nx_f32_matmul_perf.nx

nx_f32_matmul_perf.nx -- MEASURE the sovereign f32 GEMM's throughput (the math-perf rung). Times nx_f32_matmul SOVEREIGNLY -- the clock (sys_now_ms) is read INSIDE the organ around the work (per "monitor the system natively, never a host stopwatch"). Reports measured MFLOP/s + ns per MAC across a few sizes, with adaptive repetition so the timed window is meaningful. HONEST (no wave, no fabricated head-to-head): this is OUR sovereign BITS-UP f32 GEMM -- no hardware FPU, no SIMD, no BLAS. BLAS/cuBLAS are NOT run here (they are 3rd-party, bench-lane only); the literature ceiling is ~tens of GFLOP/s single-thread, so the measured number below is the sovereign BASELINE and the gap to that ceiling is exactly the simd/FPU lock the genealogy names. Sovereign: imports nx_f32_matmul (-> nx_f32) + nx_syscalls. license_tier: ORIGINAL