[Enhancement] Reduce CI Test Time

[LeiWang1999]

We can optimize the test cases by reducing test shapes and optimizing the critical path to shorten CI time.

Example CI Test Time:

=========================================================== fixture duration top ============================================================
total          name                                                                                         num med            min           
       0:00:00 grand total                                                                                    0        0:00:00        0:00:00
========================================================== test call duration top ===========================================================
total          name                                                                                         num med            min           
0:02:29.488783 gemm_sp/test_example_gemm_sp.py::test_example_gemm_sp                                          1 0:02:29.488783 0:02:29.488783
0:01:45.305204 sparse_tensorcore/test_example_sparse_tensorcore.py::test_tilelang_example_sparse_tensorcore   1 0:01:45.305204 0:01:45.305204
0:01:18.967367 dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemm_fp4_hopper          1 0:01:18.967367 0:01:18.967367
0:01:00.995997 gemm/test_example_gemm.py::test_example_gemm_schedule                                          1 0:01:00.995997 0:01:00.995997
0:00:28.190835 attention_sink/test_example_attention_sink.py::test_example_gqa_sink_bwd_bhsd                  1 0:00:28.190835 0:00:28.190835
0:00:26.493459 deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_sparse_mla_bwd                1 0:00:26.493459 0:00:26.493459
0:00:25.231914 attention_sink/test_example_attention_sink.py::test_example_mha_sink_bwd_bhsd                  1 0:00:25.231914 0:00:25.231914
0:00:24.066198 flash_attention/test_example_flash_attention.py::test_example_gqa_bwd_tma_reduce_varlen        1 0:00:24.066198 0:00:24.066198
0:00:23.609148 flash_attention/test_example_flash_attention.py::test_example_gqa_bwd                          1 0:00:23.609148 0:00:23.609148
0:00:23.046067 flash_attention/test_example_flash_attention.py::test_example_mha_bwd                          1 0:00:23.046067 0:00:23.046067
0:00:22.868651 gemv/test_example_gemv.py::test_example_gemv                                                   1 0:00:22.868651 0:00:22.868651
0:00:22.039937 flash_attention/test_example_flash_attention.py::test_example_mha_bwd_bhsd                     1 0:00:22.039937 0:00:22.039937
0:00:20.711386 flash_attention/test_example_flash_attention.py::test_example_gqa_bwd_wgmma_pipelined          1 0:00:20.711386 0:00:20.711386
0:00:18.213203 flash_attention/test_example_flash_attention.py::test_example_mha_bwd_wgmma_pipelined          1 0:00:18.213203 0:00:18.213203
0:00:16.536672 minference/test_vs_sparse_attn.py::test_vs_sparse_attn                                         1 0:00:16.536672 0:00:16.536672
0:00:16.185672 gdn/test_example_gdn_compilation.py::test_example_wy_fast_bwd_split_compilation                1 0:00:16.185672 0:00:16.185672
0:00:15.578780 elementwise/test_example_elementwise.py::test_example_elementwise_add_autotune                 1 0:00:15.578780 0:00:15.578780
0:00:15.572629 attention_sink/test_example_attention_sink.py::test_example_gqa_sink_bwd_bhsd_sliding_window   1 0:00:15.572629 0:00:15.572629
0:00:15.499994 linear_attention/test_linear_attn.py::test_example_linear_attn_bwd                             1 0:00:15.499994 0:00:15.499994
0:00:14.530373 attention_sink/test_example_attention_sink.py::test_example_mha_sink_bwd_bhsd_sliding_window   1 0:00:14.530373 0:00:14.530373
0:00:13.514484 fusedmoe/test_example_fusedmoe.py::test_example_fusedmoe_tilelang                              1 0:00:13.514484 0:00:13.514484
0:00:13.412966 gemm/test_example_gemm.py::test_example_gemm_autotune                                          1 0:00:13.412966 0:00:13.412966
0:00:13.003403 gdn/test_example_gdn_compilation.py::test_example_chunk_o_bwd_compilation                      1 0:00:13.003403 0:00:13.003403
0:00:11.722232 gemm_fp8/test_example_gemm_fp8.py::test_example_tilelang_gemm_fp8_intrinsic                    1 0:00:11.722232 0:00:11.722232
0:00:11.672637 linear_attention/test_linear_attn.py::test_example_linear_attn_fwd                             1 0:00:11.672637 0:00:11.672637
0:00:11.554361 seer_attention/test_block_sparse_attn_tilelang.py::test_block_sparse_attn_tilelang             1 0:00:11.554361 0:00:11.554361
0:00:11.096757 gemm_sp/test_example_gemm_sp.py::test_example_custom_compress                                  1 0:00:11.096757 0:00:11.096757
