code wiki / (root) / nx_conv_gemm_gate.nx

nx_conv_gemm_gate.nx

buildroot/runtime/nx_conv_gemm_gate.nx

7928 B156 linesdepth 3pulls 3 transitivereach 0 importersview sourcekind gate/prooftopic conv
docsdependenciesstructsconstsfunctions

about

nx_conv_gemm_gate.nx -- wire the COMPUTE LEVER into the diffusion workhorse: f32 Conv2D via im2col->GEMM. Convolution dominates diffusion U-Nets (what sdcpp/GGML run). The fast path (what GGML uses) is im2col then GEMM -- so a 3x3 conv becomes Weight[Cout, Cin*9] x im2col_T[H*W, Cin*9]^T, reusing the FMA-4acc matmul kernel built this session. This compares a NAIVE direct f32 conv (scalar __f32_mul/add) against im2col + FMA-4acc GEMM, BIT-EXACT (small-int f32) + MEASURES the speedup. This is the first sovereign f32 image-gen primitive accelerated by our AVX2/FMA stack -- the concrete start of the sdcpp benchmark path. No hw writes (Rule 26). expect_exit: 0 license_tier: ORIGINAL

dependencies 2 imports · 0 importers

nx_syscalls.nx nx_gate_verdict.nx nx_conv_gemm_gate.nx

imports: nx_syscalls.nxnx_gate_verdict.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main cg_puts sys_write sys_mmap pack4 sys_now_us sys_mmap ↻ sys_clock_gettime_mono fma4dot cg_num sys_mmap ↻ sys_write ↻ gv_ctr sys_mmap ↻ gv_verdict gv_puts sys_write ↻ gv_num sys_mmap ↻ sys_write ↻ sys_munmap gv_journal sys_openat_append sys_mmap ↻ gv_catn sys_mmap ↻ sys_munmap ↻ sys_now_realtime_sec sys_mmap ↻ sys_clock_gettime_real gv_cat sys_write ↻ sys_close sys_munmap ↻ sys_exit

structs

none

consts

none

functions

11func cg_puts(s: *u8) -> i64 { var n: i64=0; while s[n]!=(0 as u8){n=n+1} sys_write(1,s,n); return 0 }
called by 1: main calls 1: sys_write
12func cg_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 }
called by 1: main calls 2: sys_mmapsys_write
13func 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
16func fma4dot(wr: i64, ir: i64, K: i64, a0: *u8, a1: *u8, a2: *u8, a3: *u8) -> i64
called by 1: main
38func main() -> i64