stbi__parse_huffman_block(a)) return 0; + } + } while (!final); + return 1; +} + +static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header) +{ + a->zout_start = obuf; + a->zout = obuf; + a->zout_end = obuf + olen; + a->z_expandable = exp; + + return stbi__parse_zlib(a, parse_header); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, 1)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen) +{ + return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 1)) + return (int) (a.zout - a.zout_start); + else + return -1; +} + +STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(16384); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer+len; + if (stbi__do_zlib(&a, p, 16384, 1, 0)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 0)) + return (int) (a.zout - a.zout_start); + else + return -1; +} +#endif + +// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18 +// simple implementation +// - only 8-bit samples +// - no CRC checking +// - allocates lots of intermediate memory +// - avoids problem of streaming data between subsystems +// - avoids explicit window management +// performance +// - uses stb_zlib, a PD zlib implementation with fast huffman decoding + +#ifndef STBI_NO_PNG +typedef struct +{ + stbi__uint32 length; + stbi__uint32 type; +} stbi__pngchunk; + +static stbi__pngchunk stbi__get_chunk_header(stbi__context *s) +{ + stbi__pngchunk c; + c.length = stbi__get32be(s); + c.type = stbi__get32be(s); + return c; +} + +static int stbi__check_png_header(stbi__context *s) +{ + static const stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 }; + int i; + for (i=0; i < 8; ++i) + if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG"); + return 1; +} + +typedef struct +{ + stbi__context *s; + stbi_uc *idata, *expanded, *out; + int depth; +} stbi__png; + + +enum { + STBI__F_none=0, + STBI__F_sub=1, + STBI__F_up=2, + STBI__F_avg=3, + STBI__F_paeth=4, + // synthetic filter used for first scanline to avoid needing a dummy row of 0s + STBI__F_avg_first +}; + +static stbi_uc first_row_filter[5] = +{ + STBI__F_none, + STBI__F_sub, + STBI__F_none, + STBI__F_avg_first, + STBI__F_sub // Paeth with b=c=0 turns out to be equivalent to sub +}; + +static int stbi__paeth(int a, int b, int c) +{ + // This formulation looks very different from the reference in the PNG spec, but is + // actually equivalent and has favorable data dependencies and admits straightforward + // generation of branch-free code, which helps performance significantly. + int thresh = c*3 - (a + b); + int lo = a < b ? a : b; + int hi = a < b ? b : a; + int t0 = (hi <= thresh) ? lo : c; + int t1 = (thresh <= lo) ? hi : t0; + return t1; +} + +static const stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 }; + +// adds an extra all-255 alpha channel +// dest == src is legal +// img_n must be 1 or 3 +static void stbi__create_png_alpha_expand8(stbi_uc *dest, stbi_uc *src, stbi__uint32 x, int img_n) +{ + int i; + // must process data backwards since we allow dest==src + if (img_n == 1) { + for (i=x-1; i >= 0; --i) { + dest[i*2+1] = 255; + dest[i*2+0] = src[i]; + } + } else { + STBI_ASSERT(img_n == 3); + for (i=x-1; i >= 0; --i) { + dest[i*4+3] = 255; + dest[i*4+2] = src[i*3+2]; + dest[i*4+1] = src[i*3+1]; + dest[i*4+0] = src[i*3+0]; + } + } +} + +// create the png data from post-deflated data +static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color) +{ + int bytes = (depth == 16 ? 2 : 1); + stbi__context *s = a->s; + stbi__uint32 i,j,stride = x*out_n*bytes; + stbi__uint32 img_len, img_width_bytes; + stbi_uc *filter_buf; + int all_ok = 1; + int k; + int img_n = s->img_n; // copy it into a local for later + + int output_bytes = out_n*bytes; + int filter_bytes = img_n*bytes; + int width = x; + + STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1); + a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into + if (!a->out) return stbi__err("outofmem", "Out of memory"); + + // note: error exits here don't need to clean up a->out individually, + // stbi__do_png always does on error. + if (!stbi__mad3sizes_valid(img_n, x, depth, 7)) return stbi__err("too large", "Corrupt PNG"); + img_width_bytes = (((img_n * x * depth) + 7) >> 3); + if (!stbi__mad2sizes_valid(img_width_bytes, y, img_width_bytes)) return stbi__err("too large", "Corrupt PNG"); + img_len = (img_width_bytes + 1) * y; + + // we used to check for exact match between raw_len and img_len on non-interlaced PNGs, + // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros), + // so just check for raw_len < img_len always. + if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG"); + + // Allocate two scan lines worth of filter workspace buffer. + filter_buf = (stbi_uc *) stbi__malloc_mad2(img_width_bytes, 2, 0); + if (!filter_buf) return stbi__err("outofmem", "Out of memory"); + + // Filtering for low-bit-depth images + if (depth < 8) { + filter_bytes = 1; + width = img_width_bytes; + } + + for (j=0; j < y; ++j) { + // cur/prior filter buffers alternate + stbi_uc *cur = filter_buf + (j & 1)*img_width_bytes; + stbi_uc *prior = filter_buf + (~j & 1)*img_width_bytes; + stbi_uc *dest = a->out + stride*j; + int nk = width * filter_bytes; + int filter = *raw++; + + // check filter type + if (filter > 4) { + all_ok = stbi__err("invalid filter","Corrupt PNG"); + break; + } + + // if first row, use special filter that doesn't sample previous row + if (j == 0) filter = first_row_filter[filter]; + + // perform actual filtering + switch (filter) { + case STBI__F_none: + memcpy(cur, raw, nk); + break; + case STBI__F_sub: + memcpy(cur, raw, filter_bytes); + for (k = filter_bytes; k < nk; ++k) + cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); + break; + case STBI__F_up: + for (k = 0; k < nk; ++k) + cur[k] = STBI__BYTECAST(raw[k] + prior[k]); + break; + case STBI__F_avg: + for (k = 0; k < filter_bytes; ++k) + cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); + for (k = filter_bytes; k < nk; ++k) + cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); + break; + case STBI__F_paeth: + for (k = 0; k < filter_bytes; ++k) + cur[k] = STBI__BYTECAST(raw[k] + prior[k]); // prior[k] == stbi__paeth(0,prior[k],0) + for (k = filter_bytes; k < nk; ++k) + cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes], prior[k], prior[k-filter_bytes])); + break; + case STBI__F_avg_first: + memcpy(cur, raw, filter_bytes); + for (k = filter_bytes; k < nk; ++k) + cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); + break; + } + + raw += nk; + + // expand decoded bits in cur to dest, also adding an extra alpha channel if desired + if (depth < 8) { + stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range + stbi_uc *in = cur; + stbi_uc *out = dest; + stbi_uc inb = 0; + stbi__uint32 nsmp = x*img_n; + + // expand bits to bytes first + if (depth == 4) { + for (i=0; i < nsmp; ++i) { + if ((i & 1) == 0) inb = *in++; + *out++ = scale * (inb >> 4); + inb <<= 4; + } + } else if (depth == 2) { + for (i=0; i < nsmp; ++i) { + if ((i & 3) == 0) inb = *in++; + *out++ = scale * (inb >> 6); + inb <<= 2; + } + } else { + STBI_ASSERT(depth == 1); + for (i=0; i < nsmp; ++i) { + if ((i & 7) == 0) inb = *in++; + *out++ = scale * (inb >> 7); + inb <<= 1; + } + } + + // insert alpha=255 values if desired + if (img_n != out_n) + stbi__create_png_alpha_expand8(dest, dest, x, img_n); + } else if (depth == 8) { + if (img_n == out_n) + memcpy(dest, cur, x*img_n); + else + stbi__create_png_alpha_expand8(dest, cur, x, img_n); + } else if (depth == 16) { + // convert the image data from big-endian to platform-native + stbi__uint16 *dest16 = (stbi__uint16*)dest; + stbi__uint32 nsmp = x*img_n; + + if (img_n == out_n) { + for (i = 0; i < nsmp; ++i, ++dest16, cur += 2) + *dest16 = (cur[0] << 8) | cur[1]; + } else { + STBI_ASSERT(img_n+1 == out_n); + if (img_n == 1) { + for (i = 0; i < x; ++i, dest16 += 2, cur += 2) { + dest16[0] = (cur[0] << 8) | cur[1]; + dest16[1] = 0xffff; + } + } else { + STBI_ASSERT(img_n == 3); + for (i = 0; i < x; ++i, dest16 += 4, cur += 6) { + dest16[0] = (cur[0] << 8) | cur[1]; + dest16[1] = (cur[2] << 8) | cur[3]; + dest16[2] = (cur[4] << 8) | cur[5]; + dest16[3] = 0xffff; + } + } + } + } + } + + STBI_FREE(filter_buf); + if (!all_ok) return 0; + + return 1; +} + +static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced) +{ + int bytes = (depth == 16 ? 2 : 1); + int out_bytes = out_n * bytes; + stbi_uc *final; + int p; + if (!interlaced) + return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color); + + // de-interlacing + final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0); + if (!final) return stbi__err("outofmem", "Out of memory"); + for (p=0; p < 7; ++p) { + int xorig[] = { 0,4,0,2,0,1,0 }; + int yorig[] = { 0,0,4,0,2,0,1 }; + int xspc[] = { 8,8,4,4,2,2,1 }; + int yspc[] = { 8,8,8,4,4,2,2 }; + int i,j,x,y; + // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1 + x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p]; + y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p]; + if (x && y) { + stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y; + if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) { + STBI_FREE(final); + return 0; + } + for (j=0; j < y; ++j) { + for (i=0; i < x; ++i) { + int out_y = j*yspc[p]+yorig[p]; + int out_x = i*xspc[p]+xorig[p]; + memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes, + a->out + (j*x+i)*out_bytes, out_bytes); + } + } + STBI_FREE(a->out); + image_data += img_len; + image_data_len -= img_len; + } + } + a->out = final; + + return 1; +} + +static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + // compute color-based transparency, assuming we've + // already got 255 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i=0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 255); + p += 2; + } + } else { + for (i=0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi__uint16 *p = (stbi__uint16*) z->out; + + // compute color-based transparency, assuming we've + // already got 65535 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i = 0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 65535); + p += 2; + } + } else { + for (i = 0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n) +{ + stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y; + stbi_uc *p, *temp_out, *orig = a->out; + + p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0); + if (p == NULL) return stbi__err("outofmem", "Out of memory"); + + // between here and free(out) below, exitting would leak + temp_out = p; + + if (pal_img_n == 3) { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p += 3; + } + } else { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p[3] = palette[n+3]; + p += 4; + } + } + STBI_FREE(a->out); + a->out = temp_out; + + STBI_NOTUSED(len); + + return 1; +} + +static int stbi__unpremultiply_on_load_global = 0; +static int stbi__de_iphone_flag_global = 0; + +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply) +{ + stbi__unpremultiply_on_load_global = flag_true_if_should_unpremultiply; +} + +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert) +{ + stbi__de_iphone_flag_global = flag_true_if_should_convert; +} + +#ifndef STBI_THREAD_LOCAL +#define stbi__unpremultiply_on_load stbi__unpremultiply_on_load_global +#define stbi__de_iphone_flag stbi__de_iphone_flag_global +#else +static STBI_THREAD_LOCAL int stbi__unpremultiply_on_load_local, stbi__unpremultiply_on_load_set; +static STBI_THREAD_LOCAL int stbi__de_iphone_flag_local, stbi__de_iphone_flag_set; + +STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply) +{ + stbi__unpremultiply_on_load_local = flag_true_if_should_unpremultiply; + stbi__unpremultiply_on_load_set = 1; +} + +STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert) +{ + stbi__de_iphone_flag_local = flag_true_if_should_convert; + stbi__de_iphone_flag_set = 1; +} + +#define stbi__unpremultiply_on_load (stbi__unpremultiply_on_load_set \ + ? stbi__unpremultiply_on_load_local \ + : stbi__unpremultiply_on_load_global) +#define stbi__de_iphone_flag (stbi__de_iphone_flag_set \ + ? stbi__de_iphone_flag_local \ + : stbi__de_iphone_flag_global) +#endif // STBI_THREAD_LOCAL + +static void stbi__de_iphone(stbi__png *z) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + if (s->img_out_n == 3) { // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 3; + } + } else { + STBI_ASSERT(s->img_out_n == 4); + if (stbi__unpremultiply_on_load) { + // convert bgr to rgb and unpremultiply + for (i=0; i < pixel_count; ++i) { + stbi_uc a = p[3]; + stbi_uc t = p[0]; + if (a) { + stbi_uc half = a / 2; + p[0] = (p[2] * 255 + half) / a; + p[1] = (p[1] * 255 + half) / a; + p[2] = ( t * 255 + half) / a; + } else { + p[0] = p[2]; + p[2] = t; + } + p += 4; + } + } else { + // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 4; + } + } + } +} + +#define STBI__PNG_TYPE(a,b,c,d) (((unsigned) (a) << 24) + ((unsigned) (b) << 16) + ((unsigned) (c) << 8) + (unsigned) (d)) + +static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp) +{ + stbi_uc palette[1024], pal_img_n=0; + stbi_uc has_trans=0, tc[3]={0}; + stbi__uint16 tc16[3]; + stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0; + int first=1,k,interlace=0, color=0, is_iphone=0; + stbi__context *s = z->s; + + z->expanded = NULL; + z->idata = NULL; + z->out = NULL; + + if (!stbi__check_png_header(s)) return 0; + + if (scan == STBI__SCAN_type) return 1; + + for (;;) { + stbi__pngchunk c = stbi__get_chunk_header(s); + switch (c.type) { + case STBI__PNG_TYPE('C','g','B','I'): + is_iphone = 1; + stbi__skip(s, c.length); + break; + case STBI__PNG_TYPE('I','H','D','R'): { + int comp,filter; + if (!first) return stbi__err("multiple IHDR","Corrupt PNG"); + first = 0; + if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG"); + s->img_x = stbi__get32be(s); + s->img_y = stbi__get32be(s); + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only"); + color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG"); + comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG"); + filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG"); + interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG"); + if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG"); + if (!pal_img_n) { + s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0); + if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode"); + } else { + // if paletted, then pal_n is our final components, and + // img_n is # components to decompress/filter. + s->img_n = 1; + if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG"); + } + // even with SCAN_header, have to scan to see if we have a tRNS + break; + } + + case STBI__PNG_TYPE('P','L','T','E'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG"); + pal_len = c.length / 3; + if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG"); + for (i=0; i < pal_len; ++i) { + palette[i*4+0] = stbi__get8(s); + palette[i*4+1] = stbi__get8(s); + palette[i*4+2] = stbi__get8(s); + palette[i*4+3] = 255; + } + break; + } + + case STBI__PNG_TYPE('t','R','N','S'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG"); + if (pal_img_n) { + if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; } + if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG"); + if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG"); + pal_img_n = 4; + for (i=0; i < c.length; ++i) + palette[i*4+3] = stbi__get8(s); + } else { + if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG"); + if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG"); + has_trans = 1; + // non-paletted with tRNS = constant alpha. if header-scanning, we can stop now. + if (scan == STBI__SCAN_header) { ++s->img_n; return 1; } + if (z->depth == 16) { + for (k = 0; k < s->img_n && k < 3; ++k) // extra loop test to suppress false GCC warning + tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is + } else { + for (k = 0; k < s->img_n && k < 3; ++k) + tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger + } + } + break; + } + + case STBI__PNG_TYPE('I','D','A','T'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG"); + if (scan == STBI__SCAN_header) { + // header scan definitely stops at first IDAT + if (pal_img_n) + s->img_n = pal_img_n; + return 1; + } + if (c.length > (1u << 30)) return stbi__err("IDAT size limit", "IDAT section larger than 2^30 bytes"); + if ((int)(ioff + c.length) < (int)ioff) return 0; + if (ioff + c.length > idata_limit) { + stbi__uint32 idata_limit_old = idata_limit; + stbi_uc *p; + if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096; + while (ioff + c.length > idata_limit) + idata_limit *= 2; + STBI_NOTUSED(idata_limit_old); + p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory"); + z->idata = p; + } + if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG"); + ioff += c.length; + break; + } + + case STBI__PNG_TYPE('I','E','N','D'): { + stbi__uint32 raw_len, bpl; + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (scan != STBI__SCAN_load) return 1; + if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG"); + // initial guess for decoded data size to avoid unnecessary reallocs + bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component + raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */; + z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone); + if (z->expanded == NULL) return 0; // zlib should set error + STBI_FREE(z->idata); z->idata = NULL; + if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans) + s->img_out_n = s->img_n+1; + else + s->img_out_n = s->img_n; + if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0; + if (has_trans) { + if (z->depth == 16) { + if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0; + } else { + if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0; + } + } + if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2) + stbi__de_iphone(z); + if (pal_img_n) { + // pal_img_n == 3 or 4 + s->img_n = pal_img_n; // record the actual colors we had + s->img_out_n = pal_img_n; + if (req_comp >= 3) s->img_out_n = req_comp; + if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n)) + return 0; + } else if (has_trans) { + // non-paletted image with tRNS -> source image has (constant) alpha + ++s->img_n; + } + STBI_FREE(z->expanded); z->expanded = NULL; + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + return 1; + } + + default: + // if critical, fail + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if ((c.type & (1 << 29)) == 0) { + #ifndef STBI_NO_FAILURE_STRINGS + // not threadsafe + static char invalid_chunk[] = "XXXX PNG chunk not known"; + invalid_chunk[0] = STBI__BYTECAST(c.type >> 24); + invalid_chunk[1] = STBI__BYTECAST(c.type >> 16); + invalid_chunk[2] = STBI__BYTECAST(c.type >> 8); + invalid_chunk[3] = STBI__BYTECAST(c.type >> 0); + #endif + return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type"); + } + stbi__skip(s, c.length); + break; + } + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + } +} + +static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri) +{ + void *result=NULL; + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) { + if (p->depth <= 8) + ri->bits_per_channel = 8; + else if (p->depth == 16) + ri->bits_per_channel = 16; + else + return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color depth"); + result = p->out; + p->out = NULL; + if (req_comp && req_comp != p->s->img_out_n) { + if (ri->bits_per_channel == 8) + result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + else + result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + p->s->img_out_n = req_comp; + if (result == NULL) return result; + } + *x = p->s->img_x; + *y = p->s->img_y; + if (n) *n = p->s->img_n; + } + STBI_FREE(p->out); p->out = NULL; + STBI_FREE(p->expanded); p->expanded = NULL; + STBI_FREE(p->idata); p->idata = NULL; + + return result; +} + +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi__png p; + p.s = s; + return stbi__do_png(&p, x,y,comp,req_comp, ri); +} + +static int stbi__png_test(stbi__context *s) +{ + int r; + r = stbi__check_png_header(s); + stbi__rewind(s); + return r; +} + +static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp) +{ + if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) { + stbi__rewind( p->s ); + return 0; + } + if (x) *x = p->s->img_x; + if (y) *y = p->s->img_y; + if (comp) *comp = p->s->img_n; + return 1; +} + +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__png p; + p.s = s; + return stbi__png_info_raw(&p, x, y, comp); +} + +static int stbi__png_is16(stbi__context *s) +{ + stbi__png p; + p.s = s; + if (!stbi__png_info_raw(&p, NULL, NULL, NULL)) + return 0; + if (p.depth != 16) { + stbi__rewind(p.s); + return 0; + } + return 1; +} +#endif + +// Microsoft/Windows BMP image + +#ifndef STBI_NO_BMP +static int stbi__bmp_test_raw(stbi__context *s) +{ + int r; + int sz; + if (stbi__get8(s) != 'B') return 0; + if (stbi__get8(s) != 'M') return 0; + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + stbi__get32le(s); // discard data offset + sz = stbi__get32le(s); + r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124); + return r; +} + +static int stbi__bmp_test(stbi__context *s) +{ + int r = stbi__bmp_test_raw(s); + stbi__rewind(s); + return r; +} + + +// returns 0..31 for the highest set bit +static int stbi__high_bit(unsigned int z) +{ + int n=0; + if (z == 0) return -1; + if (z >= 0x10000) { n += 16; z >>= 16; } + if (z >= 0x00100) { n += 8; z >>= 8; } + if (z >= 0x00010) { n += 4; z >>= 4; } + if (z >= 0x00004) { n += 2; z >>= 2; } + if (z >= 0x00002) { n += 1;/* >>= 1;*/ } + return n; +} + +static int stbi__bitcount(unsigned int a) +{ + a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2 + a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4 + a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits + a = (a + (a >> 8)); // max 16 per 8 bits + a = (a + (a >> 16)); // max 32 per 8 bits + return a & 0xff; +} + +// extract an arbitrarily-aligned N-bit value (N=bits) +// from v, and then make it 8-bits long and fractionally +// extend it to full full range. +static int stbi__shiftsigned(unsigned int v, int shift, int bits) +{ + static unsigned int mul_table[9] = { + 0, + 0xff/*0b11111111*/, 0x55/*0b01010101*/, 0x49/*0b01001001*/, 0x11/*0b00010001*/, + 0x21/*0b00100001*/, 0x41/*0b01000001*/, 0x81/*0b10000001*/, 0x01/*0b00000001*/, + }; + static unsigned int shift_table[9] = { + 0, 0,0,1,0,2,4,6,0, + }; + if (shift < 0) + v <<= -shift; + else + v >>= shift; + STBI_ASSERT(v < 256); + v >>= (8-bits); + STBI_ASSERT(bits >= 0 && bits <= 8); + return (int) ((unsigned) v * mul_table[bits]) >> shift_table[bits]; +} + +typedef struct +{ + int bpp, offset, hsz; + unsigned int mr,mg,mb,ma, all_a; + int extra_read; +} stbi__bmp_data; + +static int stbi__bmp_set_mask_defaults(stbi__bmp_data *info, int compress) +{ + // BI_BITFIELDS specifies masks explicitly, don't override + if (compress == 3) + return 1; + + if (compress == 0) { + if (info->bpp == 16) { + info->mr = 31u << 10; + info->mg = 31u << 5; + info->mb = 31u << 0; + } else if (info->bpp == 32) { + info->mr = 0xffu << 16; + info->mg = 0xffu << 8; + info->mb = 0xffu << 0; + info->ma = 0xffu << 24; + info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0 + } else { + // otherwise, use defaults, which is all-0 + info->mr = info->mg = info->mb = info->ma = 0; + } + return 1; + } + return 0; // error +} + +static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info) +{ + int hsz; + if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP"); + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + info->offset = stbi__get32le(s); + info->hsz = hsz = stbi__get32le(s); + info->mr = info->mg = info->mb = info->ma = 0; + info->extra_read = 14; + + if (info->offset < 0) return stbi__errpuc("bad BMP", "bad BMP"); + + if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown"); + if (hsz == 12) { + s->img_x = stbi__get16le(s); + s->img_y = stbi__get16le(s); + } else { + s->img_x = stbi__get32le(s); + s->img_y = stbi__get32le(s); + } + if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP"); + info->bpp = stbi__get16le(s); + if (hsz != 12) { + int compress = stbi__get32le(s); + if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE"); + if (compress >= 4) return stbi__errpuc("BMP JPEG/PNG", "BMP type not supported: unsupported compression"); // this includes PNG/JPEG modes + if (compress == 3 && info->bpp != 16 && info->bpp != 32) return stbi__errpuc("bad BMP", "bad BMP"); // bitfields requires 16 or 32 bits/pixel + stbi__get32le(s); // discard sizeof + stbi__get32le(s); // discard hres + stbi__get32le(s); // discard vres + stbi__get32le(s); // discard colorsused + stbi__get32le(s); // discard max important + if (hsz == 40 || hsz == 56) { + if (hsz == 56) { + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + } + if (info->bpp == 16 || info->bpp == 32) { + if (compress == 0) { + stbi__bmp_set_mask_defaults(info, compress); + } else if (compress == 3) { + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->extra_read += 12; + // not documented, but generated by photoshop and handled by mspaint + if (info->mr == info->mg && info->mg == info->mb) { + // ?!?!? + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else { + // V4/V5 header + int i; + if (hsz != 108 && hsz != 124) + return stbi__errpuc("bad BMP", "bad BMP"); + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->ma = stbi__get32le(s); + if (compress != 3) // override mr/mg/mb unless in BI_BITFIELDS mode, as per docs + stbi__bmp_set_mask_defaults(info, compress); + stbi__get32le(s); // discard color space + for (i=0; i < 12; ++i) + stbi__get32le(s); // discard color space parameters + if (hsz == 124) { + stbi__get32le(s); // discard rendering intent + stbi__get32le(s); // discard offset of profile data + stbi__get32le(s); // discard size of profile data + stbi__get32le(s); // discard reserved + } + } + } + return (void *) 1; +} + + +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + unsigned int mr=0,mg=0,mb=0,ma=0, all_a; + stbi_uc pal[256][4]; + int psize=0,i,j,width; + int flip_vertically, pad, target; + stbi__bmp_data info; + STBI_NOTUSED(ri); + + info.all_a = 255; + if (stbi__bmp_parse_header(s, &info) == NULL) + return NULL; // error code already set + + flip_vertically = ((int) s->img_y) > 0; + s->img_y = abs((int) s->img_y); + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + mr = info.mr; + mg = info.mg; + mb = info.mb; + ma = info.ma; + all_a = info.all_a; + + if (info.hsz == 12) { + if (info.bpp < 24) + psize = (info.offset - info.extra_read - 24) / 3; + } else { + if (info.bpp < 16) + psize = (info.offset - info.extra_read - info.hsz) >> 2; + } + if (psize == 0) { + // accept some number of extra bytes after the header, but if the offset points either to before + // the header ends or implies a large amount of extra data, reject the file as malformed + int bytes_read_so_far = s->callback_already_read + (int)(s->img_buffer - s->img_buffer_original); + int header_limit = 1024; // max we actually read is below 256 bytes currently. + int extra_data_limit = 256*4; // what ordinarily goes here is a palette; 256 entries*4 bytes is its max size. + if (bytes_read_so_far <= 0 || bytes_read_so_far > header_limit) { + return stbi__errpuc("bad header", "Corrupt BMP"); + } + // we established that bytes_read_so_far is positive and sensible. + // the first half of this test rejects offsets that are either too small positives, or + // negative, and guarantees that info.offset >= bytes_read_so_far > 0. this in turn + // ensures the number computed in the second half of the test can't overflow. + if (info.offset < bytes_read_so_far || info.offset - bytes_read_so_far > extra_data_limit) { + return stbi__errpuc("bad offset", "Corrupt BMP"); + } else { + stbi__skip(s, info.offset - bytes_read_so_far); + } + } + + if (info.bpp == 24 && ma == 0xff000000) + s->img_n = 3; + else + s->img_n = ma ? 4 : 3; + if (req_comp && req_comp >= 3) // we can directly decode 3 or 4 + target = req_comp; + else + target = s->img_n; // if they want monochrome, we'll post-convert + + // sanity-check size + if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0)) + return stbi__errpuc("too large", "Corrupt BMP"); + + out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + if (info.bpp < 16) { + int z=0; + if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); } + for (i=0; i < psize; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + if (info.hsz != 12) stbi__get8(s); + pal[i][3] = 255; + } + stbi__skip(s, info.offset - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4)); + if (info.bpp == 1) width = (s->img_x + 7) >> 3; + else if (info.bpp == 4) width = (s->img_x + 1) >> 1; + else if (info.bpp == 8) width = s->img_x; + else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); } + pad = (-width)&3; + if (info.bpp == 1) { + for (j=0; j < (int) s->img_y; ++j) { + int bit_offset = 7, v = stbi__get8(s); + for (i=0; i < (int) s->img_x; ++i) { + int color = (v>>bit_offset)&0x1; + out[z++] = pal[color][0]; + out[z++] = pal[color][1]; + out[z++] = pal[color][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + if((--bit_offset) < 0) { + bit_offset = 7; + v = stbi__get8(s); + } + } + stbi__skip(s, pad); + } + } else { + for (j=0; j < (int) s->img_y; ++j) { + for (i=0; i < (int) s->img_x; i += 2) { + int v=stbi__get8(s),v2=0; + if (info.bpp == 4) { + v2 = v & 15; + v >>= 4; + } + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + v = (info.bpp == 8) ? stbi__get8(s) : v2; + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + } + stbi__skip(s, pad); + } + } + } else { + int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0; + int z = 0; + int easy=0; + stbi__skip(s, info.offset - info.extra_read - info.hsz); + if (info.bpp == 24) width = 3 * s->img_x; + else if (info.bpp == 16) width = 2*s->img_x; + else /* bpp = 32 and pad = 0 */ width=0; + pad = (-width) & 3; + if (info.bpp == 24) { + easy = 1; + } else if (info.bpp == 32) { + if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000) + easy = 2; + } + if (!easy) { + if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + // right shift amt to put high bit in position #7 + rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr); + gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg); + bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb); + ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma); + if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + } + for (j=0; j < (int) s->img_y; ++j) { + if (easy) { + for (i=0; i < (int) s->img_x; ++i) { + unsigned char a; + out[z+2] = stbi__get8(s); + out[z+1] = stbi__get8(s); + out[z+0] = stbi__get8(s); + z += 3; + a = (easy == 2 ? stbi__get8(s) : 255); + all_a |= a; + if (target == 4) out[z++] = a; + } + } else { + int bpp = info.bpp; + for (i=0; i < (int) s->img_x; ++i) { + stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s)); + unsigned int a; + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount)); + a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255); + all_a |= a; + if (target == 4) out[z++] = STBI__BYTECAST(a); + } + } + stbi__skip(s, pad); + } + } + + // if alpha channel is all 0s, replace with all 255s + if (target == 4 && all_a == 0) + for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4) + out[i] = 255; + + if (flip_vertically) { + stbi_uc t; + for (j=0; j < (int) s->img_y>>1; ++j) { + stbi_uc *p1 = out + j *s->img_x*target; + stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target; + for (i=0; i < (int) s->img_x*target; ++i) { + t = p1[i]; p1[i] = p2[i]; p2[i] = t; + } + } + } + + if (req_comp && req_comp != target) { + out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + return out; +} +#endif + +// Targa Truevision - TGA +// by Jonathan Dummer +#ifndef STBI_NO_TGA +// returns STBI_rgb or whatever, 0 on error +static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16) +{ + // only RGB or RGBA (incl. 16bit) or grey allowed + if (is_rgb16) *is_rgb16 = 0; + switch(bits_per_pixel) { + case 8: return STBI_grey; + case 16: if(is_grey) return STBI_grey_alpha; + // fallthrough + case 15: if(is_rgb16) *is_rgb16 = 1; + return STBI_rgb; + case 24: // fallthrough + case 32: return bits_per_pixel/8; + default: return 0; + } +} + +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp) +{ + int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp; + int sz, tga_colormap_type; + stbi__get8(s); // discard Offset + tga_colormap_type = stbi__get8(s); // colormap type + if( tga_colormap_type > 1 ) { + stbi__rewind(s); + return 0; // only RGB or indexed allowed + } + tga_image_type = stbi__get8(s); // image type + if ( tga_colormap_type == 1 ) { // colormapped (paletted) image + if (tga_image_type != 1 && tga_image_type != 9) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip image x and y origin + tga_colormap_bpp = sz; + } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE + if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) { + stbi__rewind(s); + return 0; // only RGB or grey allowed, +/- RLE + } + stbi__skip(s,9); // skip colormap specification and image x/y origin + tga_colormap_bpp = 0; + } + tga_w = stbi__get16le(s); + if( tga_w < 1 ) { + stbi__rewind(s); + return 0; // test width + } + tga_h = stbi__get16le(s); + if( tga_h < 1 ) { + stbi__rewind(s); + return 0; // test height + } + tga_bits_per_pixel = stbi__get8(s); // bits per pixel + stbi__get8(s); // ignore alpha bits + if (tga_colormap_bpp != 0) { + if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) { + // when using a colormap, tga_bits_per_pixel is the size of the indexes + // I don't think anything but 8 or 16bit indexes makes sense + stbi__rewind(s); + return 0; + } + tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL); + } else { + tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL); + } + if(!tga_comp) { + stbi__rewind(s); + return 0; + } + if (x) *x = tga_w; + if (y) *y = tga_h; + if (comp) *comp = tga_comp; + return 1; // seems to have passed everything +} + +static int stbi__tga_test(stbi__context *s) +{ + int res = 0; + int sz, tga_color_type; + stbi__get8(s); // discard Offset + tga_color_type = stbi__get8(s); // color type + if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed + sz = stbi__get8(s); // image type + if ( tga_color_type == 1 ) { // colormapped (paletted) image + if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9 + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + stbi__skip(s,4); // skip image x and y origin + } else { // "normal" image w/o colormap + if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE + stbi__skip(s,9); // skip colormap specification and image x/y origin + } + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height + sz = stbi__get8(s); // bits per pixel + if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + + res = 1; // if we got this far, everything's good and we can return 1 instead of 0 + +errorEnd: + stbi__rewind(s); + return res; +} + +// read 16bit value and convert to 24bit RGB +static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out) +{ + stbi__uint16 px = (stbi__uint16)stbi__get16le(s); + stbi__uint16 fiveBitMask = 31; + // we have 3 channels with 5bits each + int r = (px >> 10) & fiveBitMask; + int g = (px >> 5) & fiveBitMask; + int b = px & fiveBitMask; + // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later + out[0] = (stbi_uc)((r * 255)/31); + out[1] = (stbi_uc)((g * 255)/31); + out[2] = (stbi_uc)((b * 255)/31); + + // some people claim that the most significant bit might be used for alpha + // (possibly if an alpha-bit is set in the "image descriptor byte") + // but that only made 16bit test images completely translucent.. + // so let's treat all 15 and 16bit TGAs as RGB with no alpha. +} + +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + // read in the TGA header stuff + int tga_offset = stbi__get8(s); + int tga_indexed = stbi__get8(s); + int tga_image_type = stbi__get8(s); + int tga_is_RLE = 0; + int tga_palette_start = stbi__get16le(s); + int tga_palette_len = stbi__get16le(s); + int tga_palette_bits = stbi__get8(s); + int tga_x_origin = stbi__get16le(s); + int tga_y_origin = stbi__get16le(s); + int tga_width = stbi__get16le(s); + int tga_height = stbi__get16le(s); + int tga_bits_per_pixel = stbi__get8(s); + int tga_comp, tga_rgb16=0; + int tga_inverted = stbi__get8(s); + // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?) + // image data + unsigned char *tga_data; + unsigned char *tga_palette = NULL; + int i, j; + unsigned char raw_data[4] = {0}; + int RLE_count = 0; + int RLE_repeating = 0; + int read_next_pixel = 1; + STBI_NOTUSED(ri); + STBI_NOTUSED(tga_x_origin); // @TODO + STBI_NOTUSED(tga_y_origin); // @TODO + + if (tga_height > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (tga_width > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // do a tiny bit of precessing + if ( tga_image_type >= 8 ) + { + tga_image_type -= 8; + tga_is_RLE = 1; + } + tga_inverted = 1 - ((tga_inverted >> 5) & 1); + + // If I'm paletted, then I'll use the number of bits from the palette + if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16); + else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16); + + if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency + return stbi__errpuc("bad format", "Can't find out TGA pixelformat"); + + // tga info + *x = tga_width; + *y = tga_height; + if (comp) *comp = tga_comp; + + if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0)) + return stbi__errpuc("too large", "Corrupt TGA"); + + tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0); + if (!tga_data) return stbi__errpuc("outofmem", "Out of memory"); + + // skip to the data's starting position (offset usually = 0) + stbi__skip(s, tga_offset ); + + if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) { + for (i=0; i < tga_height; ++i) { + int row = tga_inverted ? tga_height -i - 1 : i; + stbi_uc *tga_row = tga_data + row*tga_width*tga_comp; + stbi__getn(s, tga_row, tga_width * tga_comp); + } + } else { + // do I need to load a palette? + if ( tga_indexed) + { + if (tga_palette_len == 0) { /* you have to have at least one entry! */ + STBI_FREE(tga_data); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + + // any data to skip? (offset usually = 0) + stbi__skip(s, tga_palette_start ); + // load the palette + tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0); + if (!tga_palette) { + STBI_FREE(tga_data); + return stbi__errpuc("outofmem", "Out of memory"); + } + if (tga_rgb16) { + stbi_uc *pal_entry = tga_palette; + STBI_ASSERT(tga_comp == STBI_rgb); + for (i=0; i < tga_palette_len; ++i) { + stbi__tga_read_rgb16(s, pal_entry); + pal_entry += tga_comp; + } + } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) { + STBI_FREE(tga_data); + STBI_FREE(tga_palette); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + } + // load the data + for (i=0; i < tga_width * tga_height; ++i) + { + // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk? + if ( tga_is_RLE ) + { + if ( RLE_count == 0 ) + { + // yep, get the next byte as a RLE command + int RLE_cmd = stbi__get8(s); + RLE_count = 1 + (RLE_cmd & 127); + RLE_repeating = RLE_cmd >> 7; + read_next_pixel = 1; + } else if ( !RLE_repeating ) + { + read_next_pixel = 1; + } + } else + { + read_next_pixel = 1; + } + // OK, if I need to read a pixel, do it now + if ( read_next_pixel ) + { + // load however much data we did have + if ( tga_indexed ) + { + // read in index, then perform the lookup + int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s); + if ( pal_idx >= tga_palette_len ) { + // invalid index + pal_idx = 0; + } + pal_idx *= tga_comp; + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = tga_palette[pal_idx+j]; + } + } else if(tga_rgb16) { + STBI_ASSERT(tga_comp == STBI_rgb); + stbi__tga_read_rgb16(s, raw_data); + } else { + // read in the data raw + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = stbi__get8(s); + } + } + // clear the reading flag for the next pixel + read_next_pixel = 0; + } // end of reading a pixel + + // copy data + for (j = 0; j < tga_comp; ++j) + tga_data[i*tga_comp+j] = raw_data[j]; + + // in case we're in RLE mode, keep counting down + --RLE_count; + } + // do I need to invert the image? + if ( tga_inverted ) + { + for (j = 0; j*2 < tga_height; ++j) + { + int index1 = j * tga_width * tga_comp; + int index2 = (tga_height - 1 - j) * tga_width * tga_comp; + for (i = tga_width * tga_comp; i > 0; --i) + { + unsigned char temp = tga_data[index1]; + tga_data[index1] = tga_data[index2]; + tga_data[index2] = temp; + ++index1; + ++index2; + } + } + } + // clear my palette, if I had one + if ( tga_palette != NULL ) + { + STBI_FREE( tga_palette ); + } + } + + // swap RGB - if the source data was RGB16, it already is in the right order + if (tga_comp >= 3 && !tga_rgb16) + { + unsigned char* tga_pixel = tga_data; + for (i=0; i < tga_width * tga_height; ++i) + { + unsigned char temp = tga_pixel[0]; + tga_pixel[0] = tga_pixel[2]; + tga_pixel[2] = temp; + tga_pixel += tga_comp; + } + } + + // convert to target component count + if (req_comp && req_comp != tga_comp) + tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height); + + // the things I do to get rid of an error message, and yet keep + // Microsoft's C compilers happy... [8^( + tga_palette_start = tga_palette_len = tga_palette_bits = + tga_x_origin = tga_y_origin = 0; + STBI_NOTUSED(tga_palette_start); + // OK, done + return tga_data; +} +#endif + +// ************************************************************************************************* +// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s) +{ + int r = (stbi__get32be(s) == 0x38425053); + stbi__rewind(s); + return r; +} + +static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount) +{ + int count, nleft, len; + + count = 0; + while ((nleft = pixelCount - count) > 0) { + len = stbi__get8(s); + if (len == 128) { + // No-op. + } else if (len < 128) { + // Copy next len+1 bytes literally. + len++; + if (len > nleft) return 0; // corrupt data + count += len; + while (len) { + *p = stbi__get8(s); + p += 4; + len--; + } + } else if (len > 128) { + stbi_uc val; + // Next -len+1 bytes in the dest are replicated from next source byte. + // (Interpret len as a negative 8-bit int.) + len = 257 - len; + if (len > nleft) return 0; // corrupt data + val = stbi__get8(s); + count += len; + while (len) { + *p = val; + p += 4; + len--; + } + } + } + + return 1; +} + +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + int pixelCount; + int channelCount, compression; + int channel, i; + int bitdepth; + int w,h; + stbi_uc *out; + STBI_NOTUSED(ri); + + // Check identifier + if (stbi__get32be(s) != 0x38425053) // "8BPS" + return stbi__errpuc("not PSD", "Corrupt PSD image"); + + // Check file type version. + if (stbi__get16be(s) != 1) + return stbi__errpuc("wrong version", "Unsupported version of PSD image"); + + // Skip 6 reserved bytes. + stbi__skip(s, 6 ); + + // Read the number of channels (R, G, B, A, etc). + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) + return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image"); + + // Read the rows and columns of the image. + h = stbi__get32be(s); + w = stbi__get32be(s); + + if (h > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (w > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // Make sure the depth is 8 bits. + bitdepth = stbi__get16be(s); + if (bitdepth != 8 && bitdepth != 16) + return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit"); + + // Make sure the color mode is RGB. + // Valid options are: + // 0: Bitmap + // 1: Grayscale + // 2: Indexed color + // 3: RGB color + // 4: CMYK color + // 7: Multichannel + // 8: Duotone + // 9: Lab color + if (stbi__get16be(s) != 3) + return stbi__errpuc("wrong color format", "PSD is not in RGB color format"); + + // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.) + stbi__skip(s,stbi__get32be(s) ); + + // Skip the image resources. (resolution, pen tool paths, etc) + stbi__skip(s, stbi__get32be(s) ); + + // Skip the reserved data. + stbi__skip(s, stbi__get32be(s) ); + + // Find out if the data is compressed. + // Known values: + // 0: no compression + // 1: RLE compressed + compression = stbi__get16be(s); + if (compression > 1) + return stbi__errpuc("bad compression", "PSD has an unknown compression format"); + + // Check size + if (!stbi__mad3sizes_valid(4, w, h, 0)) + return stbi__errpuc("too large", "Corrupt PSD"); + + // Create the destination image. + + if (!compression && bitdepth == 16 && bpc == 16) { + out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0); + ri->bits_per_channel = 16; + } else + out = (stbi_uc *) stbi__malloc(4 * w*h); + + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + pixelCount = w*h; + + // Initialize the data to zero. + //memset( out, 0, pixelCount * 4 ); + + // Finally, the image data. + if (compression) { + // RLE as used by .PSD and .TIFF + // Loop until you get the number of unpacked bytes you are expecting: + // Read the next source byte into n. + // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally. + // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times. + // Else if n is 128, noop. + // Endloop + + // The RLE-compressed data is preceded by a 2-byte data count for each row in the data, + // which we're going to just skip. + stbi__skip(s, h * channelCount * 2 ); + + // Read the RLE data by channel. + for (channel = 0; channel < 4; channel++) { + stbi_uc *p; + + p = out+channel; + if (channel >= channelCount) { + // Fill this channel with default data. + for (i = 0; i < pixelCount; i++, p += 4) + *p = (channel == 3 ? 