0:00:10.975390 deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_sparse_mla_fwd_pipelined      1 0:00:10.975390 0:00:10.975390
0:00:10.391669 deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_fp8_lighting_indexer          1 0:00:10.391669 0:00:10.391669
0:00:10.137347 gemm_fp8/test_example_gemm_fp8.py::test_example_tilelang_gemm_fp8                              1 0:00:10.137347 0:00:10.137347
0:19:03.438066 grand total                                                                                   84 0:00:07.041250 0:00:00.025089
========================================================== test setup duration top ==========================================================
total          name                                                                                         num med            min           
0:00:00.009047 grand total                                                                                   88 0:00:00.000092 0:00:00.000063
======================================================== test teardown duration top =========================================================
total          name                                                                                         num med            min           
0:00:00.009374 grand total                                                                                   88 0:00:00.000099 0:00:00.000054
======================================== slowest durations =========================================
149.49s call     gemm_sp/test_example_gemm_sp.py::test_example_gemm_sp
105.31s call     sparse_tensorcore/test_example_sparse_tensorcore.py::test_tilelang_example_sparse_tensorcore
78.97s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemm_fp4_hopper
61.00s call     gemm/test_example_gemm.py::test_example_gemm_schedule
28.19s call     attention_sink/test_example_attention_sink.py::test_example_gqa_sink_bwd_bhsd
26.49s call     deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_sparse_mla_bwd
25.23s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_bwd_bhsd
24.07s call     flash_attention/test_example_flash_attention.py::test_example_gqa_bwd_tma_reduce_varlen
23.61s call     flash_attention/test_example_flash_attention.py::test_example_gqa_bwd
23.05s call     flash_attention/test_example_flash_attention.py::test_example_mha_bwd
22.87s call     gemv/test_example_gemv.py::test_example_gemv
22.04s call     flash_attention/test_example_flash_attention.py::test_example_mha_bwd_bhsd
20.71s call     flash_attention/test_example_flash_attention.py::test_example_gqa_bwd_wgmma_pipelined
18.21s call     flash_attention/test_example_flash_attention.py::test_example_mha_bwd_wgmma_pipelined
16.54s call     minference/test_vs_sparse_attn.py::test_vs_sparse_attn
16.19s call     gdn/test_example_gdn_compilation.py::test_example_wy_fast_bwd_split_compilation
15.58s call     elementwise/test_example_elementwise.py::test_example_elementwise_add_autotune
15.57s call     attention_sink/test_example_attention_sink.py::test_example_gqa_sink_bwd_bhsd_sliding_window
15.50s call     linear_attention/test_linear_attn.py::test_example_linear_attn_bwd
14.53s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_bwd_bhsd_sliding_window
13.51s call     fusedmoe/test_example_fusedmoe.py::test_example_fusedmoe_tilelang
13.41s call     gemm/test_example_gemm.py::test_example_gemm_autotune
13.00s call     gdn/test_example_gdn_compilation.py::test_example_chunk_o_bwd_compilation
11.72s call     gemm_fp8/test_example_gemm_fp8.py::test_example_tilelang_gemm_fp8_intrinsic
11.67s call     linear_attention/test_linear_attn.py::test_example_linear_attn_fwd
11.55s call     seer_attention/test_block_sparse_attn_tilelang.py::test_block_sparse_attn_tilelang
11.10s call     gemm_sp/test_example_gemm_sp.py::test_example_custom_compress
10.98s call     deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_sparse_mla_fwd_pipelined
10.39s call     deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_fp8_lighting_indexer
10.14s call     gemm_fp8/test_example_gemm_fp8.py::test_example_tilelang_gemm_fp8
9.24s call     gemm_fp8/test_example_gemm_fp8.py::test_example_tilelang_gemm_fp8_2xAcc
8.81s call     deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_sparse_mla_fwd
7.91s call     flash_attention/test_example_flash_attention.py::test_example_mha_fwd_varlen
7.90s call     flash_decoding/test_example_flash_decoding.py::test_example_example_mha_inference
7.87s call     blocksparse_attention/test_example_blocksparse_attention.py::test_example_tilelang_sparse_gqa_decode_varlen_mask
7.60s call     deepseek_nsa/test_example_tilelang_nsa.py::test_example_tilelang_nsa_fwd
