biondizzle 73f38acf74 STAGE B BUG 4b FIXED: TMEM P/O overlap + FMHA V reconstruction ...

Root cause: PV output O started at TMEM column 64 (from find_tmem_tensor_col_offset),
overlapping with P at columns [32,96). PV MMA reading P while writing O to overlapping
columns corrupted the A operand mid-computation.

For (128,128) PV, O started at 128 (no overlap) so it worked by accident.
For (128,64) PV, O started at 64, overlapping P [32,96) -> NaN/garbage.

Fix: Place O at column 128 (after both S [0,128) and P [32,96)).
Also added FMHA-style V reconstruction: logical (HEAD_DIM, s_k, 1) stride (1, hd, hd*s_k)
instead of passing DLPack V directly to TMA.

test_fmha_v3.py: (128,64) PV with random V -> cosine 0.999999 PASS

2026-05-21 15:30:24 +00:00

..

cudagraph_test.py

fix: test L2 weight N dim should be hidden_size, not hidden_size//2

2026-05-16 19:07:36 +00:00

debug_output.py

Update CURRENT_BUG.md: current status, outstanding garbage output issue, hypotheses

2026-05-17 16:52:40 +00:00

debug_stages.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

diag_layouts.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

diag_tmem.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

layertest.py

Fused SwiGLU epilogue with granularity-8 weight interleave

2026-05-20 04:13:52 +00:00

requirements.txt

test: add standalone layer 0 comparison test (no vLLM, no Docker)

2026-05-16 02:13:18 +00:00

run_test.sh

fix: use setup.py install for CUTLASS extension build

2026-05-16 02:21:17 +00:00

stage_b_debug5.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_128_16_bigP.py

TMEM alias analysis: (128,16) PV broken, (128,128) PV with zero-pad works. Root cause: PV A-fragment layout differs from QK C-fragment layout for (128,16) PV, causing TMEM column mismatch. Using (128,128) PV as workaround.

2026-05-21 09:10:12 +00:00

test_128_16_debug2.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_debug3.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_debug.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_fp16.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_full.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_minimal.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_nogC.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_nopack.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_nosoftmax.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_pAtS.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_pvlayout.py

TMEM alias analysis: (128,16) PV broken, (128,128) PV with zero-pad works. Root cause: PV A-fragment layout differs from QK C-fragment layout for (128,16) PV, causing TMEM column mismatch. Using (128,128) PV as workaround.

2026-05-21 09:10:12 +00:00

test_128_16_pvpack.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_pvwrite.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_qkread.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_smem.py

Debugging TMEM alias for (128,16) PV: zero output confirmed, PV reads from wrong TMEM columns. Need to align softmax P write with PV A-fragment layout.

2026-05-21 09:00:42 +00:00

test_128_16_stepA.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_stepB.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_stepC.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_stepD.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_stepE.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_tiler.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_v8.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_128_16_zeropad.py

TMEM alias analysis: (128,16) PV broken, (128,128) PV with zero-pad works. Root cause: PV A-fragment layout differs from QK C-fragment layout for (128,16) PV, causing TMEM column mismatch. Using (128,128) PV as workaround.

2026-05-21 09:10:12 +00:00

test_128_32_ctafix2.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_128_32_ctafix.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_128_32_native.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_128_32_vdiag.py

Root cause FOUND: V SMEM only holds 1 K-tile (2048 BF16), but PV MMA iterates 8 K-phases. For non-(128,128) PV, most K-phases read wrong V data. Zero-padded V works because V is (128,128) covering all 8 K-phases. FMHA interleaves QK+PV per KV-tile to avoid this.

2026-05-21 09:56:54 +00:00

test_128_32_zeropad.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_128_128_fmha_v.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_128_128_vdiag.py

Root cause FOUND: V SMEM only holds 1 K-tile (2048 BF16), but PV MMA iterates 8 K-phases. For non-(128,128) PV, most K-phases read wrong V data. Zero-padded V works because V is (128,128) covering all 8 K-phases. FMHA interleaves QK+PV per KV-tile to avoid this.

