W0 toy-only synthetic fp16-input matmul; CPU Python loops, not Metal/GPU
captured: 2026-10-22T13:13:49+07:00
platform: macOS-27.0.1-arm64-arm-64bit; Python 3.9.6
uptime/load: 13:13  up 17 days,  4:35, 1 user, load averages: 4.89 4.59 4.51
GPU idle gate: not applicable; this W0 uses CPU-only Python
shape: 32x64 times 64x32; tiles: 8x8x16; repeats: 5
checksum: 0.085937500; C[0,0]: -0.351562500; C[31,31]: 0.172851562
max absolute difference (naive vs tiled): 0
median wall time: naive 8.519 ms; tiled 11.000 ms
Timing is interpreter overhead and a toy-only loop comparison; it is not a GPU speed claim.
