igneum/proto-cuda/packs-readwidth/w64
igneum-labs dc84789e34 read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims
Nothing changes for the default class: the pinned packs are byte-identical (tests/packs.rs), the v2 draw stream is untouched.
LoadClass {mix, load_slots, scratch}: fixed widths w16, w64, w64x4 (4 loads of 64 B), era mixes 50/35/15 and 25/50/25 drawn per load with one extra below(100) roll, and the scratch variant scr0/2/4/8 (persistent warps, 1 MiB per warp, tagged lazy fill, measurement only). A wide load reads the W-aligned address and folds every word: x = dst ^ w0; x = (rotl(x, 11) * 0x9e3779b1) ^ w[j]. Program ids carry the class. proto-opencl/host.c taken from opencl-rdna4 23810df (--memprobe, select read-back).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-10-05 19:46:54 +00:00
..
kernel.cl read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
kernel.cu read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
kernel_bound.cl read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
kernel_bound.cu read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
memhard.h read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
memhard.metal read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
program.h read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
program.json read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
program.metal read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
program_bound.metal read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
vectors.h read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00
vectors.json read-width experiment (gate 1): load classes W=4/16/64, per-load width mix, scratch RMW variant behind a generator flag; 20 packs; CPU verifier and acceptance mirror; emulator shims 2026-10-05 19:46:54 +00:00