255 : 0); + } else { + // Read the RLE data. + if (!stbi__psd_decode_rle(s, p, pixelCount)) { + STBI_FREE(out); + return stbi__errpuc("corrupt", "bad RLE data"); + } + } + } + + } else { + // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...) + // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image. + + // Read the data by channel. + for (channel = 0; channel < 4; channel++) { + if (channel >= channelCount) { + // Fill this channel with default data. + if (bitdepth == 16 && bpc == 16) { + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + stbi__uint16 val = channel == 3 ? 65535 : 0; + for (i = 0; i < pixelCount; i++, q += 4) + *q = val; + } else { + stbi_uc *p = out+channel; + stbi_uc val = channel == 3 ? 255 : 0; + for (i = 0; i < pixelCount; i++, p += 4) + *p = val; + } + } else { + if (ri->bits_per_channel == 16) { // output bpc + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + for (i = 0; i < pixelCount; i++, q += 4) + *q = (stbi__uint16) stbi__get16be(s); + } else { + stbi_uc *p = out+channel; + if (bitdepth == 16) { // input bpc + for (i = 0; i < pixelCount; i++, p += 4) + *p = (stbi_uc) (stbi__get16be(s) >> 8); + } else { + for (i = 0; i < pixelCount; i++, p += 4) + *p = stbi__get8(s); + } + } + } + } + } + + // remove weird white matte from PSD + if (channelCount >= 4) { + if (ri->bits_per_channel == 16) { + for (i=0; i < w*h; ++i) { + stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i; + if (pixel[3] != 0 && pixel[3] != 65535) { + float a = pixel[3] / 65535.0f; + float ra = 1.0f / a; + float inv_a = 65535.0f * (1 - ra); + pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a); + pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a); + pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a); + } + } + } else { + for (i=0; i < w*h; ++i) { + unsigned char *pixel = out + 4*i; + if (pixel[3] != 0 && pixel[3] != 255) { + float a = pixel[3] / 255.0f; + float ra = 1.0f / a; + float inv_a = 255.0f * (1 - ra); + pixel[0] = (unsigned char) (pixel[0]*ra + inv_a); + pixel[1] = (unsigned char) (pixel[1]*ra + inv_a); + pixel[2] = (unsigned char) (pixel[2]*ra + inv_a); + } + } + } + } + + // convert to desired output format + if (req_comp && req_comp != 4) { + if (ri->bits_per_channel == 16) + out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h); + else + out = stbi__convert_format(out, 4, req_comp, w, h); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + if (comp) *comp = 4; + *y = h; + *x = w; + + return out; +} +#endif + +// ************************************************************************************************* +// Softimage PIC loader +// by Tom Seddon +// +// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format +// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/ + +#ifndef STBI_NO_PIC +static int stbi__pic_is4(stbi__context *s,const char *str) +{ + int i; + for (i=0; i<4; ++i) + if (stbi__get8(s) != (stbi_uc)str[i]) + return 0; + + return 1; +} + +static int stbi__pic_test_core(stbi__context *s) +{ + int i; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) + return 0; + + for(i=0;i<84;++i) + stbi__get8(s); + + if (!stbi__pic_is4(s,"PICT")) + return 0; + + return 1; +} + +typedef struct +{ + stbi_uc size,type,channel; +} stbi__pic_packet; + +static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest) +{ + int mask=0x80, i; + + for (i=0; i<4; ++i, mask>>=1) { + if (channel & mask) { + if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short"); + dest[i]=stbi__get8(s); + } + } + + return dest; +} + +static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src) +{ + int mask=0x80,i; + + for (i=0;i<4; ++i, mask>>=1) + if (channel&mask) + dest[i]=src[i]; +} + +static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result) +{ + int act_comp=0,num_packets=0,y,chained; + stbi__pic_packet packets[10]; + + // this will (should...) cater for even some bizarre stuff like having data + // for the same channel in multiple packets. + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return stbi__errpuc("bad format","too many packets"); + + packet = &packets[num_packets++]; + + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + + act_comp |= packet->channel; + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)"); + if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp"); + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel? + + for(y=0; ytype) { + default: + return stbi__errpuc("bad format","packet has bad compression type"); + + case 0: {//uncompressed + int x; + + for(x=0;xchannel,dest)) + return 0; + break; + } + + case 1://Pure RLE + { + int left=width, i; + + while (left>0) { + stbi_uc count,value[4]; + + count=stbi__get8(s); + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)"); + + if (count > left) + count = (stbi_uc) left; + + if (!stbi__readval(s,packet->channel,value)) return 0; + + for(i=0; ichannel,dest,value); + left -= count; + } + } + break; + + case 2: {//Mixed RLE + int left=width; + while (left>0) { + int count = stbi__get8(s), i; + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)"); + + if (count >= 128) { // Repeated + stbi_uc value[4]; + + if (count==128) + count = stbi__get16be(s); + else + count -= 127; + if (count > left) + return stbi__errpuc("bad file","scanline overrun"); + + if (!stbi__readval(s,packet->channel,value)) + return 0; + + for(i=0;ichannel,dest,value); + } else { // Raw + ++count; + if (count>left) return stbi__errpuc("bad file","scanline overrun"); + + for(i=0;ichannel,dest)) + return 0; + } + left-=count; + } + break; + } + } + } + } + + return result; +} + +static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri) +{ + stbi_uc *result; + int i, x,y, internal_comp; + STBI_NOTUSED(ri); + + if (!comp) comp = &internal_comp; + + for (i=0; i<92; ++i) + stbi__get8(s); + + x = stbi__get16be(s); + y = stbi__get16be(s); + + if (y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)"); + if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode"); + + stbi__get32be(s); //skip `ratio' + stbi__get16be(s); //skip `fields' + stbi__get16be(s); //skip `pad' + + // intermediate buffer is RGBA + result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0); + if (!result) return stbi__errpuc("outofmem", "Out of memory"); + memset(result, 0xff, x*y*4); + + if (!stbi__pic_load_core(s,x,y,comp, result)) { + STBI_FREE(result); + result=0; + } + *px = x; + *py = y; + if (req_comp == 0) req_comp = *comp; + result=stbi__convert_format(result,4,req_comp,x,y); + + return result; +} + +static int stbi__pic_test(stbi__context *s) +{ + int r = stbi__pic_test_core(s); + stbi__rewind(s); + return r; +} +#endif + +// ************************************************************************************************* +// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb + +#ifndef STBI_NO_GIF +typedef struct +{ + stbi__int16 prefix; + stbi_uc first; + stbi_uc suffix; +} stbi__gif_lzw; + +typedef struct +{ + int w,h; + stbi_uc *out; // output buffer (always 4 components) + stbi_uc *background; // The current "background" as far as a gif is concerned + stbi_uc *history; + int flags, bgindex, ratio, transparent, eflags; + stbi_uc pal[256][4]; + stbi_uc lpal[256][4]; + stbi__gif_lzw codes[8192]; + stbi_uc *color_table; + int parse, step; + int lflags; + int start_x, start_y; + int max_x, max_y; + int cur_x, cur_y; + int line_size; + int delay; +} stbi__gif; + +static int stbi__gif_test_raw(stbi__context *s) +{ + int sz; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0; + sz = stbi__get8(s); + if (sz != '9' && sz != '7') return 0; + if (stbi__get8(s) != 'a') return 0; + return 1; +} + +static int stbi__gif_test(stbi__context *s) +{ + int r = stbi__gif_test_raw(s); + stbi__rewind(s); + return r; +} + +static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp) +{ + int i; + for (i=0; i < num_entries; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + pal[i][3] = transp == i ? 0 : 255; + } +} + +static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info) +{ + stbi_uc version; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') + return stbi__err("not GIF", "Corrupt GIF"); + + version = stbi__get8(s); + if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF"); + if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF"); + + stbi__g_failure_reason = ""; + g->w = stbi__get16le(s); + g->h = stbi__get16le(s); + g->flags = stbi__get8(s); + g->bgindex = stbi__get8(s); + g->ratio = stbi__get8(s); + g->transparent = -1; + + if (g->w > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (g->h > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + + if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments + + if (is_info) return 1; + + if (g->flags & 0x80) + stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1); + + return 1; +} + +static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif)); + if (!g) return stbi__err("outofmem", "Out of memory"); + if (!stbi__gif_header(s, g, comp, 1)) { + STBI_FREE(g); + stbi__rewind( s ); + return 0; + } + if (x) *x = g->w; + if (y) *y = g->h; + STBI_FREE(g); + return 1; +} + +static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code) +{ + stbi_uc *p, *c; + int idx; + + // recurse to decode the prefixes, since the linked-list is backwards, + // and working backwards through an interleaved image would be nasty + if (g->codes[code].prefix >= 0) + stbi__out_gif_code(g, g->codes[code].prefix); + + if (g->cur_y >= g->max_y) return; + + idx = g->cur_x + g->cur_y; + p = &g->out[idx]; + g->history[idx / 4] = 1; + + c = &g->color_table[g->codes[code].suffix * 4]; + if (c[3] > 128) { // don't render transparent pixels; + p[0] = c[2]; + p[1] = c[1]; + p[2] = c[0]; + p[3] = c[3]; + } + g->cur_x += 4; + + if (g->cur_x >= g->max_x) { + g->cur_x = g->start_x; + g->cur_y += g->step; + + while (g->cur_y >= g->max_y && g->parse > 0) { + g->step = (1 << g->parse) * g->line_size; + g->cur_y = g->start_y + (g->step >> 1); + --g->parse; + } + } +} + +static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g) +{ + stbi_uc lzw_cs; + stbi__int32 len, init_code; + stbi__uint32 first; + stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear; + stbi__gif_lzw *p; + + lzw_cs = stbi__get8(s); + if (lzw_cs > 12) return NULL; + clear = 1 << lzw_cs; + first = 1; + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + bits = 0; + valid_bits = 0; + for (init_code = 0; init_code < clear; init_code++) { + g->codes[init_code].prefix = -1; + g->codes[init_code].first = (stbi_uc) init_code; + g->codes[init_code].suffix = (stbi_uc) init_code; + } + + // support no starting clear code + avail = clear+2; + oldcode = -1; + + len = 0; + for(;;) { + if (valid_bits < codesize) { + if (len == 0) { + len = stbi__get8(s); // start new block + if (len == 0) + return g->out; + } + --len; + bits |= (stbi__int32) stbi__get8(s) << valid_bits; + valid_bits += 8; + } else { + stbi__int32 code = bits & codemask; + bits >>= codesize; + valid_bits -= codesize; + // @OPTIMIZE: is there some way we can accelerate the non-clear path? + if (code == clear) { // clear code + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + avail = clear + 2; + oldcode = -1; + first = 0; + } else if (code == clear + 1) { // end of stream code + stbi__skip(s, len); + while ((len = stbi__get8(s)) > 0) + stbi__skip(s,len); + return g->out; + } else if (code <= avail) { + if (first) { + return stbi__errpuc("no clear code", "Corrupt GIF"); + } + + if (oldcode >= 0) { + p = &g->codes[avail++]; + if (avail > 8192) { + return stbi__errpuc("too many codes", "Corrupt GIF"); + } + + p->prefix = (stbi__int16) oldcode; + p->first = g->codes[oldcode].first; + p->suffix = (code == avail) ? p->first : g->codes[code].first; + } else if (code == avail) + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + + stbi__out_gif_code(g, (stbi__uint16) code); + + if ((avail & codemask) == 0 && avail <= 0x0FFF) { + codesize++; + codemask = (1 << codesize) - 1; + } + + oldcode = code; + } else { + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + } + } + } +} + +// this function is designed to support animated gifs, although stb_image doesn't support it +// two back is the image from two frames ago, used for a very specific disposal format +static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp, stbi_uc *two_back) +{ + int dispose; + int first_frame; + int pi; + int pcount; + STBI_NOTUSED(req_comp); + + // on first frame, any non-written pixels get the background colour (non-transparent) + first_frame = 0; + if (g->out == 0) { + if (!stbi__gif_header(s, g, comp,0)) return 0; // stbi__g_failure_reason set by stbi__gif_header + if (!stbi__mad3sizes_valid(4, g->w, g->h, 0)) + return stbi__errpuc("too large", "GIF image is too large"); + pcount = g->w * g->h; + g->out = (stbi_uc *) stbi__malloc(4 * pcount); + g->background = (stbi_uc *) stbi__malloc(4 * pcount); + g->history = (stbi_uc *) stbi__malloc(pcount); + if (!g->out || !g->background || !g->history) + return stbi__errpuc("outofmem", "Out of memory"); + + // image is treated as "transparent" at the start - ie, nothing overwrites the current background; + // background colour is only used for pixels that are not rendered first frame, after that "background" + // color refers to the color that was there the previous frame. + memset(g->out, 0x00, 4 * pcount); + memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent) + memset(g->history, 0x00, pcount); // pixels that were affected previous frame + first_frame = 1; + } else { + // second frame - how do we dispose of the previous one? + dispose = (g->eflags & 0x1C) >> 2; + pcount = g->w * g->h; + + if ((dispose == 3) && (two_back == 0)) { + dispose = 2; // if I don't have an image to revert back to, default to the old background + } + + if (dispose == 3) { // use previous graphic + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &two_back[pi * 4], 4 ); + } + } + } else if (dispose == 2) { + // restore what was changed last frame to background before that frame; + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &g->background[pi * 4], 4 ); + } + } + } else { + // This is a non-disposal case eithe way, so just + // leave the pixels as is, and they will become the new background + // 1: do not dispose + // 0: not specified. + } + + // background is what out is after the undoing of the previou frame; + memcpy( g->background, g->out, 4 * g->w * g->h ); + } + + // clear my history; + memset( g->history, 0x00, g->w * g->h ); // pixels that were affected previous frame + + for (;;) { + int tag = stbi__get8(s); + switch (tag) { + case 0x2C: /* Image Descriptor */ + { + stbi__int32 x, y, w, h; + stbi_uc *o; + + x = stbi__get16le(s); + y = stbi__get16le(s); + w = stbi__get16le(s); + h = stbi__get16le(s); + if (((x + w) > (g->w)) || ((y + h) > (g->h))) + return stbi__errpuc("bad Image Descriptor", "Corrupt GIF"); + + g->line_size = g->w * 4; + g->start_x = x * 4; + g->start_y = y * g->line_size; + g->max_x = g->start_x + w * 4; + g->max_y = g->start_y + h * g->line_size; + g->cur_x = g->start_x; + g->cur_y = g->start_y; + + // if the width of the specified rectangle is 0, that means + // we may not see *any* pixels or the image is malformed; + // to make sure this is caught, move the current y down to + // max_y (which is what out_gif_code checks). + if (w == 0) + g->cur_y = g->max_y; + + g->lflags = stbi__get8(s); + + if (g->lflags & 0x40) { + g->step = 8 * g->line_size; // first interlaced spacing + g->parse = 3; + } else { + g->step = g->line_size; + g->parse = 0; + } + + if (g->lflags & 0x80) { + stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1); + g->color_table = (stbi_uc *) g->lpal; + } else if (g->flags & 0x80) { + g->color_table = (stbi_uc *) g->pal; + } else + return stbi__errpuc("missing color table", "Corrupt GIF"); + + o = stbi__process_gif_raster(s, g); + if (!o) return NULL; + + // if this was the first frame, + pcount = g->w * g->h; + if (first_frame && (g->bgindex > 0)) { + // if first frame, any pixel not drawn to gets the background color + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi] == 0) { + g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be; + memcpy( &g->out[pi * 4], &g->pal[g->bgindex], 4 ); + } + } + } + + return o; + } + + case 0x21: // Comment Extension. + { + int len; + int ext = stbi__get8(s); + if (ext == 0xF9) { // Graphic Control Extension. + len = stbi__get8(s); + if (len == 4) { + g->eflags = stbi__get8(s); + g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths. + + // unset old transparent + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 255; + } + if (g->eflags & 0x01) { + g->transparent = stbi__get8(s); + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 0; + } + } else { + // don't need transparent + stbi__skip(s, 1); + g->transparent = -1; + } + } else { + stbi__skip(s, len); + break; + } + } + while ((len = stbi__get8(s)) != 0) { + stbi__skip(s, len); + } + break; + } + + case 0x3B: // gif stream termination code + return (stbi_uc *) s; // using '1' causes warning on some compilers + + default: + return stbi__errpuc("unknown code", "Corrupt GIF"); + } + } +} + +static void *stbi__load_gif_main_outofmem(stbi__gif *g, stbi_uc *out, int **delays) +{ + STBI_FREE(g->out); + STBI_FREE(g->history); + STBI_FREE(g->background); + + if (out) STBI_FREE(out); + if (delays && *delays) STBI_FREE(*delays); + return stbi__errpuc("outofmem", "Out of memory"); +} + +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + if (stbi__gif_test(s)) { + int layers = 0; + stbi_uc *u = 0; + stbi_uc *out = 0; + stbi_uc *two_back = 0; + stbi__gif g; + int stride; + int out_size = 0; + int delays_size = 0; + + STBI_NOTUSED(out_size); + STBI_NOTUSED(delays_size); + + memset(&g, 0, sizeof(g)); + if (delays) { + *delays = 0; + } + + do { + u = stbi__gif_load_next(s, &g, comp, req_comp, two_back); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + + if (u) { + *x = g.w; + *y = g.h; + ++layers; + stride = g.w * g.h * 4; + + if (out) { + void *tmp = (stbi_uc*) STBI_REALLOC_SIZED( out, out_size, layers * stride ); + if (!tmp) + return stbi__load_gif_main_outofmem(&g, out, delays); + else { + out = (stbi_uc*) tmp; + out_size = layers * stride; + } + + if (delays) { + int *new_delays = (int*) STBI_REALLOC_SIZED( *delays, delays_size, sizeof(int) * layers ); + if (!new_delays) + return stbi__load_gif_main_outofmem(&g, out, delays); + *delays = new_delays; + delays_size = layers * sizeof(int); + } + } else { + out = (stbi_uc*)stbi__malloc( layers * stride ); + if (!out) + return stbi__load_gif_main_outofmem(&g, out, delays); + out_size = layers * stride; + if (delays) { + *delays = (int*) stbi__malloc( layers * sizeof(int) ); + if (!*delays) + return stbi__load_gif_main_outofmem(&g, out, delays); + delays_size = layers * sizeof(int); + } + } + memcpy( out + ((layers - 1) * stride), u, stride ); + if (layers >= 2) { + two_back = out - 2 * stride; + } + + if (delays) { + (*delays)[layers - 1U] = g.delay; + } + } + } while (u != 0); + + // free temp buffer; + STBI_FREE(g.out); + STBI_FREE(g.history); + STBI_FREE(g.background); + + // do the final conversion after loading everything; + if (req_comp && req_comp != 4) + out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h); + + *z = layers; + return out; + } else { + return stbi__errpuc("not GIF", "Image was not as a gif type."); + } +} + +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *u = 0; + stbi__gif g; + memset(&g, 0, sizeof(g)); + STBI_NOTUSED(ri); + + u = stbi__gif_load_next(s, &g, comp, req_comp, 0); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + if (u) { + *x = g.w; + *y = g.h; + + // moved conversion to after successful load so that the same + // can be done for multiple frames. + if (req_comp && req_comp != 4) + u = stbi__convert_format(u, 4, req_comp, g.w, g.h); + } else if (g.out) { + // if there was an error and we allocated an image buffer, free it! + STBI_FREE(g.out); + } + + // free buffers needed for multiple frame loading; + STBI_FREE(g.history); + STBI_FREE(g.background); + + return u; +} + +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp) +{ + return stbi__gif_info_raw(s,x,y,comp); +} +#endif + +// ************************************************************************************************* +// Radiance RGBE HDR loader +// originally by Nicolas Schulz +#ifndef STBI_NO_HDR +static int stbi__hdr_test_core(stbi__context *s, const char *signature) +{ + int i; + for (i=0; signature[i]; ++i) + if (stbi__get8(s) != signature[i]) + return 0; + stbi__rewind(s); + return 1; +} + +static int stbi__hdr_test(stbi__context* s) +{ + int r = stbi__hdr_test_core(s, "#?RADIANCE\n"); + stbi__rewind(s); + if(!r) { + r = stbi__hdr_test_core(s, "#?RGBE\n"); + stbi__rewind(s); + } + return r; +} + +#define STBI__HDR_BUFLEN 1024 +static char *stbi__hdr_gettoken(stbi__context *z, char *buffer) +{ + int len=0; + char c = '\0'; + + c = (char) stbi__get8(z); + + while (!stbi__at_eof(z) && c != '\n') { + buffer[len++] = c; + if (len == STBI__HDR_BUFLEN-1) { + // flush to end of line + while (!stbi__at_eof(z) && stbi__get8(z) != '\n') + ; + break; + } + c = (char) stbi__get8(z); + } + + buffer[len] = 0; + return buffer; +} + +static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp) +{ + if ( input[3] != 0 ) { + float f1; + // Exponent + f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8)); + if (req_comp <= 2) + output[0] = (input[0] + input[1] + input[2]) * f1 / 3; + else { + output[0] = input[0] * f1; + output[1] = input[1] * f1; + output[2] = input[2] * f1; + } + if (req_comp == 2) output[1] = 1; + if (req_comp == 4) output[3] = 1; + } else { + switch (req_comp) { + case 4: output[3] = 1; /* fallthrough */ + case 3: output[0] = output[1] = output[2] = 0; + break; + case 2: output[1] = 1; /* fallthrough */ + case 1: output[0] = 0; + break; + } + } +} + +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int width, height; + stbi_uc *scanline; + float *hdr_data; + int len; + unsigned char count, value; + int i, j, k, c1,c2, z; + const char *headerToken; + STBI_NOTUSED(ri); + + // Check identifier + headerToken = stbi__hdr_gettoken(s,buffer); + if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0) + return stbi__errpf("not HDR", "Corrupt HDR image"); + + // Parse header + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format"); + + // Parse width and height + // can't use sscanf() if we're not using stdio! + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + height = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + width = (int) strtol(token, NULL, 10); + + if (height > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + if (width > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + + *x = width; + *y = height; + + if (comp) *comp = 3; + if (req_comp == 0) req_comp = 3; + + if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0)) + return stbi__errpf("too large", "HDR image is too large"); + + // Read data + hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0); + if (!hdr_data) + return stbi__errpf("outofmem", "Out of memory"); + + // Load image data + // image data is stored as some number of sca + if ( width < 8 || width >= 32768) { + // Read flat data + for (j=0; j < height; ++j) { + for (i=0; i < width; ++i) { + stbi_uc rgbe[4]; + main_decode_loop: + stbi__getn(s, rgbe, 4); + stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp); + } + } + } else { + // Read RLE-encoded data + scanline = NULL; + + for (j = 0; j < height; ++j) { + c1 = stbi__get8(s); + c2 = stbi__get8(s); + len = stbi__get8(s); + if (c1 != 2 || c2 != 2 || (len & 0x80)) { + // not run-length encoded, so we have to actually use THIS data as a decoded + // pixel (note this can't be a valid pixel--one of RGB must be >= 128) + stbi_uc rgbe[4]; + rgbe[0] = (stbi_uc) c1; + rgbe[1] = (stbi_uc) c2; + rgbe[2] = (stbi_uc) len; + rgbe[3] = (stbi_uc) stbi__get8(s); + stbi__hdr_convert(hdr_data, rgbe, req_comp); + i = 1; + j = 0; + STBI_FREE(scanline); + goto main_decode_loop; // yes, this makes no sense + } + len <<= 8; + len |= stbi__get8(s); + if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); } + if (scanline == NULL) { + scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0); + if (!scanline) { + STBI_FREE(hdr_data); + return stbi__errpf("outofmem", "Out of memory"); + } + } + + for (k = 0; k < 4; ++k) { + int nleft; + i = 0; + while ((nleft = width - i) > 0) { + count = stbi__get8(s); + if (count > 128) { + // Run + value = stbi__get8(s); + count -= 128; + if ((count == 0) || (count > nleft)) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = value; + } else { + // Dump + if ((count == 0) || (count > nleft)) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = stbi__get8(s); + } + } + } + for (i=0; i < width; ++i) + stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp); + } + if (scanline) + STBI_FREE(scanline); + } + + return hdr_data; +} + +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int dummy; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (stbi__hdr_test(s) == 0) { + stbi__rewind( s ); + return 0; + } + + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) { + stbi__rewind( s ); + return 0; + } + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *y = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *x = (int) strtol(token, NULL, 10); + *comp = 3; + return 1; +} +#endif // STBI_NO_HDR + +#ifndef STBI_NO_BMP +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp) +{ + void *p; + stbi__bmp_data info; + + info.all_a = 255; + p = stbi__bmp_parse_header(s, &info); + if (p == NULL) { + stbi__rewind( s ); + return 0; + } + if (x) *x = s->img_x; + if (y) *y = s->img_y; + if (comp) { + if (info.bpp == 24 && info.ma == 0xff000000) + *comp = 3; + else + *comp = info.ma ? 4 : 3; + } + return 1; +} +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp) +{ + int channelCount, dummy, depth; + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + *y = stbi__get32be(s); + *x = stbi__get32be(s); + depth = stbi__get16be(s); + if (depth != 8 && depth != 16) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 3) { + stbi__rewind( s ); + return 0; + } + *comp = 4; + return 1; +} + +static int stbi__psd_is16(stbi__context *s) +{ + int channelCount, depth; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + STBI_NOTUSED(stbi__get32be(s)); + STBI_NOTUSED(stbi__get32be(s)); + depth = stbi__get16be(s); + if (depth != 16) { + stbi__rewind( s ); + return 0; + } + return 1; +} +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp) +{ + int act_comp=0,num_packets=0,chained,dummy; + stbi__pic_packet packets[10]; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) { + stbi__rewind(s); + return 0; + } + + stbi__skip(s, 88); + + *x = stbi__get16be(s); + *y = stbi__get16be(s); + if (stbi__at_eof(s)) { + stbi__rewind( s); + return 0; + } + if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) { + stbi__rewind( s ); + return 0; + } + + stbi__skip(s, 8); + + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return 0; + + packet = &packets[num_packets++]; + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + act_comp |= packet->channel; + + if (stbi__at_eof(s)) { + stbi__rewind( s ); + return 0; + } + if (packet->size != 8) { + stbi__rewind( s ); + return 0; + } + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); + + return 1; +} +#endif + +// ************************************************************************************************* +// Portable Gray Map and Portable Pixel Map loader +// by Ken Miller +// +// PGM: http://netpbm.sourceforge.net/doc/pgm.html +// PPM: http://netpbm.sourceforge.net/doc/ppm.html +// +// Known limitations: +// Does not support comments in the header section +// Does not support ASCII image data (formats P2 and P3) + +#ifndef STBI_NO_PNM + +static int stbi__pnm_test(stbi__context *s) +{ + char p, t; + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind( s ); + return 0; + } + return 1; +} + +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + STBI_NOTUSED(ri); + + ri->bits_per_channel = stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n); + if (ri->bits_per_channel == 0) + return 0; + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + + if (!stbi__mad4sizes_valid(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0)) + return stbi__errpuc("too large", "PNM too large"); + + out = (stbi_uc *) stbi__malloc_mad4(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + if (!stbi__getn(s, out, s->img_n * s->img_x * s->img_y * (ri->bits_per_channel / 8))) { + STBI_FREE(out); + return stbi__errpuc("bad PNM", "PNM file truncated"); + } + + if (req_comp && req_comp != s->img_n) { + if (ri->bits_per_channel == 16) { + out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, s->img_n, req_comp, s->img_x, s->img_y); + } else { + out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y); + } + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + return out; +} + +static int stbi__pnm_isspace(char c) +{ + return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r'; +} + +static void stbi__pnm_skip_whitespace(stbi__context *s, char *c) +{ + for (;;) { + while (!stbi__at_eof(s) && stbi__pnm_isspace(*c)) + *c = (char) stbi__get8(s); + + if (stbi__at_eof(s) || *c != '#') + break; + + while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' ) + *c = (char) stbi__get8(s); + } +} + +static int stbi__pnm_isdigit(char c) +{ + return c >= '0' && c <= '9'; +} + +static int stbi__pnm_getinteger(stbi__context *s, char *c) +{ + int value = 0; + + while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) { + value = value*10 + (*c - '0'); + *c = (char) stbi__get8(s); + if((value > 214748364) || (value == 214748364 && *c > '7')) + return stbi__err("integer parse overflow", "Parsing an integer in the PPM header overflowed a 32-bit int"); + } + + return value; +} + +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp) +{ + int maxv, dummy; + char c, p, t; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + stbi__rewind(s); + + // Get identifier + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind(s); + return 0; + } + + *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm + + c = (char) stbi__get8(s); + stbi__pnm_skip_whitespace(s, &c); + + *x = stbi__pnm_getinteger(s, &c); // read width + if(*x == 0) + return stbi__err("invalid width", "PPM image header had zero or overflowing width"); + stbi__pnm_skip_whitespace(s, &c); + + *y = stbi__pnm_getinteger(s, &c); // read height + if (*y == 0) + return stbi__err("invalid width", "PPM image header had zero or overflowing width"); + stbi__pnm_skip_whitespace(s, &c); + + maxv = stbi__pnm_getinteger(s, &c); // read max value + if (maxv > 65535) + return stbi__err("max value > 65535", "PPM image supports only 8-bit and 16-bit images"); + else if (maxv > 255) + return 16; + else + return 8; +} + +static int stbi__pnm_is16(stbi__context *s) +{ + if (stbi__pnm_info(s, NULL, NULL, NULL) == 16) + return 1; + return 0; +} +#endif + +static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp) +{ + #ifndef STBI_NO_JPEG + if (stbi__jpeg_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNG + if (stbi__png_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_GIF + if (stbi__gif_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_BMP + if (stbi__bmp_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PIC + if (stbi__pic_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNM + if (stbi__pnm_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_info(s, x, y, comp)) return 1; + #endif + + // test tga last because it's a crappy test! + #ifndef STBI_NO_TGA + if (stbi__tga_info(s, x, y, comp)) + return 1; + #endif + return stbi__err("unknown image type", "Image not of any known type, or corrupt"); +} + +static int stbi__is_16_main(stbi__context *s) +{ + #ifndef STBI_NO_PNG + if (stbi__png_is16(s)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_is16(s)) return 1; + #endif + + #ifndef STBI_NO_PNM + if (stbi__pnm_is16(s)) return 1; + #endif + return 0; +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_info_from_file(f, x, y, comp); + fclose(f); + return result; +} + +STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__info_main(&s,x,y,comp); + fseek(f,pos,SEEK_SET); + return r; +} + +STBIDEF int stbi_is_16_bit(char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_is_16_bit_from_file(f); + fclose(f); + return result; +} + +STBIDEF int stbi_is_16_bit_from_file(FILE *f) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__is_16_main(&s); + fseek(f,pos,SEEK_SET); + return r; +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__is_16_main(&s); +} + +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *c, void *user) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__is_16_main(&s); +} + +#endif // STB_IMAGE_IMPLEMENTATION + +/* + revision history: + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug + 1-bit BMP + *_is_16_bit api + avoid warnings + 2.16 (2017-07-23) all functions have 16-bit variants; + STBI_NO_STDIO works again; + compilation fixes; + fix rounding in unpremultiply; + optimize vertical flip; + disable raw_len validation; + documentation fixes + 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode; + warning fixes; disable run-time SSE detection on gcc; + uniform handling of optional "return" values; + thread-safe initialization of zlib tables + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) allocate large structures on the stack + remove white matting for transparent PSD + fix reported channel count for PNG & BMP + re-enable SSE2 in non-gcc 64-bit + support RGB-formatted JPEG + read 16-bit PNGs (only as 8-bit) + 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED + 2.09 (2016-01-16) allow comments in PNM files + 16-bit-per-pixel TGA (not bit-per-component) + info() for TGA could break due to .hdr handling + info() for BMP to shares code instead of sloppy parse + can use STBI_REALLOC_SIZED if allocator doesn't support realloc + code cleanup + 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA + 2.07 (2015-09-13) fix compiler warnings + partial animated GIF support + limited 16-bpc PSD support + #ifdef unused functions + bug with < 92 byte PIC,PNM,HDR,TGA + 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value + 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning + 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit + 2.03 (2015-04-12) extra corruption checking (mmozeiko) + stbi_set_flip_vertically_on_load (nguillemot) + fix NEON support; fix mingw support + 2.02 (2015-01-19) fix incorrect assert, fix warning + 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2 + 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG + 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg) + progressive JPEG (stb) + PGM/PPM support (Ken Miller) + STBI_MALLOC,STBI_REALLOC,STBI_FREE + GIF bugfix -- seemingly never worked + STBI_NO_*, STBI_ONLY_* + 1.48 (2014-12-14) fix incorrectly-named assert() + 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb) + optimize PNG (ryg) + fix bug in interlaced PNG with user-specified channel count (stb) + 1.46 (2014-08-26) + fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG + 1.45 (2014-08-16) + fix MSVC-ARM internal compiler error by wrapping malloc + 1.44 (2014-08-07) + various warning fixes from Ronny Chevalier + 1.43 (2014-07-15) + fix MSVC-only compiler problem in code changed in 1.42 + 1.42 (2014-07-09) + don't define _CRT_SECURE_NO_WARNINGS (affects user code) + fixes to stbi__cleanup_jpeg path + added STBI_ASSERT to avoid requiring assert.h + 1.41 (2014-06-25) + fix search&replace from 1.36 that messed up comments/error messages + 1.40 (2014-06-22) + fix gcc struct-initialization warning + 1.39 (2014-06-15) + fix to TGA optimization when req_comp != number of components in TGA; + fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite) + add support for BMP version 5 (more ignored fields) + 1.38 (2014-06-06) + suppress MSVC warnings on integer casts truncating values + fix accidental rename of 'skip' field of I/O + 1.37 (2014-06-04) + remove duplicate typedef + 1.36 (2014-06-03) + convert to header file single-file library + if de-iphone isn't set, load iphone images color-swapped instead of returning NULL + 1.35 (2014-05-27) + various warnings + fix broken STBI_SIMD path + fix bug where stbi_load_from_file no longer left file pointer in correct place + fix