7.55s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_fwd_bhsd_sliding_window
7.37s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemm_bf16_mxfp4_hopper_tma
7.35s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemm_w4a8
7.29s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_fwd_bhsd_full_attn
7.19s call     gdn/test_example_gdn_compilation.py::test_example_chunk_delta_h_compilation
7.16s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_fwd_bhsd_wgmma_pipelined_sliding_window
6.93s call     gdn/test_example_gdn_compilation.py::test_example_chunk_delta_bwd_compilation
6.87s call     attention_sink/test_example_attention_sink.py::test_example_gqa_sink_fwd_bhsd_wgmma_pipelined_sliding_window
6.78s call     blocksparse_attention/test_example_blocksparse_attention.py::test_example_tilelang_sparse_gqa_decode_varlen_indice
6.73s call     deepseek_mla/test_example_mla_decode.py::test_example_mla_decode
6.70s call     attention_sink/test_example_attention_sink.py::test_example_gqa_sink_fwd_bhsd_wgmma_pipelined_full_attn
6.65s call     attention_sink/test_example_attention_sink.py::test_example_mha_sink_fwd_bhsd_wgmma_pipelined_full_attn
6.48s call     gdn/test_example_gdn_compilation.py::test_example_chunk_o_compilation
6.39s call     blocksparse_attention/test_example_blocksparse_attention.py::test_example_tilelang_block_sparse_attn
6.32s call     flash_attention/test_example_flash_attention.py::test_example_gqa_fwd_bshd_wgmma_pipelined
6.14s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_groupedgemm_bf16_mxfp4_hopper
6.07s call     deepseek_v32/test_tilelang_example_deepseek_v32.py::test_example_topk_selector
5.93s call     flash_attention/test_example_flash_attention.py::test_example_mha_fwd_bhsd_wgmma_pipelined
5.78s call     flash_attention/test_example_flash_attention.py::test_example_mha_fwd_bshd
5.64s call     gdn/test_example_gdn_compilation.py::test_example_wy_fast_compilation
5.62s call     flash_attention/test_example_flash_attention.py::test_example_mha_fwd_bshd_wgmma_pipelined
5.54s call     flash_attention/test_example_flash_attention.py::test_example_gqa_fwd_bshd
5.47s call     cast/test_example_cast.py::test_example_group_per_split_token_cast_to_fp8
5.44s call     flash_attention/test_example_flash_attention.py::test_example_mha_fwd_bhsd
5.32s call     deepseek_nsa/test_example_tilelang_nsa.py::test_example_tilelang_nsa_fwd_decode
5.26s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemm_bf16_mxfp4_hopper
5.08s call     warp_specialize/test_example_warp_specialize.py::test_example_warp_specialize_gemm_copy_1_gemm_0
5.07s call     gemm_splitk/test_example_gemm_splitk.py::test_example_tilelang_gemm_splitk
5.06s call     deepseek_deepgemm/test_example_deepgemm_fp8_2xAcc.py::test_deepgemm_fp8_2xAcc
5.02s call     gdn/test_example_gdn_compilation.py::test_example_chunk_scaled_dot_kkt_compilation
5.01s call     gemm/test_example_gemm.py::test_example_gemm
4.98s call     warp_specialize/test_example_warp_specialize.py::test_example_warp_specialize_gemm_barrierpipe_stage2
4.82s call     gemm_splitk/test_example_gemm_splitk.py::test_example_tilelang_gemm_splitk_vectorize_atomicadd
4.72s call     cast/test_example_cast.py::test_example_per_token_cast_to_fp8
4.71s call     warp_specialize/test_example_warp_specialize.py::test_example_warp_specialize_gemm_copy_0_gemm_1
4.65s call     warp_specialize/test_example_warp_specialize.py::test_example_warp_specialize_gemm_softpipe_stage2
4.62s call     blocksparse_gemm/test_example_blocksparse_gemm.py::test_example_blocksparse_gemm
4.61s call     topk/test_topk_tilelang.py::test_topk_tilelang
4.30s call     blocksparse_attention/test_example_blocksparse_attention.py::test_example_triton_sparse_gqa_decode_varlen_mask
4.23s call     gemm/test_example_gemm.py::test_example_gemm_intrinsics
4.14s call     elementwise/test_example_elementwise.py::test_example_elementwise_add
4.14s call     norm/test_rms_norm.py::test_rms_norm
4.02s call     blocksparse_attention/test_example_blocksparse_attention.py::test_example_triton_sparse_gqa_decode_varlen_indice
4.01s call     dequantize_gemm/test_example_dequantize_gemm.py::test_example_dequant_gemv_fp16xint4
3.72s call     gdn/test_example_gdn_compilation.py::test_example_cumsum_compilation
0.61s call     blocksparse_attention/test_example_blocksparse_attention.py::test_block_sparse_attn_triton
0.03s call     analyze/test_example_analyze.py::test_example_conv_analyze
0.03s call     analyze/test_example_analyze.py::test_example_gemm_analyze

Testing CI Test Time:

================================================================ fixture duration top ================================================================
