Niko1221 0d2ee8defe Int3 fix (test programs only; four were already broken on MSVC in v0.2.9, gr_v3_screen is new): setenv / unsetenv / __builtin_popcount shims so a Windows build with tests ON compiles

2026-11-01 10:05:00 +01:00

..

micro

V100: operator fusions for the decode GEMV / MoE paths, bitwise

2026-10-14 07:05:13 +02:00

prompts/standard-v1

Standard benchmark: frozen fixtures (bench/prompts/standard-v1) and the tool that builds them

2026-10-19 21:23:19 +02:00

results

Merge f-022-bench: strata bench run fixes from the first real run

2026-10-20 12:39:58 +02:00

bench_vs_llama.py

bench: Strata vs llama.cpp v0.6.0 head-to-head script and results (3060, P100, 5070 one round)

2026-10-07 01:33:18 +02:00

gdn_chunk_bench.cpp

Int3 fix (test programs only; four were already broken on MSVC in v0.2.9, gr_v3_screen is new): setenv / unsetenv / __builtin_popcount shims so a Windows build with tests ON compiles

2026-11-01 10:05:00 +01:00

gfx906_qsa_query_swizzle_probe.cpp

perf(gfx906): add guarded opt-in query LDS swizzle for batched QSA

2026-10-09 16:48:03 +02:00

profile_corpus.json

tools: make_profile_mix.py (an expert profile from several routing traces, each trace counting the same, with weights), collect_profile_traces.py (the traces of bench/profile_corpus.json's 45 varied prompts through the engine)

2026-10-11 07:58:34 +02:00

qsa_prompt_attn_sm70_bench.cpp

kernels (sm_70): the prompt attention's latency pipeline (mma884 kernel)

2026-10-14 07:05:12 +02:00

test_gfx906_qsa_query_swizzle_guard.py

Fix the gfx906 query-swizzle host guard test: the dispatch expression now spans two lines and reads kernel

2026-10-21 02:57:22 +02:00