broken non-easy path for 32-bit BMP (possibly never used) + TGA optimization by Arseny Kapoulkine + 1.34 (unknown) + use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case + 1.33 (2011-07-14) + make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements + 1.32 (2011-07-13) + support for "info" function for all supported filetypes (SpartanJ) + 1.31 (2011-06-20) + a few more leak fixes, bug in PNG handling (SpartanJ) + 1.30 (2011-06-11) + added ability to load files via callbacks to accomidate custom input streams (Ben Wenger) + removed deprecated format-specific test/load functions + removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway + error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha) + fix inefficiency in decoding 32-bit BMP (David Woo) + 1.29 (2010-08-16) + various warning fixes from Aurelien Pocheville + 1.28 (2010-08-01) + fix bug in GIF palette transparency (SpartanJ) + 1.27 (2010-08-01) + cast-to-stbi_uc to fix warnings + 1.26 (2010-07-24) + fix bug in file buffering for PNG reported by SpartanJ + 1.25 (2010-07-17) + refix trans_data warning (Won Chun) + 1.24 (2010-07-12) + perf improvements reading from files on platforms with lock-heavy fgetc() + minor perf improvements for jpeg + deprecated type-specific functions so we'll get feedback if they're needed + attempt to fix trans_data warning (Won Chun) + 1.23 fixed bug in iPhone support + 1.22 (2010-07-10) + removed image *writing* support + stbi_info support from Jetro Lauha + GIF support from Jean-Marc Lienher + iPhone PNG-extensions from James Brown + warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva) + 1.21 fix use of 'stbi_uc' in header (reported by jon blow) + 1.20 added support for Softimage PIC, by Tom Seddon + 1.19 bug in interlaced PNG corruption check (found by ryg) + 1.18 (2008-08-02) + fix a threading bug (local mutable static) + 1.17 support interlaced PNG + 1.16 major bugfix - stbi__convert_format converted one too many pixels + 1.15 initialize some fields for thread safety + 1.14 fix threadsafe conversion bug + header-file-only version (#define STBI_HEADER_FILE_ONLY before including) + 1.13 threadsafe + 1.12 const qualifiers in the API + 1.11 Support installable IDCT, colorspace conversion routines + 1.10 Fixes for 64-bit (don't use "unsigned long") + optimized upsampling by Fabian "ryg" Giesen + 1.09 Fix format-conversion for PSD code (bad global variables!) + 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz + 1.07 attempt to fix C++ warning/errors again + 1.06 attempt to fix C++ warning/errors again + 1.05 fix TGA loading to return correct *comp and use good luminance calc + 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free + 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR + 1.02 support for (subset of) HDR files, float interface for preferred access to them + 1.01 fix bug: possible bug in handling right-side up bmps... not sure + fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all + 1.00 interface to zlib that skips zlib header + 0.99 correct handling of alpha in palette + 0.98 TGA loader by lonesock; dynamically add loaders (untested) + 0.97 jpeg errors on too large a file; also catch another malloc failure + 0.96 fix detection of invalid v value - particleman@mollyrocket forum + 0.95 during header scan, seek to markers in case of padding + 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same + 0.93 handle jpegtran output; verbose errors + 0.92 read 4,8,16,24,32-bit BMP files of several formats + 0.91 output 24-bit Windows 3.0 BMP files + 0.90 fix a few more warnings; bump version number to approach 1.0 + 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd + 0.60 fix compiling as c++ + 0.59 fix warnings: merge Dave Moore's -Wall fixes + 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian + 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available + 0.56 fix bug: zlib uncompressed mode len vs. nlen + 0.55 fix bug: restart_interval not initialized to 0 + 0.54 allow NULL for 'int *comp' + 0.53 fix bug in png 3->4; speedup png decoding + 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments + 0.51 obey req_comp requests, 1-component jpegs return as 1-component, + on 'test' only check type, not whether we support this variant + 0.50 (2006-11-19) + first released version +*/ + + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/third_party/stb_image_write.h b/third_party/stb_image_write.h new file mode 100644 index 0000000..e4b32ed --- /dev/null +++ b/third_party/stb_image_write.h @@ -0,0 +1,1724 @@ +/* stb_image_write - v1.16 - public domain - http://nothings.org/stb + writes out PNG/BMP/TGA/JPEG/HDR images to C stdio - Sean Barrett 2010-2015 + no warranty implied; use at your own risk + + Before #including, + + #define STB_IMAGE_WRITE_IMPLEMENTATION + + in the file that you want to have the implementation. + + Will probably not work correctly with strict-aliasing optimizations. + +ABOUT: + + This header file is a library for writing images to C stdio or a callback. + + The PNG output is not optimal; it is 20-50% larger than the file + written by a decent optimizing implementation; though providing a custom + zlib compress function (see STBIW_ZLIB_COMPRESS) can mitigate that. + This library is designed for source code compactness and simplicity, + not optimal image file size or run-time performance. + +BUILDING: + + You can #define STBIW_ASSERT(x) before the #include to avoid using assert.h. + You can #define STBIW_MALLOC(), STBIW_REALLOC(), and STBIW_FREE() to replace + malloc,realloc,free. + You can #define STBIW_MEMMOVE() to replace memmove() + You can #define STBIW_ZLIB_COMPRESS to use a custom zlib-style compress function + for PNG compression (instead of the builtin one), it must have the following signature: + unsigned char * my_compress(unsigned char *data, int data_len, int *out_len, int quality); + The returned data will be freed with STBIW_FREE() (free() by default), + so it must be heap allocated with STBIW_MALLOC() (malloc() by default), + +UNICODE: + + If compiling for Windows and you wish to use Unicode filenames, compile + with + #define STBIW_WINDOWS_UTF8 + and pass utf8-encoded filenames. Call stbiw_convert_wchar_to_utf8 to convert + Windows wchar_t filenames to utf8. + +USAGE: + + There are five functions, one for each image file format: + + int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes); + int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data); + int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data); + int stbi_write_jpg(char const *filename, int w, int h, int comp, const void *data, int quality); + int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data); + + void stbi_flip_vertically_on_write(int flag); // flag is non-zero to flip data vertically + + There are also five equivalent functions that use an arbitrary write function. You are + expected to open/close your file-equivalent before and after calling these: + + int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data, int stride_in_bytes); + int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data); + int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data); + int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data); + int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality); + + where the callback is: + void stbi_write_func(void *context, void *data, int size); + + You can configure it with these global variables: + int stbi_write_tga_with_rle; // defaults to true; set to 0 to disable RLE + int stbi_write_png_compression_level; // defaults to 8; set to higher for more compression + int stbi_write_force_png_filter; // defaults to -1; set to 0..5 to force a filter mode + + + You can define STBI_WRITE_NO_STDIO to disable the file variant of these + functions, so the library will not use stdio.h at all. However, this will + also disable HDR writing, because it requires stdio for formatted output. + + Each function returns 0 on failure and non-0 on success. + + The functions create an image file defined by the parameters. The image + is a rectangle of pixels stored from left-to-right, top-to-bottom. + Each pixel contains 'comp' channels of data stored interleaved with 8-bits + per channel, in the following order: 1=Y, 2=YA, 3=RGB, 4=RGBA. (Y is + monochrome color.) The rectangle is 'w' pixels wide and 'h' pixels tall. + The *data pointer points to the first byte of the top-left-most pixel. + For PNG, "stride_in_bytes" is the distance in bytes from the first byte of + a row of pixels to the first byte of the next row of pixels. + + PNG creates output files with the same number of components as the input. + The BMP format expands Y to RGB in the file format and does not + output alpha. + + PNG supports writing rectangles of data even when the bytes storing rows of + data are not consecutive in memory (e.g. sub-rectangles of a larger image), + by supplying the stride between the beginning of adjacent rows. The other + formats do not. (Thus you cannot write a native-format BMP through the BMP + writer, both because it is in BGR order and because it may have padding + at the end of the line.) + + PNG allows you to set the deflate compression level by setting the global + variable 'stbi_write_png_compression_level' (it defaults to 8). + + HDR expects linear float data. Since the format is always 32-bit rgb(e) + data, alpha (if provided) is discarded, and for monochrome data it is + replicated across all three channels. + + TGA supports RLE or non-RLE compressed data. To use non-RLE-compressed + data, set the global variable 'stbi_write_tga_with_rle' to 0. + + JPEG does ignore alpha channels in input data; quality is between 1 and 100. + Higher quality looks better but results in a bigger image. + JPEG baseline (no JPEG progressive). + +CREDITS: + + + Sean Barrett - PNG/BMP/TGA + Baldur Karlsson - HDR + Jean-Sebastien Guay - TGA monochrome + Tim Kelsey - misc enhancements + Alan Hickman - TGA RLE + Emmanuel Julien - initial file IO callback implementation + Jon Olick - original jo_jpeg.cpp code + Daniel Gibson - integrate JPEG, allow external zlib + Aarni Koskela - allow choosing PNG filter + + bugfixes: + github:Chribba + Guillaume Chereau + github:jry2 + github:romigrou + Sergio Gonzalez + Jonas Karlsson + Filip Wasil + Thatcher Ulrich + github:poppolopoppo + Patrick Boettcher + github:xeekworx + Cap Petschulat + Simon Rodriguez + Ivan Tikhonov + github:ignotion + Adam Schackart + Andrew Kensler + +LICENSE + + See end of file for license information. + +*/ + +#ifndef INCLUDE_STB_IMAGE_WRITE_H +#define INCLUDE_STB_IMAGE_WRITE_H + +#include + +// if STB_IMAGE_WRITE_STATIC causes problems, try defining STBIWDEF to 'inline' or 'static inline' +#ifndef STBIWDEF +#ifdef STB_IMAGE_WRITE_STATIC +#define STBIWDEF static +#else +#ifdef __cplusplus +#define STBIWDEF extern "C" +#else +#define STBIWDEF extern +#endif +#endif +#endif + +#ifndef STB_IMAGE_WRITE_STATIC // C++ forbids static forward declarations +STBIWDEF int stbi_write_tga_with_rle; +STBIWDEF int stbi_write_png_compression_level; +STBIWDEF int stbi_write_force_png_filter; +#endif + +#ifndef STBI_WRITE_NO_STDIO +STBIWDEF int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes); +STBIWDEF int stbi_write_bmp(char const *filename, int w, int h, int comp, const void *data); +STBIWDEF int stbi_write_tga(char const *filename, int w, int h, int comp, const void *data); +STBIWDEF int stbi_write_hdr(char const *filename, int w, int h, int comp, const float *data); +STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality); + +#ifdef STBIW_WINDOWS_UTF8 +STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input); +#endif +#endif + +typedef void stbi_write_func(void *context, void *data, int size); + +STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data, int stride_in_bytes); +STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data); +STBIWDEF int stbi_write_tga_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const void *data); +STBIWDEF int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int w, int h, int comp, const float *data); +STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality); + +STBIWDEF void stbi_flip_vertically_on_write(int flip_boolean); + +#endif//INCLUDE_STB_IMAGE_WRITE_H + +#ifdef STB_IMAGE_WRITE_IMPLEMENTATION + +#ifdef _WIN32 + #ifndef _CRT_SECURE_NO_WARNINGS + #define _CRT_SECURE_NO_WARNINGS + #endif + #ifndef _CRT_NONSTDC_NO_DEPRECATE + #define _CRT_NONSTDC_NO_DEPRECATE + #endif +#endif + +#ifndef STBI_WRITE_NO_STDIO +#include +#endif // STBI_WRITE_NO_STDIO + +#include +#include +#include +#include + +#if defined(STBIW_MALLOC) && defined(STBIW_FREE) && (defined(STBIW_REALLOC) || defined(STBIW_REALLOC_SIZED)) +// ok +#elif !defined(STBIW_MALLOC) && !defined(STBIW_FREE) && !defined(STBIW_REALLOC) && !defined(STBIW_REALLOC_SIZED) +// ok +#else +#error "Must define all or none of STBIW_MALLOC, STBIW_FREE, and STBIW_REALLOC (or STBIW_REALLOC_SIZED)." +#endif + +#ifndef STBIW_MALLOC +#define STBIW_MALLOC(sz) malloc(sz) +#define STBIW_REALLOC(p,newsz) realloc(p,newsz) +#define STBIW_FREE(p) free(p) +#endif + +#ifndef STBIW_REALLOC_SIZED +#define STBIW_REALLOC_SIZED(p,oldsz,newsz) STBIW_REALLOC(p,newsz) +#endif + + +#ifndef STBIW_MEMMOVE +#define STBIW_MEMMOVE(a,b,sz) memmove(a,b,sz) +#endif + + +#ifndef STBIW_ASSERT +#include +#define STBIW_ASSERT(x) assert(x) +#endif + +#define STBIW_UCHAR(x) (unsigned char) ((x) & 0xff) + +#ifdef STB_IMAGE_WRITE_STATIC +static int stbi_write_png_compression_level = 8; +static int stbi_write_tga_with_rle = 1; +static int stbi_write_force_png_filter = -1; +#else +int stbi_write_png_compression_level = 8; +int stbi_write_tga_with_rle = 1; +int stbi_write_force_png_filter = -1; +#endif + +static int stbi__flip_vertically_on_write = 0; + +STBIWDEF void stbi_flip_vertically_on_write(int flag) +{ + stbi__flip_vertically_on_write = flag; +} + +typedef struct +{ + stbi_write_func *func; + void *context; + unsigned char buffer[64]; + int buf_used; +} stbi__write_context; + +// initialize a callback-based context +static void stbi__start_write_callbacks(stbi__write_context *s, stbi_write_func *c, void *context) +{ + s->func = c; + s->context = context; +} + +#ifndef STBI_WRITE_NO_STDIO + +static void stbi__stdio_write(void *context, void *data, int size) +{ + fwrite(data,1,size,(FILE*) context); +} + +#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8) +#ifdef __cplusplus +#define STBIW_EXTERN extern "C" +#else +#define STBIW_EXTERN extern +#endif +STBIW_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide); +STBIW_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default); + +STBIWDEF int stbiw_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input) +{ + return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL); +} +#endif + +static FILE *stbiw__fopen(char const *filename, char const *mode) +{ + FILE *f; +#if defined(_WIN32) && defined(STBIW_WINDOWS_UTF8) + wchar_t wMode[64]; + wchar_t wFilename[1024]; + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename))) + return 0; + + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode))) + return 0; + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != _wfopen_s(&f, wFilename, wMode)) + f = 0; +#else + f = _wfopen(wFilename, wMode); +#endif + +#elif defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != fopen_s(&f, filename, mode)) + f=0; +#else + f = fopen(filename, mode); +#endif + return f; +} + +static int stbi__start_write_file(stbi__write_context *s, const char *filename) +{ + FILE *f = stbiw__fopen(filename, "wb"); + stbi__start_write_callbacks(s, stbi__stdio_write, (void *) f); + return f != NULL; +} + +static void stbi__end_write_file(stbi__write_context *s) +{ + fclose((FILE *)s->context); +} + +#endif // !STBI_WRITE_NO_STDIO + +typedef unsigned int stbiw_uint32; +typedef int stb_image_write_test[sizeof(stbiw_uint32)==4 ? 1 : -1]; + +static void stbiw__writefv(stbi__write_context *s, const char *fmt, va_list v) +{ + while (*fmt) { + switch (*fmt++) { + case ' ': break; + case '1': { unsigned char x = STBIW_UCHAR(va_arg(v, int)); + s->func(s->context,&x,1); + break; } + case '2': { int x = va_arg(v,int); + unsigned char b[2]; + b[0] = STBIW_UCHAR(x); + b[1] = STBIW_UCHAR(x>>8); + s->func(s->context,b,2); + break; } + case '4': { stbiw_uint32 x = va_arg(v,int); + unsigned char b[4]; + b[0]=STBIW_UCHAR(x); + b[1]=STBIW_UCHAR(x>>8); + b[2]=STBIW_UCHAR(x>>16); + b[3]=STBIW_UCHAR(x>>24); + s->func(s->context,b,4); + break; } + default: + STBIW_ASSERT(0); + return; + } + } +} + +static void stbiw__writef(stbi__write_context *s, const char *fmt, ...) +{ + va_list v; + va_start(v, fmt); + stbiw__writefv(s, fmt, v); + va_end(v); +} + +static void stbiw__write_flush(stbi__write_context *s) +{ + if (s->buf_used) { + s->func(s->context, &s->buffer, s->buf_used); + s->buf_used = 0; + } +} + +static void stbiw__putc(stbi__write_context *s, unsigned char c) +{ + s->func(s->context, &c, 1); +} + +static void stbiw__write1(stbi__write_context *s, unsigned char a) +{ + if ((size_t)s->buf_used + 1 > sizeof(s->buffer)) + stbiw__write_flush(s); + s->buffer[s->buf_used++] = a; +} + +static void stbiw__write3(stbi__write_context *s, unsigned char a, unsigned char b, unsigned char c) +{ + int n; + if ((size_t)s->buf_used + 3 > sizeof(s->buffer)) + stbiw__write_flush(s); + n = s->buf_used; + s->buf_used = n+3; + s->buffer[n+0] = a; + s->buffer[n+1] = b; + s->buffer[n+2] = c; +} + +static void stbiw__write_pixel(stbi__write_context *s, int rgb_dir, int comp, int write_alpha, int expand_mono, unsigned char *d) +{ + unsigned char bg[3] = { 255, 0, 255}, px[3]; + int k; + + if (write_alpha < 0) + stbiw__write1(s, d[comp - 1]); + + switch (comp) { + case 2: // 2 pixels = mono + alpha, alpha is written separately, so same as 1-channel case + case 1: + if (expand_mono) + stbiw__write3(s, d[0], d[0], d[0]); // monochrome bmp + else + stbiw__write1(s, d[0]); // monochrome TGA + break; + case 4: + if (!write_alpha) { + // composite against pink background + for (k = 0; k < 3; ++k) + px[k] = bg[k] + ((d[k] - bg[k]) * d[3]) / 255; + stbiw__write3(s, px[1 - rgb_dir], px[1], px[1 + rgb_dir]); + break; + } + /* FALLTHROUGH */ + case 3: + stbiw__write3(s, d[1 - rgb_dir], d[1], d[1 + rgb_dir]); + break; + } + if (write_alpha > 0) + stbiw__write1(s, d[comp - 1]); +} + +static void stbiw__write_pixels(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, void *data, int write_alpha, int scanline_pad, int expand_mono) +{ + stbiw_uint32 zero = 0; + int i,j, j_end; + + if (y <= 0) + return; + + if (stbi__flip_vertically_on_write) + vdir *= -1; + + if (vdir < 0) { + j_end = -1; j = y-1; + } else { + j_end = y; j = 0; + } + + for (; j != j_end; j += vdir) { + for (i=0; i < x; ++i) { + unsigned char *d = (unsigned char *) data + (j*x+i)*comp; + stbiw__write_pixel(s, rgb_dir, comp, write_alpha, expand_mono, d); + } + stbiw__write_flush(s); + s->func(s->context, &zero, scanline_pad); + } +} + +static int stbiw__outfile(stbi__write_context *s, int rgb_dir, int vdir, int x, int y, int comp, int expand_mono, void *data, int alpha, int pad, const char *fmt, ...) +{ + if (y < 0 || x < 0) { + return 0; + } else { + va_list v; + va_start(v, fmt); + stbiw__writefv(s, fmt, v); + va_end(v); + stbiw__write_pixels(s,rgb_dir,vdir,x,y,comp,data,alpha,pad, expand_mono); + return 1; + } +} + +static int stbi_write_bmp_core(stbi__write_context *s, int x, int y, int comp, const void *data) +{ + if (comp != 4) { + // write RGB bitmap + int pad = (-x*3) & 3; + return stbiw__outfile(s,-1,-1,x,y,comp,1,(void *) data,0,pad, + "11 4 22 4" "4 44 22 444444", + 'B', 'M', 14+40+(x*3+pad)*y, 0,0, 14+40, // file header + 40, x,y, 1,24, 0,0,0,0,0,0); // bitmap header + } else { + // RGBA bitmaps need a v4 header + // use BI_BITFIELDS mode with 32bpp and alpha mask + // (straight BI_RGB with alpha mask doesn't work in most readers) + return stbiw__outfile(s,-1,-1,x,y,comp,1,(void *)data,1,0, + "11 4 22 4" "4 44 22 444444 4444 4 444 444 444 444", + 'B', 'M', 14+108+x*y*4, 0, 0, 14+108, // file header + 108, x,y, 1,32, 3,0,0,0,0,0, 0xff0000,0xff00,0xff,0xff000000u, 0, 0,0,0, 0,0,0, 0,0,0, 0,0,0); // bitmap V4 header + } +} + +STBIWDEF int stbi_write_bmp_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data) +{ + stbi__write_context s = { 0 }; + stbi__start_write_callbacks(&s, func, context); + return stbi_write_bmp_core(&s, x, y, comp, data); +} + +#ifndef STBI_WRITE_NO_STDIO +STBIWDEF int stbi_write_bmp(char const *filename, int x, int y, int comp, const void *data) +{ + stbi__write_context s = { 0 }; + if (stbi__start_write_file(&s,filename)) { + int r = stbi_write_bmp_core(&s, x, y, comp, data); + stbi__end_write_file(&s); + return r; + } else + return 0; +} +#endif //!STBI_WRITE_NO_STDIO + +static int stbi_write_tga_core(stbi__write_context *s, int x, int y, int comp, void *data) +{ + int has_alpha = (comp == 2 || comp == 4); + int colorbytes = has_alpha ? comp-1 : comp; + int format = colorbytes < 2 ? 3 : 2; // 3 color channels (RGB/RGBA) = 2, 1 color channel (Y/YA) = 3 + + if (y < 0 || x < 0) + return 0; + + if (!stbi_write_tga_with_rle) { + return stbiw__outfile(s, -1, -1, x, y, comp, 0, (void *) data, has_alpha, 0, + "111 221 2222 11", 0, 0, format, 0, 0, 0, 0, 0, x, y, (colorbytes + has_alpha) * 8, has_alpha * 8); + } else { + int i,j,k; + int jend, jdir; + + stbiw__writef(s, "111 221 2222 11", 0,0,format+8, 0,0,0, 0,0,x,y, (colorbytes + has_alpha) * 8, has_alpha * 8); + + if (stbi__flip_vertically_on_write) { + j = 0; + jend = y; + jdir = 1; + } else { + j = y-1; + jend = -1; + jdir = -1; + } + for (; j != jend; j += jdir) { + unsigned char *row = (unsigned char *) data + j * x * comp; + int len; + + for (i = 0; i < x; i += len) { + unsigned char *begin = row + i * comp; + int diff = 1; + len = 1; + + if (i < x - 1) { + ++len; + diff = memcmp(begin, row + (i + 1) * comp, comp); + if (diff) { + const unsigned char *prev = begin; + for (k = i + 2; k < x && len < 128; ++k) { + if (memcmp(prev, row + k * comp, comp)) { + prev += comp; + ++len; + } else { + --len; + break; + } + } + } else { + for (k = i + 2; k < x && len < 128; ++k) { + if (!memcmp(begin, row + k * comp, comp)) { + ++len; + } else { + break; + } + } + } + } + + if (diff) { + unsigned char header = STBIW_UCHAR(len - 1); + stbiw__write1(s, header); + for (k = 0; k < len; ++k) { + stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin + k * comp); + } + } else { + unsigned char header = STBIW_UCHAR(len - 129); + stbiw__write1(s, header); + stbiw__write_pixel(s, -1, comp, has_alpha, 0, begin); + } + } + } + stbiw__write_flush(s); + } + return 1; +} + +STBIWDEF int stbi_write_tga_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data) +{ + stbi__write_context s = { 0 }; + stbi__start_write_callbacks(&s, func, context); + return stbi_write_tga_core(&s, x, y, comp, (void *) data); +} + +#ifndef STBI_WRITE_NO_STDIO +STBIWDEF int stbi_write_tga(char const *filename, int x, int y, int comp, const void *data) +{ + stbi__write_context s = { 0 }; + if (stbi__start_write_file(&s,filename)) { + int r = stbi_write_tga_core(&s, x, y, comp, (void *) data); + stbi__end_write_file(&s); + return r; + } else + return 0; +} +#endif + +// ************************************************************************************************* +// Radiance RGBE HDR writer +// by Baldur Karlsson + +#define stbiw__max(a, b) ((a) > (b) ? (a) : (b)) + +#ifndef STBI_WRITE_NO_STDIO + +static void stbiw__linear_to_rgbe(unsigned char *rgbe, float *linear) +{ + int exponent; + float maxcomp = stbiw__max(linear[0], stbiw__max(linear[1], linear[2])); + + if (maxcomp < 1e-32f) { + rgbe[0] = rgbe[1] = rgbe[2] = rgbe[3] = 0; + } else { + float normalize = (float) frexp(maxcomp, &exponent) * 256.0f/maxcomp; + + rgbe[0] = (unsigned char)(linear[0] * normalize); + rgbe[1] = (unsigned char)(linear[1] * normalize); + rgbe[2] = (unsigned char)(linear[2] * normalize); + rgbe[3] = (unsigned char)(exponent + 128); + } +} + +static void stbiw__write_run_data(stbi__write_context *s, int length, unsigned char databyte) +{ + unsigned char lengthbyte = STBIW_UCHAR(length+128); + STBIW_ASSERT(length+128 <= 255); + s->func(s->context, &lengthbyte, 1); + s->func(s->context, &databyte, 1); +} + +static void stbiw__write_dump_data(stbi__write_context *s, int length, unsigned char *data) +{ + unsigned char lengthbyte = STBIW_UCHAR(length); + STBIW_ASSERT(length <= 128); // inconsistent with spec but consistent with official code + s->func(s->context, &lengthbyte, 1); + s->func(s->context, data, length); +} + +static void stbiw__write_hdr_scanline(stbi__write_context *s, int width, int ncomp, unsigned char *scratch, float *scanline) +{ + unsigned char scanlineheader[4] = { 2, 2, 0, 0 }; + unsigned char rgbe[4]; + float linear[3]; + int x; + + scanlineheader[2] = (width&0xff00)>>8; + scanlineheader[3] = (width&0x00ff); + + /* skip RLE for images too small or large */ + if (width < 8 || width >= 32768) { + for (x=0; x < width; x++) { + switch (ncomp) { + case 4: /* fallthrough */ + case 3: linear[2] = scanline[x*ncomp + 2]; + linear[1] = scanline[x*ncomp + 1]; + linear[0] = scanline[x*ncomp + 0]; + break; + default: + linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0]; + break; + } + stbiw__linear_to_rgbe(rgbe, linear); + s->func(s->context, rgbe, 4); + } + } else { + int c,r; + /* encode into scratch buffer */ + for (x=0; x < width; x++) { + switch(ncomp) { + case 4: /* fallthrough */ + case 3: linear[2] = scanline[x*ncomp + 2]; + linear[1] = scanline[x*ncomp + 1]; + linear[0] = scanline[x*ncomp + 0]; + break; + default: + linear[0] = linear[1] = linear[2] = scanline[x*ncomp + 0]; + break; + } + stbiw__linear_to_rgbe(rgbe, linear); + scratch[x + width*0] = rgbe[0]; + scratch[x + width*1] = rgbe[1]; + scratch[x + width*2] = rgbe[2]; + scratch[x + width*3] = rgbe[3]; + } + + s->func(s->context, scanlineheader, 4); + + /* RLE each component separately */ + for (c=0; c < 4; c++) { + unsigned char *comp = &scratch[width*c]; + + x = 0; + while (x < width) { + // find first run + r = x; + while (r+2 < width) { + if (comp[r] == comp[r+1] && comp[r] == comp[r+2]) + break; + ++r; + } + if (r+2 >= width) + r = width; + // dump up to first run + while (x < r) { + int len = r-x; + if (len > 128) len = 128; + stbiw__write_dump_data(s, len, &comp[x]); + x += len; + } + // if there's a run, output it + if (r+2 < width) { // same test as what we break out of in search loop, so only true if we break'd + // find next byte after run + while (r < width && comp[r] == comp[x]) + ++r; + // output run up to r + while (x < r) { + int len = r-x; + if (len > 127) len = 127; + stbiw__write_run_data(s, len, comp[x]); + x += len; + } + } + } + } + } +} + +static int stbi_write_hdr_core(stbi__write_context *s, int x, int y, int comp, float *data) +{ + if (y <= 0 || x <= 0 || data == NULL) + return 0; + else { + // Each component is stored separately. Allocate scratch space for full output scanline. + unsigned char *scratch = (unsigned char *) STBIW_MALLOC(x*4); + int i, len; + char buffer[128]; + char header[] = "#?RADIANCE\n# Written by stb_image_write.h\nFORMAT=32-bit_rle_rgbe\n"; + s->func(s->context, header, sizeof(header)-1); + +#ifdef __STDC_LIB_EXT1__ + len = sprintf_s(buffer, sizeof(buffer), "EXPOSURE= 1.0000000000000\n\n-Y %d +X %d\n", y, x); +#else + len = sprintf(buffer, "EXPOSURE= 1.0000000000000\n\n-Y %d +X %d\n", y, x); +#endif + s->func(s->context, buffer, len); + + for(i=0; i < y; i++) + stbiw__write_hdr_scanline(s, x, comp, scratch, data + comp*x*(stbi__flip_vertically_on_write ? y-1-i : i)); + STBIW_FREE(scratch); + return 1; + } +} + +STBIWDEF int stbi_write_hdr_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const float *data) +{ + stbi__write_context s = { 0 }; + stbi__start_write_callbacks(&s, func, context); + return stbi_write_hdr_core(&s, x, y, comp, (float *) data); +} + +STBIWDEF int stbi_write_hdr(char const *filename, int x, int y, int comp, const float *data) +{ + stbi__write_context s = { 0 }; + if (stbi__start_write_file(&s,filename)) { + int r = stbi_write_hdr_core(&s, x, y, comp, (float *) data); + stbi__end_write_file(&s); + return r; + } else + return 0; +} +#endif // STBI_WRITE_NO_STDIO + + +////////////////////////////////////////////////////////////////////////////// +// +// PNG writer +// + +#ifndef STBIW_ZLIB_COMPRESS +// stretchy buffer; stbiw__sbpush() == vector<>::push_back() -- stbiw__sbcount() == vector<>::size() +#define stbiw__sbraw(a) ((int *) (void *) (a) - 2) +#define stbiw__sbm(a) stbiw__sbraw(a)[0] +#define stbiw__sbn(a) stbiw__sbraw(a)[1] + +#define stbiw__sbneedgrow(a,n) ((a)==0 || stbiw__sbn(a)+n >= stbiw__sbm(a)) +#define stbiw__sbmaybegrow(a,n) (stbiw__sbneedgrow(a,(n)) ? stbiw__sbgrow(a,n) : 0) +#define stbiw__sbgrow(a,n) stbiw__sbgrowf((void **) &(a), (n), sizeof(*(a))) + +#define stbiw__sbpush(a, v) (stbiw__sbmaybegrow(a,1), (a)[stbiw__sbn(a)++] = (v)) +#define stbiw__sbcount(a) ((a) ? stbiw__sbn(a) : 0) +#define stbiw__sbfree(a) ((a) ? STBIW_FREE(stbiw__sbraw(a)),0 : 0) + +static void *stbiw__sbgrowf(void **arr, int increment, int itemsize) +{ + int m = *arr ? 2*stbiw__sbm(*arr)+increment : increment+1; + void *p = STBIW_REALLOC_SIZED(*arr ? stbiw__sbraw(*arr) : 0, *arr ? (stbiw__sbm(*arr)*itemsize + sizeof(int)*2) : 0, itemsize * m + sizeof(int)*2); + STBIW_ASSERT(p); + if (p) { + if (!*arr) ((int *) p)[1] = 0; + *arr = (void *) ((int *) p + 2); + stbiw__sbm(*arr) = m; + } + return *arr; +} + +static unsigned char *stbiw__zlib_flushf(unsigned char *data, unsigned int *bitbuffer, int *bitcount) +{ + while (*bitcount >= 8) { + stbiw__sbpush(data, STBIW_UCHAR(*bitbuffer)); + *bitbuffer >>= 8; + *bitcount -= 8; + } + return data; +} + +static int stbiw__zlib_bitrev(int code, int codebits) +{ + int res=0; + while (codebits--) { + res = (res << 1) | (code & 1); + code >>= 1; + } + return res; +} + +static unsigned int stbiw__zlib_countm(unsigned char *a, unsigned char *b, int limit) +{ + int i; + for (i=0; i < limit && i < 258; ++i) + if (a[i] != b[i]) break; + return i; +} + +static unsigned int stbiw__zhash(unsigned char *data) +{ + stbiw_uint32 hash = data[0] + (data[1] << 8) + (data[2] << 16); + hash ^= hash << 3; + hash += hash >> 5; + hash ^= hash << 4; + hash += hash >> 17; + hash ^= hash << 25; + hash += hash >> 6; + return hash; +} + +#define stbiw__zlib_flush() (out = stbiw__zlib_flushf(out, &bitbuf, &bitcount)) +#define stbiw__zlib_add(code,codebits) \ + (bitbuf |= (code) << bitcount, bitcount += (codebits), stbiw__zlib_flush()) +#define stbiw__zlib_huffa(b,c) stbiw__zlib_add(stbiw__zlib_bitrev(b,c),c) +// default huffman tables +#define stbiw__zlib_huff1(n) stbiw__zlib_huffa(0x30 + (n), 8) +#define stbiw__zlib_huff2(n) stbiw__zlib_huffa(0x190 + (n)-144, 9) +#define stbiw__zlib_huff3(n) stbiw__zlib_huffa(0 + (n)-256,7) +#define stbiw__zlib_huff4(n) stbiw__zlib_huffa(0xc0 + (n)-280,8) +#define stbiw__zlib_huff(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : (n) <= 255 ? stbiw__zlib_huff2(n) : (n) <= 279 ? stbiw__zlib_huff3(n) : stbiw__zlib_huff4(n)) +#define stbiw__zlib_huffb(n) ((n) <= 143 ? stbiw__zlib_huff1(n) : stbiw__zlib_huff2(n)) + +#define stbiw__ZHASH 16384 + +#endif // STBIW_ZLIB_COMPRESS + +STBIWDEF unsigned char * stbi_zlib_compress(unsigned char *data, int data_len, int *out_len, int quality) +{ +#ifdef STBIW_ZLIB_COMPRESS + // user provided a zlib compress implementation, use that + return STBIW_ZLIB_COMPRESS(data, data_len, out_len, quality); +#else // use builtin + static unsigned short lengthc[] = { 3,4,5,6,7,8,9,10,11,13,15,17,19,23,27,31,35,43,51,59,67,83,99,115,131,163,195,227,258, 259 }; + static unsigned char lengtheb[]= { 0,0,0,0,0,0,0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0 }; + static unsigned short distc[] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193,257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577, 32768 }; + static unsigned char disteb[] = { 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13 }; + unsigned int bitbuf=0; + int i,j, bitcount=0; + unsigned char *out = NULL; + unsigned char ***hash_table = (unsigned char***) STBIW_MALLOC(stbiw__ZHASH * sizeof(unsigned char**)); + if (hash_table == NULL) + return NULL; + if (quality < 5) quality = 5; + + stbiw__sbpush(out, 0x78); // DEFLATE 32K window + stbiw__sbpush(out, 0x5e); // FLEVEL = 1 + stbiw__zlib_add(1,1); // BFINAL = 1 + stbiw__zlib_add(1,2); // BTYPE = 1 -- fixed huffman + + for (i=0; i < stbiw__ZHASH; ++i) + hash_table[i] = NULL; + + i=0; + while (i < data_len-3) { + // hash next 3 bytes of data to be compressed + int h = stbiw__zhash(data+i)&(stbiw__ZHASH-1), best=3; + unsigned char *bestloc = 0; + unsigned char **hlist = hash_table[h]; + int n = stbiw__sbcount(hlist); + for (j=0; j < n; ++j) { + if (hlist[j]-data > i-32768) { // if entry lies within window + int d = stbiw__zlib_countm(hlist[j], data+i, data_len-i); + if (d >= best) { best=d; bestloc=hlist[j]; } + } + } + // when hash table entry is too long, delete half the entries + if (hash_table[h] && stbiw__sbn(hash_table[h]) == 2*quality) { + STBIW_MEMMOVE(hash_table[h], hash_table[h]+quality, sizeof(hash_table[h][0])*quality); + stbiw__sbn(hash_table[h]) = quality; + } + stbiw__sbpush(hash_table[h],data+i); + + if (bestloc) { + // "lazy matching" - check match at *next* byte, and if it's better, do cur byte as literal + h = stbiw__zhash(data+i+1)&(stbiw__ZHASH-1); + hlist = hash_table[h]; + n = stbiw__sbcount(hlist); + for (j=0; j < n; ++j) { + if (hlist[j]-data > i-32767) { + int e = stbiw__zlib_countm(hlist[j], data+i+1, data_len-i-1); + if (e > best) { // if next match is better, bail on current match + bestloc = NULL; + break; + } + } + } + } + + if (bestloc) { + int d = (int) (data+i - bestloc); // distance back + STBIW_ASSERT(d <= 32767 && best <= 258); + for (j=0; best > lengthc[j+1]-1; ++j); + stbiw__zlib_huff(j+257); + if (lengtheb[j]) stbiw__zlib_add(best - lengthc[j], lengtheb[j]); + for (j=0; d > distc[j+1]-1; ++j); + stbiw__zlib_add(stbiw__zlib_bitrev(j,5),5); + if (disteb[j]) stbiw__zlib_add(d - distc[j], disteb[j]); + i += best; + } else { + stbiw__zlib_huffb(data[i]); + ++i; + } + } + // write out final bytes + for (;i < data_len; ++i) + stbiw__zlib_huffb(data[i]); + stbiw__zlib_huff(256); // end of block + // pad with 0 bits to byte boundary + while (bitcount) + stbiw__zlib_add(0,1); + + for (i=0; i < stbiw__ZHASH; ++i) + (void) stbiw__sbfree(hash_table[i]); + STBIW_FREE(hash_table); + + // store uncompressed instead if compression was worse + if (stbiw__sbn(out) > data_len + 2 + ((data_len+32766)/32767)*5) { + stbiw__sbn(out) = 2; // truncate to DEFLATE 32K window and FLEVEL = 1 + for (j = 0; j < data_len;) { + int blocklen = data_len - j; + if (blocklen > 32767) blocklen = 32767; + stbiw__sbpush(out, data_len - j == blocklen); // BFINAL = ?, BTYPE = 0 -- no compression + stbiw__sbpush(out, STBIW_UCHAR(blocklen)); // LEN + stbiw__sbpush(out, STBIW_UCHAR(blocklen >> 8)); + stbiw__sbpush(out, STBIW_UCHAR(~blocklen)); // NLEN + stbiw__sbpush(out, STBIW_UCHAR(~blocklen >> 8)); + memcpy(out+stbiw__sbn(out), data+j, blocklen); + stbiw__sbn(out) += blocklen; + j += blocklen; + } + } + + { + // compute adler32 on input + unsigned int s1=1, s2=0; + int blocklen = (int) (data_len % 5552); + j=0; + while (j < data_len) { + for (i=0; i < blocklen; ++i) { s1 += data[j+i]; s2 += s1; } + s1 %= 65521; s2 %= 65521; + j += blocklen; + blocklen = 5552; + } + stbiw__sbpush(out, STBIW_UCHAR(s2 >> 8)); + stbiw__sbpush(out, STBIW_UCHAR(s2)); + stbiw__sbpush(out, STBIW_UCHAR(s1 >> 8)); + stbiw__sbpush(out, STBIW_UCHAR(s1)); + } + *out_len = stbiw__sbn(out); + // make returned pointer freeable + STBIW_MEMMOVE(stbiw__sbraw(out), out, *out_len); + return (unsigned char *) stbiw__sbraw(out); +#endif // STBIW_ZLIB_COMPRESS +} + +static unsigned int stbiw__crc32(unsigned char *buffer, int len) +{ +#ifdef STBIW_CRC32 + return STBIW_CRC32(buffer, len); +#else + static unsigned int crc_table[256] = + { + 0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3, + 0x0eDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91, + 0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7, + 0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5, + 0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B, + 0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59, + 0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F, + 0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D, + 0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433, + 0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01, + 0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457, + 0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65, + 0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB, + 0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9, + 0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F, + 0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD, + 0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683, + 0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1, + 0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7, + 0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5, + 0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B, + 0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79, + 0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F, + 0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D, + 0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713, + 0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21, + 0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777, + 0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45, + 0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB, + 0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9, + 0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF, + 0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D + }; + + unsigned int crc = ~0u; + int i; + for (i=0; i < len; ++i) + crc = (crc >> 8) ^ crc_table[buffer[i] ^ (crc & 0xff)]; + return ~crc; +#endif +} + +#define stbiw__wpng4(o,a,b,c,d) ((o)[0]=STBIW_UCHAR(a),(o)[1]=STBIW_UCHAR(b),(o)[2]=STBIW_UCHAR(c),(o)[3]=STBIW_UCHAR(d),(o)+=4) +#define stbiw__wp32(data,v) stbiw__wpng4(data, (v)>>24,(v)>>16,(v)>>8,(v)); +#define stbiw__wptag(data,s) stbiw__wpng4(data, s[0],s[1],s[2],s[3]) + +static void stbiw__wpcrc(unsigned char **data, int len) +{ + unsigned int crc = stbiw__crc32(*data - len - 4, len+4); + stbiw__wp32(*data, crc); +} + +static unsigned char stbiw__paeth(int a, int b, int c) +{ + int p = a + b - c, pa = abs(p-a), pb = abs(p-b), pc = abs(p-c); + if (pa <= pb && pa <= pc) return STBIW_UCHAR(a); + if (pb <= pc) return STBIW_UCHAR(b); + return STBIW_UCHAR(c); +} + +// @OPTIMIZE: provide an option that always forces left-predict or paeth predict +static void stbiw__encode_png_line(unsigned char *pixels, int stride_bytes, int width, int height, int y, int n, int filter_type, signed char *line_buffer) +{ + static int mapping[] = { 0,1,2,3,4 }; + static int firstmap[] = { 0,1,0,5,6 }; + int *mymap = (y != 0) ? mapping : firstmap; + int i; + int type = mymap[filter_type]; + unsigned char *z = pixels + stride_bytes * (stbi__flip_vertically_on_write ? height-1-y : y); + int signed_stride = stbi__flip_vertically_on_write ? -stride_bytes : stride_bytes; + + if (type==0) { + memcpy(line_buffer, z, width*n); + return; + } + + // first loop isn't optimized since it's just one pixel + for (i = 0; i < n; ++i) { + switch (type) { + case 1: line_buffer[i] = z[i]; break; + case 2: line_buffer[i] = z[i] - z[i-signed_stride]; break; + case 3: line_buffer[i] = z[i] - (z[i-signed_stride]>>1); break; + case 4: line_buffer[i] = (signed char) (z[i] - stbiw__paeth(0,z[i-signed_stride],0)); break; + case 5: line_buffer[i] = z[i]; break; + case 6: line_buffer[i] = z[i]; break; + } + } + switch (type) { + case 1: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-n]; break; + case 2: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - z[i-signed_stride]; break; + case 3: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - ((z[i-n] + z[i-signed_stride])>>1); break; + case 4: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], z[i-signed_stride], z[i-signed_stride-n]); break; + case 5: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - (z[i-n]>>1); break; + case 6: for (i=n; i < width*n; ++i) line_buffer[i] = z[i] - stbiw__paeth(z[i-n], 0,0); break; + } +} + +STBIWDEF unsigned char *stbi_write_png_to_mem(const unsigned char *pixels, int stride_bytes, int x, int y, int n, int *out_len) +{ + int force_filter = stbi_write_force_png_filter; + int ctype[5] = { -1, 0, 4, 2, 6 }; + unsigned char sig[8] = { 137,80,78,71,13,10,26,10 }; + unsigned char *out,*o, *filt, *zlib; + signed char *line_buffer; + int j,zlen; + + if (stride_bytes == 0) + stride_bytes = x * n; + + if (force_filter >= 5) { + force_filter = -1; + } + + filt = (unsigned char *) STBIW_MALLOC((x*n+1) * y); if (!filt) return 0; + line_buffer = (signed char *) STBIW_MALLOC(x * n); if (!line_buffer) { STBIW_FREE(filt); return 0; } + for (j=0; j < y; ++j) { + int filter_type; + if (force_filter > -1) { + filter_type = force_filter; + stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, force_filter, line_buffer); + } else { // Estimate the best filter by running through all of them: + int best_filter = 0, best_filter_val = 0x7fffffff, est, i; + for (filter_type = 0; filter_type < 5; filter_type++) { + stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, filter_type, line_buffer); + + // Estimate the entropy of the line using this filter; the less, the better. + est = 0; + for (i = 0; i < x*n; ++i) { + est += abs((signed char) line_buffer[i]); + } + if (est < best_filter_val) { + best_filter_val = est; + best_filter = filter_type; + } + } + if (filter_type != best_filter) { // If the last iteration already got us the best filter, don't redo it + stbiw__encode_png_line((unsigned char*)(pixels), stride_bytes, x, y, j, n, best_filter, line_buffer); + filter_type = best_filter; + } + } + // when we get here, filter_type contains the filter type, and line_buffer contains the data + filt[j*(x*n+1)] = (unsigned char) filter_type; + STBIW_MEMMOVE(filt+j*(x*n+1)+1, line_buffer, x*n); + } + STBIW_FREE(line_buffer); + zlib = stbi_zlib_compress(filt, y*( x*n+1), &zlen, stbi_write_png_compression_level); + STBIW_FREE(filt); + if (!zlib) return 0; + + // each tag requires 12 bytes of overhead + out = (unsigned char *) STBIW_MALLOC(8 + 12+13 + 12+zlen + 12); + if (!out) return 0; + *out_len = 8 + 12+13 + 12+zlen + 12; + + o=out; + STBIW_MEMMOVE(o,sig,8); o+= 8; + stbiw__wp32(o, 13); // header length + stbiw__wptag(o, "IHDR"); + stbiw__wp32(o, x); + stbiw__wp32(o, y); + *o++ = 8; + *o++ = STBIW_UCHAR(ctype[n]); + *o++ = 0; + *o++ = 0; + *o++ = 0; + stbiw__wpcrc(&o,13); + + stbiw__wp32(o, zlen); + stbiw__wptag(o, "IDAT"); + STBIW_MEMMOVE(o, zlib, zlen); + o += zlen; + STBIW_FREE(zlib); + stbiw__wpcrc(&o, zlen); + + stbiw__wp32(o,0); + stbiw__wptag(o, "IEND"); + stbiw__wpcrc(&o,0); + + STBIW_ASSERT(o == out + *out_len); + + return out; +} + +#ifndef STBI_WRITE_NO_STDIO +STBIWDEF int stbi_write_png(char const *filename, int x, int y, int comp, const void *data, int stride_bytes) +{ + FILE *f; + int len; + unsigned char *png = stbi_write_png_to_mem((const unsigned char *) data, stride_bytes, x, y, comp, &len); + if (png == NULL) return 0; + + f = stbiw__fopen(filename, "wb"); + if (!f) { STBIW_FREE(png); return 0; } + fwrite(png, 1, len, f); + fclose(f); + STBIW_FREE(png); + return 1; +} +#endif + +STBIWDEF int stbi_write_png_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int stride_bytes) +{ + int len; + unsigned char *png = stbi_write_png_to_mem((const unsigned char *) data, stride_bytes, x, y, comp, &len); + if (png == NULL) return 0; + func(context, png, len); + STBIW_FREE(png); + return 1; +} + + +/* *************************************************************************** + * + * JPEG writer + * + * This is based on Jon Olick's jo_jpeg.cpp: + * public domain Simple, Minimalistic JPEG writer - http://www.jonolick.com/code.html + */ + +static const unsigned char stbiw__jpg_ZigZag[] = { 0,1,5,6,14,15,27,28,2,4,7,13,16,26,29,42,3,8,12,17,25,30,41,43,9,11,18, + 24,31,40,44,53,10,19,23,32,39,45,52,54,20,22,33,38,46,51,55,60,21,34,37,47,50,56,59,61,35,36,48,49,57,58,62,63 }; + +static void stbiw__jpg_writeBits(stbi__write_context *s, int *bitBufP, int *bitCntP, const unsigned short *bs) { + int bitBuf = *bitBufP, bitCnt = *bitCntP; + bitCnt += bs[1]; + bitBuf |= bs[0] << (24 - bitCnt); + while(bitCnt >= 8) { + unsigned char c = (bitBuf >> 16) & 255; + stbiw__putc(s, c); + if(c == 255) { + stbiw__putc(s, 0); + } + bitBuf <<= 8; + bitCnt -= 8; + } + *bitBufP = bitBuf; + *bitCntP = bitCnt; +} + +static void stbiw__jpg_DCT(float *d0p, float *d1p, float *d2p, float *d3p, float *d4p, float *d5p, float *d6p, float *d7p) { + float d0 = *d0p, d1 = *d1p, d2 = *d2p, d3 = *d3p, d4 = *d4p, d5 = *d5p, d6 = *d6p, d7 = *d7p; + float z1, z2, z3, z4, z5, z11, z13; + + float tmp0 = d0 + d7; + float tmp7 = d0 - d7; + float tmp1 = d1 + d6; + float tmp6 = d1 - d6; + float tmp2 = d2 + d5; + float tmp5 = d2 - d5; + float tmp3 = d3 + d4; + float tmp4 = d3 - d4; + + // Even part + float tmp10 = tmp0 + tmp3; // phase 2 + float tmp13 = tmp0 - tmp3; + float tmp11 = tmp1 + tmp2; + float tmp12 = tmp1 - tmp2; + + d0 = tmp10 + tmp11; // phase 3 + d4 = tmp10 - tmp11; + + z1 = (tmp12 + tmp13) * 0.707106781f; // c4 + d2 = tmp13 + z1; // phase 5 + d6 = tmp13 - z1; + + // Odd part + tmp10 = tmp4 + tmp5; // phase 2 + tmp11 = tmp5 + tmp6; + tmp12 = tmp6 + tmp7; + + // The rotator is modified from fig 4-8 to avoid extra negations. + z5 = (tmp10 - tmp12) * 0.382683433f; // c6 + z2 = tmp10 * 0.541196100f + z5; // c2-c6 + z4 = tmp12 * 1.306562965f + z5; // c2+c6 + z3 = tmp11 * 0.707106781f; // c4 + + z11 = tmp7 + z3; // phase 5 + z13 = tmp7 - z3; + + *d5p = z13 + z2; // phase 6 + *d3p = z13 - z2; + *d1p = z11 + z4; + *d7p = z11 - z4; + + *d0p = d0; *d2p = d2; *d4p = d4; *d6p = d6; +} + +static void stbiw__jpg_calcBits(int val, unsigned short bits[2]) { + int tmp1 = val < 0 ? -val : val; + val = val < 0 ? val-1 : val; + bits[1] = 1; + while(tmp1 >>= 1) { + ++bits[1]; + } + bits[0] = val & ((1<0)&&(DU[end0pos]==0); --end0pos) { + } + // end0pos = first element in reverse order !