total          name                                                                                        num  med            min                    
0:00:00.006439 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr                   208 0:00:00.000028          0:00:00.000020
0:00:00.005086 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr                   182 0:00:00.000025          0:00:00.000019
0:00:00.038944 grand total                                                                                 1280 0:00:00.000029          0:00:00.000018
=============================================================== test call duration top ===============================================================
total          name                                                                                        num  med            min                    
0:02:18.910773 testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast       16 0:00:08.538184          0:00:08.029851
0:01:46.428779 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90            12 0:00:09.515094          0:00:00.018699
0:01:29.868598 testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath           21 0:00:04.321056          0:00:03.693514
0:01:27.243296 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr                    16 0:00:05.474963          0:00:04.412709
0:01:15.992348 testing/python/debug/test_tilelang_debug_print.py::test_debug_print_buffer                     1 0:01:15.992348          0:01:15.992348
0:01:14.331675 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr              12 0:00:06.135772          0:00:05.781362
0:01:13.141573 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr                    14 0:00:05.253100          0:00:04.204602
0:01:06.130100 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs              10 0:00:06.671352          0:00:05.074628
0:01:04.966100 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss              11 0:00:05.631887          0:00:04.633975
0:01:02.123817 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr              10 0:00:06.325301          0:00:04.823512
0:00:37.038636 testing/python/jit/test_tilelang_jit_gemm_cython.py::test_cython_dynamic_shape                 1 0:00:37.038636          0:00:37.038636
0:00:35.672505 testing/python/language/test_tilelang_language_reduce.py::test_reduce_sum_clear                1 0:00:35.672505          0:00:35.672505
0:00:32.988014 testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions                       8 0:00:04.070887          0:00:03.906435
0:00:32.091503 testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_return                   1 0:00:32.091503          0:00:32.091503
0:00:28.219543 testing/python/math/test_math_bitwise_reduce.py::test_bitwise_reduce_ops                       1 0:00:28.219543          0:00:28.219543
0:00:26.006352 testing/python/autotune/test_tilelang_autotune.py::test_autotune_matmul                        1 0:00:26.006352          0:00:26.006352
0:00:25.474265 testing/python/language/test_tilelang_language_vectorize.py::test_vectorize_invariant_index    1 0:00:25.474265          0:00:25.474265
0:00:24.857759 testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_many_annot               1 0:00:24.857759          0:00:24.857759
0:00:24.716346 testing/python/language/test_tilelang_language_rand.py::test_rand_1d                           3 0:00:08.257283          0:00:08.135095
0:00:21.285440 testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_region_1d                1 0:00:21.285440          0:00:21.285440
0:00:18.875960 testing/python/language/test_tilelang_language_pipeline.py::test_pipeline_order_stage          1 0:00:18.875960          0:00:18.875960
0:00:18.711309 testing/python/fastmath/test_mathops_fastmath.py::test_two_arg_mathops_fastmath                2 0:00:09.355655          0:00:08.956302
0:00:17.558687 testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_smem                     1 0:00:17.558687          0:00:17.558687
0:00:17.340705 testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_region_2d                1 0:00:17.340705          0:00:17.340705
0:00:17.252213 testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_fragment                 1 0:00:17.252213          0:00:17.252213
0:00:17.227597 testing/python/math/test_math_ieee_math.py::test_ieee_sub_all_rounding_modes                   1 0:00:17.227597          0:00:17.227597
0:00:17.205029 testing/python/math/test_math_ieee_math.py::test_ieee_mul_all_rounding_modes                   1 0:00:17.205029          0:00:17.205029
