code wiki / _hdl_build / nx_f32_hw_matmul_perf.nx

nx_f32_hw_matmul_perf.nx

buildroot/runtime/_hdl_build/nx_f32_hw_matmul_perf.nx

2507 B58 linesdepth 4pulls 6 transitivereach 0 importersview sourcekind tooltopic f32
docsdependenciesstructsconstsfunctions

about

nx_f32_hw_matmul_perf.nx -- MEASURED head-to-head: software vs hardware f32 GEMM (both sovereign). Times nx_f32_matmul (bits-up SOFTWARE f32) vs nx_f32_hw_matmul (SSE addss/mulss) on the SAME inputs, clock read IN-ORGAN (sys_now_ms). Reports MFLOP/s for each + the speedup. This is a legitimate measured comparison (both are OURS -- no 3rd-party); it quantifies how much the SSE hardware rung buys, and what gap to BLAS remains (packed SIMD addps/mulps + blocking = the next rung). Sovereign: imports nx_f32_matmul + nx_f32_hw_matmul + nx_syscalls. license_tier: ORIGINAL

dependencies 3 imports · 0 importers

nx_f32_matmul.nx nx_f32_hw_matmul.nx nx_syscalls.nx nx_f32_hw_matmul_perf.nx

imports: nx_f32_matmul.nxnx_f32_hw_matmul.nxnx_syscalls.nx

imported by: nobody (leaf or entry point)

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

main hmm_puts sys_write run_pair sys_mmap f32_of time_mm sys_now_ms sys_mmap ↻ sys_clock_gettime_mono nx_f32_matmul nx_f32_hw_matmul hmm_puts ↻ hmm_putn sys_write ↻ sys_mmap ↻ sys_exit

structs

none

consts

11const K_MAGIC_32768: i64 = 32768

functions

14func time_mm(which: i64, A: *i64, B: *i64, C: *i64, N: i64) -> i64
called by 1: run_pair calls 3: sys_now_msnx_f32_matmulnx_f32_hw_matmul
33func run_pair(N: i64) -> i64
called by 1: main calls 5: sys_mmapf32_oftime_mmhmm_putshmm_putn
50func main() -> i64
calls 3: hmm_putsrun_pairsys_exit