ray_chan
/
TileOPs-Metax
forked from ccf-ai-infra/TileOPs-Metax
1
0
Fork 1
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity
TileOPs-Metax/benchmarks/ops
History
ray24777 e18a3a1a71 feat(quant_swiglu): add fused SwiGLU + per-channel FP8 cast op 
Implement QuantSwiGLUFwdChannelCastTransposeOp and flip its manifest status
to implemented: fused SwiGLU forward (silu(gate)*value) + per-token-block
FP8 e4m3 cast with optional transpose, returning a packed FP8 output plus an
fp32 per-block scaling-factor tensor.

Faithful port of swiglu_forward_and_per_channel_cast_and_transpose from
MetaX-MACA/TileKernels-Metax @ 0266ab7 (tile_kernels/quant/). The kernel
tiles tokens x hidden, stages the activation in shared memory, reduces the
per-block absmax (packed-bf16 in the non-transpose path, fp32 fragment in
the transpose path), and emits the quantized output.

- kernel: tileops/kernels/quant_swiglu_channel_cast_transpose.py
- op: tileops/ops/quant_swiglu_channel_cast_transpose.py
- workload, correctness tests, manifest-driven baseline benchmark
- manifest: status spec-only -> implemented, source.kernel_map added

Baseline (MetaX C500, 50% compute slice): 4.7-9.4x faster than the torch
reference; 0.24-0.36 TB/s achieved on roofline bytes (kernel is not yet
bandwidth-saturated). See logs/2026-08-03_baseline_notes.md.

Co-Authored-By: Claude <noreply@anthropic.com>
 		2026-08-03 15:28:30 +00:00
..
attention docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
__init__.py [Refactor] Separate benchmark profiling from unit tests (#224) 2026-02-27 16:46:35 +08:00
bench_ada_layer_norm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_argreduce.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_batch_norm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_binary_elementwise.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_bmm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_convolution.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_cumulative.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_deltanet.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_deltanet_recurrence.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_dropout.py [Refactor][Ops] Infer static metadata from forward inputs (#1734) 2026-07-25 12:05:17 +08:00
bench_elementwise_fp8.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_elementwise_manifest.py [Fix][Benchmark] stop losing results on native failures; derive workload keys from signature (#1745) 2026-07-25 22:11:21 +08:00
bench_engram.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_fft.py [Refactor][Ops] Infer static metadata from forward inputs (#1734) 2026-07-25 12:05:17 +08:00
bench_fp8_lightning_indexer.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_fp8_quant.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_fused_moe_experts.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_gated_deltanet.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_gated_deltanet_prefill.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_gated_deltanet_recurrence.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_gemm.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_gla_chunkwise.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_gla_recurrence.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_group_norm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_grouped_gemm.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_grouped_gemm_baselines.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_independent_elementwise.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_instance_norm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_logical_reduce.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_mamba.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_mamba2_e2e.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_mhc.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_moe_fused_activation.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_moe_fused_moe.py [Feat][Ops] Infer shape metadata from op inputs (#1651) 2026-07-06 17:51:47 +08:00
bench_moe_fused_topk.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_moe_grouped_gemm_nopad.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_moe_kernel_breakdown.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_moe_permute_align.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_moe_permute_nopad.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_moe_shared_fused_moe.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_moe_unpermute.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_norm.py [Chore][Cleanup] repo slimming: dead code, duplicated tests/benches, compat shims, file defragmentation (#1764) 2026-07-26 12:59:48 +08:00
bench_pool.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_quant_swiglu_channel_cast_transpose.py feat(quant_swiglu): add fused SwiGLU + per-channel FP8 cast op 2026-08-03 15:28:30 +00:00
bench_reduce.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_reduce_multidim.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_rope.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_softmax.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_topk_selector.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_unary_elementwise.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00
bench_vector_norm.py docs: add MetaX C500 summer-camp guide on top of upstream dev 2026-07-28 17:55:09 +00:00