=0 + if(end0pos == 0) { + stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB); + return DU[0]; + } + for(i = 1; i <= end0pos; ++i) { + int startpos = i; + int nrzeroes; + unsigned short bits[2]; + for (; DU[i]==0 && i<=end0pos; ++i) { + } + nrzeroes = i-startpos; + if ( nrzeroes >= 16 ) { + int lng = nrzeroes>>4; + int nrmarker; + for (nrmarker=1; nrmarker <= lng; ++nrmarker) + stbiw__jpg_writeBits(s, bitBuf, bitCnt, M16zeroes); + nrzeroes &= 15; + } + stbiw__jpg_calcBits(DU[i], bits); + stbiw__jpg_writeBits(s, bitBuf, bitCnt, HTAC[(nrzeroes<<4)+bits[1]]); + stbiw__jpg_writeBits(s, bitBuf, bitCnt, bits); + } + if(end0pos != 63) { + stbiw__jpg_writeBits(s, bitBuf, bitCnt, EOB); + } + return DU[0]; +} + +static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, int comp, const void* data, int quality) { + // Constants that don't pollute global namespace + static const unsigned char std_dc_luminance_nrcodes[] = {0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0}; + static const unsigned char std_dc_luminance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11}; + static const unsigned char std_ac_luminance_nrcodes[] = {0,0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,0x7d}; + static const unsigned char std_ac_luminance_values[] = { + 0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xa1,0x08, + 0x23,0x42,0xb1,0xc1,0x15,0x52,0xd1,0xf0,0x24,0x33,0x62,0x72,0x82,0x09,0x0a,0x16,0x17,0x18,0x19,0x1a,0x25,0x26,0x27,0x28, + 0x29,0x2a,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58,0x59, + 0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x83,0x84,0x85,0x86,0x87,0x88,0x89, + 0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4,0xb5,0xb6, + 0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xe1,0xe2, + 0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa + }; + static const unsigned char std_dc_chrominance_nrcodes[] = {0,0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0}; + static const unsigned char std_dc_chrominance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11}; + static const unsigned char std_ac_chrominance_nrcodes[] = {0,0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,0x77}; + static const unsigned char std_ac_chrominance_values[] = { + 0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91, + 0xa1,0xb1,0xc1,0x09,0x23,0x33,0x52,0xf0,0x15,0x62,0x72,0xd1,0x0a,0x16,0x24,0x34,0xe1,0x25,0xf1,0x17,0x18,0x19,0x1a,0x26, + 0x27,0x28,0x29,0x2a,0x35,0x36,0x37,0x38,0x39,0x3a,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x53,0x54,0x55,0x56,0x57,0x58, + 0x59,0x5a,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x82,0x83,0x84,0x85,0x86,0x87, + 0x88,0x89,0x8a,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xb2,0xb3,0xb4, + 0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda, + 0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa + }; + // Huffman tables + static const unsigned short YDC_HT[256][2] = { {0,2},{2,3},{3,3},{4,3},{5,3},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9}}; + static const unsigned short UVDC_HT[256][2] = { {0,2},{1,2},{2,2},{6,3},{14,4},{30,5},{62,6},{126,7},{254,8},{510,9},{1022,10},{2046,11}}; + static const unsigned short YAC_HT[256][2] = { + {10,4},{0,2},{1,2},{4,3},{11,4},{26,5},{120,7},{248,8},{1014,10},{65410,16},{65411,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {12,4},{27,5},{121,7},{502,9},{2038,11},{65412,16},{65413,16},{65414,16},{65415,16},{65416,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {28,5},{249,8},{1015,10},{4084,12},{65417,16},{65418,16},{65419,16},{65420,16},{65421,16},{65422,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {58,6},{503,9},{4085,12},{65423,16},{65424,16},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {59,6},{1016,10},{65430,16},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {122,7},{2039,11},{65438,16},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {123,7},{4086,12},{65446,16},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {250,8},{4087,12},{65454,16},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {504,9},{32704,15},{65462,16},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {505,9},{65470,16},{65471,16},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {506,9},{65479,16},{65480,16},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {1017,10},{65488,16},{65489,16},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {1018,10},{65497,16},{65498,16},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {2040,11},{65506,16},{65507,16},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {65515,16},{65516,16},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{0,0},{0,0},{0,0},{0,0},{0,0}, + {2041,11},{65525,16},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0} + }; + static const unsigned short UVAC_HT[256][2] = { + {0,2},{1,2},{4,3},{10,4},{24,5},{25,5},{56,6},{120,7},{500,9},{1014,10},{4084,12},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {11,4},{57,6},{246,8},{501,9},{2038,11},{4085,12},{65416,16},{65417,16},{65418,16},{65419,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {26,5},{247,8},{1015,10},{4086,12},{32706,15},{65420,16},{65421,16},{65422,16},{65423,16},{65424,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {27,5},{248,8},{1016,10},{4087,12},{65425,16},{65426,16},{65427,16},{65428,16},{65429,16},{65430,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {58,6},{502,9},{65431,16},{65432,16},{65433,16},{65434,16},{65435,16},{65436,16},{65437,16},{65438,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {59,6},{1017,10},{65439,16},{65440,16},{65441,16},{65442,16},{65443,16},{65444,16},{65445,16},{65446,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {121,7},{2039,11},{65447,16},{65448,16},{65449,16},{65450,16},{65451,16},{65452,16},{65453,16},{65454,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {122,7},{2040,11},{65455,16},{65456,16},{65457,16},{65458,16},{65459,16},{65460,16},{65461,16},{65462,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {249,8},{65463,16},{65464,16},{65465,16},{65466,16},{65467,16},{65468,16},{65469,16},{65470,16},{65471,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {503,9},{65472,16},{65473,16},{65474,16},{65475,16},{65476,16},{65477,16},{65478,16},{65479,16},{65480,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {504,9},{65481,16},{65482,16},{65483,16},{65484,16},{65485,16},{65486,16},{65487,16},{65488,16},{65489,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {505,9},{65490,16},{65491,16},{65492,16},{65493,16},{65494,16},{65495,16},{65496,16},{65497,16},{65498,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {506,9},{65499,16},{65500,16},{65501,16},{65502,16},{65503,16},{65504,16},{65505,16},{65506,16},{65507,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {2041,11},{65508,16},{65509,16},{65510,16},{65511,16},{65512,16},{65513,16},{65514,16},{65515,16},{65516,16},{0,0},{0,0},{0,0},{0,0},{0,0},{0,0}, + {16352,14},{65517,16},{65518,16},{65519,16},{65520,16},{65521,16},{65522,16},{65523,16},{65524,16},{65525,16},{0,0},{0,0},{0,0},{0,0},{0,0}, + {1018,10},{32707,15},{65526,16},{65527,16},{65528,16},{65529,16},{65530,16},{65531,16},{65532,16},{65533,16},{65534,16},{0,0},{0,0},{0,0},{0,0},{0,0} + }; + static const int YQT[] = {16,11,10,16,24,40,51,61,12,12,14,19,26,58,60,55,14,13,16,24,40,57,69,56,14,17,22,29,51,87,80,62,18,22, + 37,56,68,109,103,77,24,35,55,64,81,104,113,92,49,64,78,87,103,121,120,101,72,92,95,98,112,100,103,99}; + static const int UVQT[] = {17,18,24,47,99,99,99,99,18,21,26,66,99,99,99,99,24,26,56,99,99,99,99,99,47,66,99,99,99,99,99,99, + 99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99,99}; + static const float aasf[] = { 1.0f * 2.828427125f, 1.387039845f * 2.828427125f, 1.306562965f * 2.828427125f, 1.175875602f * 2.828427125f, + 1.0f * 2.828427125f, 0.785694958f * 2.828427125f, 0.541196100f * 2.828427125f, 0.275899379f * 2.828427125f }; + + int row, col, i, k, subsample; + float fdtbl_Y[64], fdtbl_UV[64]; + unsigned char YTable[64], UVTable[64]; + + if(!data || !width || !height || comp > 4 || comp < 1) { + return 0; + } + + quality = quality ? quality : 90; + subsample = quality <= 90 ? 1 : 0; + quality = quality < 1 ? 1 : quality > 100 ? 100 : quality; + quality = quality < 50 ? 5000 / quality : 200 - quality * 2; + + for(i = 0; i < 64; ++i) { + int uvti, yti = (YQT[i]*quality+50)/100; + YTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (yti < 1 ? 1 : yti > 255 ? 255 : yti); + uvti = (UVQT[i]*quality+50)/100; + UVTable[stbiw__jpg_ZigZag[i]] = (unsigned char) (uvti < 1 ? 1 : uvti > 255 ? 255 : uvti); + } + + for(row = 0, k = 0; row < 8; ++row) { + for(col = 0; col < 8; ++col, ++k) { + fdtbl_Y[k] = 1 / (YTable [stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]); + fdtbl_UV[k] = 1 / (UVTable[stbiw__jpg_ZigZag[k]] * aasf[row] * aasf[col]); + } + } + + // Write Headers + { + static const unsigned char head0[] = { 0xFF,0xD8,0xFF,0xE0,0,0x10,'J','F','I','F',0,1,1,0,0,1,0,1,0,0,0xFF,0xDB,0,0x84,0 }; + static const unsigned char head2[] = { 0xFF,0xDA,0,0xC,3,1,0,2,0x11,3,0x11,0,0x3F,0 }; + const unsigned char head1[] = { 0xFF,0xC0,0,0x11,8,(unsigned char)(height>>8),STBIW_UCHAR(height),(unsigned char)(width>>8),STBIW_UCHAR(width), + 3,1,(unsigned char)(subsample?0x22:0x11),0,2,0x11,1,3,0x11,1,0xFF,0xC4,0x01,0xA2,0 }; + s->func(s->context, (void*)head0, sizeof(head0)); + s->func(s->context, (void*)YTable, sizeof(YTable)); + stbiw__putc(s, 1); + s->func(s->context, UVTable, sizeof(UVTable)); + s->func(s->context, (void*)head1, sizeof(head1)); + s->func(s->context, (void*)(std_dc_luminance_nrcodes+1), sizeof(std_dc_luminance_nrcodes)-1); + s->func(s->context, (void*)std_dc_luminance_values, sizeof(std_dc_luminance_values)); + stbiw__putc(s, 0x10); // HTYACinfo + s->func(s->context, (void*)(std_ac_luminance_nrcodes+1), sizeof(std_ac_luminance_nrcodes)-1); + s->func(s->context, (void*)std_ac_luminance_values, sizeof(std_ac_luminance_values)); + stbiw__putc(s, 1); // HTUDCinfo + s->func(s->context, (void*)(std_dc_chrominance_nrcodes+1), sizeof(std_dc_chrominance_nrcodes)-1); + s->func(s->context, (void*)std_dc_chrominance_values, sizeof(std_dc_chrominance_values)); + stbiw__putc(s, 0x11); // HTUACinfo + s->func(s->context, (void*)(std_ac_chrominance_nrcodes+1), sizeof(std_ac_chrominance_nrcodes)-1); + s->func(s->context, (void*)std_ac_chrominance_values, sizeof(std_ac_chrominance_values)); + s->func(s->context, (void*)head2, sizeof(head2)); + } + + // Encode 8x8 macroblocks + { + static const unsigned short fillBits[] = {0x7F, 7}; + int DCY=0, DCU=0, DCV=0; + int bitBuf=0, bitCnt=0; + // comp == 2 is grey+alpha (alpha is ignored) + int ofsG = comp > 2 ? 1 : 0, ofsB = comp > 2 ? 2 : 0; + const unsigned char *dataR = (const unsigned char *)data; + const unsigned char *dataG = dataR + ofsG; + const unsigned char *dataB = dataR + ofsB; + int x, y, pos; + if(subsample) { + for(y = 0; y < height; y += 16) { + for(x = 0; x < width; x += 16) { + float Y[256], U[256], V[256]; + for(row = y, pos = 0; row < y+16; ++row) { + // row >= height => use last input row + int clamped_row = (row < height) ? row : height - 1; + int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp; + for(col = x; col < x+16; ++col, ++pos) { + // if col >= width => use pixel from last input column + int p = base_p + ((col < width) ? col : (width-1))*comp; + float r = dataR[p], g = dataG[p], b = dataB[p]; + Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128; + U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b; + V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b; + } + } + DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+0, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT); + DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+8, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT); + DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+128, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT); + DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y+136, 16, fdtbl_Y, DCY, YDC_HT, YAC_HT); + + // subsample U,V + { + float subU[64], subV[64]; + int yy, xx; + for(yy = 0, pos = 0; yy < 8; ++yy) { + for(xx = 0; xx < 8; ++xx, ++pos) { + int j = yy*32+xx*2; + subU[pos] = (U[j+0] + U[j+1] + U[j+16] + U[j+17]) * 0.25f; + subV[pos] = (V[j+0] + V[j+1] + V[j+16] + V[j+17]) * 0.25f; + } + } + DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subU, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT); + DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, subV, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT); + } + } + } + } else { + for(y = 0; y < height; y += 8) { + for(x = 0; x < width; x += 8) { + float Y[64], U[64], V[64]; + for(row = y, pos = 0; row < y+8; ++row) { + // row >= height => use last input row + int clamped_row = (row < height) ? row : height - 1; + int base_p = (stbi__flip_vertically_on_write ? (height-1-clamped_row) : clamped_row)*width*comp; + for(col = x; col < x+8; ++col, ++pos) { + // if col >= width => use pixel from last input column + int p = base_p + ((col < width) ? col : (width-1))*comp; + float r = dataR[p], g = dataG[p], b = dataB[p]; + Y[pos]= +0.29900f*r + 0.58700f*g + 0.11400f*b - 128; + U[pos]= -0.16874f*r - 0.33126f*g + 0.50000f*b; + V[pos]= +0.50000f*r - 0.41869f*g - 0.08131f*b; + } + } + + DCY = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, Y, 8, fdtbl_Y, DCY, YDC_HT, YAC_HT); + DCU = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, U, 8, fdtbl_UV, DCU, UVDC_HT, UVAC_HT); + DCV = stbiw__jpg_processDU(s, &bitBuf, &bitCnt, V, 8, fdtbl_UV, DCV, UVDC_HT, UVAC_HT); + } + } + } + + // Do the bit alignment of the EOI marker + stbiw__jpg_writeBits(s, &bitBuf, &bitCnt, fillBits); + } + + // EOI + stbiw__putc(s, 0xFF); + stbiw__putc(s, 0xD9); + + return 1; +} + +STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x, int y, int comp, const void *data, int quality) +{ + stbi__write_context s = { 0 }; + stbi__start_write_callbacks(&s, func, context); + return stbi_write_jpg_core(&s, x, y, comp, (void *) data, quality); +} + + +#ifndef STBI_WRITE_NO_STDIO +STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality) +{ + stbi__write_context s = { 0 }; + if (stbi__start_write_file(&s,filename)) { + int r = stbi_write_jpg_core(&s, x, y, comp, data, quality); + stbi__end_write_file(&s); + return r; + } else + return 0; +} +#endif + +#endif // STB_IMAGE_WRITE_IMPLEMENTATION + +/* Revision history + 1.16 (2021-07-11) + make Deflate code emit uncompressed blocks when it would otherwise expand + support writing BMPs with alpha channel + 1.15 (2020-07-13) unknown + 1.14 (2020-02-02) updated JPEG writer to downsample chroma channels + 1.13 + 1.12 + 1.11 (2019-08-11) + + 1.10 (2019-02-07) + support utf8 filenames in Windows; fix warnings and platform ifdefs + 1.09 (2018-02-11) + fix typo in zlib quality API, improve STB_I_W_STATIC in C++ + 1.08 (2018-01-29) + add stbi__flip_vertically_on_write, external zlib, zlib quality, choose PNG filter + 1.07 (2017-07-24) + doc fix + 1.06 (2017-07-23) + writing JPEG (using Jon Olick's code) + 1.05 ??? + 1.04 (2017-03-03) + monochrome BMP expansion + 1.03 ??? + 1.02 (2016-04-02) + avoid allocating large structures on the stack + 1.01 (2016-01-16) + STBIW_REALLOC_SIZED: support allocators with no realloc support + avoid race-condition in crc initialization + minor compile issues + 1.00 (2015-09-14) + installable file IO function + 0.99 (2015-09-13) + warning fixes; TGA rle support + 0.98 (2015-04-08) + added STBIW_MALLOC, STBIW_ASSERT etc + 0.97 (2015-01-18) + fixed HDR asserts, rewrote HDR rle logic + 0.96 (2015-01-17) + add HDR output + fix monochrome BMP + 0.95 (2014-08-17) + add monochrome TGA output + 0.94 (2014-05-31) + rename private functions to avoid conflicts with stb_image.h + 0.93 (2014-05-27) + warning fixes + 0.92 (2010-08-01) + casts to unsigned char to fix warnings + 0.91 (2010-07-17) + first public release + 0.90 first internal release +*/ + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/tools/build_release.sh b/tools/build_release.sh new file mode 100755 index 0000000..1f78f67 --- /dev/null +++ b/tools/build_release.sh @@ -0,0 +1,58 @@ +#!/usr/bin/env bash +# Builds every release download into dist/release/ — for maintainers. +# +# tools/build_release.sh +# +# Needs: zig (cross-compiles all six platforms from one machine), python3, +# and node + npx for the VS Code package. Outputs: +# yapf-.zip command-line tool + shared/static library + header +# yapf_image-blender.zip Blender extension with libraries for all platforms +# yapf-preview.vsix VS Code extension +# yapf-figma.zip Figma plugin (import via manifest.json) +# yapf-photoshop.zip Photoshop scripts +# yapf-gimp-source.zip GIMP plugin source (build with gimptool) +set -euo pipefail +cd "$(dirname "$0")/.." +OUT=dist/release +LIBS=dist/libs +rm -rf "$OUT" "$LIBS" +mkdir -p "$OUT" "$LIBS" + +command -v zig >/dev/null || { echo "zig not found: https://ziglang.org/download/" >&2; exit 1; } +export ZIG_GLOBAL_CACHE_DIR=${ZIG_GLOBAL_CACHE_DIR:-$PWD/dist/.zig-cache} + +# zig target folder library tool +targets="x86_64-linux-gnu.2.17 linux-x64 libyapf.so yapf + aarch64-linux-gnu.2.17 linux-arm64 libyapf.so yapf + x86_64-windows-gnu windows-x64 yapf.dll yapf.exe + aarch64-windows-gnu windows-arm64 yapf.dll yapf.exe + x86_64-macos macos-x64 libyapf.dylib yapf + aarch64-macos macos-arm64 libyapf.dylib yapf" + +echo "$targets" | while read -r tgt dir lib tool; do + pkg=dist/pkg/yapf-$dir + rm -rf "$pkg" && mkdir -p "$pkg" "$LIBS/$dir" + zig cc -target "$tgt" -O2 -std=c99 -shared -fPIC -s yapf.c -o "$LIBS/$dir/$lib" + zig cc -target "$tgt" -O2 -std=c99 -s tools/yapf_cli.c yapf.c -o "$pkg/$tool" -lm + zig cc -target "$tgt" -O2 -std=c99 -c yapf.c -o "$pkg/yapf.o" + zig ar rcs "$pkg/libyapf.a" "$pkg/yapf.o" && rm "$pkg/yapf.o" + cp "$LIBS/$dir/$lib" yapf.h LICENSE "$pkg/" + rm -f "$pkg"/*.lib "$pkg"/*.pdb "$LIBS/$dir"/*.lib "$LIBS/$dir"/*.pdb + (cd dist/pkg && python3 -m zipfile -c "../../$OUT/yapf-$dir.zip" "yapf-$dir") + echo "built yapf-$dir.zip" +done + +python3 extensions/blender/build.py "$LIBS" +mv dist/yapf_image-blender.zip "$OUT/" + +if command -v npx >/dev/null; then + (cd extensions/vscode && npm run --silent package && mv yapf-preview-*.vsix "../../$OUT/yapf-preview.vsix") +else + echo "node/npx not found: skipped the VS Code package" >&2 +fi + +(cd extensions/figma && node build.js 2>/dev/null || true) +python3 -m zipfile -c "$OUT/yapf-figma.zip" extensions/figma/manifest.json extensions/figma/code.js extensions/figma/ui.html +python3 -m zipfile -c "$OUT/yapf-photoshop.zip" extensions/photoshop +python3 -m zipfile -c "$OUT/yapf-gimp-source.zip" extensions/gimp yapf.c yapf.h +ls -la "$OUT" diff --git a/tools/yapf_cli.c b/tools/yapf_cli.c new file mode 100644 index 0000000..7571436 --- /dev/null +++ b/tools/yapf_cli.c @@ -0,0 +1,345 @@ +/* + * yapf_cli.c — the `yapf` command-line tool. + * + * Copyright 2026 Nexoniarz + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * ------------------------------------------------------------------------- + * + * yapf [options] convert; formats chosen by extension + * yapf info print header information + * yapf bench time decoding + * yapf thumbnail [size] small preview (default 256) + * + * Reads PNG, JPEG, BMP, TGA, GIF (first frame), PSD (composite), PNM, HDR + * (tone-clamped) and YAPF. Writes YAPF, PNG, BMP, TGA and JPEG. + * + * Options for writing YAPF: + * --mips also store a full mip chain (2×2 box filter) + * --linear mark pixels as linear instead of sRGB + * --premult mark alpha as premultiplied + * Options for reading YAPF: + * --mip N export mip level N instead of the base image + * + * Build: cc -O2 -std=c99 -Ithird_party tools/yapf_cli.c yapf.c -o yapf -lm -pthread + */ + +#if !defined(_WIN32) && !defined(_POSIX_C_SOURCE) +# define _POSIX_C_SOURCE 200809L +#endif + +#include "../yapf.h" + +/* Third-party, public domain; keep their warnings out of our build. */ +#if defined(__GNUC__) || defined(__clang__) +# pragma GCC diagnostic push +# pragma GCC diagnostic ignored "-Wunused-function" +# pragma GCC diagnostic ignored "-Wunused-but-set-variable" +# pragma GCC diagnostic ignored "-Wsign-compare" +#endif +#define STB_IMAGE_IMPLEMENTATION +#define STBI_FAILURE_USERMSG +#include "../third_party/stb_image.h" +#define STB_IMAGE_WRITE_IMPLEMENTATION +#include "../third_party/stb_image_write.h" +#if defined(__GNUC__) || defined(__clang__) +# pragma GCC diagnostic pop +#endif + +#include +#include +#include +#include +#include + +static int ends_with(const char *s, const char *ext) { + size_t n = strlen(s), m = strlen(ext); + if (n < m) return 0; + for (size_t i = 0; i < m; i++) + if (tolower((unsigned char)s[n - m + i]) != ext[i]) return 0; + return 1; +} + +static int is_yapf(const char *path) { return ends_with(path, ".yapf"); } + +static double now_s(void) { +#if defined(_WIN32) + return (double)clock() / CLOCKS_PER_SEC; +#else + struct timespec t; + clock_gettime(CLOCK_MONOTONIC, &t); + return (double)t.tv_sec + (double)t.tv_nsec * 1e-9; +#endif +} + +static uint8_t gpu_hint(int ch, int srgb) { + switch (ch) { + case 1: return YAPF_GPU_R8; + case 2: return YAPF_GPU_RG8; + case 3: return srgb ? YAPF_GPU_SRGB8 : YAPF_GPU_RGB8; + default: return srgb ? YAPF_GPU_SRGB8_A8 : YAPF_GPU_RGBA8; + } +} + +/* Full mip chain with a 2×2 box filter (edges clamp for odd sizes). */ +static int build_mips(yapf_image_t *img) { + int levels = 1; + while (levels < (int)YAPF_MAX_MIPS && + ((img->width >> levels) > 0 || (img->height >> levels) > 0)) + levels++; + img->mips = (uint8_t **)calloc((size_t)levels, sizeof(uint8_t *)); + if (!img->mips) return -1; + img->mips[0] = img->pixels; + img->mip_levels = (uint8_t)levels; + + int ch = img->channels; + for (int m = 1; m < levels; m++) { + uint32_t pw = img->width >> (m - 1), ph = img->height >> (m - 1); + uint32_t w = img->width >> m, h = img->height >> m; + if (pw < 1) pw = 1; + if (ph < 1) ph = 1; + if (w < 1) w = 1; + if (h < 1) h = 1; + uint8_t *src = img->mips[m - 1]; + uint8_t *dst = (uint8_t *)malloc((size_t)w * h * ch); + if (!dst) return -1; + for (uint32_t y = 0; y < h; y++) + for (uint32_t x = 0; x < w; x++) + for (int c = 0; c < ch; c++) { + uint32_t x0 = 2 * x < pw ? 2 * x : pw - 1, x1 = 2 * x + 1 < pw ? 2 * x + 1 : pw - 1; + uint32_t y0 = 2 * y < ph ? 2 * y : ph - 1, y1 = 2 * y + 1 < ph ? 2 * y + 1 : ph - 1; + unsigned s = src[((size_t)y0 * pw + x0) * ch + c] + src[((size_t)y0 * pw + x1) * ch + c] + + src[((size_t)y1 * pw + x0) * ch + c] + src[((size_t)y1 * pw + x1) * ch + c]; + dst[((size_t)y * w + x) * ch + c] = (uint8_t)((s + 2) / 4); + } + img->mips[m] = dst; + } + return 0; +} + +static int write_image(const char *path, const uint8_t *px, int w, int h, int ch) { + if (ends_with(path, ".png")) return stbi_write_png(path, w, h, ch, px, w * ch); + if (ends_with(path, ".bmp")) return stbi_write_bmp(path, w, h, ch, px); + if (ends_with(path, ".tga")) return stbi_write_tga(path, w, h, ch, px); + if (ends_with(path, ".jpg") || ends_with(path, ".jpeg")) + return stbi_write_jpg(path, w, h, ch, px, 95); + fprintf(stderr, "yapf: unsupported output format: %s (use .yapf .png .bmp .tga .jpg)\n", path); + return 0; +} + +static const char *gpu_name(int g) { + static const char *n[] = { "RGBA8", "RGB8", "RG8", "R8", "SRGB8_A8", "SRGB8" }; + return g >= 0 && g < 6 ? n[g] : "unknown"; +} + +static int cmd_info(const char *path) { + FILE *f = fopen(path, "rb"); + uint8_t hdr[20]; + if (!f || fread(hdr, 1, 20, f) != 20 || memcmp(hdr, "YAPF", 4) != 0) { + fprintf(stderr, "yapf: %s is not a YAPF file\n", path); + if (f) fclose(f); + return 1; + } + fseek(f, 0, SEEK_END); + long size = ftell(f); + fclose(f); + unsigned w = hdr[12] | hdr[13] << 8 | (unsigned)hdr[14] << 16 | (unsigned)hdr[15] << 24; + unsigned h = hdr[16] | hdr[17] << 8 | (unsigned)hdr[18] << 16 | (unsigned)hdr[19] << 24; + printf("%s\n", path); + printf(" format version : %u\n", hdr[4]); + printf(" size : %u x %u, %u channel(s)\n", w, h, hdr[5]); + printf(" mip levels : %u\n", hdr[7]); + printf(" gpu format : %s\n", gpu_name(hdr[6])); + printf(" flags : %s%s\n", hdr[8] & YAPF_FLAG_SRGB ? "sRGB " : "linear ", + hdr[8] & YAPF_FLAG_PREMULT_ALPHA ? "premultiplied-alpha" : ""); + printf(" file size : %ld bytes (%.1f%% of raw)\n", size, + 100.0 * (double)size / ((double)w * h * (hdr[5] ? hdr[5] : 1))); + yapf_image_t *img = yapf_load(path); + printf(" decodes : %s\n", img ? "ok" : "FAILED (corrupt or unsupported)"); + yapf_free(img); + return img ? 0 : 1; +} + +static int cmd_bench(const char *path) { + FILE *f = fopen(path, "rb"); + if (!f) { fprintf(stderr, "yapf: cannot open %s\n", path); return 1; } + fseek(f, 0, SEEK_END); + long n = ftell(f); + rewind(f); + uint8_t *buf = (uint8_t *)malloc((size_t)n); + if (!buf || fread(buf, 1, (size_t)n, f) != (size_t)n) { fclose(f); free(buf); return 1; } + fclose(f); + yapf_image_t *img = yapf_load_memory(buf, (size_t)n); + if (!img) { fprintf(stderr, "yapf: %s does not decode\n", path); free(buf); return 1; } + double raw = (double)img->width * img->height * img->channels; + yapf_free(img); + for (int mt = 0; mt < 2; mt++) { + /* Best of 15 rounds; each round repeats enough decodes to last + * ~20 ms, so short images are measured warm and not timer-bound. */ + double one = now_s(); + yapf_free(mt ? yapf_load_memory_mt(buf, (size_t)n, 0) : yapf_load_memory(buf, (size_t)n)); + one = now_s() - one; + int reps = (int)(0.02 / (one > 1e-6 ? one : 1e-6)) + 1; + double best = 1e9; + for (int r = 0; r < 15; r++) { + double t = now_s(); + for (int k = 0; k < reps; k++) + yapf_free(mt ? yapf_load_memory_mt(buf, (size_t)n, 0) + : yapf_load_memory(buf, (size_t)n)); + t = (now_s() - t) / reps; + if (t < best) best = t; + } + printf(" %-12s %8.3f ms %6.2f GB/s\n", mt ? "all cores" : "1 thread", best * 1e3, raw / best / 1e9); + } + free(buf); + return 0; +} + +/* Preview no larger than size×size: start from the smallest mip that is + * still big enough, then average boxes of source pixels. */ +static int cmd_thumbnail(const char *in, const char *out, int size) { + yapf_image_t *img = yapf_load_mt(in, 0); + if (!img) { fprintf(stderr, "yapf: cannot read %s\n", in); return 1; } + if (size < 1) size = 256; + int m = 0; + while (m + 1 < img->mip_levels && + (int)(img->width >> (m + 1)) >= size && (int)(img->height >> (m + 1)) >= size) + m++; + uint32_t sw = img->width >> m ? img->width >> m : 1, sh = img->height >> m ? img->height >> m : 1; + const uint8_t *src = img->mips[m]; + double k = (double)size / (double)(sw > sh ? sw : sh); + if (k > 1.0) k = 1.0; + uint32_t dw = (uint32_t)(sw * k + 0.5), dh = (uint32_t)(sh * k + 0.5); + if (dw < 1) dw = 1; + if (dh < 1) dh = 1; + int ch = img->channels; + uint8_t *dst = (uint8_t *)malloc((size_t)dw * dh * ch); + if (!dst) { yapf_free(img); return 1; } + for (uint32_t y = 0; y < dh; y++) { + uint32_t y0 = (uint32_t)((uint64_t)y * sh / dh), y1 = (uint32_t)((uint64_t)(y + 1) * sh / dh); + if (y1 <= y0) y1 = y0 + 1; + for (uint32_t x = 0; x < dw; x++) { + uint32_t x0 = (uint32_t)((uint64_t)x * sw / dw), x1 = (uint32_t)((uint64_t)(x + 1) * sw / dw); + if (x1 <= x0) x1 = x0 + 1; + for (int c = 0; c < ch; c++) { + uint64_t sum = 0; + for (uint32_t yy = y0; yy < y1; yy++) + for (uint32_t xx = x0; xx < x1; xx++) + sum += src[((size_t)yy * sw + xx) * ch + c]; + uint64_t n = (uint64_t)(y1 - y0) * (x1 - x0); + dst[((size_t)y * dw + x) * ch + c] = (uint8_t)((sum + n / 2) / n); + } + } + } + int ok = stbi_write_png(out, (int)dw, (int)dh, ch, dst, (int)dw * ch); + free(dst); + yapf_free(img); + if (!ok) fprintf(stderr, "yapf: cannot write %s\n", out); + return ok ? 0 : 1; +} + +static void usage(void) { + fprintf(stderr, + "usage: yapf [options] convert (formats by extension)\n" + " yapf info show header information\n" + " yapf bench time decoding\n" + " yapf thumbnail [size] preview, default 256 px\n" + "\n" + "reads: .yapf .png .jpg .bmp .tga .gif .psd .pnm .hdr\n" + "writes: .yapf .png .jpg .bmp .tga\n" + "\n" + "options when writing .yapf:\n" + " --mips store a full mip chain\n" + " --linear pixels are linear, not sRGB\n" + " --premult alpha is premultiplied\n" + "options when reading .yapf:\n" + " --mip N export mip level N\n"); +} + +int main(int argc, char **argv) { + if (argc == 3 && strcmp(argv[1], "info") == 0) return cmd_info(argv[2]); + if (argc == 3 && strcmp(argv[1], "bench") == 0) return cmd_bench(argv[2]); + if ((argc == 4 || argc == 5) && strcmp(argv[1], "thumbnail") == 0) + return cmd_thumbnail(argv[2], argv[3], argc == 5 ? atoi(argv[4]) : 256); + if (argc < 3) { usage(); return 2; } + + const char *in = argv[1], *out = argv[2]; + int mips = 0, srgb = 1, premult = 0, mip_out = 0; + for (int i = 3; i < argc; i++) { + if (strcmp(argv[i], "--mips") == 0) mips = 1; + else if (strcmp(argv[i], "--linear") == 0) srgb = 0; + else if (strcmp(argv[i], "--premult") == 0) premult = 1; + else if (strcmp(argv[i], "--mip") == 0 && i + 1 < argc) mip_out = atoi(argv[++i]); + else { usage(); return 2; } + } + + /* Read */ + yapf_image_t src; + yapf_image_t *loaded = NULL; + uint8_t *stb_px = NULL; + memset(&src, 0, sizeof(src)); + if (is_yapf(in)) { + loaded = yapf_load_mt(in, 0); + if (!loaded) { fprintf(stderr, "yapf: cannot read %s (missing, corrupt or not YAPF v1)\n", in); return 1; } + src = *loaded; + } else { + int w, h, ch; + if (stbi_is_16_bit(in)) + fprintf(stderr, "yapf: note: %s has 16-bit channels; YAPF stores 8 bits per channel\n", in); + stb_px = stbi_load(in, &w, &h, &ch, 0); + if (!stb_px) { fprintf(stderr, "yapf: cannot read %s: %s\n", in, stbi_failure_reason()); return 1; } + src.width = (uint32_t)w; src.height = (uint32_t)h; src.channels = (uint8_t)ch; + src.pixels = stb_px; src.mip_levels = 1; + src.flags = srgb ? YAPF_FLAG_SRGB : 0; + src.gpu_format = gpu_hint(ch, srgb); + } + + /* Write */ + int rc = 0; + if (is_yapf(out)) { + yapf_image_t dst = src; + if (!loaded) { + if (premult) dst.flags |= YAPF_FLAG_PREMULT_ALPHA; + } + dst.mips = NULL; + dst.mip_levels = 1; + if (mips && build_mips(&dst) != 0) { fprintf(stderr, "yapf: out of memory\n"); rc = 1; } + if (!rc && yapf_save(out, &dst) != YAPF_OK) { fprintf(stderr, "yapf: cannot write %s\n", out); rc = 1; } + if (dst.mips) { + for (int m = 1; m < dst.mip_levels; m++) free(dst.mips[m]); + free(dst.mips); + } + } else { + const uint8_t *px = src.pixels; + uint32_t w = src.width, h = src.height; + if (mip_out > 0) { + if (!loaded || mip_out >= src.mip_levels) { + fprintf(stderr, "yapf: %s has no mip level %d\n", in, mip_out); + rc = 1; + } else { + px = src.mips[mip_out]; + w = src.width >> mip_out ? src.width >> mip_out : 1; + h = src.height >> mip_out ? src.height >> mip_out : 1; + } + } + if (!rc && !write_image(out, px, (int)w, (int)h, src.channels)) { + fprintf(stderr, "yapf: cannot write %s\n", out); + rc = 1; + } + } + + yapf_free(loaded); + stbi_image_free(stb_px); + return rc; +} diff --git a/yapf.c b/yapf.c index 04b584b..f27a52f 100644 --- a/yapf.c +++ b/yapf.c @@ -17,6 +17,13 @@ * limitations under the License. */ +#if !defined(_WIN32) && !defined(_POSIX_C_SOURCE) +# define _POSIX_C_SOURCE 200809L +#endif +#if defined(__APPLE__) && !defined(_DARWIN_C_SOURCE) +# define _DARWIN_C_SOURCE /* sysconf(_SC_NPROCESSORS_ONLN) on macOS */ +#endif + #include "yapf.h" #include @@ -25,21 +32,53 @@ /* ── Constants ───────────────────────────────────────────────────────── */ -#define YAPF__MAGIC "YAPF" -#define YAPF__VERSION 2 -#define YAPF__HDR_LEN 20 -#define YAPF__MIP_ENTRY 8 -#define YAPF__MAX_TILE_PX (YAPF_TILE_SIZE * YAPF_TILE_SIZE) /* 1024 */ +#define YAPF__MAGIC "YAPF" +#define YAPF__VERSION 1 +#define YAPF__HDR_LEN 20 +#define YAPF__MIP_ENTRY 8 +#define YAPF__TILE_PX (YAPF_TILE_SIZE * YAPF_TILE_SIZE) -/* ── Tile pixel type ─────────────────────────────────────────────────── */ -/* - * After YCoCg-R, channels 1 (Co) and 2 (Cg) span [-255, 255] and need - * 9 bits, so we use int16_t for all channels internally within a tile. - * Grayscale and alpha channels stay in [0, 255]. - */ -typedef struct { int16_t v[4]; } yapf__tpx; +#define YAPF__GROUP 8 /* residuals per group: w bits each = w bytes */ +#define YAPF__MAX_GROUPS (YAPF__TILE_PX / YAPF__GROUP) +#define YAPF__BAND 8 /* rows sharing one filter byte */ +#define YAPF__STORED 0x80 /* tile header: raw pixels follow */ -/* ── I/O helpers ─────────────────────────────────────────────────────── */ +/* Control nibble → number of groups it covers. + * 0..8: one group of that width; 9..15: that many groups repeating the + * width of the last 0..8 nibble (0 at the start of a plane). */ +static const uint8_t yapf__nib_groups[16] = { + 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 3, 4, 8, 16, 32, 64 +}; +#define YAPF__NIB_REPEAT 9 + +/* Row filters (one byte per band of 8 rows, per plane). L / A / D / AR + * are the left, above, above-left and above-right samples. Every filter + * depends on the previous sample through at most one add; MED and + * average-with-left predictors were measured and dropped (~1 % smaller, + * half the decode speed). */ +enum { + YAPF__F_LEFT = 0, /* L */ + YAPF__F_GRAD = 1, /* L + A - D */ + YAPF__F_UPRIGHT = 2, /* AR */ + YAPF__F_UPAVG = 3, /* (A + AR) >> 1 */ + YAPF__F_UP = 16 /* 16 + k - 1: the sample k rows above, */ + /* k = 1..8 (repeating textures, dither) */ +}; +#define YAPF__UP_MAX 8 +#define YAPF__RING 16 /* decoder row ring, >= UP_MAX + 1, power of 2 */ + +static int yapf__filter_valid(uint8_t f) { + return f <= YAPF__F_UPAVG || (f >= YAPF__F_UP && f < YAPF__F_UP + YAPF__UP_MAX); +} + +/* Filters the encoder tries, cheapest to decode first: the up-row filters + * vectorise, Left and Gradient carry a serial dependency. On equal cost + * the earlier one wins. */ +static const uint8_t yapf__filters[] = { + YAPF__F_UP + 0, YAPF__F_UP + 1, YAPF__F_UP + 2, YAPF__F_UP + 3, + YAPF__F_UP + 4, YAPF__F_UP + 5, YAPF__F_UP + 6, YAPF__F_UP + 7, + YAPF__F_UPRIGHT, YAPF__F_UPAVG, YAPF__F_LEFT, YAPF__F_GRAD +}; static uint32_t yapf__r32le(const uint8_t *p) { return (uint32_t)p[0] @@ -55,434 +94,1318 @@ static void yapf__w32le(uint8_t *p, uint32_t v) { p[3] = (uint8_t)(v >> 24); } -/* ── Bit-writer ──────────────────────────────────────────────────────── */ +/* Little-endian 64-bit load; compilers turn this into a single load. */ +static uint64_t yapf__r64le(const uint8_t *p) { + return (uint64_t)p[0] + | ((uint64_t)p[1] << 8) + | ((uint64_t)p[2] << 16) + | ((uint64_t)p[3] << 24) + | ((uint64_t)p[4] << 32) + | ((uint64_t)p[5] << 40) + | ((uint64_t)p[6] << 48) + | ((uint64_t)p[7] << 56); +} + +/* ── Growable output buffer (encoder) ────────────────────────────────── */ typedef struct { uint8_t *buf; - size_t cap; size_t len; - uint8_t acc; /* accumulator */ - int bits; /* bits filled in acc (0-7) */ -} yapf__bw; + size_t cap; +} yapf__out; -static int yapf__bw_init(yapf__bw *w, size_t cap) { - w->buf = (uint8_t *)malloc(cap ? cap : 64); - if (!w->buf) return -1; - w->cap = cap ? cap : 64; - w->len = 0; - w->acc = 0; - w->bits = 0; - return 0; -} - -static int yapf__bw_grow(yapf__bw *w, size_t need) { - if (w->len + need <= w->cap) return 0; - size_t nc = w->cap * 2 + need; - uint8_t *nb = (uint8_t *)realloc(w->buf, nc); +static int yapf__reserve(yapf__out *o, size_t n) { + if (o->len + n <= o->cap) return 0; + size_t nc = o->cap ? o->cap : 4096; + while (nc < o->len + n) nc *= 2; + uint8_t *nb = (uint8_t *)realloc(o->buf, nc); if (!nb) return -1; - w->buf = nb; - w->cap = nc; + o->buf = nb; + o->cap = nc; return 0; } -/* Write n bits from v, MSB first; n in [1, 32]. */ -static int yapf__bw_put(yapf__bw *w, uint32_t v, int n) { - if (yapf__bw_grow(w, (size_t)(n + 7) / 8 + 2) != 0) return -1; - for (int i = n - 1; i >= 0; i--) { - w->acc = (uint8_t)((w->acc << 1) | ((v >> i) & 1u)); - if (++w->bits == 8) { - w->buf[w->len++] = w->acc; - w->acc = 0; - w->bits = 0; - } - } - return 0; -} - -/* Flush partial byte (zero-padded) and reset to byte boundary. */ -static void yapf__bw_flush(yapf__bw *w) { - if (w->bits > 0) { - w->buf[w->len++] = (uint8_t)(w->acc << (8 - w->bits)); - w->acc = 0; - w->bits = 0; - } -} - -/* ── Bit-reader ──────────────────────────────────────────────────────── */ - -typedef struct { - const uint8_t *buf; - size_t len; - size_t pos; - uint8_t acc; - int bits; -} yapf__br; - -static void yapf__br_init(yapf__br *r, const uint8_t *buf, size_t len) { - r->buf = buf; - r->len = len; - r->pos = 0; - r->acc = 0; - r->bits = 0; -} - -/* Read n bits MSB-first into *out. Returns 0 on success, -1 on underflow. */ -static int yapf__br_get(yapf__br *r, int n, uint32_t *out) { - uint32_t v = 0; - for (int i = 0; i < n; i++) { - if (r->bits == 0) { - if (r->pos >= r->len) return -1; - r->acc = r->buf[r->pos++]; - r->bits = 8; - } - v = (v << 1) | ((r->acc >> 7) & 1u); - r->acc = (uint8_t)(r->acc << 1); - r->bits--; - } - *out = v; - return 0; -} - -/* Discard remaining bits in current byte (tile boundary alignment). */ -static void yapf__br_align(yapf__br *r) { r->acc = 0; r->bits = 0; } - -/* ── YCoCg-R transform ───────────────────────────────────────────────── */ +/* ── YCoCg-R, 8-bit modular form ─────────────────────────────────────── */ /* - * Lossless, integer, fully reversible. - * Y ∈ [0,255], Co ∈ [-255,255], Cg ∈ [-255,255]. - * Alpha passes through unchanged. + * The usual YCoCg-R lifting steps, computed mod 256, with Co and Cg biased + * by 128 so neutral colours sit mid-range. Every lifting step adds a + * function of the other channel, so the transform stays exactly + * reversible while every plane remains one byte per sample. */ -static void yapf__forward(yapf__tpx *dst, const uint8_t *src, int ch) { - if (ch >= 3) { - int r = src[0], g = src[1], b = src[2]; - int co = r - b; - int t = b + (co >> 1); - int cg = g - t; - int y = t + (cg >> 1); - dst->v[0] = (int16_t)y; - dst->v[1] = (int16_t)co; - dst->v[2] = (int16_t)cg; - if (ch == 4) dst->v[3] = (int16_t)src[3]; - } else { - for (int c = 0; c < ch; c++) dst->v[c] = (int16_t)src[c]; +static void yapf__forward(uint8_t *y, uint8_t *co, uint8_t *cg, + uint8_t r, uint8_t g, uint8_t b) { + uint8_t o = (uint8_t)(r - b); + uint8_t t = (uint8_t)(b + ((int8_t)o >> 1)); + uint8_t q = (uint8_t)(g - t); + *y = (uint8_t)(t + ((int8_t)q >> 1)); + *co = (uint8_t)(o + 128); + *cg = (uint8_t)(q + 128); +} + +/* ── Prediction ──────────────────────────────────────────────────────── */ +/* Samples outside the tile (left of column 0, above row 0) read as 0. */ + + +/* Residuals are folded so small magnitudes become small codes: + * 0, -1, 1, -2, 2, ... → 0, 1, 2, 3, 4, ... */ +static uint8_t yapf__zz(uint8_t r) { + return (uint8_t)((r << 1) ^ (uint8_t)((int8_t)r >> 7)); +} + +static uint8_t yapf__unzz(uint8_t z) { + return (uint8_t)((z >> 1) ^ (uint8_t)-(z & 1)); +} + +/* Prediction for one row: R = the row, U = the row above, K = the row k + * rows above for YAPF__F_UP + k - 1 (rows outside the tile are all 0). + * Writes pred[0..tw). The encoder knows every sample, so this has no + * serial dependency. */ +static void yapf__predict_row(uint8_t f, const uint8_t *R, const uint8_t *U, + const uint8_t *K, uint8_t *pred, uint32_t tw) +{ + uint32_t x; + switch (f) { + case YAPF__F_LEFT: + pred[0] = 0; + for (x = 1; x < tw; x++) pred[x] = R[x - 1]; + break; + case YAPF__F_GRAD: + pred[0] = U[0]; + for (x = 1; x < tw; x++) pred[x] = (uint8_t)(R[x - 1] + U[x] - U[x - 1]); + break; + case YAPF__F_UPRIGHT: + for (x = 0; x + 1 < tw; x++) pred[x] = U[x + 1]; + pred[tw - 1] = U[tw - 1]; + break; + case YAPF__F_UPAVG: + for (x = 0; x + 1 < tw; x++) pred[x] = (uint8_t)((U[x] + U[x + 1]) >> 1); + pred[tw - 1] = U[tw - 1]; + break; + default: + memcpy(pred, K, tw); + break; } } -static void yapf__inverse(uint8_t *dst, const yapf__tpx *src, int ch) { - if (ch >= 3) { - int y = src->v[0]; - int co = src->v[1]; - int cg = src->v[2]; - int t = y - (cg >> 1); - int g = cg + t; - int b = t - (co >> 1); - int r = b + co; - dst[0] = (uint8_t)(r < 0 ? 