2026-05-21 09:56:54 +00:00

test_afrag_roundtrip.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_attention_path_b200.py

Add attention path test: pinpoint FlashMLA failure

2026-05-19 07:54:01 +00:00

test_attention.py

Add NVFP4 linear runner + attention projection test

2026-05-18 20:14:03 +00:00

test_b_afrag2.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_b_layout.py

cleanup: move useful tests to tests/, nuke stale debug tests

2026-05-16 02:14:37 +00:00

test_bf16_elemwise.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_bf16_pack.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_bf16_recast_full.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_bf16_recast_simple.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_blackwell_attn_b200.py

Add blackwell_attention module and comprehensive test

2026-05-19 15:30:29 +00:00

test_compile_custom_op.py

Fix compile test: add warmup for activation global scales

2026-05-19 01:57:16 +00:00

test_csa_attention_b200.py

Add CSA/HCA attention kernel (PyTorch SDPA, Blackwell-safe)

2026-05-19 07:58:10 +00:00

test_csa_sparse_attn_b200.py

Fix N for C128A (need 128 tokens)

2026-05-19 16:04:53 +00:00

test_custom_op.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

test_cutedsl.py

fix: B tensor K-major strides, scale_b axis swap

2026-05-16 03:04:31 +00:00

test_decode_attention_b200.py

Fix attention for decode (1 query vs N cached KVs)

2026-05-19 15:28:52 +00:00

test_decode_pipeline.py

feat: GPU-native SWA + sparse decode attention kernels (CuTeDSL)

2026-05-20 05:46:15 +00:00

test_decode_vs_prefill_b200.py

Add decode vs prefill consistency test

2026-05-19 16:00:33 +00:00

test_diag_layout.py

more stuff

2026-05-21 05:08:57 +00:00

test_diag_permute.py

more stuff

2026-05-21 05:08:57 +00:00

test_diag_smem_layout.py

more stuff

2026-05-21 05:08:57 +00:00

test_diag_v_mma128.py

more stuff

2026-05-21 05:08:57 +00:00

test_diag_v_ones.py

more stuff

2026-05-21 05:08:57 +00:00

test_diag_v_truncid.py

more stuff

2026-05-21 05:08:57 +00:00

test_e2e_decode_b200.py

Test with all 61 layers (shared experts only)

2026-05-19 15:55:41 +00:00

test_error_pattern.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_fmha_pipeline.py

FMHA v1: pv_mma_tiler=(128,64,128) works with V=I, fails with real V (SMEM layout bug)

2026-05-21 10:47:46 +00:00

test_fmha_v1.py

stuff and stuff

2026-05-21 10:50:30 +00:00

test_fmha_v2_fixed.py

Stage B: PV(128,64) test + v2 pipeline fixes

2026-05-21 11:49:06 +00:00

test_fmha_v2.py

FMHA v1: pv_mma_tiler=(128,64,128) works with V=I, fails with real V (SMEM layout bug)

2026-05-21 10:47:46 +00:00

test_fmha_v3_debug.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_fmha_v3.py

STAGE B BUG 4b FIXED: TMEM P/O overlap + FMHA V reconstruction

2026-05-21 15:30:24 +00:00

test_fp4_roundtrip.py

chore: deprecate prepare_weights_from_dequantized and prepare_weights_direct

2026-05-20 02:11:40 +00:00

test_full_layer_b200.py

Fix checkpoint keys: attn_hc.*, compressor.*, q_a_proj/q_b_proj/kv_proj

2026-05-19 07:17:37 +00:00

test_full_layer_nan_b200.py

Add full layer NaN test (attention + MoE, multi-layer chain)

2026-05-19 18:36:49 +00:00

test_full_model_b200.py

Add full model forward test (WIP), sparse attention test passes

2026-05-19 09:04:19 +00:00

test_fused_step1.py

wip: add run_fused_swiglu_grouped_gemm bridge + step1 test

2026-05-20 03:10:56 +00:00

test_inspect_types.py

Stage B progress: PV works for square (128,128), broken for (128,64)

2026-05-21 04:40:28 +00:00

test_interleave_gemm.py

wip: fused SwiGLU kernel scaffold + bridge interleave + plan

2026-05-20 03:04:38 +00:00

test_interleave.py

wip: fused SwiGLU kernel scaffold + bridge interleave + plan

2026-05-20 03:04:38 +00:00

test_inv_rope.py

Add unit tests for NVFP4 weight mapper and inverse RoPE BF16

2026-05-19 03:22:00 +00:00

test_kv_cache_b200.py

Fix kv_ref transpose in KV cache test

2026-05-19 08:58:46 +00:00

test_layout_compare.py

Stage B progress: PV works for square (128,128), broken for (128,64)

2026-05-21 04:40:28 +00:00

test_mma_si_only.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_mma_si_pv.py

Stage B progress: PV works for square (128,128), broken for (128,64)

