mirror of
https://github.com/Nexoniarz/Finch.git
synced 2026-11-03 00:44:27 +00:00
04f1f94a21- comparisons of vectors give masks (boolxN): select(m, a, b), m.any/all/count, & | ^ !;
f32x8(list, i, fill) and v.store_part(list, i) for the end of a list (masked load/store)
- v.shuffle(constant lanes), v.lo(), v.hi(), concat(a, b), table.lookup(indices):
pshufb / vpshufb on x86 (SSSE3 and up), tbl on ARM64 (checked in the assembly, LLVM 18 and 21)
- acc.dot_add(a, b): i32 lanes gain 4 byte products each, exactly; vpdpbusd with AVX-VNNI or AVX-512 VNNI,
usdot/sdot/udot on ARM64, the sum written out elsewhere (pmaddwd); run here with all three paths agreeing
- @cpu("avx2,fma") fn: LLVM target-features for one function (+evex512 for AVX-512, which LLVM 18+ needs);
cpu_has("avx2"): CPUID + XGETBV on x86, hwcaps on Linux ARM64, sysctl on macOS; a wide vector can't be passed
between functions with different @cpu (its registers differ), and the compiler says so
- the compiler asks clang which features --cpu means (its predefined macros, cached), so dot_add picks the instruction
- Nix's clang wrapper dropped -march=native in nix-shell (reproduced): finch tells it NIX_ENFORCE_NO_NATIVE=0, shell.nix too,
and warns if --cpu native still made no difference
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Psc42moPLwP43Cmb9oZqSR
27 lines
1.4 KiB
Text
27 lines
1.4 KiB
Text
// Moving lanes: shuffle, lo/hi, concat, and a byte table lookup (how 4-bit weights are unpacked)
|
|
|
|
// sixteen 4-bit numbers, two to a byte, made into sixteen bytes through a table of what each nibble means
|
|
fn unpack4(u8x8 packed, i8x16 table) -> i8x16 {
|
|
pairs := packed.shuffle(0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7) // u8x16: each byte twice
|
|
odd := u8x16(0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1) == 1
|
|
nibbles := select(odd, pairs >> 4, pairs & 15) // low nibble first, then the high one
|
|
return table.lookup(nibbles)
|
|
}
|
|
|
|
fn main() {
|
|
v := f32x4(1.0, 2.0, 3.0, 4.0)
|
|
print(v.shuffle(3, 2, 1, 0), v.shuffle(0, 0, 1, 1, 2, 2, 3, 3), v.shuffle(1, 2))
|
|
w := i32x8(0, 1, 2, 3, 4, 5, 6, 7)
|
|
print(w.lo(), w.hi(), concat(w.hi(), w.lo()))
|
|
print(concat(w, w).len, concat(v, v * 10.0).shuffle(0, 4, 1, 5, 2, 6, 3, 7))
|
|
|
|
table := u8x16(10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25)
|
|
print(table.lookup(u8x16(15, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)))
|
|
print(table.lookup(u8x16(16, 17, 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)).lo())
|
|
print(table.lookup(u8x32(3)).hi().lo().lo())
|
|
|
|
q4 := i8x16(-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7) // what each 4-bit number means
|
|
print(unpack4(u8x8(0x10, 0x32, 0x54, 0x76, 0x98, 0xba, 0xdc, 0xfe), q4))
|
|
m := f32x4(1.0, -1.0, 2.0, -2.0) > 0.0
|
|
print(m.shuffle(1, 0, 3, 2), m.lo())
|
|
}
|