0 : r > 255 ? 255 : r); - dst[1] = (uint8_t)(g < 0 ? 0 : g > 255 ? 255 : g); - dst[2] = (uint8_t)(b < 0 ? 0 : b > 255 ? 255 : b); - if (ch == 4) dst[3] = (uint8_t)src->v[3]; - } else { - for (int c = 0; c < ch; c++) dst[c] = (uint8_t)src->v[c]; +static int yapf__width(uint8_t z) { + int w = 0; + while (z) { z >>= 1; w++; } + return w; +} + +static const uint8_t yapf__zero_row[YAPF_TILE_SIZE + 1]; + +/* ── SIMD ────────────────────────────────────────────────────────────── */ +/* + * On by default; define YAPF_NO_SIMD for plain C everywhere. + * x86 / x64 (GCC, Clang, MSVC, MinGW): SSSE3 kernels, chosen at run time + * with CPUID, so the same binary still runs on CPUs without SSSE3. + * ARM64 (GCC, Clang, MSVC): NEON kernels; NEON is always present. + * Every kernel produces exactly the same bytes as the scalar code. Row + * kernels work in whole 16-sample chunks: they may read and write up to + * the end of the 64 (+1) sample row buffers, never past them, and pixels + * of narrow edge tiles go through a scratch row before being copied out. + * + * Group unpack, both ISAs: lane i of 16 bits gathers the two bytes holding + * value i (table yapf__simd_shuf), shifts right by that value's bit offset + * and masks to w bits. SSSE3 has no per-lane shift, so it multiplies: + * (2u * 2^(15 - s)) >> 16 = u >> s; overflow of 2u only touches bits >= 8. + */ +#if !defined(YAPF_NO_SIMD) && \ + (defined(__x86_64__) || defined(__i386__) || defined(_M_X64) || defined(_M_IX86)) +# define YAPF__SIMD_X86 1 +#elif !defined(YAPF_NO_SIMD) && (defined(__aarch64__) || defined(_M_ARM64)) +# define YAPF__SIMD_NEON 1 +#endif + +#if defined(YAPF__SIMD_X86) || defined(YAPF__SIMD_NEON) +# define YAPF__SIMD 1 +static const uint8_t yapf__simd_shuf[9][16] = { + { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 }, + { 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01 }, + { 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x01, 0x02, 0x01, 0x02, 0x01, 0x02 }, + { 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x01, 0x02, 0x01, 0x02, 0x02, 0x03, 0x02, 0x03 }, + { 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x01, 0x02, 0x02, 0x03, 0x02, 0x03, 0x03, 0x04, 0x03, 0x04 }, + { 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x03, 0x04, 0x04, 0x05 }, + { 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x03, 0x04, 0x04, 0x05, 0x05, 0x06 }, + { 0x00, 0x01, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06, 0x07 }, + { 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06, 0x07, 0x07, 0x80 } +}; +#else +# define YAPF__SIMD 0 +#endif + +#if defined(YAPF__SIMD_X86) +# if defined(_MSC_VER) +# include +# else +# include +# endif +# include +# if defined(__GNUC__) || defined(__clang__) +# define YAPF__SIMD_FN static __attribute__((target("ssse3"))) +# else +# define YAPF__SIMD_FN static /* MSVC allows intrinsics anywhere */ +# endif + +static const uint16_t yapf__simd_mul[9][8] = { + { 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000 }, + { 0x8000, 0x4000, 0x2000, 0x1000, 0x0800, 0x0400, 0x0200, 0x0100 }, + { 0x8000, 0x2000, 0x0800, 0x0200, 0x8000, 0x2000, 0x0800, 0x0200 }, + { 0x8000, 0x1000, 0x0200, 0x4000, 0x0800, 0x0100, 0x2000, 0x0400 }, + { 0x8000, 0x0800, 0x8000, 0x0800, 0x8000, 0x0800, 0x8000, 0x0800 }, + { 0x8000, 0x0400, 0x2000, 0x0100, 0x0800, 0x4000, 0x0200, 0x1000 }, + { 0x8000, 0x0200, 0x0800, 0x2000, 0x8000, 0x0200, 0x0800, 0x2000 }, + { 0x8000, 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 }, + { 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000 } +}; + +/* Shuffle masks: R, G, B rows of 16 → 48 interleaved bytes. */ +static const uint8_t yapf__simd_rgb[3][3][16] = { + { + { 0x00, 0x80, 0x80, 0x01, 0x80, 0x80, 0x02, 0x80, 0x80, 0x03, 0x80, 0x80, 0x04, 0x80, 0x80, 0x05 }, + { 0x80, 0x00, 0x80, 0x80, 0x01, 0x80, 0x80, 0x02, 0x80, 0x80, 0x03, 0x80, 0x80, 0x04, 0x80, 0x80 }, + { 0x80, 0x80, 0x00, 0x80, 0x80, 0x01, 0x80, 0x80, 0x02, 0x80, 0x80, 0x03, 0x80, 0x80, 0x04, 0x80 } + }, + { + { 0x80, 0x80, 0x06, 0x80, 0x80, 0x07, 0x80, 0x80, 0x08, 0x80, 0x80, 0x09, 0x80, 0x80, 0x0A, 0x80 }, + { 0x05, 0x80, 0x80, 0x06, 0x80, 0x80, 0x07, 0x80, 0x80, 0x08, 0x80, 0x80, 0x09, 0x80, 0x80, 0x0A }, + { 0x80, 0x05, 0x80, 0x80, 0x06, 0x80, 0x80, 0x07, 0x80, 0x80, 0x08, 0x80, 0x80, 0x09, 0x80, 0x80 } + }, + { + { 0x80, 0x0B, 0x80, 0x80, 0x0C, 0x80, 0x80, 0x0D, 0x80, 0x80, 0x0E, 0x80, 0x80, 0x0F, 0x80, 0x80 }, + { 0x80, 0x80, 0x0B, 0x80, 0x80, 0x0C, 0x80, 0x80, 0x0D, 0x80, 0x80, 0x0E, 0x80, 0x80, 0x0F, 0x80 }, + { 0x0A, 0x80, 0x80, 0x0B, 0x80, 0x80, 0x0C, 0x80, 0x80, 0x0D, 0x80, 0x80, 0x0E, 0x80, 0x80, 0x0F } + } +}; + +/* CPUID leaf 1, ECX bit 9 = SSSE3. Called once per load. */ +static int yapf__simd_available(void) { +# if defined(_MSC_VER) + int r[4]; + __cpuid(r, 1); + return (r[2] >> 9) & 1; +# else + unsigned a, b, c, d; + if (!__get_cpuid(1, &a, &b, &c, &d)) return 0; + return (int)((c >> 9) & 1u); +# endif +} + +YAPF__SIMD_FN const uint8_t *yapf__simd_unpack(const uint8_t *p, + const uint8_t *end, + const uint8_t *gw, uint32_t ng, + uint8_t *z) +{ + for (uint32_t g = 0; g < ng; g++) { + int w = gw[g]; + __m128i x; + if ((size_t)(end - p) >= 8) { + x = _mm_loadl_epi64((const __m128i *)(const void *)p); + } else { + uint8_t tmp[8] = {0}; + memcpy(tmp, p, (size_t)w); + x = _mm_loadl_epi64((const __m128i *)(const void *)tmp); + } + __m128i u = _mm_shuffle_epi8(x, _mm_loadu_si128( + (const __m128i *)(const void *)yapf__simd_shuf[w])); + u = _mm_add_epi16(u, u); + __m128i v = _mm_mulhi_epu16(u, _mm_loadu_si128( + (const __m128i *)(const void *)yapf__simd_mul[w])); + v = _mm_and_si128(v, _mm_set1_epi16((short)((1 << w) - 1))); + _mm_storel_epi64((__m128i *)(void *)(z + g * YAPF__GROUP), + _mm_packus_epi16(v, v)); + p += w; + } + return p; +} + +/* Left / Gradient for a full 64-sample row: the per-sample increments are + * independent (r, or r + A - D), so compute them 16 at a time and turn the + * serial chain into a prefix sum. */ +YAPF__SIMD_FN void yapf__simd_serial_row(uint8_t f, uint8_t *R, + const uint8_t *U, const uint8_t *Z, + uint32_t tw) +{ + const __m128i zero = _mm_setzero_si128(); + __m128i carry = zero, uprev = zero; + for (uint32_t k = 0; k < tw; k += 16) { + __m128i zz = _mm_loadu_si128((const __m128i *)(const void *)(Z + k)); + __m128i t = _mm_xor_si128( + _mm_and_si128(_mm_srli_epi16(zz, 1), _mm_set1_epi8(0x7F)), + _mm_sub_epi8(zero, _mm_and_si128(zz, _mm_set1_epi8(1)))); + if (f == YAPF__F_GRAD) { + __m128i u = _mm_loadu_si128((const __m128i *)(const void *)(U + k)); + t = _mm_add_epi8(t, _mm_sub_epi8(u, _mm_alignr_epi8(u, uprev, 15))); + uprev = u; + } + t = _mm_add_epi8(t, _mm_slli_si128(t, 1)); + t = _mm_add_epi8(t, _mm_slli_si128(t, 2)); + t = _mm_add_epi8(t, _mm_slli_si128(t, 4)); + t = _mm_add_epi8(t, _mm_slli_si128(t, 8)); + t = _mm_add_epi8(t, carry); + _mm_storeu_si128((__m128i *)(void *)(R + k), t); + carry = _mm_shuffle_epi8(t, _mm_set1_epi8(15)); } } -/* ── Predictor ───────────────────────────────────────────────────────── */ -/* left + above - diag, unclipped (works for signed int16 channels). */ -static int32_t yapf__predict(int16_t L, int16_t A, int16_t D) { - return (int32_t)L + (int32_t)A - (int32_t)D; +/* Bytes as int8, arithmetic shift right by one: ((v ^ 0x80) >> 1) - 64. */ +YAPF__SIMD_FN __m128i yapf__sse_half_s8(__m128i v) { + __m128i x = _mm_xor_si128(v, _mm_set1_epi8((char)0x80)); + x = _mm_and_si128(_mm_srli_epi16(x, 1), _mm_set1_epi8(0x7F)); + return _mm_sub_epi8(x, _mm_set1_epi8(64)); } -/* ── Signed bit width ────────────────────────────────────────────────── */ -/* Minimum two's-complement bits to represent v. Returns 0 if v == 0. */ -static int yapf__sbits(int v) { - if (v == 0) return 0; - if (v < 0) v = ~v; /* largest positive needing same width */ - int b = 1; /* sign bit */ - while (v > 0) { v >>= 1; b++; } - return b; +YAPF__SIMD_FN __m128i yapf__sse_unzz(__m128i z) { + return _mm_xor_si128(_mm_and_si128(_mm_srli_epi16(z, 1), _mm_set1_epi8(0x7F)), + _mm_sub_epi8(_mm_setzero_si128(), _mm_and_si128(z, _mm_set1_epi8(1)))); } -/* First-pixel bit widths: Co (ch1) and Cg (ch2) need 9 bits each because - * YCoCg-R maps them to [-255, 255]; all other channels stay at 8 bits. */ -static int yapf__first_bits(int c, int ch) { - return (ch >= 3 && (c == 1 || c == 2)) ? 9 : 8; +/* Vertical filters for a full 64-sample row: R = ((a + b) >> 1) + r. */ +YAPF__SIMD_FN void yapf__simd_vertical_row(uint8_t *R, const uint8_t *A, + const uint8_t *B, const uint8_t *Z, + uint32_t tw) +{ + for (uint32_t k = 0; k < tw; k += 16) { + __m128i a = _mm_loadu_si128((const __m128i *)(const void *)(A + k)); + __m128i b = _mm_loadu_si128((const __m128i *)(const void *)(B + k)); + __m128i avg = _mm_sub_epi8(_mm_avg_epu8(a, b), /* avg rounds up */ + _mm_and_si128(_mm_xor_si128(a, b), _mm_set1_epi8(1))); + __m128i z = _mm_loadu_si128((const __m128i *)(const void *)(Z + k)); + _mm_storeu_si128((__m128i *)(void *)(R + k), _mm_add_epi8(avg, yapf__sse_unzz(z))); + } } -/* ── Tile encode ─────────────────────────────────────────────────────── */ +/* Planes of one 64-pixel row → interleaved output pixels. */ +YAPF__SIMD_FN void yapf__simd_inverse_row(uint8_t *d, const uint8_t *P0, + const uint8_t *P1, const uint8_t *P2, + const uint8_t *P3, int ch, int xf) +{ + const __m128i bias = _mm_set1_epi8((char)0x80); + for (int k = 0; k < 64; k += 16) { + __m128i p0 = _mm_loadu_si128((const __m128i *)(const void *)(P0 + k)); + if (ch == 1) { + _mm_storeu_si128((__m128i *)(void *)(d + k), p0); + continue; + } + __m128i p1 = _mm_loadu_si128((const __m128i *)(const void *)(P1 + k)); + if (ch == 2) { + _mm_storeu_si128((__m128i *)(void *)(d + 2 * k), _mm_unpacklo_epi8(p0, p1)); + _mm_storeu_si128((__m128i *)(void *)(d + 2 * k + 16), _mm_unpackhi_epi8(p0, p1)); + continue; + } + __m128i p2 = _mm_loadu_si128((const __m128i *)(const void *)(P2 + k)); + __m128i r, g, b; + if (xf) { /* subtract-green */ + g = p0; + r = _mm_add_epi8(_mm_sub_epi8(p1, bias), g); + b = _mm_add_epi8(_mm_sub_epi8(p2, bias), g); + } else { /* YCoCg-R */ + __m128i co = _mm_sub_epi8(p1, bias), cg = _mm_sub_epi8(p2, bias); + __m128i t = _mm_sub_epi8(p0, yapf__sse_half_s8(cg)); + b = _mm_sub_epi8(t, yapf__sse_half_s8(co)); + r = _mm_add_epi8(b, co); + g = _mm_add_epi8(cg, t); + } + if (ch == 3) { + uint8_t *o = d + 3 * k; + for (int j = 0; j < 3; j++) { + const __m128i *m = (const __m128i *)(const void *)yapf__simd_rgb[j]; + __m128i v = _mm_or_si128( + _mm_or_si128(_mm_shuffle_epi8(r, _mm_loadu_si128(m)), + _mm_shuffle_epi8(g, _mm_loadu_si128(m + 1))), + _mm_shuffle_epi8(b, _mm_loadu_si128(m + 2))); + _mm_storeu_si128((__m128i *)(void *)(o + 16 * j), v); + } + } else { + __m128i a = _mm_loadu_si128((const __m128i *)(const void *)(P3 + k)); + __m128i rg0 = _mm_unpacklo_epi8(r, g), rg1 = _mm_unpackhi_epi8(r, g); + __m128i ba0 = _mm_unpacklo_epi8(b, a), ba1 = _mm_unpackhi_epi8(b, a); + uint8_t *o = d + 4 * k; + _mm_storeu_si128((__m128i *)(void *)(o), _mm_unpacklo_epi16(rg0, ba0)); + _mm_storeu_si128((__m128i *)(void *)(o + 16), _mm_unpackhi_epi16(rg0, ba0)); + _mm_storeu_si128((__m128i *)(void *)(o + 32), _mm_unpacklo_epi16(rg1, ba1)); + _mm_storeu_si128((__m128i *)(void *)(o + 48), _mm_unpackhi_epi16(rg1, ba1)); + } + } +} + +#elif defined(YAPF__SIMD_NEON) +# include +# define YAPF__SIMD_FN static + +static const int16_t yapf__simd_nshift[9][8] = { + { 0, 0, 0, 0, 0, 0, 0, 0 }, + { 0, -1, -2, -3, -4, -5, -6, -7 }, + { 0, -2, -4, -6, 0, -2, -4, -6 }, + { 0, -3, -6, -1, -4, -7, -2, -5 }, + { 0, -4, 0, -4, 0, -4, 0, -4 }, + { 0, -5, -2, -7, -4, -1, -6, -3 }, + { 0, -6, -4, -2, 0, -6, -4, -2 }, + { 0, -7, -6, -5, -4, -3, -2, -1 }, + { 0, 0, 0, 0, 0, 0, 0, 0 } +}; + +static int yapf__simd_available(void) { return 1; } + +YAPF__SIMD_FN const uint8_t *yapf__simd_unpack(const uint8_t *p, + const uint8_t *end, + const uint8_t *gw, uint32_t ng, + uint8_t *z) +{ + for (uint32_t g = 0; g < ng; g++) { + int w = gw[g]; + uint8_t tmp[8] = {0}; + const uint8_t *src = p; + if ((size_t)(end - p) < 8) { memcpy(tmp, p, (size_t)w); src = tmp; } + uint8x16_t x = vcombine_u8(vld1_u8(src), vdup_n_u8(0)); + uint16x8_t u = vreinterpretq_u16_u8(vqtbl1q_u8(x, vld1q_u8(yapf__simd_shuf[w]))); + uint16x8_t v = vandq_u16(vshlq_u16(u, vld1q_s16(yapf__simd_nshift[w])), + vdupq_n_u16((uint16_t)((1u << w) - 1))); + vst1_u8(z + g * YAPF__GROUP, vmovn_u16(v)); + p += w; + } + return p; +} + +YAPF__SIMD_FN void yapf__simd_serial_row(uint8_t f, uint8_t *R, + const uint8_t *U, const uint8_t *Z, + uint32_t tw) +{ + const uint8x16_t zero = vdupq_n_u8(0); + uint8x16_t carry = zero, uprev = zero; + for (uint32_t k = 0; k < tw; k += 16) { + uint8x16_t zz = vld1q_u8(Z + k); + uint8x16_t t = veorq_u8(vshrq_n_u8(zz, 1), + vsubq_u8(zero, vandq_u8(zz, vdupq_n_u8(1)))); + if (f == YAPF__F_GRAD) { + uint8x16_t u = vld1q_u8(U + k); + t = vaddq_u8(t, vsubq_u8(u, vextq_u8(uprev, u, 15))); + uprev = u; + } + t = vaddq_u8(t, vextq_u8(zero, t, 15)); + t = vaddq_u8(t, vextq_u8(zero, t, 14)); + t = vaddq_u8(t, vextq_u8(zero, t, 12)); + t = vaddq_u8(t, vextq_u8(zero, t, 8)); + t = vaddq_u8(t, carry); + vst1q_u8(R + k, t); + carry = vdupq_laneq_u8(t, 15); + } +} + +YAPF__SIMD_FN void yapf__simd_vertical_row(uint8_t *R, const uint8_t *A, + const uint8_t *B, const uint8_t *Z, + uint32_t tw) +{ + for (uint32_t k = 0; k < tw; k += 16) { + uint8x16_t z = vld1q_u8(Z + k); + uint8x16_t r = veorq_u8(vshrq_n_u8(z, 1), + vsubq_u8(vdupq_n_u8(0), vandq_u8(z, vdupq_n_u8(1)))); + vst1q_u8(R + k, vaddq_u8(vhaddq_u8(vld1q_u8(A + k), vld1q_u8(B + k)), r)); + } +} + +YAPF__SIMD_FN void yapf__simd_inverse_row(uint8_t *d, const uint8_t *P0, + const uint8_t *P1, const uint8_t *P2, + const uint8_t *P3, int ch, int xf) +{ + const uint8x16_t bias = vdupq_n_u8(0x80); + for (int k = 0; k < 64; k += 16) { + uint8x16_t p0 = vld1q_u8(P0 + k); + if (ch == 1) { vst1q_u8(d + k, p0); continue; } + uint8x16_t p1 = vld1q_u8(P1 + k); + if (ch == 2) { + uint8x16x2_t v; v.val[0] = p0; v.val[1] = p1; + vst2q_u8(d + 2 * k, v); + continue; + } + uint8x16_t p2 = vld1q_u8(P2 + k), r, g, b; + if (xf) { + g = p0; + r = vaddq_u8(vsubq_u8(p1, bias), g); + b = vaddq_u8(vsubq_u8(p2, bias), g); + } else { + uint8x16_t co = vsubq_u8(p1, bias), cg = vsubq_u8(p2, bias); + uint8x16_t t = vsubq_u8(p0, vreinterpretq_u8_s8(vshrq_n_s8(vreinterpretq_s8_u8(cg), 1))); + b = vsubq_u8(t, vreinterpretq_u8_s8(vshrq_n_s8(vreinterpretq_s8_u8(co), 1))); + r = vaddq_u8(b, co); + g = vaddq_u8(cg, t); + } + if (ch == 3) { + uint8x16x3_t v; v.val[0] = r; v.val[1] = g; v.val[2] = b; + vst3q_u8(d + 3 * k, v); + } else { + uint8x16x4_t v; v.val[0] = r; v.val[1] = g; v.val[2] = b; v.val[3] = vld1q_u8(P3 + k); + vst4q_u8(d + 4 * k, v); + } + } +} +#endif + + +/* ── Plane coding ────────────────────────────────────────────────────── */ +/* + * Plane payload (all byte aligned): + * filter[ceil(th / 8)] one YAPF__F_* byte per band of 8 rows + * control nibbles low nibble first; one per group of 8 folded + * residuals, in raster order: + * 0..8 width w of this group + * 9..15 2, 3, 4, 8, 16, 32 or 64 groups repeating + * the last 0..8 width (0 at plane start) + * an odd count is padded with a zero nibble + * group data w bytes per group: the 8 values packed + * little-endian, value i at bit i*w + * The last group of a plane is padded with zeros. + */ + +/* Chooses a filter per band and writes the folded residuals to z (zero + * padded to a whole group) and the filters to filt. Returns the estimated + * coded size in 1/16 byte units. */ +static uint32_t yapf__choose_filters(const uint8_t *P, uint32_t tw, uint32_t th, + uint8_t *z, uint8_t *filt) +{ + uint32_t total = tw * th; + uint32_t nb = (th + YAPF__BAND - 1) / YAPF__BAND; + uint32_t sum = 0; + + /* Per band, pick the filter whose residuals code smallest: group + * widths plus control nibbles, in 1/16 byte units (zero groups mostly + * end up in runs, so they are counted as cheap). Left and Gradient + * decode serially, so they must win by more than 4 %. The previous + * band's winner is tried first and the others stop as soon as they + * cannot win; ties go to the earlier filter in yapf__filters. */ + size_t prev_fi = 0; + for (uint32_t b = 0; b < nb; b++) { + uint32_t y0 = b * YAPF__BAND; + uint32_t y1 = y0 + YAPF__BAND < th ? y0 + YAPF__BAND : th; + uint32_t n = (y1 - y0) * tw; + uint8_t *Z = z + y0 * tw; + uint8_t tmp[YAPF__BAND * YAPF_TILE_SIZE + YAPF__GROUP]; + uint32_t best_cost = UINT32_MAX; + size_t best_fi = 0; + + for (size_t t = 0; t < sizeof(yapf__filters); t++) { + size_t fi = t == 0 ? prev_fi : (t <= prev_fi ? t - 1 : t); + uint8_t f = yapf__filters[fi]; + uint32_t k = f >= YAPF__F_UP ? (uint32_t)(f - YAPF__F_UP + 1) : 1; + int serial = f == YAPF__F_LEFT || f == YAPF__F_GRAD; + uint32_t cost = 0, done = 0; + int lost = 0; + + for (uint32_t y = y0; y < y1 && !lost; y++) { + const uint8_t *R = P + y * tw; + const uint8_t *U = y ? R - tw : yapf__zero_row; + const uint8_t *K = y >= k ? R - k * tw : yapf__zero_row; + uint8_t *T = tmp + (y - y0) * tw; + uint8_t pred[YAPF_TILE_SIZE]; + yapf__predict_row(f, R, U, K, pred, tw); + for (uint32_t x = 0; x < tw; x++) + T[x] = yapf__zz((uint8_t)(R[x] - pred[x])); + + /* cost of the groups completed so far */ + uint32_t end = y + 1 == y1 ? n : (y + 1 - y0) * tw; + if (y + 1 == y1) memset(tmp + n, 0, YAPF__GROUP); + for (; done + YAPF__GROUP <= end || (y + 1 == y1 && done < end); + done += YAPF__GROUP) { + uint8_t m = 0; + for (int i = 0; i < YAPF__GROUP; i++) m |= tmp[done + i]; + int w = yapf__width(m); + cost += w ? (uint32_t)(16 * w + 8) : 2; + } + uint32_t c = serial ? cost + cost / 25 : cost; + if (c > best_cost || (c == best_cost && fi > best_fi)) lost = 1; + } + if (lost) continue; + if (serial) cost += cost / 25; + if (cost < best_cost || (cost == best_cost && fi < best_fi)) { + best_cost = cost; + best_fi = fi; + memcpy(Z, tmp, n); + } + } + prev_fi = best_fi; + filt[b] = yapf__filters[best_fi]; + sum += best_cost; + } + memset(z + total, 0, YAPF__GROUP); + return sum; +} + +/* Writes a plane payload from residuals and filters made by + * yapf__choose_filters. */ +static int yapf__emit_plane(yapf__out *o, const uint8_t *z, const uint8_t *filt, + uint32_t tw, uint32_t th) +{ + uint32_t total = tw * th; + uint32_t ng = (total + YAPF__GROUP - 1) / YAPF__GROUP; + uint32_t nb = (th + YAPF__BAND - 1) / YAPF__BAND; + uint8_t gw[YAPF__MAX_GROUPS]; + uint8_t nib[YAPF__MAX_GROUPS]; + uint32_t nn = 0; + + /* worst case: filters + a nibble and 8 bytes per group */ + if (yapf__reserve(o, nb + ng + (size_t)ng * 8) != 0) return -1; + memcpy(o->buf + o->len, filt, nb); + o->len += nb; + + /* Group widths → control nibbles. A group may be stored wider than + * it needs if that lets a repeat code cover it; a small dynamic + * program over (group, current width) finds the cheapest sequence, + * counted in nibbles (1 per code, 2 per data byte). */ + uint8_t need[YAPF__MAX_GROUPS]; + for (uint32_t g = 0; g < ng; g++) { + uint8_t m = 0; + for (int i = 0; i < YAPF__GROUP; i++) m |= z[g * YAPF__GROUP + i]; + need[g] = (uint8_t)yapf__width(m); + } + { + static const uint8_t lens[7] = { 2, 3, 4, 8, 16, 32, 64 }; + uint32_t cost[YAPF__MAX_GROUPS + 1][9]; + uint8_t pick[YAPF__MAX_GROUPS][9]; /* nibble code */ + + /* span[j][g] = widest need in groups [g, g + 2^j) */ + static const int levels = 7; + uint8_t span[7][YAPF__MAX_GROUPS]; + memcpy(span[0], need, ng); + for (int j = 1; j < levels; j++) + for (uint32_t g = 0; g + (1u << j) <= ng; g++) { + uint8_t a = span[j - 1][g], c = span[j - 1][g + (1u << (j - 1))]; + span[j][g] = a > c ? a : c; + } + + for (int pw = 0; pw <= 8; pw++) cost[ng][pw] = 0; + for (uint32_t g = ng; g-- > 0; ) { + /* Best literal: independent of the current width. */ + uint32_t lc = UINT32_MAX; + uint8_t lk = 0; + for (int w = need[g]; w <= 8; w++) { + uint32_t c = 1 + 2u * (uint32_t)w + cost[g + 1][w]; + if (c < lc) { lc = c; lk = (uint8_t)w; } + } + /* Widest need over each repeat length that fits, from the + * power-of-two max table (length 3 = 2 + one more). */ + uint8_t mx[7]; + int nl = 0; + for (; nl < 7 && g + lens[nl] <= ng; nl++) { + uint32_t L = lens[nl]; + if (L == 3) { + uint8_t a = span[1][g], c = need[g + 2]; + mx[nl] = a > c ? a : c; + } else { + int j = 0; + while ((1u << j) < L) j++; + mx[nl] = span[j][g]; + } + } + for (int pw = 0; pw <= 8; pw++) { + uint32_t bc = lc; + uint8_t bk = lk; + for (int li = 0; li < nl && mx[li] <= pw; li++) { + uint32_t c = 1 + 2u * (uint32_t)pw * lens[li] + cost[g + lens[li]][pw]; + if (c < bc) { bc = c; bk = (uint8_t)(YAPF__NIB_REPEAT + li); } + } + cost[g][pw] = bc; + pick[g][pw] = bk; + } + } + uint8_t pw = 0; + for (uint32_t g = 0; g < ng; ) { + uint8_t k = pick[g][pw]; + nib[nn++] = k; + if (k < YAPF__NIB_REPEAT) { + pw = k; + gw[g++] = k; + } else { + for (uint32_t n = 0; n < yapf__nib_groups[k]; n++) gw[g++] = pw; + } + } + } + for (uint32_t k = 0; k < nn; k += 2) + o->buf[o->len++] = (uint8_t)(nib[k] | (k + 1 < nn ? nib[k + 1] << 4 : 0)); + + /* Packed group data: 8 values of w bits = w bytes */ + for (uint32_t g = 0; g < ng; g++) { + int w = gw[g]; + uint64_t acc = 0; + for (int i = 0; i < YAPF__GROUP; i++) + acc |= (uint64_t)z[g * YAPF__GROUP + i] << (i * w); + for (int k = 0; k < w; k++) o->buf[o->len++] = (uint8_t)(acc >> (8 * k)); + } + return 0; +} + +/* Parses one plane payload from [p, end) and unpacks its folded residuals + * into z; *filt receives the band filters. Returns the end of the payload + * or NULL on corrupt input. All sizes are validated before the unpack. */ +static const uint8_t *yapf__unpack_plane(const uint8_t *p, const uint8_t *end, + uint8_t *z, const uint8_t **filt_out, + uint32_t tw, uint32_t th, int simd) +{ + uint32_t total = tw * th; + uint32_t ng = (total + YAPF__GROUP - 1) / YAPF__GROUP; + uint32_t nb = (th + YAPF__BAND - 1) / YAPF__BAND; + uint8_t gw[YAPF__MAX_GROUPS + 64]; /* a final run may overshoot */ + + if ((size_t)(end - p) < nb) return NULL; + const uint8_t *filt = p; + *filt_out = filt; + p += nb; + for (uint32_t b = 0; b < nb; b++) + if (!yapf__filter_valid(filt[b])) return NULL; + + /* Control nibbles → group widths, and the total data size. Every + * nibble stores its width into the next 8 entries with one 64-bit + * write (the array has room); only runs longer than 8 groups, which + * are rare, take a branch to fill the rest. */ + size_t data = 0; + uint32_t g = 0; + int hi = 0; + uint8_t w = 0; + while (g < ng) { + if (p >= end) return NULL; + uint8_t c = (uint8_t)((*p >> (hi * 4)) & 15); + uint32_t n = yapf__nib_groups[c]; + p += hi; + hi ^= 1; + w = c < YAPF__NIB_REPEAT ? c : w; + { + uint64_t w8 = (uint64_t)w * 0x0101010101010101ull; + memcpy(gw + g, &w8, 8); + if (n > 8) memset(gw + g + 8, w, 56); + } + data += (size_t)w * n; + g += n; + } + if (g != ng) return NULL; /* run past the last group */ + if (hi) { /* padding nibble must be 0 */ + if (*p >> 4) return NULL; + p++; + } + if ((size_t)(end - p) < data) return NULL; + +#if YAPF__SIMD + if (simd) return yapf__simd_unpack(p, end, gw, ng, z); +#else + (void)simd; +#endif + + /* Unpack. Branch-free per group: one 64-bit load, 8 shifts. */ + for (g = 0; g < ng; g++) { + int w = gw[g]; + uint64_t x; + if ((size_t)(end - p) >= 8) { + x = yapf__r64le(p); + } else { + uint8_t tmp[8] = {0}; + memcpy(tmp, p, (size_t)w); + x = yapf__r64le(tmp); + } + uint64_t m = (1u << w) - 1; + uint8_t *o = z + g * YAPF__GROUP; + for (int i = 0; i < YAPF__GROUP; i++) + o[i] = (uint8_t)((x >> (i * w)) & m); + p += w; + } + + return p; +} + +/* Undo filter f for one row. R = reconstructed row, U = row above, + * K = row k above (for YAPF__F_UP + k - 1), Z = folded residuals. Rows + * carry one extra sample, R[tw] = R[tw - 1], so "above-right" needs no + * edge case. All vertical filters share one loop, pred = (a + b) >> 1: + * Up k: a = b = K Up-right: a = b = U + 1 + * Up-avg: a = U, b = U + 1 + * which keeps the per-row dispatch down to three targets. */ +static void yapf__unfilter_row(uint8_t f, uint8_t *R, const uint8_t *U, + const uint8_t *K, const uint8_t *Z, uint32_t tw) +{ + uint32_t x; + if (f == YAPF__F_LEFT) { + uint8_t L = 0; + for (x = 0; x < tw; x++) R[x] = L = (uint8_t)(L + yapf__unzz(Z[x])); + } else if (f == YAPF__F_GRAD) { + /* L + (A - D + r): only one add depends on the previous sample. */ + uint8_t L = (uint8_t)(U[0] + yapf__unzz(Z[0])); + R[0] = L; + for (x = 1; x < tw; x++) { + uint8_t t = (uint8_t)(U[x] - U[x - 1] + yapf__unzz(Z[x])); + R[x] = L = (uint8_t)(L + t); + } + } else { + const uint8_t *a = f >= YAPF__F_UP ? K : f == YAPF__F_UPAVG ? U : U + 1; + const uint8_t *b = f >= YAPF__F_UP ? K : U + 1; + for (x = 0; x < tw; x++) + R[x] = (uint8_t)(((a[x] + b[x]) >> 1) + yapf__unzz(Z[x])); + } + R[tw] = R[tw - 1]; +} + +/* ── Tile encode / decode ────────────────────────────────────────────── */ +/* + * Tile layout (byte aligned): + * header byte: 0x80 stored: tw*th*ch raw interleaved bytes follow + * bits 0..3 coded: bit c set = plane c is constant + * bit 4 coded RGB(A): planes are G, R-G, B-G (+ A) + * instead of Y, Co, Cg (+ A) + * coded: one value byte per constant plane, in plane order, + * then a plane payload per non-constant plane, in plane order + * The encoder tries both colour transforms and keeps the smaller tile, and + * stores the tile raw whenever coding would not make it smaller. + */ +#define YAPF__SUBGREEN 0x10 + +/* Encoder scratch, heap allocated once per level (too big for small + * thread stacks). */ +typedef struct { + uint8_t plane[2][4][YAPF__TILE_PX]; + uint8_t z[2][4][YAPF__TILE_PX + YAPF__GROUP]; + uint8_t filt[2][4][YAPF_TILE_SIZE / YAPF__BAND]; +} yapf__scratch; static int yapf__encode_tile( - yapf__bw *w, + yapf__out *o, yapf__scratch *sc, const uint8_t *pixels, uint32_t img_w, uint32_t tx, uint32_t ty, uint32_t tw, uint32_t th, int ch) { - yapf__tpx raw[YAPF__MAX_TILE_PX]; - int max_abs[4] = {0, 0, 0, 0}; - int bd[4]; - size_t total = (size_t)tw * th; + uint32_t total = tw * th; + size_t start = o->len; + size_t raw = (size_t)total * ch; + int tries = ch >= 3 ? 2 : 1; + uint8_t mask[2] = { 0, 0 }; + uint32_t est[2] = { 0, 0 }; - /* Gather + transform */ - for (uint32_t y = 0; y < th; y++) { - for (uint32_t x = 0; x < tw; x++) { - const uint8_t *src = - pixels + ((size_t)(ty + y) * img_w + (tx + x)) * (size_t)ch; - yapf__forward(&raw[y * tw + x], src, ch); + /* For each colour transform: planes, constant mask, filters, and an + * estimate of the coded size. Only the cheaper one is written. */ + for (int t = 0; t < tries; t++) { + uint8_t (*plane)[YAPF__TILE_PX] = sc->plane[t]; + for (uint32_t y = 0; y < th; y++) { + const uint8_t *row = pixels + ((size_t)(ty + y) * img_w + tx) * (size_t)ch; + for (uint32_t x = 0; x < tw; x++) { + const uint8_t *s = row + (size_t)x * ch; + uint32_t i = y * tw + x; + if (ch < 3) { + for (int c = 0; c < ch; c++) plane[c][i] = s[c]; + continue; + } + if (t == 0) { + yapf__forward(&plane[0][i], &plane[1][i], &plane[2][i], + s[0], s[1], s[2]); + } else { + plane[0][i] = s[1]; + plane[1][i] = (uint8_t)(s[0] - s[1] + 128); + plane[2][i] = (uint8_t)(s[2] - s[1] + 128); + } + if (ch == 4) plane[3][i] = s[3]; + } } - } - - /* Compute residuals and track max |residual| per channel. - * Pixel (0,0) is stored raw; all others use 2-D prediction. */ - for (size_t i = 1; i < total; i++) { - uint32_t x = (uint32_t)(i % tw); - uint32_t y = (uint32_t)(i / tw); - for (int c = 0; c < ch; c++) { - int16_t L = (x > 0) ? raw[y * tw + (x-1)].v[c] : 0; - int16_t A = (y > 0) ? raw[(y-1)*tw + x].v[c] : 0; - int16_t D = (x > 0 && y > 0) ? raw[(y-1)*tw + (x-1)].v[c] : 0; - int32_t pred = yapf__predict(L, A, D); - int res = (int)raw[i].v[c] - (int)pred; - int ab = res < 0 ? -res : res; - if (ab > max_abs[c]) max_abs[c] = ab; + uint32_t i = 1; + while (i < total && plane[c][i] == plane[c][0]) i++; + if (i == total) { mask[t] |= (uint8_t)(1u << c); est[t] += 16; continue; } + est[t] += yapf__choose_filters(plane[c], tw, th, sc->z[t][c], sc->filt[t][c]); } } + int t = (tries == 2 && est[1] < est[0]) ? 1 : 0; - /* Write per-channel residual bit depths */ - for (int c = 0; c < ch; c++) { - bd[c] = yapf__sbits(max_abs[c]); - if (yapf__bw_put(w, (uint32_t)bd[c], 8) != 0) return -1; - } + if (yapf__reserve(o, 1 + 4) != 0) return -1; + o->buf[o->len++] = (uint8_t)(mask[t] | (t ? YAPF__SUBGREEN : 0)); + for (int c = 0; c < ch; c++) + if (mask[t] & (1u << c)) o->buf[o->len++] = sc->plane[t][c][0]; + for (int c = 0; c < ch; c++) + if (!(mask[t] & (1u << c)) && + yapf__emit_plane(o, sc->z[t][c], sc->filt[t][c], tw, th) != 0) return -1; - /* Write first pixel with channel-appropriate width */ - for (int c = 0; c < ch; c++) { - int nbits = yapf__first_bits(c, ch); - uint32_t bits = (uint32_t)((int32_t)raw[0].v[c] & ((1 << nbits) - 1)); - if (yapf__bw_put(w, bits, nbits) != 0) return -1; - } - - /* Write packed residuals for pixels 1 .. total-1 */ - for (size_t i = 1; i < total; i++) { - uint32_t x = (uint32_t)(i % tw); - uint32_t y = (uint32_t)(i / tw); - - for (int c = 0; c < ch; c++) { - if (bd[c] == 0) continue; - int16_t L = (x > 0) ? raw[y * tw + (x-1)].v[c] : 0; - int16_t A = (y > 0) ? raw[(y-1)*tw + x].v[c] : 0; - int16_t D = (x > 0 && y > 0) ? raw[(y-1)*tw + (x-1)].v[c] : 0; - int32_t pred = yapf__predict(L, A, D); - int res = (int)raw[i].v[c] - (int)pred; - uint32_t bits = (uint32_t)(res & ((1 << bd[c]) - 1)); - if (yapf__bw_put(w, bits, bd[c]) != 0) return -1; + /* Fall back to a stored tile if coding did not pay off. */ + if (o->len - start > raw + 1) { + o->len = start; + if (yapf__reserve(o, raw + 1) != 0) return -1; + o->buf[o->len++] = YAPF__STORED; + for (uint32_t y = 0; y < th; y++) { + const uint8_t *row = pixels + ((size_t)(ty + y) * img_w + tx) * (size_t)ch; + memcpy(o->buf + o->len, row, (size_t)tw * ch); + o->len += (size_t)tw * ch; } } - - yapf__bw_flush(w); return 0; } -/* ── Tile decode ─────────────────────────────────────────────────────── */ - static int yapf__decode_tile( - yapf__br *r, + const uint8_t *data, size_t len, uint8_t *pixels, uint32_t img_w, uint32_t tx, uint32_t ty, uint32_t tw, uint32_t th, - int ch) + int ch, int simd) { - yapf__tpx raw[YAPF__MAX_TILE_PX]; - int bd[4]; - uint32_t bits; - size_t total = (size_t)tw * th; + uint8_t rows[4][YAPF__RING][YAPF_TILE_SIZE + 1]; /* last rows, ring */ + uint8_t z[4][YAPF__TILE_PX + YAPF__GROUP]; + const uint8_t *filt[4] = { NULL, NULL, NULL, NULL }; + size_t rowb = (size_t)tw * ch; + const uint8_t *p = data, *end = data + len; - /* Read residual bit depths */ - for (int c = 0; c < ch; c++) { - if (yapf__br_get(r, 8, &bits) != 0) return -1; - bd[c] = (int)bits; + if (len == 0) return -1; + uint8_t hdr = *p++; + + if (hdr == YAPF__STORED) { + if (len != 1 + rowb * th) return -1; + for (uint32_t y = 0; y < th; y++) + memcpy(pixels + ((size_t)(ty + y) * img_w + tx) * (size_t)ch, + p + rowb * y, rowb); + return 0; } + if (hdr & 0xE0) return -1; /* reserved bits set */ + uint8_t xf = hdr & YAPF__SUBGREEN; + hdr &= 0x0F; + if ((hdr >> ch) || (xf && ch < 3)) return -1; /* unknown bits set */ - /* Read first pixel. - * 8-bit channels (Y, A, grayscale) are unsigned [0,255] — no sign-extend. - * 9-bit channels (Co, Cg) are signed [-255,255] — two's complement. */ for (int c = 0; c < ch; c++) { - int nbits = yapf__first_bits(c, ch); - if (yapf__br_get(r, nbits, &bits) != 0) return -1; - int v; - if (nbits == 9) { - v = (int)(bits & 0xFFu); - if (bits & 0x100u) v -= 256; - } else { - v = (int)bits; /* unsigned: 0–255 */ - } - raw[0].v[c] = (int16_t)v; + if (!(hdr & (1u << c))) continue; + if (p >= end) return -1; + memset(rows[c], *p++, sizeof(rows[c])); } - - /* Decode pixels 1 .. total-1 */ - for (size_t i = 1; i < total; i++) { - uint32_t x = (uint32_t)(i % tw); - uint32_t y = (uint32_t)(i / tw); - - for (int c = 0; c < ch; c++) { - int16_t L = (x > 0) ? raw[y * tw + (x-1)].v[c] : 0; - int16_t A = (y > 0) ? raw[(y-1)*tw + x].v[c] : 0; - int16_t D = (x > 0 && y > 0) ? raw[(y-1)*tw + (x-1)].v[c] : 0; - int32_t pred = yapf__predict(L, A, D); - int res = 0; - - if (bd[c] > 0) { - if (yapf__br_get(r, bd[c], &bits) != 0) return -1; - int sb = 1 << (bd[c] - 1); - res = (int)(bits & (uint32_t)(sb - 1)); - if (bits & (uint32_t)sb) res -= sb; - } - raw[i].v[c] = (int16_t)((int)pred + res); + for (int c = 0; c < ch; c++) { + if (hdr & (1u << c)) continue; + p = yapf__unpack_plane(p, end, z[c], &filt[c], tw, th, simd); + if (!p) return -1; + /* SIMD row kernels read whole 16-sample chunks; keep the unused + * tail of narrow tiles defined. */ + if (tw != YAPF_TILE_SIZE) { + size_t used = ((size_t)tw * th + YAPF__GROUP - 1) / YAPF__GROUP * YAPF__GROUP; + memset(z[c] + used, 0, sizeof(z[c]) - used); } } + if (p != end) return -1; - yapf__br_align(r); - - /* Inverse transform and write to output buffer */ + /* Row by row: undo the filters of every plane (independent dependency + * chains the CPU can overlap), then the colour transform while the + * row is still in L1. */ for (uint32_t y = 0; y < th; y++) { - for (uint32_t x = 0; x < tw; x++) { - uint8_t *dst = - pixels + ((size_t)(ty + y) * img_w + (tx + x)) * (size_t)ch; - yapf__inverse(dst, &raw[y * tw + x], ch); + for (int c = 0; c < ch; c++) { + if (!filt[c]) continue; + uint8_t f = filt[c][y / YAPF__BAND]; + uint32_t k = f >= YAPF__F_UP ? (uint32_t)(f - YAPF__F_UP + 1) : 1; + uint8_t *R = rows[c][y % YAPF__RING]; + const uint8_t *U = y ? rows[c][(y - 1) % YAPF__RING] : yapf__zero_row; + const uint8_t *K = y >= k ? rows[c][(y - k) % YAPF__RING] : yapf__zero_row; + const uint8_t *Z = z[c] + y * tw; +#if YAPF__SIMD + if (simd) { + if (f == YAPF__F_LEFT || f == YAPF__F_GRAD) + yapf__simd_serial_row(f, R, U, Z, tw); + else if (f >= YAPF__F_UP) + yapf__simd_vertical_row(R, K, K, Z, tw); + else if (f == YAPF__F_UPAVG) + yapf__simd_vertical_row(R, U, U + 1, Z, tw); + else + yapf__simd_vertical_row(R, U + 1, U + 1, Z, tw); + R[tw] = R[tw - 1]; + continue; + } +#endif + yapf__unfilter_row(f, R, U, K, Z, tw); } + + uint8_t *d = pixels + ((size_t)(ty + y) * img_w + tx) * (size_t)ch; + const uint8_t *P0 = rows[0][y % YAPF__RING], *P1 = rows[1][y % YAPF__RING]; + const uint8_t *P2 = rows[2][y % YAPF__RING], *P3 = rows[3][y % YAPF__RING]; +#if YAPF__SIMD + if (simd) { + if (tw == YAPF_TILE_SIZE) { + yapf__simd_inverse_row(d, P0, P1, P2, P3, ch, xf); + } else { + uint8_t tmp[YAPF_TILE_SIZE * 4]; + yapf__simd_inverse_row(tmp, P0, P1, P2, P3, ch, xf); + memcpy(d, tmp, (size_t)tw * ch); + } + continue; + } +#endif + +#define YAPF__INVERSE(N, ALPHA) \ + for (uint32_t x = 0; x < tw; x++, d += N) { \ + uint8_t co = (uint8_t)(P1[x] - 128), cg = (uint8_t)(P2[x] - 128); \ + uint8_t t = (uint8_t)(P0[x] - ((int8_t)cg >> 1)); \ + uint8_t b = (uint8_t)(t - ((int8_t)co >> 1)); \ + d[0] = (uint8_t)(b + co); \ + d[1] = (uint8_t)(cg + t); \ + d[2] = b; \ + ALPHA; \ + } +#define YAPF__INVERSE_SG(N, ALPHA) \ + for (uint32_t x = 0; x < tw; x++, d += N) { \ + uint8_t g = P0[x]; \ + d[0] = (uint8_t)(P1[x] + g - 128); \ + d[1] = g; \ + d[2] = (uint8_t)(P2[x] + g - 128); \ + ALPHA; \ + } + if (ch == 3) { + /* Planes → separate R, G, B rows (vectorises), then interleave. */ + uint8_t rgb[3][YAPF_TILE_SIZE]; + if (xf) { + for (uint32_t x = 0; x < tw; x++) { + uint8_t g = P0[x]; + rgb[0][x] = (uint8_t)(P1[x] + g - 128); + rgb[1][x] = g; + rgb[2][x] = (uint8_t)(P2[x] + g - 128); + } + } else { + for (uint32_t x = 0; x < tw; x++) { + uint8_t co = (uint8_t)(P1[x] - 128), cg = (uint8_t)(P2[x] - 128); + uint8_t t = (uint8_t)(P0[x] - ((int8_t)cg >> 1)); + uint8_t b = (uint8_t)(t - ((int8_t)co >> 1)); + rgb[0][x] = (uint8_t)(b + co); + rgb[1][x] = (uint8_t)(cg + t); + rgb[2][x] = b; + } + } + d = pixels + ((size_t)(ty + y) * img_w + tx) * 3; + for (uint32_t x = 0; x < tw; x++, d += 3) { + d[0] = rgb[0][x]; d[1] = rgb[1][x]; d[2] = rgb[2][x]; + } + } else if (xf) { + YAPF__INVERSE_SG(4, d[3] = P3[x]) + } else if (ch == 4) { + YAPF__INVERSE(4, d[3] = P3[x]) + } else if (ch == 2) { + for (uint32_t x = 0; x < tw; x++, d += 2) { d[0] = P0[x]; d[1] = P1[x]; } + } else { + memcpy(d, P0, tw); + } +#undef YAPF__INVERSE +#undef YAPF__INVERSE_SG } return 0; } /* ── Level encode / decode ───────────────────────────────────────────── */ +/* + * Level layout: + * uint32 tile_size[tile_count] byte size of each tile, row-major order + * tile data, back to back + * Tiles are independent, so they can be decoded in any order or in + * parallel, and a sub-rectangle can be decoded without the rest. + */ + +static uint32_t yapf__tiles(uint32_t n) { + return (n + YAPF_TILE_SIZE - 1) / YAPF_TILE_SIZE; +} static int yapf__encode_level( - yapf__bw *w, const uint8_t *pixels, uint32_t mw, uint32_t mh, int ch) + yapf__out *o, const uint8_t *pixels, uint32_t mw, uint32_t mh, int ch) { - uint32_t tx_count = (mw + YAPF_TILE_SIZE - 1) / YAPF_TILE_SIZE; - uint32_t ty_count = (mh + YAPF_TILE_SIZE - 1) / YAPF_TILE_SIZE; + uint32_t tcx = yapf__tiles(mw), tcy = yapf__tiles(mh); + size_t tcount = (size_t)tcx * tcy; - for (uint32_t ty = 0; ty < ty_count; ty++) { - for (uint32_t tx = 0; tx < tx_count; tx++) { - uint32_t tw = (tx + 1) * YAPF_TILE_SIZE <= mw - ? YAPF_TILE_SIZE : mw - tx * YAPF_TILE_SIZE; - uint32_t th = (ty + 1) * YAPF_TILE_SIZE <= mh - ? YAPF_TILE_SIZE : mh - ty * YAPF_TILE_SIZE; - if (yapf__encode_tile(w, pixels, mw, - tx * YAPF_TILE_SIZE, - ty * YAPF_TILE_SIZE, - tw, th, ch) != 0) + /* Each level has its own buffer, so the table starts at offset 0. + * Reserve the size table, fill it after each tile is written. */ + if (yapf__reserve(o, tcount * 4) != 0) return -1; + memset(o->buf, 0, tcount * 4); + o->len = tcount * 4; + + yapf__scratch *sc = (yapf__scratch *)malloc(sizeof(yapf__scratch)); + if (!sc) return -1; + + size_t t = 0; + int rc = 0; + for (uint32_t ty = 0; ty < tcy && rc == 0; ty++) { + for (uint32_t tx = 0; tx < tcx && rc == 0; tx++, t++) { + uint32_t x0 = tx * YAPF_TILE_SIZE, y0 = ty * YAPF_TILE_SIZE; + uint32_t tw = mw - x0 < YAPF_TILE_SIZE ? mw - x0 : YAPF_TILE_SIZE; + uint32_t th = mh - y0 < YAPF_TILE_SIZE ? mh - y0 : YAPF_TILE_SIZE; + size_t start = o->len; + rc = yapf__encode_tile(o, sc, pixels, mw, x0, y0, tw, th, ch); + if (rc == 0) yapf__w32le(o->buf + t * 4, (uint32_t)(o->len - start)); + } + } + free(sc); + return rc; +} + +/* A level whose tile table has been validated; tiles can then be decoded + * in any order, from any thread. */ +typedef struct { + const uint8_t *data; + size_t *offs; /* tcount + 1 byte offsets into data */ + uint8_t *pixels; + uint32_t mw, mh, tcx, tcy; + int ch; + int simd; /* SIMD kernels usable on this CPU */ +} yapf__level; + +static int yapf__level_init(yapf__level *lv, const uint8_t *data, size_t len, + uint8_t *pixels, uint32_t mw, uint32_t mh, int ch) +{ + lv->data = data; + lv->pixels = pixels; + lv->mw = mw; + lv->mh = mh; + lv->tcx = yapf__tiles(mw); + lv->tcy = yapf__tiles(mh); + lv->ch = ch; + + size_t tcount = (size_t)lv->tcx * lv->tcy; + if (len / 4 < tcount) return -1; + lv->offs = (size_t *)malloc((tcount + 1) * sizeof(size_t)); + if (!lv->offs) return -1; + + size_t pos = tcount * 4; + for (size_t t = 0; t < tcount; t++) { + uint32_t sz = yapf__r32le(data + t * 4); + if (sz > len - pos) return -1; + lv->offs[t] = pos; + pos += sz; + } + lv->offs[tcount] = pos; + return 0; +} + +/* Decode tile rows first_row, first_row + step, ... */ +static int yapf__level_decode_rows(const yapf__level *lv, + uint32_t first_row, uint32_t step) +{ + for (uint32_t ty = first_row; ty < lv->tcy; ty += step) { + for (uint32_t tx = 0; tx < lv->tcx; tx++) { + size_t t = (size_t)ty * lv->tcx + tx; + uint32_t x0 = tx * YAPF_TILE_SIZE, y0 = ty * YAPF_TILE_SIZE; + uint32_t tw = lv->mw - x0 < YAPF_TILE_SIZE ? lv->mw - x0 : YAPF_TILE_SIZE; + uint32_t th = lv->mh - y0 < YAPF_TILE_SIZE ? lv->mh - y0 : YAPF_TILE_SIZE; + if (yapf__decode_tile(lv->data + lv->offs[t], + lv->offs[t + 1] - lv->offs[t], + lv->pixels, lv->mw, x0, y0, tw, th, lv->ch, + lv->simd) != 0) return -1; } } return 0; } -static int yapf__decode_level( - const uint8_t *data, size_t data_len, - uint8_t *pixels, uint32_t mw, uint32_t mh, int ch) +/* ── Threads ─────────────────────────────────────────────────────────── */ +/* + * Worker t of n decodes tile rows t, t + n, t + 2n, ... of every level. + * Define YAPF_NO_THREADS to build without any threading support. + */ + +typedef struct { + const yapf__level *levels; + int nlevels; + uint32_t index, count; + int rc; +} yapf__job; + +static void yapf__job_run(yapf__job *j) { + j->rc = 0; + for (int m = 0; m < j->nlevels && j->rc == 0; m++) + j->rc = yapf__level_decode_rows(&j->levels[m], j->index, j->count); +} + +#if !defined(YAPF_NO_THREADS) && defined(_WIN32) +# include +# define YAPF__THREADS 1 +typedef HANDLE yapf__thread; +static DWORD WINAPI yapf__thread_main(LPVOID arg) { + yapf__job_run((yapf__job *)arg); + return 0; +} +static int yapf__thread_start(yapf__thread *t, yapf__job *j) { + *t = CreateThread(NULL, 0, yapf__thread_main, j, 0, NULL); + return *t ? 