0:00:17.176580 testing/python/language/test_tilelang_language_vectorize.py::test_vectorize                    1 0:00:17.176580          0:00:17.176580
0:00:17.060915 testing/python/math/test_math_ieee_math.py::test_ieee_fdiv_all_rounding_modes                  1 0:00:17.060915          0:00:17.060915
0:00:17.003212 testing/python/language/test_tilelang_language_view.py::test_reshape_view                      1 0:00:17.003212          0:00:17.003212
0:41:22.062133 grand total                                                                                  527 0:00:03.975783          0:00:00.000128
============================================================== test setup duration top ===============================================================
total          name                                                                                        num  med            min                    
0:00:00.016333 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr                    16 0:00:00.001036          0:00:00.000817
0:00:00.013137 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr                    14 0:00:00.000928          0:00:00.000803
0:00:00.012077 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss              11 0:00:00.000874          0:00:00.000822
0:00:00.010821 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr              12 0:00:00.000838          0:00:00.000811
0:00:00.010257 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs              10 0:00:00.001038          0:00:00.000864
0:00:00.010016 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90            12 0:00:00.000828          0:00:00.000763
0:00:00.008982 testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr              10 0:00:00.000879          0:00:00.000818
0:00:00.008018 testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath           21 0:00:00.000345          0:00:00.000323
0:00:00.007218 testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast       16 0:00:00.000460          0:00:00.000379
0:00:00.135999 grand total                                                                                  613 0:00:00.000092 -1 day, 23:59:59.999441
============================================================= test teardown duration top =============================================================
total          name                                                                                        num  med            min                    
0:00:00.074007 grand total                                                                                  613 0:00:00.000099          0:00:00.000053
======================================== slowest durations =========================================
75.99s call     testing/python/debug/test_tilelang_debug_print.py::test_debug_print_buffer
37.04s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_cython_dynamic_shape
35.67s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_sum_clear
32.09s call     testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_return
28.22s call     testing/python/math/test_math_bitwise_reduce.py::test_bitwise_reduce_ops
26.01s call     testing/python/autotune/test_tilelang_autotune.py::test_autotune_matmul
25.47s call     testing/python/language/test_tilelang_language_vectorize.py::test_vectorize_invariant_index
24.86s call     testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_many_annot
21.29s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_region_1d
18.88s call     testing/python/language/test_tilelang_language_pipeline.py::test_pipeline_order_stage
17.56s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_smem
17.34s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_region_2d
17.25s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_fragment
17.23s call     testing/python/math/test_math_ieee_math.py::test_ieee_sub_all_rounding_modes
17.21s call     testing/python/math/test_math_ieee_math.py::test_ieee_mul_all_rounding_modes
17.18s call     testing/python/language/test_tilelang_language_vectorize.py::test_vectorize
17.06s call     testing/python/math/test_math_ieee_math.py::test_ieee_fdiv_all_rounding_modes
17.00s call     testing/python/language/test_tilelang_language_view.py::test_reshape_view
16.61s call     testing/python/math/test_math_ieee_math.py::test_ieee_frcp_all_rounding_modes
16.58s call     testing/python/math/test_math_ieee_math.py::test_ieee_add_all_rounding_modes
16.56s call     testing/python/language/test_tilelang_language_clamp.py::test_clamp
16.51s call     testing/python/math/test_math_ieee_math.py::test_ieee_fmaf_all_rounding_modes
