nx_f32x8_mt_matmul.nx
buildroot/runtime/nx_f32x8_mt_matmul.nx
about
nx_f32x8_mt_matmul.nx -- the FULL sovereign compute stack toward physics: AVX2 8-wide x MULTICORE.
__f32x8_dot (256-bit vmovups+vmulps via .byte VEX) inside forked workers (sys_fork+mmap_shared+wait4).
Measures scalar-ST -> AVX2-ST -> AVX2-MT and the gap to the derived silicon f32 peak (~1.84 TFLOP/s).
This is the deepest "get to exceed" climb with the levers built this session; what remains = FMA (2x,
vfmadd231ps, VEX encoder already proven) + better multicore scaling. Bit-exact vs scalar.
Sovereign: nx_cc AVX2 + raw fork, NO BLAS/CUDA/OpenMP. No hw writes (Rule 26). expect_exit: 0 tier: ORIGINAL
dependencies 1 imports · 0 importers
imports: nx_syscalls.nx
imported by: nobody (leaf or entry point)
call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown
structs
| none |
consts
| 8 | const K_MAGIC_1842560: i64 = 1842560 |
functions
| 10 | func mp_puts(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 } |
| 11 | func mp_num(v: i64) -> i64 { let b: *u8=sys_mmap(28); var m: i64=v; if m<0{m=0-m;sys_write(1,"-" as *u8,1)} let t: *u8=sys_mmap(28); var k: i64=0; if m==0{t[0]=48 as u8;k=1} while m>0{t[k]=(48+(m%10)) as u8;m=m/10;k=k+1} var i: i64=0; while i<k{b[i]=t[k-1-i];i=i+1} sys_write(1,b,k); return 0 } |
| 12 | func pack4(buf: *u8, idx: i64, bits: i64) -> i64 { buf[idx*4+0]=(bits) as u8; buf[idx*4+1]=(bits>>8) as u8; buf[idx*4+2]=(bits>>16) as u8; buf[idx*4+3]=(bits>>24) as u8; return 0 } called by 1: main |
| 14 | func band_scalar(a: *i64, b: *i64, c: *i64, M: i64, N: i64, K: i64, w: i64, nw: i64) -> i64 called by 1: main |
| 25 | func band_avx2(pa: *u8, pbt: *u8, c: *i64, M: i64, N: i64, K: i64, w: i64, nw: i64) -> i64 called by 1: main |
| 39 | func main() -> i64 |