2026-05-21 04:40:28 +00:00

test_model_forward_b200.py

Rewrite test: diagnose whether warmup gs matters at inference time

2026-05-19 07:49:41 +00:00

test_moe_nan_b200.py

Fix intermediate size: 3072 not 18432

2026-05-19 18:34:12 +00:00

test_moe_runner_nan_b200.py

Use 16 experts for MoE runner test (fits in memory)

2026-05-19 18:35:40 +00:00

test_multilayer.py

Add MoE scale ratio output

2026-05-17 22:58:27 +00:00

test_nvfp4_attention_b200.py

Fix cos_sin cache shape in NVFP4 attention test

2026-05-19 08:38:55 +00:00

test_nvfp4_attn_gemm_b200.py

Fix NVFP4 attention: slice output to actual N after 128-padding

2026-05-19 08:55:31 +00:00

test_nvfp4_mapper.py

Fix hc_head mapping: checkpoint uses hc_head.hc_fn, model params are flat hc_head_fn

2026-05-19 03:58:25 +00:00

test_o_projection_b200.py

Fix dims: o_groups=16, o_lora_rank=1024 from config

2026-05-19 06:37:25 +00:00

test_o_projection.py

Patch attention forward: BF16 inv RoPE + BMM wo_a + NVFP4 wo_b

2026-05-19 06:30:18 +00:00

test_packing_diag.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_pair_swap2.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_pair_swap.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_pipeline_real_weights.py

Pipeline test: use max_num_tokens=8192 matching vLLM

2026-05-17 23:04:44 +00:00

test_pv64_fmha_v.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_pv64_kmajor_v.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_pv64_no_softmax.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_pv64_nosoftmax_fmha_v.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_pv64_with_softmax.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_pv64.py

Stage B: PV(128,64) test + v2 pipeline fixes

2026-05-21 11:49:06 +00:00

test_pv_diag.py

FOOTGUN #0: num_tma_load_bytes MUST include V bytes. Fix v27, v29, comment all. Update README.

2026-05-21 07:13:14 +00:00

test_pv_mma_mn_major.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_qk_softmax.py

FMHA v3: KV-tile interleaving pipeline - QK works, Bug 4b blocks PV

2026-05-21 12:52:29 +00:00

test_qkonly.py

FMHA v3: KV-tile interleaving pipeline - QK works, Bug 4b blocks PV

2026-05-21 12:52:29 +00:00

test_quick_rand.py

cleanup: move useful tests to tests/, nuke stale debug tests

2026-05-16 02:14:37 +00:00

test_recast_minimal.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_rope_kv_b200.py

Fix syntax in RoPE KV test

2026-05-19 10:31:07 +00:00

test_runner_vs_pipeline.py

test: runner vs pipeline comparison + scale assembly comparison

2026-05-17 07:33:20 +00:00

test_scale_assembly.py

fix: separate L1/L2 scale buffers (different K_sf), fix assembly calls

2026-05-17 07:43:05 +00:00

test_scale_debug.py

test: scale assembly debug

2026-05-17 07:37:47 +00:00

test_shared_expert.py

Fix hidden_size: shared expert uses 7168, not HC_DIM 28672

2026-05-18 20:10:32 +00:00

test_silu_step1.py

wip: add run_fused_swiglu_grouped_gemm bridge + step1 test

2026-05-20 03:10:56 +00:00

test_softmax_only.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_softmax_store_debug.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_sparse_attn_b200.py

Add CSA/HCA sparse attention kernel test

2026-05-19 09:02:12 +00:00

test_sparse_decode.py

feat: GPU-native SWA + sparse decode attention kernels (CuTeDSL)

2026-05-20 05:46:15 +00:00

test_stage_a_copy.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_minimal.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_pv_created.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_pv_param.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_qk.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_v2.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_a_with_pv_mma.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_afrag2.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_afrag.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_debug2.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_debug3.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_debug4.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_debug.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_diag.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_final.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_identity.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_minimal.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_ntile_v1.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_ntile_v3.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_ntile_v5.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_ntile_v6.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_ntile_v7.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_ntile_v8.py

Stage B N-tiling: (128,16) PV MMA compiles and runs, cosine 0.36 (TMEM alias mismatch bug). FMHA head_dim=64 passes. Debugging TMEM layout alignment.

2026-05-21 08:45:49 +00:00

test_stage_b_pipeline_only.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v1.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v2.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v3.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v4.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v5.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v6.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_stage_b_v7_rep8.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v7_rep16.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v7_rep64.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v7_rep128.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v7.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v8.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v8b.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v9.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v10.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v11.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v11b.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v12.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_stage_b_v13.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v14.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v16.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v17.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v18.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v19.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v20.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v22_bug1fix.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v22.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v23.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v24.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v25.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v26.py

Stage B: pipeline deadlock fixed, V MN-major applied, PV output garbage

2026-05-21 04:10:07 +00:00

test_stage_b_v27.py

FOOTGUN #0: num_tma_load_bytes MUST include V bytes. Fix v27, v29, comment all. Update README.