16.41s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_sum
16.20s call     testing/python/jit/test_tilelang_jit_tvm_ffi.py::test_tvm_ffi_dynamic_shape
16.18s call     testing/python/math/test_math_ieee_math.py::test_ieee_fsqrt_all_rounding_modes
16.17s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_max
16.13s call     testing/python/language/test_tilelang_language_infinity.py::test_infinity
15.18s call     testing/python/language/test_tilelang_language_pipeline.py::test_blocksparse_matmul
13.90s call     testing/python/kernel/test_tilelang_kernel_gemm_simt.py::test_assert_tl_matmul
13.70s call     testing/python/kernel/test_tilelang_kernel_gemm_mma_intrinsic.py::test_assert_tl_matmul
13.67s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_max_clear
13.05s call     testing/python/language/test_tilelang_language_ptr.py::test_matmul
13.03s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_cython_dynamic_shape_with_out_idx
12.82s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_different_memory_orders
12.73s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_cython_kernel_multi_stream
12.57s call     testing/python/language/test_tilelang_language_copy.py::test_tilelang_copy
12.37s call     testing/python/analysis/test_tilelang_fragment_loop_checker.py::test_valid_loop
12.25s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_mixed_sp
12.24s call     testing/python/language/test_tilelang_language_composable_index.py::test_tilelang_copy
12.19s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_cython_kernel_do_bench
12.10s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_gemm_f16f16f16_nn
12.05s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_matmul_int_variable
11.95s call     testing/python/language/test_tilelang_language_ceildiv.py::test_ceildiv
11.62s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-128-128-128-0-128-False-False]
11.56s call     testing/python/jit/test_tilelang_jit_gemm_cython.py::test_matmul_float_variable
10.88s call     testing/python/jit/test_tilelang_jit_nvrtc.py::test_nvrtc_dynamic_shape
10.41s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-128-128-128-2-128-False-False]
10.33s call     testing/python/components/test_tilelang_pass_config_disable_warp_specialized.py::test_gemm_f16f16f16_nn
10.33s call     testing/python/kernel/test_tilelang_kernel_int4_gemm_mma.py::test_assert_tl_matmul_correctness
10.23s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-128-256-2-128-False-False]
10.01s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-128-256-0-128-False-False]
9.93s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float8_e4m3fn-float32-__tl_cvt_fp8x2_to_float2-2]
9.76s call     testing/python/fastmath/test_mathops_fastmath.py::test_two_arg_mathops_fastmath[pow-pow]
9.74s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-64-32-2-128-False-False]
9.64s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float8_e5m2-float32-__tl_cvt_fp8x2_to_float2-4]
9.58s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-64-64-2-128-False-False]
9.46s call     testing/python/kernel/test_tilelang_kernel_fp8_gemm_mma.py::test_assert_tl_matmul
9.45s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-64-32-0-256-False-False]
9.40s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float8_e5m2-float32-__tl_cvt_fp8x2_to_float2-2]
9.33s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-64-64-0-128-False-False0]
9.30s call     testing/python/kernel/test_tilelang_kernel_fp8_gemm.py::test_assert_matmul
9.27s call     testing/python/language/test_tilelang_language_reshape.py::test_reshape_smem_2d_2_1d
9.21s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float16-float32-__half22float2-4]
9.15s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float16-float32-float32-64-64-64-0-128-False-True]
9.06s call     testing/python/language/test_tilelang_language_copy.py::test_tilelang_copy_with_stride
8.98s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float8_e5m2-__nv_cvt_float2_to_fp8x2-4]
8.97s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float16-__float22half2_rn-4]
8.96s call     testing/python/fastmath/test_mathops_fastmath.py::test_two_arg_mathops_fastmath[fmod-fmod]
8.93s call     testing/python/language/test_tilelang_language_reshape.py::test_reshape_smem
8.84s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_fragment_1d
8.81s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_tiled_op_with_parallel
8.69s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float8_e4m3fn-__nv_cvt_float2_to_fp8x2-2]
8.69s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-int8-int32-int32-64-64-64-2-128-False-True]