0 : -1; +} +static void yapf__thread_join(yapf__thread t) { + WaitForSingleObject(t, INFINITE); + CloseHandle(t); +} +static int yapf__cpu_count(void) { + SYSTEM_INFO si; + GetSystemInfo(&si); + return (int)si.dwNumberOfProcessors; +} +#elif !defined(YAPF_NO_THREADS) && (defined(__unix__) || defined(__APPLE__)) +# include +# include +# define YAPF__THREADS 1 +typedef pthread_t yapf__thread; +static void *yapf__thread_main(void *arg) { + yapf__job_run((yapf__job *)arg); + return NULL; +} +static int yapf__thread_start(yapf__thread *t, yapf__job *j) { + return pthread_create(t, NULL, yapf__thread_main, j) == 0 ? 0 : -1; +} +static void yapf__thread_join(yapf__thread t) { + pthread_join(t, NULL); +} +static int yapf__cpu_count(void) { + long n = sysconf(_SC_NPROCESSORS_ONLN); + return n > 0 ? (int)n : 1; +} +#else +# define YAPF__THREADS 0 +#endif + +#define YAPF__MAX_THREADS 64 +#define YAPF__TILES_PER_THREAD 32 + +/* Decode all levels with up to `threads` threads (0 = one per core). */ +static int yapf__decode_levels(const yapf__level *levels, int nlevels, + int threads) { - yapf__br r; - yapf__br_init(&r, data, data_len); + yapf__job jobs[YAPF__MAX_THREADS]; + int n = 1; - uint32_t tx_count = (mw + YAPF_TILE_SIZE - 1) / YAPF_TILE_SIZE; - uint32_t ty_count = (mh + YAPF_TILE_SIZE - 1) / YAPF_TILE_SIZE; - - for (uint32_t ty = 0; ty < ty_count; ty++) { - for (uint32_t tx = 0; tx < tx_count; tx++) { - uint32_t tw = (tx + 1) * YAPF_TILE_SIZE <= mw - ? YAPF_TILE_SIZE : mw - tx * YAPF_TILE_SIZE; - uint32_t th = (ty + 1) * YAPF_TILE_SIZE <= mh - ? YAPF_TILE_SIZE : mh - ty * YAPF_TILE_SIZE; - if (yapf__decode_tile(&r, pixels, mw, - tx * YAPF_TILE_SIZE, - ty * YAPF_TILE_SIZE, - tw, th, ch) != 0) - return -1; - } +#if YAPF__THREADS + yapf__thread th[YAPF__MAX_THREADS]; + n = threads > 0 ? threads : yapf__cpu_count(); + if (n > YAPF__MAX_THREADS) n = YAPF__MAX_THREADS; + if ((uint32_t)n > levels[0].tcy) n = (int)levels[0].tcy; + /* Starting a thread costs about as much as decoding a few tiles, so + * give each one at least YAPF__TILES_PER_THREAD tiles of work. */ + { + size_t tiles = 0; + for (int m = 0; m < nlevels; m++) tiles += (size_t)levels[m].tcx * levels[m].tcy; + if ((size_t)n > tiles / YAPF__TILES_PER_THREAD) + n = (int)(tiles / YAPF__TILES_PER_THREAD); } + if (n < 1) n = 1; +#else + (void)threads; +#endif + + for (int i = 0; i < n; i++) { + jobs[i].levels = levels; + jobs[i].nlevels = nlevels; + jobs[i].index = (uint32_t)i; + jobs[i].count = (uint32_t)n; + } + +#if YAPF__THREADS + int started = 1; + for (; started < n; started++) + if (yapf__thread_start(&th[started], &jobs[started]) != 0) break; + /* If a thread failed to start, the calling thread takes its rows. */ + yapf__job_run(&jobs[0]); + for (int i = started; i < n; i++) yapf__job_run(&jobs[i]); + for (int i = 1; i < started; i++) yapf__thread_join(th[i]); +#else + yapf__job_run(&jobs[0]); +#endif + + for (int i = 0; i < n; i++) + if (jobs[i].rc != 0) return -1; return 0; } /* ======================================================================== - * yapf_load + * yapf_load_memory / yapf_load * ======================================================================== */ -yapf_image_t *yapf_load(const char *filename) { - if (!filename) return NULL; +static yapf_image_t *yapf__load(const uint8_t *buf, size_t size, int threads) { + if (!buf || size < YAPF__HDR_LEN) return NULL; + if (memcmp(buf, YAPF__MAGIC, 4) != 0) return NULL; - FILE *f = fopen(filename, "rb"); - if (!f) return NULL; + uint8_t version = buf[4]; + uint8_t channels = buf[5]; + uint8_t mip_levels = buf[7]; + uint32_t base_w = yapf__r32le(buf + 12); + uint32_t base_h = yapf__r32le(buf + 16); - uint8_t hdr[YAPF__HDR_LEN]; - if (fread(hdr, 1, YAPF__HDR_LEN, f) != YAPF__HDR_LEN) goto fail_f; - if (memcmp(hdr, YAPF__MAGIC, 4) != 0 || hdr[4] != YAPF__VERSION) goto fail_f; + if (version != YAPF__VERSION) return NULL; + if (channels < 1 || channels > 4) return NULL; + if (mip_levels < 1 || mip_levels > YAPF_MAX_MIPS) return NULL; + if (base_w == 0 || base_w > YAPF_MAX_DIM) return NULL; + if (base_h == 0 || base_h > YAPF_MAX_DIM) return NULL; - uint8_t channels = hdr[5]; - uint8_t gpu_fmt = hdr[6]; - uint8_t mip_levels = hdr[7]; - uint8_t flags = hdr[8]; - uint32_t base_w = yapf__r32le(hdr + 12); - uint32_t base_h = yapf__r32le(hdr + 16); + size_t idx_end = YAPF__HDR_LEN + (size_t)mip_levels * YAPF__MIP_ENTRY; + if (size < idx_end) return NULL; - if (channels < 1 || channels > 4) goto fail_f; - if (mip_levels < 1 || mip_levels > YAPF_MAX_MIPS) goto fail_f; - if (base_w == 0 || base_w > YAPF_MAX_DIM) goto fail_f; - if (base_h == 0 || base_h > YAPF_MAX_DIM) goto fail_f; - - /* Mip index */ - size_t idx_bytes = (size_t)mip_levels * YAPF__MIP_ENTRY; - uint8_t *mip_idx = (uint8_t *)malloc(idx_bytes); - if (!mip_idx) goto fail_f; - if (fread(mip_idx, 1, idx_bytes, f) != idx_bytes) { free(mip_idx); goto fail_f; } + yapf__level levels[YAPF_MAX_MIPS]; + memset(levels, 0, sizeof(levels)); + int ok = 0; yapf_image_t *img = (yapf_image_t *)calloc(1, sizeof(yapf_image_t)); - if (!img) { free(mip_idx); goto fail_f; } - + if (!img) return NULL; img->width = base_w; img->height = base_h; img->channels = channels; - img->gpu_format = gpu_fmt; - img->flags = flags; + img->gpu_format = buf[6]; + img->flags = buf[8]; img->mip_levels = mip_levels; - img->mips = (uint8_t **)calloc(mip_levels, sizeof(uint8_t *)); - if (!img->mips) { free(mip_idx); goto fail_img; } + if (!img->mips) goto done; for (uint8_t m = 0; m < mip_levels; m++) { - uint32_t offset = yapf__r32le(mip_idx + m * YAPF__MIP_ENTRY); - uint32_t size = yapf__r32le(mip_idx + m * YAPF__MIP_ENTRY + 4); - uint32_t mw = base_w >> m; if (mw < 1) mw = 1; - uint32_t mh = base_h >> m; if (mh < 1) mh = 1; + const uint8_t *e = buf + YAPF__HDR_LEN + (size_t)m * YAPF__MIP_ENTRY; + uint32_t off = yapf__r32le(e); + uint32_t len = yapf__r32le(e + 4); + uint32_t mw = base_w >> m; if (mw < 1) mw = 1; + uint32_t mh = base_h >> m; if (mh < 1) mh = 1; - uint8_t *data = (uint8_t *)malloc(size); - if (!data) { free(mip_idx); goto fail_mips; } + if (off < idx_end || off > size || len > size - off) goto done; - if (fseek(f, (long)offset, SEEK_SET) != 0 || - fread(data, 1, size, f) != size) { - free(data); free(mip_idx); goto fail_mips; - } + img->mips[m] = (uint8_t *)malloc((size_t)mw * mh * channels); + if (!img->mips[m]) goto done; - size_t px_bytes = (size_t)mw * mh * channels; - img->mips[m] = (uint8_t *)malloc(px_bytes); - if (!img->mips[m]) { free(data); free(mip_idx); goto fail_mips; } - - if (yapf__decode_level(data, size, img->mips[m], mw, mh, channels) != 0) { - free(data); free(mip_idx); goto fail_mips; - } - free(data); + if (yapf__level_init(&levels[m], buf + off, len, + img->mips[m], mw, mh, channels) != 0) + goto done; } - free(mip_idx); - img->pixels = img->mips[0]; - fclose(f); - return img; +#if YAPF__SIMD + { + int simd = yapf__simd_available(); + for (uint8_t m = 0; m < mip_levels; m++) levels[m].simd = simd; + } +#endif + ok = yapf__decode_levels(levels, mip_levels, threads) == 0; -fail_mips: - for (uint8_t m = 0; m < mip_levels; m++) free(img->mips[m]); - free(img->mips); -fail_img: - free(img); -fail_f: +done: + for (unsigned m = 0; m < YAPF_MAX_MIPS; m++) free(levels[m].offs); + if (!ok) { yapf_free(img); return NULL; } + img->pixels = img->mips[0]; + return img; +} + +yapf_image_t *yapf_load_memory(const void *buffer, size_t size) { + return yapf__load((const uint8_t *)buffer, size, 1); +} + +yapf_image_t *yapf_load_memory_mt(const void *buffer, size_t size, int threads) { + return yapf__load((const uint8_t *)buffer, size, threads); +} + +static yapf_image_t *yapf__load_file(const char *filename, int threads) { + if (!filename) return NULL; + FILE *f = fopen(filename, "rb"); + if (!f) return NULL; + + /* Read the whole file in chunks; avoids ftell() limits on 32-bit long. */ + size_t cap = 1 << 16, len = 0; + uint8_t *buf = (uint8_t *)malloc(cap); + while (buf) { + size_t n = fread(buf + len, 1, cap - len, f); + len += n; + if (len < cap) break; + uint8_t *nb = (uint8_t *)realloc(buf, cap * 2); + if (!nb) { free(buf); buf = NULL; break; } + buf = nb; + cap *= 2; + } + int err = ferror(f); fclose(f); - return NULL; + if (!buf) return NULL; + if (err) { free(buf); return NULL; } + + yapf_image_t *img = yapf__load(buf, len, threads); + free(buf); + return img; +} + +yapf_image_t *yapf_load(const char *filename) { + return yapf__load_file(filename, 1); +} + +yapf_image_t *yapf_load_mt(const char *filename, int threads) { + return yapf__load_file(filename, threads); } /* ======================================================================== @@ -498,78 +1421,63 @@ int yapf_save(const char *filename, const yapf_image_t *img) { int ch = (int)img->channels; uint8_t mip_levels = img->mip_levels; + int err = YAPF_OK; - /* Encode each mip level */ - yapf__bw bws[YAPF_MAX_MIPS]; + yapf__out bws[YAPF_MAX_MIPS]; memset(bws, 0, sizeof(bws)); - for (uint8_t m = 0; m < mip_levels; m++) { + for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) { uint32_t mw = img->width >> m; if (mw < 1) mw = 1; uint32_t mh = img->height >> m; if (mh < 1) mh = 1; - const uint8_t *src = (m == 0) - ? img->pixels - : (img->mips ? img->mips[m] : NULL); - if (!src) { - for (uint8_t j = 0; j < m; j++) free(bws[j].buf); - return YAPF_ERR_INVALID; - } + const uint8_t *src = (m == 0) ? img->pixels + : (img->mips ? img->mips[m] : NULL); + if (!src) { err = YAPF_ERR_INVALID; break; } - size_t est = (size_t)mw * mh * (size_t)ch + 256; - if (yapf__bw_init(&bws[m], est) != 0) { - for (uint8_t j = 0; j < m; j++) free(bws[j].buf); - return YAPF_ERR_OOM; - } - if (yapf__encode_level(&bws[m], src, mw, mh, ch) != 0) { - for (uint8_t j = 0; j <= m; j++) free(bws[j].buf); - return YAPF_ERR_OOM; - } + if (yapf__reserve(&bws[m], (size_t)mw * mh * (size_t)ch / 2 + 1024) != 0 || + yapf__encode_level(&bws[m], src, mw, mh, ch) != 0) + err = YAPF_ERR_OOM; } - /* Compute mip index offsets */ + /* Offsets are 32-bit: refuse files that would not fit. */ uint32_t offsets[YAPF_MAX_MIPS]; - uint32_t cur = YAPF__HDR_LEN + (uint32_t)mip_levels * YAPF__MIP_ENTRY; - for (uint8_t m = 0; m < mip_levels; m++) { - offsets[m] = cur; - cur += (uint32_t)bws[m].len; + uint64_t cur = YAPF__HDR_LEN + (uint64_t)mip_levels * YAPF__MIP_ENTRY; + for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) { + offsets[m] = (uint32_t)cur; + cur += bws[m].len; + if (cur > 0xFFFFFFFFull) err = YAPF_ERR_INVALID; } - FILE *f = fopen(filename, "wb"); - if (!f) { - for (uint8_t m = 0; m < mip_levels; m++) free(bws[m].buf); - return YAPF_ERR_IO; + FILE *f = NULL; + if (err == YAPF_OK && !(f = fopen(filename, "wb"))) err = YAPF_ERR_IO; + + if (err == YAPF_OK) { + uint8_t hdr[YAPF__HDR_LEN]; + memset(hdr, 0, sizeof(hdr)); + memcpy(hdr, YAPF__MAGIC, 4); + hdr[4] = YAPF__VERSION; + hdr[5] = img->channels; + hdr[6] = img->gpu_format; + hdr[7] = mip_levels; + hdr[8] = img->flags; + yapf__w32le(hdr + 12, img->width); + yapf__w32le(hdr + 16, img->height); + if (fwrite(hdr, 1, YAPF__HDR_LEN, f) != YAPF__HDR_LEN) err = YAPF_ERR_IO; + + for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) { + uint8_t entry[YAPF__MIP_ENTRY]; + yapf__w32le(entry + 0, offsets[m]); + yapf__w32le(entry + 4, (uint32_t)bws[m].len); + if (fwrite(entry, 1, YAPF__MIP_ENTRY, f) != YAPF__MIP_ENTRY) + err = YAPF_ERR_IO; + } + for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) + if (fwrite(bws[m].buf, 1, bws[m].len, f) != bws[m].len) + err = YAPF_ERR_IO; + if (fclose(f) != 0 && err == YAPF_OK) err = YAPF_ERR_IO; } - /* Header */ - uint8_t hdr[YAPF__HDR_LEN]; - memset(hdr, 0, sizeof(hdr)); - memcpy(hdr, YAPF__MAGIC, 4); - hdr[4] = YAPF__VERSION; - hdr[5] = img->channels; - hdr[6] = img->gpu_format; - hdr[7] = mip_levels; - hdr[8] = img->flags; - yapf__w32le(hdr + 12, img->width); - yapf__w32le(hdr + 16, img->height); - fwrite(hdr, 1, YAPF__HDR_LEN, f); - - /* Mip index */ - for (uint8_t m = 0; m < mip_levels; m++) { - uint8_t entry[YAPF__MIP_ENTRY]; - yapf__w32le(entry + 0, offsets[m]); - yapf__w32le(entry + 4, (uint32_t)bws[m].len); - fwrite(entry, 1, YAPF__MIP_ENTRY, f); - } - - /* Payload */ - int err = YAPF_OK; - for (uint8_t m = 0; m < mip_levels; m++) { - if (fwrite(bws[m].buf, 1, bws[m].len, f) != bws[m].len) - err = YAPF_ERR_IO; - free(bws[m].buf); - } - - fclose(f); + for (uint8_t m = 0; m < mip_levels; m++) free(bws[m].buf); return err; } diff --git a/yapf.h b/yapf.h index ebcac09..463d689 100644 --- a/yapf.h +++ b/yapf.h @@ -20,21 +20,30 @@ * * YAPF — lossless, GPU-ready image format. * - * Space savings come from two cheap mathematical steps — no entropy coder, - * no lookup tables, no opcode dispatch. Decode is straight arithmetic: + * Designed for single-core decode speed. Everything is byte aligned: + * no bit reader, no entropy coder, no context model. Per row the decoder + * dispatches one filter; per 8 samples it reads a 4-bit control code and + * exactly w bytes of payload, unpacked with one 64-bit load and shifts. + * The encoder does the hard work (filter search, colour transform choice, + * optimal control codes); the decoder only follows instructions. * - * 1. YCoCg-R color transform (RGB↔Y/Co/Cg, fully reversible integers) - * 2. 2-D neighbor prediction + per-tile adaptive bit-packing + * 1. Per tile: YCoCg-R or subtract-green colour transform, 8-bit modular + * 2. Per band of 8 rows: one of 12 row filters + * 3. Residuals in groups of 8, each group at its bit width (0–8); + * 4-bit control codes, with codes that repeat the previous width + * 4. Independent 64×64 tiles with a size table → random access and + * optional parallel decode * * The file carries GPU metadata (format hint, mip levels, flags) so a * loader can upload the decoded buffer straight to a texture without any * additional transcoding step. * * ========================================================================= - * FILE FORMAT (version 2) + * FILE FORMAT (version 1) * ========================================================================= * - * All multi-byte integers are little-endian. + * All multi-byte integers are little-endian. All arithmetic on samples + * is mod 256. * * ┌─────────────────────────────────────────────────────────────────────┐ * │ HEADER — 20 bytes │ @@ -42,7 +51,7 @@ * │ Offset │ Size │ Field │ * ├────────┼──────┼────────────────────────────────────────────────────┤ * │ 0x00 │ 4 │ Magic: "YAPF" │ - * │ 0x04 │ 1 │ Version: 2 │ + * │ 0x04 │ 1 │ Version: 1 │ * │ 0x05 │ 1 │ Channels: 1 gray / 2 gray+A / 3 RGB / 4 RGBA │ * │ 0x06 │ 1 │ GPU format hint (YAPF_GPU_* constant) │ * │ 0x07 │ 1 │ Mip levels (1 = base only, max 16) │ @@ -58,53 +67,64 @@ * │ +0 │ 4 │ Byte offset of mip data from start of file (uint32) │ * │ +4 │ 4 │ Byte size of mip data (uint32) │ * └────────┴──────┴────────────────────────────────────────────────────┘ + * Mip m has size max(1, width >> m) × max(1, height >> m). * * ┌─────────────────────────────────────────────────────────────────────┐ - * │ PIXEL DATA — per mip level, referenced by the mip index │ + * │ MIP LEVEL DATA │ * ├─────────────────────────────────────────────────────────────────────┤ - * │ Pixels are divided into 32×32 tiles, stored in row-major order. │ - * │ The last column / row of tiles may be smaller (no padding). │ + * │ Pixels are divided into 64×64 tiles in row-major order; the last │ + * │ column / row of tiles may be smaller (no padding). │ * │ │ - * │ Encoding pipeline (encoder): │ - * │ a) RGB→YCoCg-R (skipped for gray/gray-alpha channels) │ - * │ b) 2-D prediction residuals (see below) │ - * │ c) Determine per-channel bit depth from max |residual| in tile │ - * │ d) Write tile header + packed residuals │ + * │ uint32 tile_size[tile_count] byte size of each tile │ + * │ tile data, back to back │ * │ │ - * │ Tile layout: │ - * │ [bit_depth per channel] 1 byte each, channels bytes total │ - * │ [first pixel, raw] 1 byte per channel │ - * │ [packed residuals] bit_depth[c] bits per sample, │ - * │ stored MSB-first, zero-padded to the │ - * │ next byte boundary at the end of tile │ + * │ TILE │ + * │ 1 byte header │ + * │ 0x80 stored: tile_w × tile_h × channels raw bytes │ + * │ follow, interleaved, in the original channel order │ + * │ otherwise coded: │ + * │ bits 0–3 bit c set = plane c is constant (only bits below │ + * │ the channel count may be set) │ + * │ bit 4 colour transform: 0 = YCoCg-R, 1 = subtract-green │ + * │ (RGB / RGBA only; must be 0 for gray) │ + * │ bits 5–7 must be 0 │ + * │ coded tiles then hold one value byte per constant plane, and a │ + * │ plane payload per other plane, both in plane order. The tile │ + * │ must end exactly where its last payload ends. │ * │ │ - * │ Prediction (lossless 2-D): │ - * │ pixel (0,0) in tile → stored raw, no residual │ - * │ pixel (x,0) → predict = left │ - * │ pixel (0,y) → predict = above │ - * │ pixel (x,y) → predict = left + above − diag │ - * │ residual = (actual − predict) wrapped mod 256, stored signed │ + * │ PLANES │ + * │ gray / gray+A: the channels as stored. │ + * │ RGB(A), YCoCg-R (>> is an arithmetic shift of the value as int8):│ + * │ planes Y, Co, Cg (+ A) │ + * │ forward: o = R − B; t = B + (o >> 1); q = G − t; │ + * │ Y = t + (q >> 1); Co = o + 128; Cg = q + 128 │ + * │ inverse: o = Co − 128; q = Cg − 128; t = Y − (q >> 1); │ + * │ G = q + t; B = t − (o >> 1); R = B + o │ + * │ RGB(A), subtract-green: │ + * │ planes G, R − G + 128, B − G + 128 (+ A) │ * │ │ - * │ Bit packing: │ - * │ bits_needed = bit_depth[c] (0 means all residuals are 0) │ - * │ Each residual is stored as a (bits_needed)-bit two's-complement │ - * │ signed integer, MSB first, packed across byte boundaries. │ - * │ bit_depth 1 → only the sign bit (value is −1 or 0) │ - * │ bit_depth 0 → no bits written; all samples in channel = 0 │ - * │ │ - * │ YCoCg-R forward transform (applied before prediction): │ - * │ Co = R − B │ - * │ t = B + (Co >> 1) │ - * │ Cg = G − t │ - * │ Y = t + (Cg >> 1) │ - * │ Stored channel order: Y, Co, Cg (replaces R, G, B) │ - * │ Alpha is left unchanged. │ - * │ │ - * │ YCoCg-R inverse transform (applied after prediction): │ - * │ t = Y − (Cg >> 1) │ - * │ G = Cg + t │ - * │ B = t − (Co >> 1) │ - * │ R = B + Co │ + * │ PLANE PAYLOAD │ + * │ filter[ceil(tile_h / 8)] one byte per band of 8 rows. │ + * │ L / A / D / AR = left, above, above-left, above-right sample; │ + * │ samples outside the tile read as 0, except that AR in the │ + * │ last column reads as A. │ + * │ 0 Left pred = L │ + * │ 1 Gradient pred = L + A − D │ + * │ 2 Up-right pred = AR │ + * │ 3 Up-average pred = (A + AR) >> 1 (not mod 256) │ + * │ 16 + k−1 Up k pred = sample k rows above, k = 1..8 │ + * │ control nibbles, low nibble of each byte first, covering │ + * │ exactly ceil(tile_w × tile_h / 8) groups of 8 consecutive │ + * │ residuals in raster order: │ + * │ 0–8 one group of width w │ + * │ 9–15 2, 3, 4, 8, 16, 32 or 64 groups at the width of │ + * │ the last 0–8 code in this plane (0 if none yet) │ + * │ an odd nibble count is padded with a zero high nibble │ + * │ group data: w bytes per group, the 8 values packed little- │ + * │ endian, value i in bits [i·w, i·w + w) │ + * │ Residual r = (sample − pred) mod 256 is stored folded: │ + * │ v = (r << 1) ^ (r >> 7) (0, −1, 1, −2, … → 0, 1, 2, 3, …) │ + * │ The last group is padded with zero values. │ * └─────────────────────────────────────────────────────────────────────┘ * * GPU FORMAT HINTS (YAPF_GPU_* values for the gpu_format field) @@ -161,7 +181,7 @@ #define YAPF_MAX_DIM 65535u #define YAPF_MAX_MIPS 16u -#define YAPF_TILE_SIZE 32u +#define YAPF_TILE_SIZE 64u /* ── Image descriptor ────────────────────────────────────────────────── */ @@ -189,6 +209,21 @@ extern "C" { */ yapf_image_t *yapf_load(const char *filename); +/* + * yapf_load_memory — decode a .yapf file already in memory (e.g. from a + * game archive). Same ownership and failure rules as yapf_load(). + */ +yapf_image_t *yapf_load_memory(const void *buffer, size_t size); + +/* + * yapf_load_mt / yapf_load_memory_mt — same as above, but decode tiles on + * up to `threads` threads (0 = one per CPU core). Output is identical. + * Without thread support (YAPF_NO_THREADS, or an unknown platform) these + * decode on the calling thread. + */ +yapf_image_t *yapf_load_mt(const char *filename, int threads); +yapf_image_t *yapf_load_memory_mt(const void *buffer, size_t size, int threads); + /* * yapf_save — encode an image to disk. * img->mip_levels controls how many levels are written; set to 1 to write From d7f494a0c91a3c6acc4dd0cb7531ef6c0f068a65 Mon Sep 17 00:00:00 2001 From: Nexoniarz Date: Sat, 3 Oct 2026 14:39:56 +0200 Subject: [PATCH 2/2] Language bindings, KDE thumbnails, yapf compare, in-memory encoding - yapf_encode() / yapf_free_buffer(): encode to memory; yapf_save uses it - yapf.c also compiles as C++ - `yapf compare`: size and decode speed of the original format vs YAPF, with a lossless check - bindings with tested examples: C++ (header-only), C# (P/Invoke), Rust (crate, builds the C code), Go (cgo), Java (FFM API, Java 22+), Zig (@cImport), LuaJIT (FFI), TypeScript types for the JS port; Python now encodes in memory - docs/LANGUAGES.md and examples/ with measured results - KDE Dolphin thumbnail plugin (KF6, CMake) - README: honest status for Photoshop / VS Code / Figma (not run in the apps yet) and macOS (cross-compiled only), platform table Co-Authored-By: Claude Opus 5.5 --- .gitignore | 8 ++ README.md | 104 ++++++++++------- bindings/cpp/README.md | 18 +++ bindings/cpp/example.cpp | 33 ++++++ bindings/cpp/yapf.hpp | 98 ++++++++++++++++ bindings/csharp/Example/Example.csproj | 11 ++ bindings/csharp/Example/Program.cs | 25 ++++ bindings/csharp/README.md | 19 +++ bindings/csharp/Yapf.cs | 121 +++++++++++++++++++ bindings/go/README.md | 17 +++ bindings/go/example/main.go | 47 ++++++++ bindings/go/go.mod | 3 + bindings/go/yapf.go | 111 ++++++++++++++++++ bindings/go/yapf_c.c | 2 + bindings/go/yapf_test.go | 46 ++++++++ bindings/java/Example.java | 29 +++++ bindings/java/README.md | 16 +++ bindings/java/Yapf.java | 121 +++++++++++++++++++ bindings/js/README.md | 4 +- bindings/js/example.mjs | 19 +++ bindings/js/package.json | 3 +- bindings/js/yapf.d.ts | 25 ++++ bindings/lua/README.md | 13 +++ bindings/lua/example.lua | 17 +++ bindings/lua/yapf.lua | 71 +++++++++++ bindings/python/README.md | 2 + bindings/python/example.py | 33 ++++++ bindings/python/yapf.py | 38 +++--- bindings/rust/Cargo.toml | 12 ++ bindings/rust/README.md | 18 +++ bindings/rust/build.rs | 12 ++ bindings/rust/examples/demo.rs | 26 +++++ bindings/rust/src/lib.rs | 156 +++++++++++++++++++++++++ bindings/zig/README.md | 13 +++ bindings/zig/example.zig | 33 ++++++ build.bat | 2 + docs/LANGUAGES.md | 144 +++++++++++++++++++++++ examples/README.md | 62 ++++++++++ examples/c/example.c | 61 ++++++++++ extensions/blender/README.md | 4 +- extensions/figma/README.md | 7 ++ extensions/kde/CMakeLists.txt | 26 +++++ extensions/kde/README.md | 50 ++++++++ extensions/kde/kde-shell.nix | 6 + extensions/kde/yapfthumbnail.cpp | 79 +++++++++++++ extensions/kde/yapfthumbnail.json | 7 ++ extensions/vscode/README.md | 9 +- tests/test_yapf.c | 42 +++++-- tools/build_release.sh | 2 + tools/yapf_cli.c | 109 +++++++++++++++++ yapf.c | 81 ++++++++----- yapf.h | 11 ++ 52 files changed, 1923 insertions(+), 103 deletions(-) create mode 100644 bindings/cpp/README.md create mode 100644 bindings/cpp/example.cpp create mode 100644 bindings/cpp/yapf.hpp create mode 100644 bindings/csharp/Example/Example.csproj create mode 100644 bindings/csharp/Example/Program.cs create mode 100644 bindings/csharp/README.md create mode 100644 bindings/csharp/Yapf.cs create mode 100644 bindings/go/README.md create mode 100644 bindings/go/example/main.go create mode 100644 bindings/go/go.mod create mode 100644 bindings/go/yapf.go create mode 100644 bindings/go/yapf_c.c create mode 100644 bindings/go/yapf_test.go create mode 100644 bindings/java/Example.java create mode 100644 bindings/java/README.md create mode 100644 bindings/java/Yapf.java create mode 100644 bindings/js/example.mjs create mode 100644 bindings/js/yapf.d.ts create mode 100644 bindings/lua/README.md create mode 100644 bindings/lua/example.lua create mode 100644 bindings/lua/yapf.lua create mode 100644 bindings/python/example.py create mode 100644 bindings/rust/Cargo.toml create mode 100644 bindings/rust/README.md create mode 100644 bindings/rust/build.rs create mode 100644 bindings/rust/examples/demo.rs create mode 100644 bindings/rust/src/lib.rs create mode 100644 bindings/zig/README.md create mode 100644 bindings/zig/example.zig create mode 100644 docs/LANGUAGES.md create mode 100644 examples/README.md create mode 100644 examples/c/example.c create mode 100644 extensions/kde/CMakeLists.txt create mode 100644 extensions/kde/README.md create mode 100644 extensions/kde/kde-shell.nix create mode 100644 extensions/kde/yapfthumbnail.cpp create mode 100644 extensions/kde/yapfthumbnail.json diff --git a/.gitignore b/.gitignore index d93af31..7de58a5 100644 --- a/.gitignore +++ b/.gitignore @@ -15,3 +15,11 @@ test_yapf_tmp.yapf node_modules/ __pycache__/ *.pyc + +# Rust +/bindings/rust/target/ +/bindings/rust/Cargo.lock +# C# / Go / Java build outputs +bin/ +obj/ +*.class diff --git a/README.md b/README.md index e9bf4a0..2480e4d 100644 --- a/README.md +++ b/README.md @@ -8,8 +8,8 @@ model: every structure is byte aligned, and the decoder runs short loops that modern CPUs execute at memory speed — with SSE / NEON where available. The encoder does the hard work, so the decoder only follows instructions. -- **6–10× faster decoding than PNG** (stb_image), single-threaded; a 4K image - decodes in about 6 ms (5.6 GB/s). Multithreaded decoding is optional. +- **5–12× faster decoding than PNG** (stb_image), single-threaded; a 4K image + decodes in about 6–7 ms (around 5 GB/s). Multithreaded decoding is optional. - **Smaller than PNG** on UI, icons and screenshots; 20–45 % larger on photos and natural textures — the price of having no entropy coder. - **Made for games and tools too**: mipmaps, GPU format hint, sRGB / @@ -25,22 +25,32 @@ This is **YAPF version 1**, the first stable format. | Where | What you get | Guide | Status | | :---- | :----------- | :---- | :----- | -| **C / C++** | `yapf.h` + `yapf.c`, drop into any project | [below](#c--c) | ✅ tested on Linux x64, Windows x64, Linux ARM64 | -| **Command line** | convert PNG/JPG/TGA/BMP/PSD/GIF ↔ YAPF, `info`, `bench`, `thumbnail` | [below](#command-line) | ✅ tested on Linux, Windows | -| **GIMP 3** | open, export (with mipmaps) | [extensions/gimp](extensions/gimp) | ✅ tested in GIMP 3.0.8 | +| **C / C++** | `yapf.h` + `yapf.c`, drop into any project (compiles as C or C++) | [below](#c--c) | ✅ tested — see [platforms](#platforms) | +| **Command line** | convert PNG/JPG/TGA/BMP/PSD/GIF ↔ YAPF, `compare`, `info`, `bench`, `thumbnail` | [below](#command-line) | ✅ tested on Linux and Windows | +| **Your language** | Python, C#, Rust, Go, Java, Zig, Lua, JavaScript / TypeScript, … | [docs/LANGUAGES.md](docs/LANGUAGES.md) | ✅ every binding tested | +| **GIMP 3** | open, export (flattened, optional mipmaps) | [extensions/gimp](extensions/gimp) | ✅ tested in GIMP 3.0.8 | | **Blender 4.2+** | import, export, drag & drop, image as plane | [extensions/blender](extensions/blender) | ✅ tested in Blender 5.1 | -| **VS Code** | image preview with zoom and mip picker | [extensions/vscode](extensions/vscode) | ✅ packaged, preview tested in Chromium | -| **Figma** | import `.yapf`, export selection as `.yapf` | [extensions/figma](extensions/figma) | ✅ plugin UI tested in Chromium | -| **Photoshop** | File → Scripts → Open YAPF / Save as YAPF | [extensions/photoshop](extensions/photoshop) | ⚠️ not yet tested inside Photoshop | -| **Linux desktop** | file type + thumbnails in GNOME Files, Nemo, Caja, Thunar | [extensions/linux](extensions/linux) | ⚠️ thumbnailer tested, not in a file manager | -| **Python** | `yapf.load()` / `yapf.save()` | [bindings/python](bindings/python) | ✅ tested | -| **JavaScript** | browsers, Node.js, workers | [bindings/js](bindings/js) | ✅ tested, byte-identical to C | +| **KDE Dolphin** | thumbnails for `.yapf` files | [extensions/kde](extensions/kde) | ⚠️ tested through KDE's plugin loader (KF6); not yet inside a running Dolphin | +| **Linux desktop** | file type + thumbnails in GNOME Files, Nemo, Caja, Thunar | [extensions/linux](extensions/linux) | ⚠️ thumbnail command tested; not yet in a file manager | +| **VS Code** | image preview with zoom and mip picker | [extensions/vscode](extensions/vscode) | ⚠️ packaged; preview code tested in a browser engine (Chromium); **not yet run inside VS Code** | +| **Figma** | import `.yapf`, export selection as `.yapf` | [extensions/figma](extensions/figma) | ⚠️ plugin UI tested in a browser engine with a simulated Figma; **not yet run inside Figma** | +| **Photoshop** | File → Scripts → Open YAPF / Save as YAPF | [extensions/photoshop](extensions/photoshop) | ⚠️ **syntax-checked only; not yet run in Photoshop** | Pre-built downloads (command-line tool and libraries for Windows, macOS and Linux, plus the plugin packages) are published on the [Releases](https://github.com/Nexoniarz/YAPF/releases) page; everything can also be built from source as described below. +### Platforms + +| Platform | Status | +| :------- | :----- | +| Linux x64 | ✅ built and fully tested | +| Windows x64 | ✅ built and fully tested (MinGW and Zig builds, run under Wine) | +| Linux ARM64 | ✅ built and fully tested (NEON, run under qemu) | +| Windows ARM64 | ⚠️ built only, not run | +| macOS x64 / ARM64 | ⚠️ **cross-compiled only, not yet run on a Mac** | + --- ## Command line @@ -53,6 +63,7 @@ yapf texture.yapf mip2.png --mip 2 # export one mip level yapf info photo.yapf # size, channels, mips, flags, ratio yapf bench photo.yapf # decode speed, 1 thread and all cores yapf thumbnail photo.yapf thumb.png 256 +yapf compare photo.png ui.png # size and decode speed vs the original format ``` Options when writing `.yapf`: `--mips`, `--linear` (pixels are not sRGB), @@ -62,25 +73,24 @@ Options when writing `.yapf`: `--mips`, `--linear` (pixels are not sRGB), ## Benchmark -Single core, AMD Ryzen 7 5700G, `gcc -O2`. PNG is decoded with `stb_image`; -the PNG files are the ones shipped with the respective projects. "Plain C" -is the library built with `-DYAPF_NO_SIMD`; the default build uses SSE / NEON. +`yapf compare` on a mixed set of images: one binary (`make`, gcc -O2) for +both formats, PNG decoded with stb_image, one core, AMD Ryzen 7 5700G. +Every result is checked to decode back to the exact original pixels. -| Image | Raw | PNG | **YAPF** | vs PNG | PNG decode | YAPF, plain C | **YAPF (default)** | -| :--------------------- | ------: | -----: | -------: | -----: | ---------: | ------------: | -----------------: | -| Wallpaper 3840×2160 | 32400 K | 1812 K | 2267 K | 125 % | 44.7 ms¹ | 10.3 ms | **5.9 ms** (5.6 GB/s) | -| UI screenshot 261×430 | 438 K | 176 K | 143 K | 81 % | 1.6 ms | 0.31 ms | **0.17 ms** | -| UI screenshot 305×529 | 630 K | 201 K | 185 K | 92 % | 2.1 ms | 0.45 ms | **0.24 ms** | -| App icon 128×128 | 64 K | 24 K | 23 K | 93 % | 0.26 ms | 0.06 ms | **0.03 ms** | -| Rock texture 512×512 | 768 K | 182 K | 221 K | 121 % | 1.4 ms | 0.35 ms | **0.16 ms** | -| Wood texture 512×512 | 768 K | 185 K | 234 K | 126 % | 1.4 ms | 0.46 ms | **0.16 ms** | -| Canvas texture 512×512 | 768 K | 171 K | 245 K | 143 % | 1.0 ms | 0.35 ms | **0.15 ms** | -| Pixel art 512×512 | 1024 K | — | 66 K | — | — | 0.74 ms | **0.32 ms** | -| Random noise 512×512 | 1024 K | — | 1024 K | — | — | 0.1 ms | 0.1 ms | +| Image | PNG | YAPF | YAPF / PNG size | PNG decode | YAPF decode | Faster | +| :---- | --: | ---: | --------------: | ---------: | ----------: | -----: | +| UI screenshot 261×430 | 176.6 KB | 143.7 KB | **81 %** | 2.05 ms | 0.17 ms | **12×** | +| UI screenshot 305×529 | 201.5 KB | 185.0 KB | **92 %** | 2.55 ms | 0.24 ms | **11×** | +| App icon 128×128 | 24.9 KB | 23.1 KB | **93 %** | 0.30 ms | 0.026 ms | **11×** | +| Wallpaper 3840×2160 | 1.77 MB | 2.21 MB | 125 % | 73.7 ms | 6.8 ms | **11×** | +| Rock texture 512×512 | 182.6 KB | 221.8 KB | 121 % | 1.64 ms | 0.17 ms | **10×** | +| Wood texture 512×512 | 185.8 KB | 234.1 KB | 126 % | 1.64 ms | 0.17 ms | **10×** | +| Canvas texture 512×512 | 171.5 KB | 245.6 KB | 143 % | 1.36 ms | 0.16 ms | **9×** | +| YAPF logo 512×512 | 60.5 KB | 89.3 KB | 148 % | 1.25 ms | 0.25 ms | **5×** | -¹ PNG decode of the wallpaper measured between 44.7 and 66 ms depending on -the harness; the table uses the fastest. `-O2` and `-O3` builds of YAPF -perform the same. +Smaller than PNG on UI-style images, 20–50 % larger on photos and textures +(the price of having no entropy coder), and 5–12× faster to decode. More +in [examples/](examples) — and run `yapf compare` on your own images. --- @@ -176,6 +186,13 @@ Encodes an image to disk. Writes `img->mip_levels` levels; `img->mips` may be `NULL` when `mip_levels == 1`. Returns `YAPF_OK` (0) on success or a negative `YAPF_ERR_*` code on failure. +```c +int yapf_encode(const yapf_image_t *img, void **out_data, size_t *out_size); +void yapf_free_buffer(void *data); +``` +Encodes into memory: `*out_data` receives a complete `.yapf` file of +`*out_size` bytes, to be released with `yapf_free_buffer()`. + ```c void yapf_free(yapf_image_t *img); ``` @@ -256,22 +273,27 @@ TILE (64×64) header byte: stored raw, or constant-plane mask + colour tr ## Testing ```sh -make test # C suite: 784 checks, plus JavaScript if node is installed +make test # C suite: 786 checks, plus JavaScript if node is installed node bindings/js/test.js # JavaScript: round-trips, sample file, corrupt input ``` -What has been verified for this release: +What has been verified: - exact round-trips for 1–4 channels, odd sizes from 1×1 up, mip chains, - single- and multi-threaded decoding; -- the same tests pass on **Linux x64**, **Windows x64** (MinGW, run under - Wine), **Linux ARM64** (NEON, run under qemu), with and without SIMD, and - under AddressSanitizer, UndefinedBehaviorSanitizer and ThreadSanitizer; + single- and multi-threaded decoding, in-memory and file encoding; +- the same tests pass on Linux x64, Windows x64 (run under Wine) and Linux + ARM64 (NEON, run under qemu), with and without SIMD, compiled as C and as + C++, and under AddressSanitizer, UndefinedBehaviorSanitizer and + ThreadSanitizer; - thousands of corrupted and truncated files are rejected without crashing; -- the JavaScript and Python versions produce byte-identical files to C; -- the GIMP plugin and the Blender extension in the real applications. +- every language binding runs its example; JavaScript, Python, C++, C#, + Rust, Go, Java and Lua re-encode the sample to byte-identical output; +- the GIMP plugin in GIMP 3.0.8 and the Blender extension in Blender 5.1; +- the KDE thumbnailer through KDE's own plugin loader. -The macOS builds are cross-compiled and not yet run on a Mac. +**Not yet verified:** running on a Mac (macOS builds are cross-compiled +only), Windows on ARM, and running inside Photoshop, VS Code, Figma or a +live Dolphin / GNOME file manager. --- @@ -282,14 +304,16 @@ yapf.h, yapf.c the library (format spec in yapf.h) tools/yapf_cli.c the `yapf` command-line tool tools/build_release.sh builds every download for all platforms tests/test_yapf.c C test suite -bindings/js/ JavaScript encoder / decoder -bindings/python/ Python module +examples/ speed and size comparison, C example +docs/LANGUAGES.md using YAPF from other languages +bindings/ cpp, python, csharp, rust, go, java, zig, lua, js extensions/gimp/ GIMP 3 plugin extensions/blender/ Blender extension +extensions/kde/ Dolphin thumbnail plugin (KF6) +extensions/linux/ MIME type and freedesktop thumbnailer extensions/vscode/ VS Code preview extensions/figma/ Figma plugin extensions/photoshop/ Photoshop scripts -extensions/linux/ MIME type and thumbnailer third_party/ stb_image / stb_image_write (public domain, CLI only) other/YAPF.YAPF sample image (the YAPF logo) ``` diff --git a/bindings/cpp/README.md b/bindings/cpp/README.md new file mode 100644 index 0000000..e6d9a77 --- /dev/null +++ b/bindings/cpp/README.md @@ -0,0 +1,18 @@ +# YAPF for C++ + +`yapf.hpp` is a header-only C++17 wrapper: RAII, exceptions, `std::vector`. +Add `../../yapf.c` to your project and include the header. + +```cpp +#include "yapf.hpp" + +yapf::Image img = yapf::Image::load("texture.yapf"); // or Image::decode(bytes) +upload(img.width(), img.height(), img.channels(), img.pixels()); + +std::vector file = yapf::Image(256, 256, 4, rgba).encode(); +``` + +`yapf.c` compiles as C or as C++, so you can simply add it to a C++ project +(Visual Studio, CMake, …). + +Example: `g++ -O2 -std=c++17 example.cpp -x c++ ../../yapf.c -o example -pthread && ./example` diff --git a/bindings/cpp/example.cpp b/bindings/cpp/example.cpp new file mode 100644 index 0000000..633114e --- /dev/null +++ b/bindings/cpp/example.cpp @@ -0,0 +1,33 @@ +// g++ -O2 -std=c++17 example.cpp ../../yapf.c -o example -pthread && ./example ../../other/YAPF.YAPF +#include "yapf.hpp" + +#include +#include +#include +#include +#include + +int main(int argc, char **argv) { + const char *path = argc > 1 ? argv[1] : "../../other/YAPF.YAPF"; + std::ifstream f(path, std::ios::binary); + std::vector bytes((std::istreambuf_iterator(f)), {}); + + yapf::Image img = yapf::Image::decode(bytes); + double ms = 1e9; // best of 10, as in a running program + for (int i = 0; i < 10; i++) { + auto t0 = std::chrono::steady_clock::now(); + img = yapf::Image::decode(bytes); + ms = std::min(ms, std::chrono::duration(std::chrono::steady_clock::now() - t0).count()); + } + std::printf("%s: %ux%u, %u channels, decoded in %.2f ms, file is %.1f%% of raw\n", path, + img.width(), img.height(), img.channels(), ms, 100.0 * bytes.size() / img.size()); + + std::vector again = img.encode(); + std::printf("re-encoded: %zu bytes, identical to the file: %s\n", again.size(), + again == bytes ? "yes" : "no"); + + std::vector gray(64 * 64); + for (size_t i = 0; i < gray.size(); i++) gray[i] = uint8_t(i % 64 * 4); + yapf::Image(64, 64, 1, gray.data()).save("gray.yapf"); + std::printf("wrote gray.yapf\n"); +} diff --git a/bindings/cpp/yapf.hpp b/bindings/cpp/yapf.hpp new file mode 100644 index 0000000..ffd09e5 --- /dev/null +++ b/bindings/cpp/yapf.hpp @@ -0,0 +1,98 @@ +// yapf.hpp — header-only C++17 wrapper for YAPF. Copyright 2026 Nexoniarz — Apache 2.0. +// +// #include "yapf.hpp" // and compile/link ../../yapf.c +// yapf::Image img = yapf::Image::load("texture.yapf"); +// img.width(); img.height(); img.channels(); img.pixels(); // span of bytes +// std::vector file = img.encode(); +// yapf::Image mine(256, 256, 4, my_rgba_bytes); mine.save("out.yapf"); +#pragma once + +extern "C" { +#include "../../yapf.h" +} + +#include +#include +#include +#include +#include +#include + +namespace yapf { + +struct Error : std::runtime_error { + using std::runtime_error::runtime_error; +}; + +class Image { +public: + /* Decode a file or a memory buffer; threads = 0 uses every core, 1 the calling thread. */ + static Image load(const std::string &path, int threads = 1) { + yapf_image_t *p = yapf_load_mt(path.c_str(), threads); + if (!p) throw Error("YAPF: cannot load " + path); + return Image(p); + } + static Image decode(const void *data, size_t size, int threads = 1) { + yapf_image_t *p = yapf_load_memory_mt(data, size, threads); + if (!p) throw Error("YAPF: invalid or corrupt data"); + return Image(p); + } + static Image decode(const std::vector &bytes, int threads = 1) { + return decode(bytes.data(), bytes.size(), threads); + } + + /* A new image from your pixels (copied). channels: 1 gray, 2 gray+A, 3 RGB, 4 RGBA. */ + Image(uint32_t width, uint32_t height, uint8_t channels, const uint8_t *pixels, + uint8_t flags = YAPF_FLAG_SRGB) + : own_(pixels, pixels + size_t(width) * height * channels) { + std::memset(&view_, 0, sizeof(view_)); + view_.width = width; view_.height = height; view_.channels = channels; + view_.flags = flags; view_.mip_levels = 1; + view_.gpu_format = channels == 4 ? YAPF_GPU_SRGB8_A8 : channels == 3 ? YAPF_GPU_SRGB8 + : channels == 2 ? YAPF_GPU_RG8 : YAPF_GPU_R8; + view_.pixels = own_.data(); + } + + Image(Image &&o) noexcept : img_(std::exchange(o.img_, nullptr)), own_(std::move(o.own_)), view_(o.view_) { + if (!img_) view_.pixels = own_.data(); + } + Image &operator=(Image &&o) noexcept { + if (this != &o) { yapf_free(img_); img_ = std::exchange(o.img_, nullptr); own_ = std::move(o.own_); + view_ = o.view_; if (!img_) view_.pixels = own_.data(); } + return *this; + } + Image(const Image &) = delete; + Image &operator=(const Image &) = delete; + ~Image() { yapf_free(img_); } + + uint32_t width() const { return raw().width; } + uint32_t height() const { return raw().height; } + uint8_t channels() const { return raw().channels; } + uint8_t flags() const { return raw().flags; } + uint8_t mipLevels() const { return raw().mip_levels; } + const uint8_t *pixels() const { return raw().pixels; } + size_t size() const { return size_t(width()) * height() * channels(); } + const uint8_t *mip(int level) const { return raw().mips ? raw().mips[level] : raw().pixels; } + + std::vector encode() const { + void *data = nullptr; + size_t n = 0; + if (yapf_encode(&raw(), &data, &n) != YAPF_OK) throw Error("YAPF: encoding failed"); + std::vector out(static_cast(data), static_cast(data) + n); + yapf_free_buffer(data); + return out; + } + void save(const std::string &path) const { + if (yapf_save(path.c_str(), &raw()) != YAPF_OK) throw Error("YAPF: cannot write " + path); + } + + const yapf_image_t &raw() const { return img_ ? *img_ : view_; } + +private: + explicit Image(yapf_image_t *p) : img_(p) { std::memset(&view_, 0, sizeof(view_)); } + yapf_image_t *img_ = nullptr; // decoded by the library + std::vector own_; // or pixels we own + yapf_image_t view_{}; +}; + +} // namespace yapf diff --git a/bindings/csharp/Example/Example.csproj b/bindings/csharp/Example/Example.csproj new file mode 100644 index 0000000..f0d0e5a --- /dev/null +++ b/bindings/csharp/Example/Example.csproj @@ -0,0 +1,11 @@ + + + Exe + net8.0 + enable + enable + + + + + diff --git a/bindings/csharp/Example/Program.cs b/bindings/csharp/Example/Program.cs new file mode 100644 index 0000000..d6f1bcc --- /dev/null +++ b/bindings/csharp/Example/Program.cs @@ -0,0 +1,25 @@ +// dotnet run -- ../../../other/YAPF.YAPF (with yapf.dll / libyapf.so on the library path) +using System.Diagnostics; +using Yapf; + +string path = args.Length > 0 ? args[0] : "../../../other/YAPF.YAPF"; +byte[] file = File.ReadAllBytes(path); + +var img = YapfImage.Decode(file); // first call also loads the library / JIT +double ms = double.MaxValue; // best of 10, as in a running program +for (int i = 0; i < 10; i++) +{ + var sw = Stopwatch.StartNew(); + img = YapfImage.Decode(file); + ms = Math.Min(ms, sw.Elapsed.TotalMilliseconds); +} +Console.WriteLine($"{path}: {img.Width}x{img.Height}, {img.Channels} channels, decoded in {ms:F2} ms"); +Console.WriteLine($"file is {file.Length} bytes, {100.0 * file.Length / img.Pixels.Length:F1}% of the raw pixels"); + +byte[] again = img.Encode(); +Console.WriteLine($"re-encoded: {again.Length} bytes, identical: {again.AsSpan().SequenceEqual(file)}"); + +var rgba = new byte[64 * 64 * 4]; +for (int i = 0; i < rgba.Length; i += 4) { rgba[i] = (byte)(i / 4 % 64 * 4); rgba[i + 1] = 100; rgba[i + 2] = 200; rgba[i + 3] = 255; } +new YapfImage(64, 64, 4, rgba).Save("csharp.yapf"); +Console.WriteLine("wrote csharp.yapf"); diff --git a/bindings/csharp/README.md b/bindings/csharp/README.md new file mode 100644 index 0000000..0aeb9dc --- /dev/null +++ b/bindings/csharp/README.md @@ -0,0 +1,19 @@ +# YAPF for C# / .NET + +`Yapf.cs` is a single-file P/Invoke wrapper for .NET 6 or later (tested with +.NET 8). It should also work in Unity with the native library in +`Assets/Plugins`, but that is not tested yet. Add it to your project and put +the native library next to your program: `yapf.dll` (Windows), +`libyapf.so` (Linux) or `libyapf.dylib` (macOS) — from the +[Releases](https://github.com/Nexoniarz/YAPF/releases) or `make`. + +```csharp +using Yapf; + +var img = YapfImage.Load("texture.yapf"); // or YapfImage.Decode(bytes) +byte[] pixels = img.Pixels; // Width * Height * Channels +new YapfImage(256, 256, 4, rgba).Save("out.yapf"); // or .Encode() → byte[] +``` + +Example: `cd Example && dotnet run -- ../../../other/YAPF.YAPF` +(Linux: `LD_LIBRARY_PATH=../../../build dotnet run …` after `make`). diff --git a/bindings/csharp/Yapf.cs b/bindings/csharp/Yapf.cs new file mode 100644 index 0000000..64d23ef --- /dev/null +++ b/bindings/csharp/Yapf.cs @@ -0,0 +1,121 @@ +// Yapf.cs — C# bindings for YAPF (P/Invoke). Copyright 2026 Nexoniarz — Apache 2.0. +// +// Needs the native library next to your program or on the library path: +// yapf.dll (Windows), libyapf.so (Linux), libyapf.dylib (macOS). +// +// var img = YapfImage.Load("texture.yapf"); +// byte[] pixels = img.Pixels; // Width * Height * Channels, rows top to bottom +// byte[] file = new YapfImage(256, 256, 4, myRgba).Encode(); +using System; +using System.IO; +using System.Runtime.InteropServices; + +namespace Yapf +{ + public sealed class YapfException : Exception + { + public YapfException(string message) : base(message) { } + } + + public sealed class YapfImage + { + public const byte FlagPremultipliedAlpha = 1, FlagSrgb = 2; + + public int Width { get; } + public int Height { get; } + public int Channels { get; } // 1 gray, 2 gray+alpha, 3 RGB, 4 RGBA + public byte Flags { get; } + public byte GpuFormat { get; } + public byte[] Pixels => Mips[0]; + public byte[][] Mips { get; } // Mips[0] is the full-size image + + public YapfImage(int width, int height, int channels, byte[] pixels, + byte flags = FlagSrgb, byte[][]? mips = null) + { + if (pixels.Length != width * height * channels) + throw new ArgumentException("pixels must hold width * height * channels bytes"); + Width = width; Height = height; Channels = channels; Flags = flags; + GpuFormat = channels switch { 4 => (byte)4, 3 => (byte)5, 2 => (byte)2, _ => (byte)3 }; + Mips = mips ?? new[] { pixels }; + Mips[0] = pixels; + } + + private YapfImage(Native n) + { + Width = (int)n.width; Height = (int)n.height; Channels = n.channels; + Flags = n.flags; GpuFormat = n.gpu_format; + Mips = new byte[n.mip_levels][]; + for (int m = 0; m < n.mip_levels; m++) + { + int w = Math.Max(1, Width >> m), h = Math.Max(1, Height >> m); + Mips[m] = new byte[w * h * Channels]; + Marshal.Copy(Marshal.ReadIntPtr(n.mips, m * IntPtr.Size), Mips[m], 0, Mips[m].Length); + } + } + + ///

Decode YAPF bytes. threads: 1 = calling thread, 0 = all cores. + public static YapfImage Decode(byte[] data, int threads = 1) + { + IntPtr p = NativeMethods.yapf_load_memory_mt(data, (UIntPtr)data.Length, threads); + if (p == IntPtr.Zero) throw new YapfException("not a valid YAPF file"); + try { return new YapfImage(Marshal.PtrToStructure(p)); } + finally { NativeMethods.yapf_free(p); } + } + + public static YapfImage Load(string path, int threads = 1) => Decode(File.ReadAllBytes(path), threads); + + /// Encode to the bytes of a .yapf file. + public byte[] Encode() + { + var handles = new GCHandle[Mips.Length]; + IntPtr table = Marshal.AllocHGlobal(IntPtr.Size * Mips.Length); + try + { + for (int m = 0; m < Mips.Length; m++) + { + handles[m] = GCHandle.Alloc(Mips[m], GCHandleType.Pinned); + Marshal.WriteIntPtr(table, m * IntPtr.Size, handles[m].AddrOfPinnedObject()); + } + var n = new Native + { + width = (uint)Width, height = (uint)Height, channels = (byte)Channels, + gpu_format = GpuFormat, flags = Flags, mip_levels = (byte)Mips.Length, + pixels = handles[0].AddrOfPinnedObject(), mips = table, + }; + int rc = NativeMethods.yapf_encode(ref n, out IntPtr data, out UIntPtr size); + if (rc != 0) throw new YapfException($"yapf_encode failed ({rc})"); + try + { + var bytes = new byte[(int)size]; + Marshal.Copy(data, bytes, 0, bytes.Length); + return bytes; + } + finally { NativeMethods.yapf_free_buffer(data); } + } + finally + { + foreach (var h in handles) if (h.IsAllocated) h.Free(); + Marshal.FreeHGlobal(table); + } + } + + public void Save(string path) => File.WriteAllBytes(path, Encode()); + + [StructLayout(LayoutKind.Sequential)] + private struct Native + { + public uint width, height; + public byte channels, gpu_format, flags, mip_levels; + public IntPtr pixels, mips; + } + + private static class NativeMethods + { + private const string Lib = "yapf"; + [DllImport(Lib)] public static extern IntPtr yapf_load_memory_mt(byte[] data, UIntPtr size, int threads); + [DllImport(Lib)] public static extern void yapf_free(IntPtr img); + [DllImport(Lib)] public static extern int yapf_encode(ref Native img, out IntPtr data, out UIntPtr size); + [DllImport(Lib)] public static extern void yapf_free_buffer(IntPtr data); + } + } +} diff --git a/bindings/go/README.md b/bindings/go/README.md new file mode 100644 index 0000000..f889865 --- /dev/null +++ b/bindings/go/README.md @@ -0,0 +1,17 @@ +# YAPF for Go + +A cgo package that builds the C library itself (needs a C compiler, as any +cgo package). + +```sh +go get github.com/Nexoniarz/YAPF/bindings/go +``` + +```go +import yapf "github.com/Nexoniarz/YAPF/bindings/go" + +img, err := yapf.Load("texture.yapf") // or yapf.Decode(bytes) +data, err := yapf.Encode(&yapf.Image{Width: 256, Height: 256, Channels: 4, Pixels: rgba}) +``` + +Example: `go run ./example ../../other/YAPF.YAPF` · tests: `go test`. diff --git a/bindings/go/example/main.go b/bindings/go/example/main.go new file mode 100644 index 0000000..5fac6ea --- /dev/null +++ b/bindings/go/example/main.go @@ -0,0 +1,47 @@ +// go run ./example ../../other/YAPF.YAPF +package main + +import ( + "bytes" + "fmt" + "os" + "time" + + yapf "github.com/Nexoniarz/YAPF/bindings/go" +) + +func main() { + path := "../../other/YAPF.YAPF" + if len(os.Args) > 1 { + path = os.Args[1] + } + file, err := os.ReadFile(path) + if err != nil { + panic(err) + } + img, err := yapf.Decode(file) + if err != nil { + panic(err) + } + best := time.Hour // best of 10, as in a running program + for i := 0; i < 10; i++ { + t := time.Now() + img, _ = yapf.Decode(file) + best = min(best, time.Since(t)) + } + fmt.Printf("%s: %dx%d, %d channels, decoded in %.2f ms\n", path, img.Width, img.Height, + img.Channels, float64(best.Microseconds())/1000) + fmt.Printf("file is %d bytes, %.1f%% of the raw pixels\n", len(file), 100*float64(len(file))/float64(len(img.Pixels))) + + again, _ := yapf.Encode(img) + fmt.Printf("re-encoded: %d bytes, identical: %v\n", len(again), bytes.Equal(again, file)) + + px := make([]byte, 128*128) + for i := range px { + px[i] = byte(i % 128 * 2) + } + if err := yapf.Save("go.yapf", &yapf.Image{Width: 128, Height: 128, Channels: 1, Flags: yapf.FlagSRGB, Pixels: px}); err != nil { + panic(err) + } + fmt.Println("wrote go.yapf") +} diff --git a/bindings/go/go.mod b/bindings/go/go.mod new file mode 100644 index 0000000..13f58b7 --- /dev/null +++ b/bindings/go/go.mod @@ -0,0 +1,3 @@ +module github.com/Nexoniarz/YAPF/bindings/go + +go 1.21 diff --git a/bindings/go/yapf.go b/bindings/go/yapf.go new file mode 100644 index 0000000..5869d5b --- /dev/null +++ b/bindings/go/yapf.go @@ -0,0 +1,111 @@ +// Package yapf reads and writes YAPF lossless images (cgo bindings to the C library). +// +// img, err := yapf.Load("texture.yapf") +// data, err := yapf.Encode(&yapf.Image{Width: 256, Height: 256, Channels: 4, Pixels: rgba}) +package yapf + +/* +#cgo CFLAGS: -O2 -std=c99 -I${SRCDIR}/../.. +#cgo !windows LDFLAGS: -lm -lpthread +#include +#include "yapf.h" +*/ +import "C" + +import ( + "errors" + "os" + "unsafe" +) + +const ( + FlagPremultAlpha = 1 + FlagSRGB = 2 +) + +var ErrInvalid = errors.New("yapf: not a valid YAPF file") + +// Image holds Width*Height*Channels bytes in Pixels, rows top to bottom. +// Channels: 1 gray, 2 gray+alpha, 3 RGB, 4 RGBA. Mips holds the smaller +// mip levels (1, 2, …), if any. +type Image struct { + Width, Height int + Channels int + Flags uint8 + GPUFormat uint8 + Pixels []byte + Mips [][]byte +} + +// Decode decodes YAPF bytes on the calling thread; DecodeThreads(data, 0) uses every core. +func Decode(data []byte) (*Image, error) { return DecodeThreads(data, 1) } + +func DecodeThreads(data []byte, threads int) (*Image, error) { + if len(data) == 0 { + return nil, ErrInvalid + } + p := C.yapf_load_memory_mt(unsafe.Pointer(&data[0]), C.size_t(len(data)), C.int(threads)) + if p == nil { + return nil, ErrInvalid + } + defer C.yapf_free(p) + img := &Image{Width: int(p.width), Height: int(p.height), Channels: int(p.channels), + Flags: uint8(p.flags), GPUFormat: uint8(p.gpu_format)} + mips := unsafe.Slice(p.mips, int(p.mip_levels)) + for m := range mips { + w, h := max(1, img.Width>>m), max(1, img.Height>>m) + level := C.GoBytes(unsafe.Pointer(mips[m]), C.int(w*h*img.Channels)) + if m == 0 { + img.Pixels = level + } else { + img.Mips = append(img.Mips, level) + } + } + return img, nil +} + +func Load(path string) (*Image, error) { + data, err := os.ReadFile(path) + if err != nil { + return nil, err + } + return Decode(data) +} + +// Encode returns the bytes of a .yapf file. +func Encode(img *Image) ([]byte, error) { + if img.Channels < 1 || img.Channels > 4 || len(img.Pixels) != img.Width*img.Height*img.Channels { + return nil, errors.New("yapf: pixels must hold Width*Height*Channels bytes") + } + // Copy the levels into C memory: cgo forbids passing Go pointers inside C structs. + levels := append([][]byte{img.Pixels}, img.Mips...) + table := (**C.uint8_t)(C.malloc(C.size_t(len(levels)) * C.size_t(unsafe.Sizeof(uintptr(0))))) + defer C.free(unsafe.Pointer(table)) + ptrs := unsafe.Slice(table, len(levels)) + for i, l := range levels { + ptrs[i] = (*C.uint8_t)(C.CBytes(l)) + defer C.free(unsafe.Pointer(ptrs[i])) + } + gpu := img.GPUFormat + if gpu == 0 { + gpu = map[int]uint8{4: 4, 3: 5, 2: 2, 1: 3}[img.Channels] + } + raw := C.yapf_image_t{width: C.uint32_t(img.Width), height: C.uint32_t(img.Height), + channels: C.uint8_t(img.Channels), gpu_format: C.uint8_t(gpu), flags: C.uint8_t(img.Flags), + mip_levels: C.uint8_t(len(levels)), pixels: ptrs[0], mips: table} + var data unsafe.Pointer + var size C.size_t + if rc := C.yapf_encode(&raw, &data, &size); rc != 0 { + return nil, errors.New("yapf: encoding failed") + } + defer C.yapf_free_buffer(data) + return C.GoBytes(data, C.int(size)), nil +} + +func Save(path string, img *Image) error { + data, err := Encode(img) + if err != nil { + return err + } + return os.WriteFile(path, data, 0o644) +} diff --git a/bindings/go/yapf_c.c b/bindings/go/yapf_c.c new file mode 100644 index 0000000..006b23c --- /dev/null +++ b/bindings/go/yapf_c.c @@ -0,0 +1,2 @@ +// Builds the C library as part of this Go package (cgo compiles .c files here). +#include "../../yapf.c" diff --git a/bindings/go/yapf_test.go b/bindings/go/yapf_test.go new file mode 100644 index 0000000..7910b1c --- /dev/null +++ b/bindings/go/yapf_test.go @@ -0,0 +1,46 @@ +package yapf + +import ( + "bytes" + "os" + "testing" +) + +func TestRoundtrip(t *testing.T) { + for ch := 1; ch <= 4; ch++ { + w, h := 97, 61 + px := make([]byte, w*h*ch) + for i := range px { + px[i] = byte(i*7 ^ i>>5) + } + data, err := Encode(&Image{Width: w, Height: h, Channels: ch, Flags: FlagSRGB, Pixels: px}) + if err != nil { + t.Fatal(err) + } + back, err := Decode(data) + if err != nil || !bytes.Equal(back.Pixels, px) { + t.Fatalf("roundtrip failed for %d channels", ch) + } + } +} + +func TestSampleReencodesIdentically(t *testing.T) { + file, err := os.ReadFile("../../other/YAPF.YAPF") + if err != nil { + t.Skip("sample not found") + } + img, err := Decode(file) + if err != nil { + t.Fatal(err) + } + again, _ := Encode(img) + if !bytes.Equal(again, file) { + t.Fatal("re-encoded sample differs from the C encoder's output") + } +} + +func TestRejectsGarbage(t *testing.T) { + if _, err := Decode([]byte("not an image")); err == nil { + t.Fatal("garbage was accepted") + } +} diff --git a/bindings/java/Example.java b/bindings/java/Example.java new file mode 100644 index 0000000..d12f7ca --- /dev/null +++ b/bindings/java/Example.java @@ -0,0 +1,29 @@ +// java --enable-native-access=ALL-UNNAMED -Dyapf.library=../../build/libyapf.so Example.java ../../other/YAPF.YAPF +import java.nio.file.Files; +import java.nio.file.Path; +import java.util.Arrays; + +public class Example { + public static void main(String[] args) throws Exception { + Path path = Path.of(args.length > 0 ? args[0] : "../../other/YAPF.YAPF"); + byte[] file = Files.readAllBytes(path); + + Yapf.Image img = Yapf.decode(file); // first call also loads the library + double ms = Double.MAX_VALUE; // best of 10, as in a running program + for (int i = 0; i < 10; i++) { + long t = System.nanoTime(); + img = Yapf.decode(file); + ms = Math.min(ms, (System.nanoTime() - t) / 1e6); + } + System.out.printf("%s: %dx%d, %d channels, decoded in %.2f ms%n", path, img.width(), img.height(), img.channels(), ms); + System.out.printf("file is %d bytes, %.1f%% of the raw pixels%n", file.length, 100.0 * file.length / img.pixels().length); + + byte[] again = Yapf.encode(img); + System.out.printf("re-encoded: %d bytes, identical: %b%n", again.length, Arrays.equals(again, file)); + + byte[] rgb = new byte[64 * 64 * 3]; + for (int i = 0; i < rgb.length; i++) rgb[i] = (byte) (i * 3); + Yapf.save(Path.of("java.yapf"), new Yapf.Image(64, 64, 3, rgb)); + System.out.println("wrote java.yapf"); + } +} diff --git a/bindings/java/README.md b/bindings/java/README.md new file mode 100644 index 0000000..605ce11 --- /dev/null +++ b/bindings/java/README.md @@ -0,0 +1,16 @@ +# YAPF for Java (and Kotlin) + +`Yapf.java` uses the Foreign Function & Memory API of **Java 22+** — no JNI, +no native glue to build; just the YAPF shared library (`yapf.dll`, +`libyapf.so`, `libyapf.dylib`) on the library path or passed with +`-Dyapf.library=/path/to/lib`. + +```java +Yapf.Image img = Yapf.load(Path.of("texture.yapf")); // or Yapf.decode(bytes) +byte[] file = Yapf.encode(new Yapf.Image(256, 256, 4, rgba)); +``` + +Kotlin can call the same class directly. Run with +`--enable-native-access=ALL-UNNAMED` to silence the native-access warning. + +Example: `java --enable-native-access=ALL-UNNAMED -Dyapf.library=../../build/libyapf.so Example.java` diff --git a/bindings/java/Yapf.java b/bindings/java/Yapf.java new file mode 100644 index 0000000..80e97c5 --- /dev/null +++ b/bindings/java/Yapf.java @@ -0,0 +1,121 @@ +// Yapf.java — YAPF for Java 22+ via the Foreign Function & Memory API (no JNI). +// Copyright 2026 Nexoniarz — Apache License 2.0. +// +// Needs the native library: yapf.dll / libyapf.so / libyapf.dylib on the +// library path, or its full path in the system property -Dyapf.library=... +// Run with --enable-native-access=ALL-UNNAMED to avoid the warning. +// +// Yapf.Image img = Yapf.load(Path.of("texture.yapf")); +// byte[] file = Yapf.encode(new Yapf.Image(256, 256, 4, rgba)); + +import java.io.IOException; +import java.lang.foreign.*; +import java.lang.invoke.MethodHandle; +import java.nio.file.Files; +import java.nio.file.Path; + +import static java.lang.foreign.ValueLayout.*; + +public final class Yapf { + public static final int FLAG_PREMULT_ALPHA = 1, FLAG_SRGB = 2; + + /** pixels: width*height*channels bytes, rows top to bottom. mips[0] == pixels. */ + public record Image(int width, int height, int channels, int flags, int gpuFormat, byte[][] mips) { + public Image(int width, int height, int channels, byte[] pixels) { + this(width, height, channels, FLAG_SRGB, switch (channels) { case 4 -> 4; case 3 -> 5; case 2 -> 2; default -> 3; }, + new byte[][] { pixels }); + } + public byte[] pixels() { return mips[0]; } + } + + // yapf_image_t: u32 width, u32 height, u8 channels, gpu_format, flags, mip_levels, ptr pixels, ptr mips + private static final StructLayout IMAGE = MemoryLayout.structLayout( + JAVA_INT.withName("width"), JAVA_INT.withName("height"), + JAVA_BYTE.withName("channels"), JAVA_BYTE.withName("gpu_format"), + JAVA_BYTE.withName("flags"), JAVA_BYTE.withName("mip_levels"), + MemoryLayout.paddingLayout(4), ADDRESS.withName("pixels"), ADDRESS.withName("mips")); + + private static final MethodHandle LOAD, FREE, ENCODE, FREE_BUFFER; + static { + Linker linker = Linker.nativeLinker(); + String custom = System.getProperty("yapf.library"); + SymbolLookup lib = custom != null + ? SymbolLookup.libraryLookup(Path.of(custom), Arena.global()) + : SymbolLookup.libraryLookup(System.mapLibraryName("yapf"), Arena.global()); + LOAD = linker.downcallHandle(lib.find("yapf_load_memory_mt").orElseThrow(), + FunctionDescriptor.of(ADDRESS, ADDRESS, JAVA_LONG, JAVA_INT)); + FREE = linker.downcallHandle(lib.find("yapf_free").orElseThrow(), FunctionDescriptor.ofVoid(ADDRESS)); + ENCODE = linker.downcallHandle(lib.find("yapf_encode").orElseThrow(), + FunctionDescriptor.of(JAVA_INT, ADDRESS, ADDRESS, ADDRESS)); + FREE_BUFFER = linker.downcallHandle(lib.find("yapf_free_buffer").orElseThrow(), FunctionDescriptor.ofVoid(ADDRESS)); + } + + private Yapf() {} + + /** Decode YAPF bytes; threads = 1 decodes on this thread, 0 uses every core. */ + public static Image decode(byte[] data, int threads) { + try (Arena arena = Arena.ofConfined()) { + MemorySegment in = arena.allocateFrom(JAVA_BYTE, data); + MemorySegment p = (MemorySegment) LOAD.invokeExact(in, (long) data.length, threads); + if (p.equals(MemorySegment.NULL)) throw new IllegalArgumentException("not a valid YAPF file"); + try { + MemorySegment s = p.reinterpret(IMAGE.byteSize()); + int w = s.get(JAVA_INT, 0), h = s.get(JAVA_INT, 4); + int ch = Byte.toUnsignedInt(s.get(JAVA_BYTE, 8)), levels = Byte.toUnsignedInt(s.get(JAVA_BYTE, 11)); + MemorySegment table = s.get(ADDRESS, 24).reinterpret(ADDRESS.byteSize() * levels); + byte[][] mips = new byte[levels][]; + for (int m = 0; m < levels; m++) { + long n = (long) Math.max(1, w >> m) * Math.max(1, h >> m) * ch; + mips[m] = table.getAtIndex(ADDRESS, m).reinterpret(n).toArray(JAVA_BYTE); + } + return new Image(w, h, ch, Byte.toUnsignedInt(s.get(JAVA_BYTE, 10)), + Byte.toUnsignedInt(s.get(JAVA_BYTE, 9)), mips); + } finally { + FREE.invokeExact(p); + } + } catch (RuntimeException e) { + throw e; + } catch (Throwable t) { + throw new IllegalStateException(t); + } + } + + public static Image decode(byte[] data) { return decode(data, 1); } + + public static Image load(Path path) throws IOException { return decode(Files.readAllBytes(path)); } + + /** Encode to the bytes of a .yapf file. */ + public static byte[] encode(Image img) { + if (img.pixels().length != img.width() * img.height() * img.channels()) + throw new IllegalArgumentException("pixels must hold width*height*channels bytes"); + try (Arena arena = Arena.ofConfined()) { + byte[][] mips = img.mips(); + MemorySegment table = arena.allocate(ADDRESS, mips.length); + for (int m = 0; m < mips.length; m++) table.setAtIndex(ADDRESS, m, arena.allocateFrom(JAVA_BYTE, mips[m])); + MemorySegment s = arena.allocate(IMAGE); + s.set(JAVA_INT, 0, img.width()); + s.set(JAVA_INT, 4, img.height()); + s.set(JAVA_BYTE, 8, (byte) img.channels()); + s.set(JAVA_BYTE, 9, (byte) img.gpuFormat()); + s.set(JAVA_BYTE, 10, (byte) img.flags()); + s.set(JAVA_BYTE, 11, (byte) mips.length); + s.set(ADDRESS, 16, table.getAtIndex(ADDRESS, 0)); + s.set(ADDRESS, 24, table); + MemorySegment outData = arena.allocate(ADDRESS), outSize = arena.allocate(JAVA_LONG); + int rc = (int) ENCODE.invokeExact(s, outData, outSize); + if (rc != 0) throw new IllegalStateException("yapf_encode failed (" + rc + ")"); + MemorySegment data = outData.get(ADDRESS, 0); + try { + return data.reinterpret(outSize.get(JAVA_LONG, 0)).toArray(JAVA_BYTE); + } finally { + FREE_BUFFER.invokeExact(data); + } + } catch (RuntimeException e) { + throw e; + } catch (Throwable t) { + throw new IllegalStateException(t); + } + } + + public static void save(Path path, Image img) throws IOException { Files.write(path, encode(img)); } +} diff --git a/bindings/js/README.md b/bindings/js/README.md index 849f640..309c0c7 100644 --- a/bindings/js/README.md +++ b/bindings/js/README.md @@ -19,4 +19,6 @@ ctx.putImageData(new ImageData(YAPF.toRGBA(img), img.width, img.height), 0, 0); `decode` throws an `Error` for corrupt or unsupported data. Speed: a 4K image decodes in about 150 ms in Node / Chrome (the C library: 6 ms). -Tests: `node test.js`. +TypeScript types are in `yapf.d.ts`. + +Example: `node example.mjs ../../other/YAPF.YAPF` · tests: `node test.js`. diff --git a/bindings/js/example.mjs b/bindings/js/example.mjs new file mode 100644 index 0000000..4974404 --- /dev/null +++ b/bindings/js/example.mjs @@ -0,0 +1,19 @@ +// node example.mjs ../../other/YAPF.YAPF +import { readFileSync } from 'node:fs'; +import { createRequire } from 'node:module'; +const YAPF = createRequire(import.meta.url)('./yapf.js'); + +const path = process.argv[2] ?? '../../other/YAPF.YAPF'; +const file = readFileSync(path); + +let img = YAPF.decode(file); +let best = Infinity; // best of 10 decodes +for (let i = 0; i < 10; i++) { + const t = performance.now(); + img = YAPF.decode(file); + best = Math.min(best, performance.now() - t); +} +console.log(`${path}: ${img.width}x${img.height}, ${img.channels} channels, decoded in ${best.toFixed(2)} ms (pure JavaScript)`); +console.log(`file is ${file.length} bytes, ${(100 * file.length / img.pixels.length).toFixed(1)}% of the raw pixels`); +const again = YAPF.encode(img); +console.log(`re-encoded: ${again.length} bytes, identical: ${Buffer.compare(Buffer.from(again), file) === 0}`); diff --git a/bindings/js/package.json b/bindings/js/package.json index 81c885c..bb4c1d7 100644 --- a/bindings/js/package.json +++ b/bindings/js/package.json @@ -3,7 +3,8 @@ "version": "1.0.0", "description": "YAPF v1 lossless image format: encoder and decoder in plain JavaScript, no dependencies", "main": "yapf.js", - "files": ["yapf.js", "README.md"], + "types": "yapf.d.ts", + "files": ["yapf.js", "yapf.d.ts", "README.md"], "scripts": { "test": "node test.js" }, "keywords": ["image", "codec", "lossless", "yapf", "texture"], "license": "Apache-2.0", diff --git a/bindings/js/yapf.d.ts b/bindings/js/yapf.d.ts new file mode 100644 index 0000000..1e52691 --- /dev/null +++ b/bindings/js/yapf.d.ts @@ -0,0 +1,25 @@ +// TypeScript declarations for yapf.js +export interface YapfImage { + width: number; + height: number; + /** 1 gray, 2 gray + alpha, 3 RGB, 4 RGBA */ + channels: 1 | 2 | 3 | 4; + /** width * height * channels bytes, rows top to bottom */ + pixels: Uint8Array; + gpuFormat?: number; + flags?: number; + mipLevels?: number; + /** mips[0] is the full-size image */ + mips?: Uint8Array[]; +} + +/** Decode a .yapf file; throws on corrupt or unsupported data. */ +export function decode(bytes: Uint8Array | ArrayBuffer): Required; +/** Encode an image to the bytes of a .yapf file (identical to the C encoder). */ +export function encode(image: YapfImage): Uint8Array; +/** Any channel count → RGBA, ready for `new ImageData(rgba, w, h)`. */ +export function toRGBA(image: YapfImage, level?: number): Uint8ClampedArray; + +export const GPU: { RGBA8: 0; RGB8: 1; RG8: 2; R8: 3; SRGB8_A8: 4; SRGB8: 5 }; +export const FLAG: { PREMULT_ALPHA: 1; SRGB: 2 }; +export const VERSION: 1; diff --git a/bindings/lua/README.md b/bindings/lua/README.md new file mode 100644 index 0000000..5548cb9 --- /dev/null +++ b/bindings/lua/README.md @@ -0,0 +1,13 @@ +# YAPF for Lua (LuaJIT) + +`yapf.lua` uses LuaJIT's FFI with the YAPF shared library (`libyapf.so`, +`yapf.dll`, `libyapf.dylib`, or the path in `$YAPF_LIBRARY`). Works in +LÖVE (love2d), which runs on LuaJIT. + +```lua +local yapf = require("yapf") +local img = yapf.load("texture.yapf") -- width, height, channels, pixels (string) +yapf.save("out.yapf", img) +``` + +Example: `YAPF_LIBRARY=../../build/libyapf.so luajit example.lua` diff --git a/bindings/lua/example.lua b/bindings/lua/example.lua new file mode 100644 index 0000000..35d4fe0 --- /dev/null +++ b/bindings/lua/example.lua @@ -0,0 +1,17 @@ +-- YAPF_LIBRARY=../../build/libyapf.so luajit example.lua ../../other/YAPF.YAPF +package.path = "./?.lua;" .. package.path +local yapf = require("yapf") +local path = arg[1] or "../../other/YAPF.YAPF" +local f = assert(io.open(path, "rb")); local data = f:read("*a"); f:close() + +local img = assert(yapf.decode(data)) +local best = math.huge -- best of 10 decodes +for _ = 1, 10 do + local t = os.clock() + img = yapf.decode(data) + best = math.min(best, os.clock() - t) +end +print(string.format("%s: %dx%d, %d channels, decoded in %.2f ms", path, img.width, img.height, img.channels, best * 1000)) +print(string.format("file is %d bytes, %.1f%% of the raw pixels", #data, 100 * #data / #img.pixels)) +local again = yapf.encode(img) +print(string.format("re-encoded: %d bytes, identical: %s", #again, tostring(again == data))) diff --git a/bindings/lua/yapf.lua b/bindings/lua/yapf.lua new file mode 100644 index 0000000..39c0a12 --- /dev/null +++ b/bindings/lua/yapf.lua @@ -0,0 +1,71 @@ +-- yapf.lua — YAPF for LuaJIT via its FFI. Copyright 2026 Nexoniarz — Apache 2.0. +-- local yapf = require("yapf") +-- local img = yapf.load("texture.yapf") -- img.width, img.height, img.channels, img.pixels (Lua string) +-- local bytes = yapf.encode(img) -- or yapf.save(path, img) +-- Loads libyapf.so / yapf.dll / libyapf.dylib, or the path in $YAPF_LIBRARY. +local ffi = require("ffi") + +ffi.cdef [[ +typedef struct { + uint32_t width, height; + uint8_t channels, gpu_format, flags, mip_levels; + uint8_t *pixels; + uint8_t **mips; +} yapf_image_t; +yapf_image_t *yapf_load_memory_mt(const void *buffer, size_t size, int threads); +void yapf_free(yapf_image_t *img); +int yapf_encode(const yapf_image_t *img, void **out_data, size_t *out_size); +void yapf_free_buffer(void *data); +]] + +local lib = ffi.load(os.getenv("YAPF_LIBRARY") or "yapf") +local M = { FLAG_PREMULT_ALPHA = 1, FLAG_SRGB = 2 } + +-- Decode a Lua string holding a .yapf file. threads: 1 (default) or 0 for all cores. +function M.decode(data, threads) + local p = lib.yapf_load_memory_mt(data, #data, threads or 1) + if p == nil then return nil, "not a valid YAPF file" end + local img = { + width = p.width, height = p.height, channels = p.channels, + flags = p.flags, gpu_format = p.gpu_format, + pixels = ffi.string(p.pixels, p.width * p.height * p.channels), + } + lib.yapf_free(p) + return img +end + +function M.load(path) + local f, err = io.open(path, "rb") + if not f then return nil, err end + local data = f:read("*a") + f:close() + return M.decode(data) +end + +-- img: { width, height, channels, pixels = string } → string with the .yapf bytes +function M.encode(img) + local raw = ffi.new("yapf_image_t") + local px = ffi.new("uint8_t[?]", #img.pixels) + ffi.copy(px, img.pixels, #img.pixels) + local table = ffi.new("uint8_t*[1]", px) + raw.width, raw.height, raw.channels = img.width, img.height, img.channels + raw.flags = img.flags or M.FLAG_SRGB + raw.gpu_format = img.gpu_format or ({ 3, 2, 5, 4 })[img.channels] + raw.mip_levels, raw.pixels, raw.mips = 1, px, table + local out, size = ffi.new("void*[1]"), ffi.new("size_t[1]") + if lib.yapf_encode(raw, out, size) ~= 0 then return nil, "encoding failed" end + local bytes = ffi.string(out[0], size[0]) + lib.yapf_free_buffer(out[0]) + return bytes +end + +function M.save(path, img) + local bytes, err = M.encode(img) + if not bytes then return nil, err end + local f = assert(io.open(path, "wb")) + f:write(bytes) + f:close() + return true +end + +return M diff --git a/bindings/python/README.md b/bindings/python/README.md index 3c93be7..33e52c8 100644 --- a/bindings/python/README.md +++ b/bindings/python/README.md @@ -21,3 +21,5 @@ It uses the C library through `ctypes` when it finds one: Without the library it falls back to a pure-Python decoder that needs `numpy` (decoding only; about 1.4 s for a 4K image). `yapf.backend()` tells you which one is in use. + +Example: `YAPF_LIBRARY=../../build/libyapf.so python3 example.py ../../other/YAPF.YAPF` diff --git a/bindings/python/example.py b/bindings/python/example.py new file mode 100644 index 0000000..1b13fae --- /dev/null +++ b/bindings/python/example.py @@ -0,0 +1,33 @@ +""" +example.py — YAPF from Python. + + make # builds build/libyapf.so (or get it from Releases) + YAPF_LIBRARY=../../build/libyapf.so python3 example.py ../../other/YAPF.YAPF +""" +import sys +import time + +import yapf + +path = sys.argv[1] if len(sys.argv) > 1 else "../../other/YAPF.YAPF" +data = open(path, "rb").read() + +img = yapf.decode(data) +ms = float("inf") # best of 10, as in a running program +for _ in range(10): + t = time.perf_counter() + img = yapf.decode(data) + ms = min(ms, (time.perf_counter() - t) * 1000) +raw = img.width * img.height * img.channels +print("%s: %dx%d, %d channels, decoded in %.2f ms (%s backend)" + % (path, img.width, img.height, img.channels, ms, yapf.backend())) +print("file is %d bytes, %.1f%% of the raw pixels" % (len(data), 100.0 * len(data) / raw)) + +if yapf.backend() == "c": + again = yapf.encode(img) + print("re-encoded:", len(again), "bytes, identical:", again == data) + + w = h = 128 # your own pixels → .yapf + rgb = bytes(v for y in range(h) for x in range(w) for v in (x * 2, y * 2, 128)) + yapf.save("gradient.yapf", yapf.Image(w, h, 3, rgb)) + print("wrote gradient.yapf") diff --git a/bindings/python/yapf.py b/bindings/python/yapf.py index 86472d1..9f8f3a1 100644 --- a/bindings/python/yapf.py +++ b/bindings/python/yapf.py @@ -94,6 +94,9 @@ def _find_library(): lib.yapf_load_memory_mt.argtypes = [ctypes.c_char_p, ctypes.c_size_t, ctypes.c_int] lib.yapf_free.argtypes = [ctypes.POINTER(_CImage)] lib.yapf_save.argtypes = [ctypes.c_char_p, ctypes.POINTER(_CImage)] + lib.yapf_encode.argtypes = [ctypes.POINTER(_CImage), ctypes.POINTER(ctypes.c_void_p), + ctypes.POINTER(ctypes.c_size_t)] + lib.yapf_free_buffer.argtypes = [ctypes.c_void_p] return lib except (OSError, AttributeError): continue @@ -124,15 +127,20 @@ def _c_decode(data): _lib.yapf_free(p) -def _c_save(path, img): +def _c_encode(img): bufs = [ctypes.create_string_buffer(m, len(m)) for m in img.mips] ptrs = (ctypes.POINTER(ctypes.c_uint8) * len(bufs))( *[ctypes.cast(b, ctypes.POINTER(ctypes.c_uint8)) for b in bufs]) c = _CImage(img.width, img.height, img.channels, img.gpu_format, img.flags, len(bufs), ptrs[0], ptrs) - rc = _lib.yapf_save(os.fsencode(path), ctypes.byref(c)) + data, size = ctypes.c_void_p(), ctypes.c_size_t() + rc = _lib.yapf_encode(ctypes.byref(c), ctypes.byref(data), ctypes.byref(size)) if rc != 0: - raise YapfError("yapf_save failed with code %d" % rc) + raise YapfError("yapf_encode failed with code %d" % rc) + try: + return ctypes.string_at(data, size.value) + finally: + _lib.yapf_free_buffer(data) # ── pure-Python fallback decoder (numpy) ────────────────────────────── @@ -280,9 +288,9 @@ def load(path): return decode(f.read()) -def save(path, img): - """Encode an Image (or anything with width/height/channels/pixels) to a - .yapf file. Needs the C library.""" +def encode(img): + """Encode an Image (or anything with width/height/channels/pixels) to + YAPF bytes. Needs the C library.""" if not _lib: raise YapfError("saving .yapf needs the YAPF native library, which was not found " "for this platform; see https://github.com/Nexoniarz/YAPF#python") @@ -291,20 +299,14 @@ def save(path, img): expected = img.width * img.height * img.channels if len(img.pixels) != expected: raise YapfError("expected %d pixel bytes, got %d" % (expected, len(img.pixels))) - _c_save(path, img) + return _c_encode(img) -def encode(img): - """Encode an Image to YAPF bytes. Needs the C library.""" - import tempfile - fd, tmp = tempfile.mkstemp(suffix=".yapf") - os.close(fd) - try: - save(tmp, img) - with open(tmp, "rb") as f: - return f.read() - finally: - os.remove(tmp) +def save(path, img): + """Encode an Image to a .yapf file. Needs the C library.""" + data = encode(img) + with open(path, "wb") as f: + f.write(data) if __name__ == "__main__": diff --git a/bindings/rust/Cargo.toml b/bindings/rust/Cargo.toml new file mode 100644 index 0000000..759099c --- /dev/null +++ b/bindings/rust/Cargo.toml @@ -0,0 +1,12 @@ +[package] +name = "yapf" +version = "1.0.0" +edition = "2021" +description = "YAPF lossless images: fast byte-aligned decoder and encoder (bindings to the C library)" +license = "Apache-2.0" +repository = "https://github.com/Nexoniarz/YAPF" +links = "yapf" +build = "build.rs" + +[build-dependencies] +cc = "1" diff --git a/bindings/rust/README.md b/bindings/rust/README.md new file mode 100644 index 0000000..5eef983 --- /dev/null +++ b/bindings/rust/README.md @@ -0,0 +1,18 @@ +# YAPF for Rust + +A crate that compiles the C library with the `cc` crate, so `cargo build` +works on Windows, macOS and Linux with no extra steps. + +```toml +[dependencies] +yapf = { git = "https://github.com/Nexoniarz/YAPF" } +``` + +```rust +let img = yapf::Image::load("texture.yapf")?; // or Image::decode(&bytes) +println!("{}x{}, {} channels", img.width, img.height, img.channels); +yapf::Image::new(256, 256, 4, rgba).save("out.yapf")?; // or .encode() → Vec +``` + +Example: `cargo run --release --example demo -- ../../other/YAPF.YAPF` · +tests: `cargo test`. diff --git a/bindings/rust/build.rs b/bindings/rust/build.rs new file mode 100644 index 0000000..8d56bac --- /dev/null +++ b/bindings/rust/build.rs @@ -0,0 +1,12 @@ +// Compiles the C library that lives two folders up. +fn main() { + let root = std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("../.."); + cc::Build::new() + .file(root.join("yapf.c")) + .include(&root) + .flag_if_supported("-std=c99") + .opt_level(2) + .compile("yapf"); + println!("cargo:rerun-if-changed={}", root.join("yapf.c").display()); + println!