2026-05-21 07:13:14 +00:00

test_stage_b_v28.py

v28 attempt: PV MMA (128,64) - cosine 0.004, debugging

2026-05-21 05:41:44 +00:00

test_stage_b_v29.py

FOOTGUN #0: num_tma_load_bytes MUST include V bytes. Fix v27, v29, comment all. Update README.

2026-05-21 07:13:14 +00:00

test_stage_b_v30.py

v29 (padded V, deadlocks), v30 (diag copy, works) — debugging epilogue deadlock with (128,128) PV

2026-05-21 06:40:27 +00:00

test_step2_subtile_v2.py

wip: Step 2 gate/up pairing — SiLU validated, runtime conditionals blocked by CuTeDSL

2026-05-20 03:26:20 +00:00

test_step2_subtile.py

wip: Step 2 gate/up pairing — SiLU validated, runtime conditionals blocked by CuTeDSL

2026-05-20 03:26:20 +00:00

test_store_verify2.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_store_verify.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_tmem_addressing.py

Stage B: two MMAs + identity softmax — crash fixed, softmax output still wrong

2026-05-20 20:26:25 +00:00

test_tmem_col2.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col3.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col4.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col5_16.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col5_32.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col5.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_col_offset.py

Key finding: PV A-fragment layout is IDENTICAL for (128,128)/(128,32)/(128,16) PV. Bug is NOT TMEM alias. cta_tile_shape_mnk wrong for non-(128,128) PV. V SMEM and O C-fragment sizes look correct. Debugging V/epilogue paths.

2026-05-21 09:44:22 +00:00

test_tmem_copy_roundtrip.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_tmem_debug2.py

TMEM alias analysis: (128,16) PV broken, (128,128) PV with zero-pad works. Root cause: PV A-fragment layout differs from QK C-fragment layout for (128,16) PV, causing TMEM column mismatch. Using (128,128) PV as workaround.

2026-05-21 09:10:12 +00:00

test_tmem_debug.py

TMEM alias analysis: (128,16) PV broken, (128,128) PV with zero-pad works. Root cause: PV A-fragment layout differs from QK C-fragment layout for (128,16) PV, causing TMEM column mismatch. Using (128,128) PV as workaround.

2026-05-21 09:10:12 +00:00

test_tmem_fp32_roundtrip.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_tmem_layout_diag.py

Update README: Bug 4 status, (128,16) PV zero output, (128,128) PV zero-pad workaround (cosine 1.0)

2026-05-21 09:20:09 +00:00

test_tmem_pure_fp32.py

Stage B: C-fragment vs A-fragment TMEM layout mismatch diagnosed

2026-05-21 00:12:47 +00:00

test_uniform_fp4.py

cleanup: move useful tests to tests/, nuke stale debug tests

2026-05-16 02:14:37 +00:00

test_v4_attention_b200.py

Add DeepSeek-V4 CSA/HCA attention pipeline test (not MLA)

2026-05-19 08:51:16 +00:00

test_v28c_noepi.py

v29 (padded V, deadlocks), v30 (diag copy, works) — debugging epilogue deadlock with (128,128) PV

2026-05-21 06:40:27 +00:00

test_v_mode_fix.py

Stage B Bug 4b debugging: P/A alias proven working, V layout issue for (128,64) PV

2026-05-21 15:20:14 +00:00

test_vllm_codepaths_b200.py

Fix imports in vLLM codepaths test

2026-05-19 17:26:50 +00:00

test_vsmem_diag.py

Root cause FOUND: V SMEM only holds 1 K-tile (2048 BF16), but PV MMA iterates 8 K-phases. For non-(128,128) PV, most K-phases read wrong V data. Zero-padded V works because V is (128,128) covering all 8 K-phases. FMHA interleaves QK+PV per KV-tile to avoid this.

2026-05-21 09:56:54 +00:00

test_warmup_gs.py

test: use runner's built-in warmup method

2026-05-17 08:24:27 +00:00

test_wo_a_bmm.py

Fix BF16 wo_a: per-group BMM instead of flat linear

2026-05-19 04:10:02 +00:00

test_wo_a.py

Fix test: cos_sin_cache on CUDA device

2026-05-19 02:37:50 +00:00