0bf66740df- CI: macOS on Apple Silicon and Intel (macos-15, macos-15-intel); a tag v* attaches every package (Linux x64/ARM64, macOS ARM64/x64, Windows x64, the VS Code extension) to its release - vectors: fma is llvm.fmuladd (llvm.fma was a libm call per lane without FMA hardware: 3.5x slower) - examples/simd.fch: a dot product eight at a time, timed against the plain loop - README and the English guides: installing on Linux, macOS and Windows; vectors (SIMD); --cpu; -g is DWARF on Windows too (lldb from llvm-mingw); no more mention of the C++ compiler - comments left from the C++ compiler (src/target.cpp, src/codegen.cpp, "never frees") brought up to date Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Psc42moPLwP43Cmb9oZqSR
2.5 KiB
The seed
Finch's compiler is written in Finch. To build it you need something that already runs Finch: the seed, which is the compiler's own LLVM IR, one file per system:
| File | System | Built and tested in CI on |
|---|---|---|
linux-x86_64.ll |
Linux x86-64 | ubuntu-24.04 |
linux-aarch64.ll |
Linux ARM64 | ubuntu-24.04-arm |
macos-arm64.ll |
macOS, Apple Silicon | macos-15 |
macos-x86_64.ll |
macOS, Intel | macos-15-intel |
windows-x86_64.ll |
Windows x86-64 (MinGW: llvm-mingw's clang + libclang.dll) | windows-2022 |
Each CI job builds the compiler from its seed, then from itself, runs every test, and packs the result; a tag v* attaches
the packages to the release.
bootstrap/build.sh compiles the seed with clang (and the runtime, and libclang) into build/finch0; finch0 builds
src/ into finch1; finch1 builds src/ again into build/finch; the two must make the same IR, or the script stops.
Why one per system
The seed includes how this system's C compiler passes structs by value, because the compiler calls libclang with
CXCursor, CXType and CXString (the calling convention is different on x86-64, on ARM64 and on Windows). So a seed
made for one system doesn't run on another, and a seed needs the target system's C headers to be made.
When to make a new one
When the compiler's own source starts to use a feature that the committed seed doesn't have, build.sh fails at the first
step. Make the seed from a working compiler:
bootstrap/build.sh seed # writes bootstrap/<this system>.ll
For another system, from a machine that has that system's headers and a cross compiler (what tests/cross.sh uses):
FINCH_CC=aarch64-linux-gnu-gcc build/finch ir src/main.fch --target arm64 > bootstrap/linux-aarch64.ll
Windows and macOS
Windows: windows-x86_64.ll is made by bootstrap/winseed.sh on Linux (llvm-mingw's MinGW headers and the headers of an
official LLVM for Windows); bootstrap/winbuild.sh turns it into finch.exe for trying under Wine. On a Windows machine
bootstrap/build.sh does the same from Git Bash: put llvm-mingw's bin in PATH and set FINCH_LIBCLANG (the path of
libclang.dll), FINCH_LLVM_INCLUDE (its include folder) and FINCH_LDFLAGS=-L<folder of libclang.dll>.
macOS: the seeds were made with the macOS 11.3 SDK. brew install llvm@21
gives clang and libclang; set FINCH_CLANG, FINCH_LIBCLANG and FINCH_LLVM_INCLUDE as the CI job does (the
technicians' guide has the lines).