("cargo:rerun-if-changed={}", root.join("yapf.h").display()); +} diff --git a/bindings/rust/examples/demo.rs b/bindings/rust/examples/demo.rs new file mode 100644 index 0000000..3be07e1 --- /dev/null +++ b/bindings/rust/examples/demo.rs @@ -0,0 +1,26 @@ +// cargo run --release --example demo -- ../../other/YAPF.YAPF +use std::time::Instant; + +fn main() -> Result<(), yapf::Error> { + let path = std::env::args().nth(1).unwrap_or_else(|| "../../other/YAPF.YAPF".into()); + let file = std::fs::read(&path)?; + + let mut img = yapf::Image::decode(&file)?; + let mut best = f64::MAX; // best of 10, as in a running program + for _ in 0..10 { + let t = Instant::now(); + img = yapf::Image::decode(&file)?; + best = best.min(t.elapsed().as_secs_f64() * 1e3); + } + println!("{path}: {}x{}, {} channels, decoded in {best:.2} ms", img.width, img.height, img.channels); + println!("file is {} bytes, {:.1}% of the raw pixels", + file.len(), 100.0 * file.len() as f64 / img.pixels.len() as f64); + + let again = img.encode()?; + println!("re-encoded: {} bytes, identical: {}", again.len(), again == file); + + let px: Vec = (0..128 * 128).flat_map(|i| [(i % 128 * 2) as u8, (i / 128 * 2) as u8, 128]).collect(); + yapf::Image::new(128, 128, 3, px).save("rust.yapf")?; + println!("wrote rust.yapf"); + Ok(()) +} diff --git a/bindings/rust/src/lib.rs b/bindings/rust/src/lib.rs new file mode 100644 index 0000000..7c4802e --- /dev/null +++ b/bindings/rust/src/lib.rs @@ -0,0 +1,156 @@ +//! YAPF lossless images for Rust. +//! +//! ```no_run +//! let img = yapf::Image::load("texture.yapf")?; +//! println!("{}x{} with {} channels", img.width, img.height, img.channels); +//! let bytes = img.encode()?; +//! # Ok::<(), yapf::Error>(()) +//! ``` +use std::ffi::{c_int, c_void}; +use std::fmt; +use std::path::Path; + +#[repr(C)] +struct RawImage { + width: u32, + height: u32, + channels: u8, + gpu_format: u8, + flags: u8, + mip_levels: u8, + pixels: *mut u8, + mips: *mut *mut u8, +} + +extern "C" { + fn yapf_load_memory_mt(buffer: *const c_void, size: usize, threads: c_int) -> *mut RawImage; + fn yapf_free(img: *mut RawImage); + fn yapf_encode(img: *const RawImage, out_data: *mut *mut c_void, out_size: *mut usize) -> c_int; + fn yapf_free_buffer(data: *mut c_void); +} + +pub const FLAG_PREMULT_ALPHA: u8 = 1; +pub const FLAG_SRGB: u8 = 2; + +#[derive(Debug)] +pub enum Error { + /// Not a YAPF v1 file, or corrupt. + Invalid, + /// The encoder rejected the image (bad size or channel count) or ran out of memory. + Encode(i32), + Io(std::io::Error), +} + +impl fmt::Display for Error { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match self { + Error::Invalid => write!(f, "not a valid YAPF file"), + Error::Encode(c) => write!(f, "YAPF encoding failed (code {c})"), + Error::Io(e) => write!(f, "{e}"), + } + } +} +impl std::error::Error for Error {} +impl From for Error { + fn from(e: std::io::Error) -> Self { Error::Io(e) } +} + +/// A decoded image: `pixels` holds `width * height * channels` bytes, rows top to bottom. +#[derive(Clone, Debug, PartialEq, Eq)] +pub struct Image { + pub width: u32, + pub height: u32, + /// 1 gray, 2 gray + alpha, 3 RGB, 4 RGBA + pub channels: u8, + pub flags: u8, + pub gpu_format: u8, + pub pixels: Vec, + /// Smaller mip levels (1, 2, …); empty when the file stores only the base image. + pub mips: Vec>, +} + +impl Image { + pub fn new(width: u32, height: u32, channels: u8, pixels: Vec) -> Image { + let gpu_format = match channels { 4 => 4, 3 => 5, 2 => 2, _ => 3 }; + Image { width, height, channels, flags: FLAG_SRGB, gpu_format, pixels, mips: Vec::new() } + } + + /// Decode YAPF bytes; `threads` = 1 decodes on this thread, 0 uses every core. + pub fn decode_with_threads(data: &[u8], threads: i32) -> Result { + unsafe { + let p = yapf_load_memory_mt(data.as_ptr().cast(), data.len(), threads); + if p.is_null() { + return Err(Error::Invalid); + } + let r = &*p; + let level = |m: usize| { + let w = (r.width >> m).max(1) as usize; + let h = (r.height >> m).max(1) as usize; + std::slice::from_raw_parts(*r.mips.add(m), w * h * r.channels as usize).to_vec() + }; + let img = Image { + width: r.width, height: r.height, channels: r.channels, flags: r.flags, + gpu_format: r.gpu_format, pixels: level(0), + mips: (1..r.mip_levels as usize).map(level).collect(), + }; + yapf_free(p); + Ok(img) + } + } + + pub fn decode(data: &[u8]) -> Result { Image::decode_with_threads(data, 1) } + + pub fn load>(path: P) -> Result { + Image::decode(&std::fs::read(path)?) + } + + /// Encode to the bytes of a `.yapf` file. + pub fn encode(&self) -> Result, Error> { + let mut levels: Vec<*mut u8> = std::iter::once(self.pixels.as_ptr() as *mut u8) + .chain(self.mips.iter().map(|m| m.as_ptr() as *mut u8)) + .collect(); + let raw = RawImage { + width: self.width, height: self.height, channels: self.channels, + gpu_format: self.gpu_format, flags: self.flags, mip_levels: levels.len() as u8, + pixels: levels[0], mips: levels.as_mut_ptr(), + }; + if self.pixels.len() != self.width as usize * self.height as usize * self.channels as usize { + return Err(Error::Encode(-1)); + } + let mut data: *mut c_void = std::ptr::null_mut(); + let mut size = 0usize; + let rc = unsafe { yapf_encode(&raw, &mut data, &mut size) }; + if rc != 0 { + return Err(Error::Encode(rc)); + } + let out = unsafe { std::slice::from_raw_parts(data as *const u8, size).to_vec() }; + unsafe { yapf_free_buffer(data) }; + Ok(out) + } + + pub fn save>(&self, path: P) -> Result<(), Error> { + std::fs::write(path, self.encode()?)?; + Ok(()) + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn roundtrip() { + for ch in 1..=4u8 { + let (w, h) = (97u32, 61u32); + let px: Vec = (0..w * h * ch as u32).map(|i| (i * 7 ^ i >> 5) as u8).collect(); + let img = Image::new(w, h, ch, px); + let back = Image::decode(&img.encode().unwrap()).unwrap(); + assert_eq!(back, img); + } + } + + #[test] + fn rejects_garbage() { + assert!(Image::decode(b"not an image").is_err()); + } +} diff --git a/bindings/zig/README.md b/bindings/zig/README.md new file mode 100644 index 0000000..09c7271 --- /dev/null +++ b/bindings/zig/README.md @@ -0,0 +1,13 @@ +# YAPF for Zig + +Zig imports `yapf.h` directly — no bindings needed: + +```zig +const c = @cImport(@cInclude("yapf.h")); +const img = c.yapf_load("texture.yapf") orelse return error.CannotLoad; +defer c.yapf_free(img); +``` + +Add `yapf.c` to your build (`exe.addCSourceFile(.{ .file = b.path("yapf.c") })`, +`exe.linkLibC()`), or run the example: +`zig run example.zig ../../yapf.c -I../.. -lc -O ReleaseFast`. diff --git a/bindings/zig/example.zig b/bindings/zig/example.zig new file mode 100644 index 0000000..5a99540 --- /dev/null +++ b/bindings/zig/example.zig @@ -0,0 +1,33 @@ +// YAPF from Zig: the C header is imported directly. +// zig run example.zig ../../yapf.c -I../.. -lc -O ReleaseFast +// (Uses only the C API plus std.debug.print, so it works across Zig versions.) +const std = @import("std"); +const c = @cImport({ + @cInclude("yapf.h"); + @cInclude("time.h"); +}); + +pub fn main() !void { + const path: [:0]const u8 = "../../other/YAPF.YAPF"; + + var img: *c.yapf_image_t = c.yapf_load(path.ptr) orelse return error.CannotLoad; + var best: c.clock_t = std.math.maxInt(c.clock_t); // best of 10 decodes + for (0..10) |_| { + const t = c.clock(); + const again = c.yapf_load(path.ptr) orelse return error.CannotLoad; + best = @min(best, c.clock() - t); + c.yapf_free(img); + img = again; + } + defer c.yapf_free(img); + std.debug.print("{s}: {d}x{d}, {d} channels, decoded in {d:.2} ms\n", .{ + path, img.width, img.height, img.channels, @as(f64, @floatFromInt(best)) * 1000.0 / @as(f64, @floatFromInt(c.CLOCKS_PER_SEC)), + }); + + var data: ?*anyopaque = null; + var size: usize = 0; + if (c.yapf_encode(img, &data, &size) != c.YAPF_OK) return error.EncodeFailed; + defer c.yapf_free_buffer(data); + const raw = @as(f64, @floatFromInt(img.width * img.height * img.channels)); + std.debug.print("encoded: {d} bytes, {d:.1}% of the raw pixels\n", .{ size, 100.0 * @as(f64, @floatFromInt(size)) / raw }); +} diff --git a/build.bat b/build.bat index 5c3d268..801253c 100644 --- a/build.bat +++ b/build.bat @@ -145,7 +145,9 @@ echo !C_GRN![ ok ]!C_RST! Header ^→ dist\include\yapf.h echo yapf_load_mt echo yapf_load_memory_mt echo yapf_save + echo yapf_encode echo yapf_free + echo yapf_free_buffer ) > dist\_yapf.def :: ==================================================================== diff --git a/docs/LANGUAGES.md b/docs/LANGUAGES.md new file mode 100644 index 0000000..7029222 --- /dev/null +++ b/docs/LANGUAGES.md @@ -0,0 +1,144 @@ +# Using YAPF from your language + +YAPF is a small C library with six functions, so every language that can +call C can use it. Ready-made bindings, each with a runnable example: + +| Language | Folder | How it works | Tested | Decode¹ | +| :------- | :----- | :----------- | :----- | ------: | +| **C** | [`yapf.h`](../yapf.h), [`examples/c`](../examples/c) | the library itself | ✅ | 0.26 ms | +| **C++** | [`bindings/cpp`](../bindings/cpp) | header-only RAII wrapper (C++17) | ✅ g++ | 0.25 ms | +| **Python** | [`bindings/python`](../bindings/python) | ctypes; numpy fallback without the library | ✅ CPython 3.13 | 0.23 ms | +| **C# / .NET** | [`bindings/csharp`](../bindings/csharp) | P/Invoke, single file | ✅ .NET 8 | 0.30 ms | +| **Rust** | [`bindings/rust`](../bindings/rust) | crate, builds the C code with `cc` | ✅ Rust 1.95 | 0.28 ms | +| **Go** | [`bindings/go`](../bindings/go) | cgo package | ✅ Go 1.26 | 0.31 ms | +| **Java / Kotlin** | [`bindings/java`](../bindings/java) | Foreign Function & Memory API (Java 22+) | ✅ JDK 25 | 0.42 ms | +| **Zig** | [`bindings/zig`](../bindings/zig) | `@cImport` of `yapf.h` | ✅ Zig 0.16 | 0.28 ms | +| **Lua** | [`bindings/lua`](../bindings/lua) | LuaJIT FFI (also LÖVE) | ✅ LuaJIT 2.1 | 0.28 ms | +| **JavaScript / TypeScript** | [`bindings/js`](../bindings/js) | pure JS port, `.d.ts` types | ✅ Node 24, Chromium | 5.9 ms | + +¹ Decoding the 512×512 RGBA YAPF logo (`other/YAPF.YAPF`), best of 10 after +a warm-up, on an AMD Ryzen 7 5700G. Every binding except JavaScript runs +the same native code, so they are all about equally fast; the JavaScript +port needs no native library at all. Every example also re-encodes the +image and checks that the result is byte-identical to the original file. + +## Getting the native library + +Bindings other than C, C++, Rust, Go, Zig and JavaScript load the shared +library at run time: + +| System | File | Get it | +| :------ | :-------------- | :----- | +| Windows | `yapf.dll` | [Releases](https://github.com/Nexoniarz/YAPF/releases) or `build.bat` | +| Linux | `libyapf.so` | Releases, `make`, or `sudo make install` | +| macOS | `libyapf.dylib` | Releases or `make` | + +Put it next to your program, on the library path (`PATH`, +`LD_LIBRARY_PATH`, `DYLD_LIBRARY_PATH`), or pass its path explicitly +(`YAPF_LIBRARY` for Python and Lua, `-Dyapf.library=` for Java). + +## Quick look + +**C** +```c +yapf_image_t *img = yapf_load("texture.yapf"); +/* img->width, img->height, img->channels, img->pixels */ +yapf_free(img); +``` + +**C++** +```cpp +auto img = yapf::Image::load("texture.yapf"); +auto bytes = img.encode(); +``` + +**Python** +```python +img = yapf.load("texture.yapf") +yapf.save("copy.yapf", img) +``` + +**C#** +```csharp +var img = YapfImage.Load("texture.yapf"); +File.WriteAllBytes("copy.yapf", img.Encode()); +``` + +**Rust** +```rust +let img = yapf::Image::load("texture.yapf")?; +img.save("copy.yapf")?; +``` + +**Go** +```go +img, err := yapf.Load("texture.yapf") +data, err := yapf.Encode(img) +``` + +**Java** +```java +Yapf.Image img = Yapf.load(Path.of("texture.yapf")); +Yapf.save(Path.of("copy.yapf"), img); +``` + +**Zig** +```zig +const img = c.yapf_load("texture.yapf") orelse return error.CannotLoad; +defer c.yapf_free(img); +``` + +**Lua** +```lua +local img = yapf.load("texture.yapf") +yapf.save("copy.yapf", img) +``` + +**JavaScript** +```js +const img = YAPF.decode(bytes); +const file = YAPF.encode(img); +``` + +## Any other language + +Swift, Kotlin/Native, Dart, PHP, Ruby, Julia, Nim, D, Odin, Haskell, … — +anything with a C foreign-function interface can call the library +directly. These are the whole API: + +```c +typedef struct { + uint32_t width, height; + uint8_t channels; /* 1 gray, 2 gray+alpha, 3 RGB, 4 RGBA */ + uint8_t gpu_format; /* hint, see yapf.h */ + uint8_t flags; /* 1 premultiplied alpha, 2 sRGB */ + uint8_t mip_levels; + uint8_t *pixels; /* width*height*channels bytes, rows top to bottom */ + uint8_t **mips; /* mips[0] == pixels */ +} yapf_image_t; /* 32 bytes on 64-bit systems: pixels at offset 16, mips at 24 */ + +yapf_image_t *yapf_load_memory_mt(const void *data, size_t size, int threads); /* threads: 1 or 0 = all cores */ +void yapf_free(yapf_image_t *img); +int yapf_encode(const yapf_image_t *img, void **out_data, size_t *out_size); +void yapf_free_buffer(void *data); +yapf_image_t *yapf_load(const char *path); +int yapf_save(const char *path, const yapf_image_t *img); +``` + +Returned images and buffers must be released with `yapf_free` / +`yapf_free_buffer` from the same library. Ready-made bindings for these +languages are not included yet; the snippets below show the starting point +and are **not tested**. + +**Swift** — add a module map for `yapf.h` (or a bridging header) and compile +`yapf.c` in the target; then `let img = yapf_load(path)` works as is. + +**Dart** — `dart:ffi`: `DynamicLibrary.open('libyapf.so')` and +`lookupFunction Function(Pointer, IntPtr, Int32), …>('yapf_load_memory_mt')`. + +**PHP 7.4+** — `$ffi = FFI::cdef(file_get_contents('yapf_api.h'), 'libyapf.so');` +with the declarations above in `yapf_api.h`. + +**Ruby** — `Fiddle::Function.new(lib['yapf_load'], [Fiddle::TYPE_VOIDP], Fiddle::TYPE_VOIDP)`. + +**Julia** — `ccall((:yapf_load, "libyapf"), Ptr{Cvoid}, (Cstring,), path)`. diff --git a/examples/README.md b/examples/README.md new file mode 100644 index 0000000..57fd093 --- /dev/null +++ b/examples/README.md @@ -0,0 +1,62 @@ +# Examples: how fast, how small + +## Try it on your own images + +```sh +make # builds build/yapf +build/yapf compare photo.png ui.png # any PNG / JPG / TGA / BMP +``` + +`yapf compare` encodes each image to YAPF in memory, checks that it decodes +back to exactly the same pixels, and times decoding the original format +against YAPF on one core and on all cores: + +``` +ui_reverb.png (261 x 430, RGBA) + raw 438.4 KB + PNG 176.6 KB 40.3% of raw decode 2.049 ms + YAPF 143.7 KB 32.8% of raw decode 0.170 ms 12.1x faster, 81% of the PNG size + YAPF on all cores: decode 0.176 ms (2.5 GB/s), encode 8 ms, lossless: yes +``` + +## Results on a mixed set of images + +AMD Ryzen 7 5700G, `make` (gcc -O2), PNG decoded with stb_image. The test +images (a 4K wallpaper, UI screenshots, an app icon and seamless textures +from open-source projects) are not redistributed here; run `yapf compare` +on your own files to reproduce. + +| Image | PNG | YAPF | YAPF / PNG size | PNG decode | YAPF decode | Faster | +| :---- | --: | ---: | --------------: | ---------: | ----------: | -----: | +| UI screenshot 261×430 | 176.6 KB | 143.7 KB | **81 %** | 2.05 ms | 0.17 ms | **12×** | +| UI screenshot 305×529 | 201.5 KB | 185.0 KB | **92 %** | 2.55 ms | 0.24 ms | **11×** | +| App icon 128×128 | 24.9 KB | 23.1 KB | **93 %** | 0.30 ms | 0.026 ms | **11×** | +| Wallpaper 3840×2160 | 1.77 MB | 2.21 MB | 125 % | 73.7 ms | 6.8 ms | **11×** | +| Rock texture 512×512 | 182.6 KB | 221.8 KB | 121 % | 1.64 ms | 0.17 ms | **10×** | +| Wood texture 512×512 | 185.8 KB | 234.1 KB | 126 % | 1.64 ms | 0.17 ms | **10×** | +| Canvas texture 512×512 | 171.5 KB | 245.6 KB | 143 % | 1.36 ms | 0.16 ms | **9×** | +| YAPF logo 512×512 | 60.5 KB | 89.3 KB | 148 % | 1.25 ms | 0.25 ms | **5×** | + +In short: **smaller than PNG on UI-style images, 20–50 % larger on photos +and textures, and 5–12× faster to decode** — always lossless. The decoder +reaches 5–6 GB/s on large images on one core, and more with +`yapf_load_mt`. + +## Code examples + +| Language | Example | +| :------- | :------ | +| C | [`examples/c/example.c`](c/example.c) | +| C++ | [`bindings/cpp/example.cpp`](../bindings/cpp/example.cpp) | +| Python | [`bindings/python/example.py`](../bindings/python/example.py) | +| C# | [`bindings/csharp/Example`](../bindings/csharp/Example) | +| Rust | [`bindings/rust/examples/demo.rs`](../bindings/rust/examples/demo.rs) | +| Go | [`bindings/go/example`](../bindings/go/example) | +| Java | [`bindings/java/Example.java`](../bindings/java/Example.java) | +| Zig | [`bindings/zig/example.zig`](../bindings/zig/example.zig) | +| Lua | [`bindings/lua/example.lua`](../bindings/lua/example.lua) | +| JavaScript | [`bindings/js/example.mjs`](../bindings/js/example.mjs) | + +Each one loads a `.yapf` file, prints its size and decode time, re-encodes +it (byte-identical to the file) and writes a new image from its own pixels. +See [docs/LANGUAGES.md](../docs/LANGUAGES.md) for setup in each language. diff --git a/examples/c/example.c b/examples/c/example.c new file mode 100644 index 0000000..cf07d13 --- /dev/null +++ b/examples/c/example.c @@ -0,0 +1,61 @@ +/* + * example.c — load, inspect, time and write YAPF images from C. + * + * cc -O2 -std=c99 -I../.. example.c ../../yapf.c -o example -pthread + * ./example ../../other/YAPF.YAPF + */ +#include "yapf.h" + +#include +#include +#include +#include + +static double seconds(void) { return (double)clock() / CLOCKS_PER_SEC; } + +int main(int argc, char **argv) { + const char *path = argc > 1 ? argv[1] : "../../other/YAPF.YAPF"; + + /* 1. Load a file (yapf_load_memory works the same on a buffer). */ + yapf_image_t *img = yapf_load(path); + if (!img) { fprintf(stderr, "cannot load %s\n", path); return 1; } + double t = 1e9; /* decode time: best of 10 */ + for (int i = 0; i < 10; i++) { + double t0 = seconds(); + yapf_free(yapf_load(path)); + t0 = seconds() - t0; + if (t0 < t) t = t0; + } + printf("%s: %u x %u, %u channels, %u mip level(s), decoded in %.2f ms\n", + path, img->width, img->height, img->channels, img->mip_levels, t * 1e3); + + /* 2. Pixels: width * height * channels bytes, rows top to bottom. */ + const uint8_t *p = img->pixels; + printf("top-left pixel: %u %u %u %u\n", p[0], p[1], p[2], img->channels == 4 ? p[3] : 255); + + /* 3. Encode to memory and to a file. */ + void *data = NULL; + size_t size = 0; + if (yapf_encode(img, &data, &size) == YAPF_OK) { + printf("re-encoded: %zu bytes (%.1f%% of raw)\n", size, + 100.0 * size / ((double)img->width * img->height * img->channels)); + yapf_free_buffer(data); + } + + /* 4. Create an image from your own pixels. */ + enum { W = 256, H = 256 }; + static uint8_t rgba[W * H * 4]; + for (int y = 0; y < H; y++) + for (int x = 0; x < W; x++) { + uint8_t *q = rgba + (y * W + x) * 4; + q[0] = (uint8_t)x; q[1] = (uint8_t)y; q[2] = 128; q[3] = 255; + } + yapf_image_t mine; + memset(&mine, 0, sizeof(mine)); + mine.width = W; mine.height = H; mine.channels = 4; mine.mip_levels = 1; + mine.pixels = rgba; mine.flags = YAPF_FLAG_SRGB; mine.gpu_format = YAPF_GPU_SRGB8_A8; + if (yapf_save("gradient.yapf", &mine) == YAPF_OK) printf("wrote gradient.yapf\n"); + + yapf_free(img); + return 0; +} diff --git a/extensions/blender/README.md b/extensions/blender/README.md index cacec89..046d5af 100644 --- a/extensions/blender/README.md +++ b/extensions/blender/README.md @@ -20,7 +20,9 @@ the .blend** and survive saving and reopening. and pick the zip. The zip contains the native YAPF library for Windows, macOS and Linux (x64 -and ARM64). If no library matches your system, the extension still +and ARM64). Tested in Blender 5.1 on Linux; the Windows library is tested +outside Blender, and the macOS libraries are cross-compiled and **not yet run +on a Mac**. If no library matches your system, the extension still **imports** using a pure-Python decoder (numpy, bundled with Blender), but cannot export. diff --git a/extensions/figma/README.md b/extensions/figma/README.md index 379cfb1..487853f 100644 --- a/extensions/figma/README.md +++ b/extensions/figma/README.md @@ -30,3 +30,10 @@ node build.js ``` Figma limits images to 4096 × 4096; larger `.yapf` files report an error. + +## Status + +The plugin UI was tested in a browser engine (headless Chromium) with a +simulated Figma main thread: import hands Figma a pixel-exact PNG, and +export downloads a pixel-exact `.yapf`. `code.js` is syntax-checked. +**It has not yet been run inside Figma itself.** diff --git a/extensions/kde/CMakeLists.txt b/extensions/kde/CMakeLists.txt new file mode 100644 index 0000000..453c78e --- /dev/null +++ b/extensions/kde/CMakeLists.txt @@ -0,0 +1,26 @@ +# YAPF thumbnails for KDE Dolphin (KF6 KIO thumbnail plugin). +# cmake -B build -DCMAKE_BUILD_TYPE=Release +# cmake --build build +# sudo cmake --install build (or see README.md for a per-user install) +cmake_minimum_required(VERSION 3.16) +project(yapfthumbnail LANGUAGES C CXX) + +find_package(ECM 6.0 REQUIRED NO_MODULE) +set(CMAKE_MODULE_PATH ${ECM_MODULE_PATH}) +include(KDEInstallDirs) +include(KDECMakeSettings) +include(KDECompilerSettings NO_POLICY_SCOPE) + +find_package(Qt6 6.5 REQUIRED COMPONENTS Core Gui) +find_package(KF6 6.0 REQUIRED COMPONENTS KIO) +find_package(Threads REQUIRED) + +kcoreaddons_add_plugin(yapfthumbnail + SOURCES yapfthumbnail.cpp ../../yapf.c + INSTALL_NAMESPACE "kf6/thumbcreator") +set_source_files_properties(../../yapf.c PROPERTIES LANGUAGE C) +target_include_directories(yapfthumbnail PRIVATE ../..) +target_link_libraries(yapfthumbnail PRIVATE KF6::KIOGui Qt6::Gui Threads::Threads) + +# Register the image/x-yapf file type so Dolphin recognises .yapf files. +install(FILES ../linux/yapf-mime.xml DESTINATION ${KDE_INSTALL_MIMEDIR} RENAME yapf.xml) diff --git a/extensions/kde/README.md b/extensions/kde/README.md new file mode 100644 index 0000000..6283557 --- /dev/null +++ b/extensions/kde/README.md @@ -0,0 +1,50 @@ +# YAPF thumbnails for KDE (Dolphin) + +Shows previews of `.yapf` files in **Dolphin**, the KDE file dialogs and +anything else that uses KIO thumbnails. KDE Frameworks 6 / Plasma 6. +Files with mipmaps are thumbnailed from the nearest mip level, so even huge +textures preview instantly. + +## Build and install + +You need CMake, a C/C++ compiler and the KDE development packages: + +| System | Packages | +| :-------------- | :------- | +| Arch / Manjaro | `sudo pacman -S cmake extra-cmake-modules kio` | +| Debian / Ubuntu | `sudo apt install cmake extra-cmake-modules libkf6kio-dev qt6-base-dev` | +| Fedora | `sudo dnf install cmake extra-cmake-modules kf6-kio-devel qt6-qtbase-devel` | +| openSUSE | `sudo zypper install cmake kf6-extra-cmake-modules kf6-kio-devel qt6-base-devel` | +| NixOS | `nix-shell kde-shell.nix` (in this folder) | + +```sh +cd extensions/kde +cmake -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr +cmake --build build +sudo cmake --install build +update-mime-database /usr/share/mime +``` + +This installs the plugin (`kf6/thumbcreator/yapfthumbnail.so`) and the +`image/x-yapf` file type. Then in Dolphin: **Settings → Configure Dolphin → +Interface → Previews**, tick **YAPF Images**, and turn previews on. + +### Without root (per user, also NixOS) + +```sh +cmake -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=$HOME/.local +cmake --build build && cmake --install build +update-mime-database ~/.local/share/mime +mkdir -p ~/.config/plasma-workspace/env +echo 'export QT_PLUGIN_PATH="$HOME/.local/lib/plugins:$QT_PLUGIN_PATH"' > ~/.config/plasma-workspace/env/yapf.sh +``` + +Log out and back in, then enable **YAPF Images** in Dolphin's preview settings. +(Some distributions use `lib64/plugins` or `lib/qt6/plugins`; `cmake +--install` prints the exact folder.) + +## Status + +Built against KF6 / Qt 6 and tested by loading the plugin through KDE's +plugin factory and creating thumbnails (RGBA, RGB, mipmapped, wide images; +corrupt files are rejected). Not yet tried inside a running Dolphin. diff --git a/extensions/kde/kde-shell.nix b/extensions/kde/kde-shell.nix new file mode 100644 index 0000000..bcb1544 --- /dev/null +++ b/extensions/kde/kde-shell.nix @@ -0,0 +1,6 @@ +# Development shell for building the plugin on NixOS: nix-shell kde-shell.nix +with import {}; +mkShell { + nativeBuildInputs = [ cmake kdePackages.extra-cmake-modules pkg-config ]; + buildInputs = [ kdePackages.kio kdePackages.qtbase kdePackages.kcoreaddons ]; +} diff --git a/extensions/kde/yapfthumbnail.cpp b/extensions/kde/yapfthumbnail.cpp new file mode 100644 index 0000000..6842a65 --- /dev/null +++ b/extensions/kde/yapfthumbnail.cpp @@ -0,0 +1,79 @@ +// YAPF thumbnails for KDE Dolphin and other KIO file dialogs. +// Copyright 2026 Nexoniarz — Apache License 2.0. + +#include +#include +#include + +extern "C" { +#include "yapf.h" +} + +class YapfThumbnailCreator : public KIO::ThumbnailCreator +{ + Q_OBJECT +public: + YapfThumbnailCreator(QObject *parent, const QVariantList &args) + : KIO::ThumbnailCreator(parent, args) + { + } + + KIO::ThumbnailResult create(const KIO::ThumbnailRequest &request) override + { + const QByteArray path = request.url().toLocalFile().toLocal8Bit(); + yapf_image_t *img = yapf_load(path.constData()); + if (!img) { + return KIO::ThumbnailResult::fail(); + } + + // Start from the smallest stored mip level that still covers the + // requested size; Qt scales the rest. + const QSize want = request.targetSize(); + int m = 0; + while (m + 1 < img->mip_levels && + int(img->width >> (m + 1)) >= want.width() && + int(img->height >> (m + 1)) >= want.height()) { + ++m; + } + const int w = qMax(1, int(img->width >> m)); + const int h = qMax(1, int(img->height >> m)); + const uchar *px = img->mips[m]; + + QImage image; + switch (img->channels) { + case 1: + image = QImage(px, w, h, w, QImage::Format_Grayscale8).copy(); + break; + case 2: { // gray + alpha → RGBA + image = QImage(w, h, QImage::Format_RGBA8888); + for (int y = 0; y < h; ++y) { + const uchar *s = px + size_t(y) * w * 2; + uchar *d = image.scanLine(y); + for (int x = 0; x < w; ++x, s += 2, d += 4) { + d[0] = d[1] = d[2] = s[0]; + d[3] = s[1]; + } + } + break; + } + case 3: + image = QImage(px, w, h, w * 3, QImage::Format_RGB888).copy(); + break; + default: + image = QImage(px, w, h, w * 4, + (img->flags & YAPF_FLAG_PREMULT_ALPHA) ? QImage::Format_RGBA8888_Premultiplied + : QImage::Format_RGBA8888).copy(); + break; + } + yapf_free(img); + + if (image.width() > want.width() || image.height() > want.height()) { + image = image.scaled(want, Qt::KeepAspectRatio, Qt::SmoothTransformation); + } + return KIO::ThumbnailResult::pass(image); + } +}; + +K_PLUGIN_CLASS_WITH_JSON(YapfThumbnailCreator, "yapfthumbnail.json") + +#include "yapfthumbnail.moc" diff --git a/extensions/kde/yapfthumbnail.json b/extensions/kde/yapfthumbnail.json new file mode 100644 index 0000000..4135ab1 --- /dev/null +++ b/extensions/kde/yapfthumbnail.json @@ -0,0 +1,7 @@ +{ + "CacheThumbnail": true, + "KPlugin": { + "MimeTypes": ["image/x-yapf"], + "Name": "YAPF Images" + } +} diff --git a/extensions/vscode/README.md b/extensions/vscode/README.md index d004122..b0f3da0 100644 --- a/extensions/vscode/README.md +++ b/extensions/vscode/README.md @@ -24,4 +24,11 @@ Decoding runs in plain JavaScript inside the preview; no native code. code --install-extension yapf-preview-1.0.0.vsix ``` -Works in VS Code, VSCodium and other VS Code–based editors (1.75+). +Should work in VS Code 1.75+ and VS Code–based editors such as VSCodium. + +## Status + +The `.vsix` packages cleanly, and the preview code was tested in a browser +engine (headless Chromium) with real files: correct pixels, mip picker, +error message for corrupt files. **It has not yet been run inside VS Code +itself.** diff --git a/tests/test_yapf.c b/tests/test_yapf.c index ed49421..2412684 100644 --- a/tests/test_yapf.c +++ b/tests/test_yapf.c @@ -44,18 +44,39 @@ static void fill(uint8_t *p, int w, int h, int ch, int kind) { } static uint8_t *encode_mem(const yapf_image_t *img, size_t *n) { + void *data = NULL; + if (yapf_encode(img, &data, n) != YAPF_OK) return NULL; + uint8_t *copy = (uint8_t *)malloc(*n); + if (copy) memcpy(copy, data, *n); + yapf_free_buffer(data); + return copy; +} + +/* yapf_save() writes exactly what yapf_encode() returns. */ +static void save_matches_encode(void) { const char *tmp = "test_yapf_tmp.yapf"; - if (yapf_save(tmp, img) != YAPF_OK) return NULL; - FILE *f = fopen(tmp, "rb"); - if (!f) return NULL; - fseek(f, 0, SEEK_END); - *n = (size_t)ftell(f); - rewind(f); - uint8_t *b = (uint8_t *)malloc(*n); - if (b && fread(b, 1, *n, f) != *n) { free(b); b = NULL; } - fclose(f); + uint8_t px[37 * 23 * 3]; + fill(px, 37, 23, 3, 3); + yapf_image_t img; + memset(&img, 0, sizeof(img)); + img.width = 37; img.height = 23; img.channels = 3; img.mip_levels = 1; img.pixels = px; + size_t len = 0; + uint8_t *mem = encode_mem(&img, &len); + int ok = mem && yapf_save(tmp, &img) == YAPF_OK; + FILE *f = ok ? fopen(tmp, "rb") : NULL; + if (f) { + uint8_t *disk = (uint8_t *)malloc(len + 1); + ok = fread(disk, 1, len + 1, f) == len && memcmp(disk, mem, len) == 0; + free(disk); + fclose(f); + } remove(tmp); - return b; + free(mem); + CHECK(ok, "yapf_save writes the same bytes as yapf_encode"); + void *d = (void *)1; + size_t n = 1; + CHECK(yapf_encode(NULL, &d, &n) == YAPF_ERR_INVALID && d == NULL && n == 0, + "yapf_encode rejects a NULL image"); } static void roundtrip(int w, int h, int ch, int kind) { @@ -162,6 +183,7 @@ int main(int argc, char **argv) { roundtrip(sizes[s][0], sizes[s][1], ch, kind); mips(); corruption(); + save_matches_encode(); const char *sample = argc > 1 ? argv[1] : "other/YAPF.YAPF"; yapf_image_t *img = yapf_load(sample); diff --git a/tools/build_release.sh b/tools/build_release.sh index 1f78f67..540c332 100755 --- a/tools/build_release.sh +++ b/tools/build_release.sh @@ -11,6 +11,7 @@ # yapf-figma.zip Figma plugin (import via manifest.json) # yapf-photoshop.zip Photoshop scripts # yapf-gimp-source.zip GIMP plugin source (build with gimptool) +# yapf-kde-source.zip Dolphin thumbnail plugin source (build with CMake) set -euo pipefail cd "$(dirname "$0")/.." OUT=dist/release @@ -55,4 +56,5 @@ fi python3 -m zipfile -c "$OUT/yapf-figma.zip" extensions/figma/manifest.json extensions/figma/code.js extensions/figma/ui.html python3 -m zipfile -c "$OUT/yapf-photoshop.zip" extensions/photoshop python3 -m zipfile -c "$OUT/yapf-gimp-source.zip" extensions/gimp yapf.c yapf.h +python3 -m zipfile -c "$OUT/yapf-kde-source.zip" extensions/kde extensions/linux/yapf-mime.xml yapf.c yapf.h ls -la "$OUT" diff --git a/tools/yapf_cli.c b/tools/yapf_cli.c index 7571436..effec05 100644 --- a/tools/yapf_cli.c +++ b/tools/yapf_cli.c @@ -21,6 +21,7 @@ * yapf info print header information * yapf bench time decoding * yapf thumbnail [size] small preview (default 256) + * yapf compare [more…] size and decode speed: original vs YAPF * * Reads PNG, JPEG, BMP, TGA, GIF (first frame), PSD (composite), PNM, HDR * (tone-clamped) and YAPF. Writes YAPF, PNG, BMP, TGA and JPEG. @@ -249,12 +250,119 @@ static int cmd_thumbnail(const char *in, const char *out, int size) { return ok ? 0 : 1; } +/* Best time over several rounds of a decode call, for fair small-image timing. */ +#define YAPF_TIME_BEST(result, expr) \ + do { \ + double one_ = now_s(); { expr; } one_ = now_s() - one_; \ + int reps_ = (int)(0.02 / (one_ > 1e-6 ? one_ : 1e-6)) + 1; \ + double best_ = 1e9; \ + for (int r_ = 0; r_ < 9; r_++) { \ + double t_ = now_s(); \ + for (int k_ = 0; k_ < reps_; k_++) { expr; } \ + t_ = (now_s() - t_) / reps_; \ + if (t_ < best_) best_ = t_; \ + } \ + (result) = best_; \ + } while (0) + +static const char *fmt_kb(double bytes, char *buf) { + if (bytes >= 1024.0 * 1024.0) sprintf(buf, "%.2f MB", bytes / (1024.0 * 1024.0)); + else sprintf(buf, "%.1f KB", bytes / 1024.0); + return buf; +} + +static const char *ext_name(const char *path) { + const char *dot = strrchr(path, '.'); + static char name[8]; + size_t i = 0; + if (!dot) return "file"; + for (dot++; *dot && i < sizeof(name) - 1; dot++) name[i++] = (char)toupper((unsigned char)*dot); + name[i] = 0; + return name; +} + +/* Original format vs YAPF: size, decode time (1 thread / all cores), + * and a lossless check. */ +static int cmd_compare(int n, char **paths) { + int rc = 0; + double tot_src = 0, tot_yapf = 0, tot_src_t = 0, tot_yapf_t = 0; + for (int i = 0; i < n; i++) { + const char *path = paths[i]; + if (is_yapf(path)) { + fprintf(stderr, "yapf: compare takes the original image (PNG, JPG, ...), not %s\n", path); + rc = 1; + continue; + } + FILE *f = fopen(path, "rb"); + if (!f) { fprintf(stderr, "yapf: cannot open %s\n", path); rc = 1; continue; } + fseek(f, 0, SEEK_END); + long fsz = ftell(f); + rewind(f); + uint8_t *src = (uint8_t *)malloc((size_t)fsz); + if (!src || fread(src, 1, (size_t)fsz, f) != (size_t)fsz) { fclose(f); free(src); rc = 1; continue; } + fclose(f); + + int w, h, ch; + uint8_t *px = stbi_load_from_memory(src, (int)fsz, &w, &h, &ch, 0); + if (!px) { fprintf(stderr, "yapf: cannot read %s: %s\n", path, stbi_failure_reason()); free(src); rc = 1; continue; } + + yapf_image_t img; + memset(&img, 0, sizeof(img)); + img.width = (uint32_t)w; img.height = (uint32_t)h; img.channels = (uint8_t)ch; + img.mip_levels = 1; img.pixels = px; img.flags = YAPF_FLAG_SRGB; + img.gpu_format = gpu_hint(ch, 1); + + void *enc = NULL; + size_t esz = 0; + double t_enc = now_s(); + int erc = yapf_encode(&img, &enc, &esz); + t_enc = now_s() - t_enc; + if (erc != YAPF_OK) { fprintf(stderr, "yapf: cannot encode %s\n", path); stbi_image_free(px); free(src); rc = 1; continue; } + + yapf_image_t *chk = yapf_load_memory(enc, esz); + int lossless = chk && memcmp(chk->pixels, px, (size_t)w * h * ch) == 0; + yapf_free(chk); + + double t_src, t_one, t_all; + int dw, dh, dc; + YAPF_TIME_BEST(t_src, stbi_image_free(stbi_load_from_memory(src, (int)fsz, &dw, &dh, &dc, 0))); + YAPF_TIME_BEST(t_one, yapf_free(yapf_load_memory(enc, esz))); + YAPF_TIME_BEST(t_all, yapf_free(yapf_load_memory_mt(enc, esz, 0))); + + double raw = (double)w * h * ch; + char a[32], b[32], c[32]; + static const char *chn[] = { "", "gray", "gray+alpha", "RGB", "RGBA" }; + printf("%s (%d x %d, %s)\n", path, w, h, chn[ch]); + printf(" %-5s %10s\n", "raw", fmt_kb(raw, c)); + printf(" %-5s %10s %5.1f%% of raw decode %8.3f ms\n", ext_name(path), fmt_kb((double)fsz, a), + 100.0 * fsz / raw, t_src * 1e3); + printf(" %-5s %10s %5.1f%% of raw decode %8.3f ms %.1fx faster, %.0f%% of the %s size\n", + "YAPF", fmt_kb((double)esz, b), 100.0 * esz / raw, t_one * 1e3, + t_src / t_one, 100.0 * esz / fsz, ext_name(path)); + printf(" YAPF on all cores: decode %.3f ms (%.1f GB/s), encode %.0f ms, %s\n\n", + t_all * 1e3, raw / t_all / 1e9, t_enc * 1e3, + lossless ? "lossless: yes" : "LOSSLESS CHECK FAILED"); + if (!lossless) rc = 1; + tot_src += (double)fsz; tot_yapf += (double)esz; tot_src_t += t_src; tot_yapf_t += t_one; + yapf_free_buffer(enc); + stbi_image_free(px); + free(src); + } + if (n > 1 && tot_yapf_t > 0) { + char a[32], b[32]; + printf("Total: originals %s, YAPF %s (%.0f%%); decoding %.1fx faster on one core\n", + fmt_kb(tot_src, a), fmt_kb(tot_yapf, b), 100.0 * tot_yapf / tot_src, tot_src_t / tot_yapf_t); + } + return rc; +} + static void usage(void) { fprintf(stderr, "usage: yapf [options] convert (formats by extension)\n" " yapf info show header information\n" " yapf bench time decoding\n" " yapf thumbnail [size] preview, default 256 px\n" + " yapf compare [more...] size and speed vs the original format\n" "\n" "reads: .yapf .png .jpg .bmp .tga .gif .psd .pnm .hdr\n" "writes: .yapf .png .jpg .bmp .tga\n" @@ -270,6 +378,7 @@ static void usage(void) { int main(int argc, char **argv) { if (argc == 3 && strcmp(argv[1], "info") == 0) return cmd_info(argv[2]); if (argc == 3 && strcmp(argv[1], "bench") == 0) return cmd_bench(argv[2]); + if (argc >= 3 && strcmp(argv[1], "compare") == 0) return cmd_compare(argc - 2, argv + 2); if ((argc == 4 || argc == 5) && strcmp(argv[1], "thumbnail") == 0) return cmd_thumbnail(argv[2], argv[3], argc == 5 ? atoi(argv[4]) : 256); if (argc < 3) { usage(); return 2; } diff --git a/yapf.c b/yapf.c index f27a52f..d926109 100644 --- a/yapf.c +++ b/yapf.c @@ -193,7 +193,7 @@ static int yapf__width(uint8_t z) { return w; } -static const uint8_t yapf__zero_row[YAPF_TILE_SIZE + 1]; +static const uint8_t yapf__zero_row[YAPF_TILE_SIZE + 1] = { 0 }; /* ── SIMD ────────────────────────────────────────────────────────────── */ /* @@ -1409,11 +1409,14 @@ yapf_image_t *yapf_load_mt(const char *filename, int threads) { } /* ======================================================================== - * yapf_save + * yapf_encode / yapf_save * ======================================================================== */ -int yapf_save(const char *filename, const yapf_image_t *img) { - if (!filename || !img || !img->pixels) return YAPF_ERR_INVALID; +int yapf_encode(const yapf_image_t *img, void **out_data, size_t *out_size) { + if (!out_data || !out_size) return YAPF_ERR_INVALID; + *out_data = NULL; + *out_size = 0; + if (!img || !img->pixels) return YAPF_ERR_INVALID; if (img->width == 0 || img->width > YAPF_MAX_DIM) return YAPF_ERR_INVALID; if (img->height == 0 || img->height > YAPF_MAX_DIM) return YAPF_ERR_INVALID; if (img->channels < 1 || img->channels > 4) return YAPF_ERR_INVALID; @@ -1440,47 +1443,63 @@ int yapf_save(const char *filename, const yapf_image_t *img) { } /* Offsets are 32-bit: refuse files that would not fit. */ - uint32_t offsets[YAPF_MAX_MIPS]; - uint64_t cur = YAPF__HDR_LEN + (uint64_t)mip_levels * YAPF__MIP_ENTRY; + uint64_t total = YAPF__HDR_LEN + (uint64_t)mip_levels * YAPF__MIP_ENTRY; for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) { - offsets[m] = (uint32_t)cur; - cur += bws[m].len; - if (cur > 0xFFFFFFFFull) err = YAPF_ERR_INVALID; + total += bws[m].len; + if (total > 0xFFFFFFFFull) err = YAPF_ERR_INVALID; } - FILE *f = NULL; - if (err == YAPF_OK && !(f = fopen(filename, "wb"))) err = YAPF_ERR_IO; + uint8_t *file = NULL; + if (err == YAPF_OK && !(file = (uint8_t *)malloc((size_t)total))) err = YAPF_ERR_OOM; if (err == YAPF_OK) { - uint8_t hdr[YAPF__HDR_LEN]; - memset(hdr, 0, sizeof(hdr)); - memcpy(hdr, YAPF__MAGIC, 4); - hdr[4] = YAPF__VERSION; - hdr[5] = img->channels; - hdr[6] = img->gpu_format; - hdr[7] = mip_levels; - hdr[8] = img->flags; - yapf__w32le(hdr + 12, img->width); - yapf__w32le(hdr + 16, img->height); - if (fwrite(hdr, 1, YAPF__HDR_LEN, f) != YAPF__HDR_LEN) err = YAPF_ERR_IO; + memset(file, 0, YAPF__HDR_LEN); + memcpy(file, YAPF__MAGIC, 4); + file[4] = YAPF__VERSION; + file[5] = img->channels; + file[6] = img->gpu_format; + file[7] = mip_levels; + file[8] = img->flags; + yapf__w32le(file + 12, img->width); + yapf__w32le(file + 16, img->height); - for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) { - uint8_t entry[YAPF__MIP_ENTRY]; - yapf__w32le(entry + 0, offsets[m]); + size_t pos = YAPF__HDR_LEN + (size_t)mip_levels * YAPF__MIP_ENTRY; + for (uint8_t m = 0; m < mip_levels; m++) { + uint8_t *entry = file + YAPF__HDR_LEN + (size_t)m * YAPF__MIP_ENTRY; + yapf__w32le(entry + 0, (uint32_t)pos); yapf__w32le(entry + 4, (uint32_t)bws[m].len); - if (fwrite(entry, 1, YAPF__MIP_ENTRY, f) != YAPF__MIP_ENTRY) - err = YAPF_ERR_IO; + memcpy(file + pos, bws[m].buf, bws[m].len); + pos += bws[m].len; } - for (uint8_t m = 0; m < mip_levels && err == YAPF_OK; m++) - if (fwrite(bws[m].buf, 1, bws[m].len, f) != bws[m].len) - err = YAPF_ERR_IO; - if (fclose(f) != 0 && err == YAPF_OK) err = YAPF_ERR_IO; + *out_data = file; + *out_size = (size_t)total; } for (uint8_t m = 0; m < mip_levels; m++) free(bws[m].buf); return err; } +void yapf_free_buffer(void *data) { + free(data); +} + +int yapf_save(const char *filename, const yapf_image_t *img) { + if (!filename) return YAPF_ERR_INVALID; + void *data = NULL; + size_t size = 0; + int err = yapf_encode(img, &data, &size); + if (err != YAPF_OK) return err; + + FILE *f = fopen(filename, "wb"); + if (!f) err = YAPF_ERR_IO; + else { + if (fwrite(data, 1, size, f) != size) err = YAPF_ERR_IO; + if (fclose(f) != 0 && err == YAPF_OK) err = YAPF_ERR_IO; + } + yapf_free_buffer(data); + return err; +} + /* ======================================================================== * yapf_free * ======================================================================== */ diff --git a/yapf.h b/yapf.h index 463d689..4e60939 100644 --- a/yapf.h +++ b/yapf.h @@ -232,6 +232,17 @@ yapf_image_t *yapf_load_memory_mt(const void *buffer, size_t size, int threads); */ int yapf_save(const char *filename, const yapf_image_t *img); +/* + * yapf_encode — encode an image into memory. + * On success *out_data points to *out_size bytes of a complete .yapf file; + * release it with yapf_free_buffer(). Same rules and return codes as + * yapf_save() (no YAPF_ERR_IO). + */ +int yapf_encode(const yapf_image_t *img, void **out_data, size_t *out_size); + +/* yapf_free_buffer — release memory returned by yapf_encode(). NULL-safe. */ +void yapf_free_buffer(void *data); + /* * yapf_free — release all memory owned by an image returned by yapf_load(). * NULL-safe.