8.67s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-64-2-128]
8.66s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-bfloat16-__float22bfloat162_rn-4]
8.66s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-64-2-128]
8.55s call     testing/python/language/test_tilelang_language_reshape.py::test_reshape_fragment
8.51s call     testing/python/language/test_tilelang_language_reshape.py::test_reduce_after_reshape
8.50s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_nested_parallels
8.50s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_nested_serials
8.41s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[bfloat16-float32-__bfloat1622float2-4]
8.41s call     testing/python/language/test_tilelang_language_reshape.py::test_reshape_layout_transform_shared
8.36s call     testing/python/jit/test_tilelang_jit_parcompile.py::test_par_compile
8.32s call     testing/python/language/test_tilelang_language_rand.py::test_rand_1d[128-0]
8.32s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-bfloat16-__float22bfloat162_rn-2]
8.27s call     testing/python/issue/test_tilelang_issue_830.py::test_empty_kernel_with_binding_variants
8.27s call     testing/python/language/test_tilelang_language_reshape.py::test_reshape_smem_1d_2_2d
8.26s call     testing/python/language/test_tilelang_language_rand.py::test_rand_1d[512-123]
8.23s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float8_e5m2-__nv_cvt_float2_to_fp8x2-2]
8.21s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[bfloat16-float32-__bfloat1622float2-2]
8.20s call     testing/python/kernel/test_tilelang_kernel_gemv_simt.py::test_gemv_simt
8.20s call     testing/python/components/test_storage_rewrite_detect_inplace.py::test_storage_rewrite_detect_inplace_toggle
8.19s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py::test_gemm_sp_sm90[512-1024-768-float8_e4m3fn-float16-float16-64-64-64-2-128-False-True]
8.14s call     testing/python/language/test_tilelang_language_rand.py::test_rand_1d[1024-42]
8.10s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float16-__float22half2_rn-2]
8.09s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float8_e4m3fn-float32-__tl_cvt_fp8x2_to_float2-4]
8.09s call     testing/python/language/test_tilelang_language_frontend_v2.py::test_serial_for_with_step
8.04s call     testing/python/language/test_tilelang_language_cumsum.py::test_cumsum_smem_1d
8.04s call     testing/python/kernel/test_tilelang_kernel_fp8_gemv_simt.py::test_gemv_simt
8.03s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float16-float32-__half22float2-2]
8.03s call     testing/python/language/test_tilelang_language_vectorized_cast.py::test_vectorized_cast[float32-float8_e4m3fn-__nv_cvt_float2_to_fp8x2-4]
7.87s call     testing/python/issue/test_tilelang_issue_1115.py::test_int64_address
7.54s call     testing/python/kernel/test_tilelang_kernel_gemv_simt.py::test_gemv_simt_fp8
7.53s call     testing/python/language/test_tilelang_language_frontend_v2.py::test_swap_logic
7.09s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[512-1024-768-True-False-float16-float16-float32-128-256-32-2-128]
7.05s call     testing/python/cpu/test_tilelang_cpu_gemm.py::test_matmul_compile
6.95s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-64-2-128]
6.87s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-128-128-False-True-int8-int8-int32-128-128-64-2-128]
6.78s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[512-1024-768-False-True-float16-float16-float32-128-256-32-2-128]
6.77s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-64-2-128]
6.76s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-False-True-float8_e5m2-float8_e5m2-float32-128-128-64-2-128]
6.76s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-128-128-False-False-int8-int8-int32-128-128-64-2-128]
6.62s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-128-128-False-False-int8-int8-int32-128-128-128-2-128]
6.58s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-128-128-True-True-int8-int8-int32-128-128-64-2-128]
6.55s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[512-1024-768-True-False-float16-float16-float32-128-256-32-2-128]
6.55s call     testing/python/language/test_tilelang_language_atomic_add.py::test_tile_atomic_add
6.52s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[512-1024-768-False-True-float16-float16-float32-128-256-32-2-128]
6.50s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-128-128-True-False-int8-int8-int32-128-128-64-2-128]
6.49s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_add
6.45s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[512-1024-768-True-True-float16-float16-float32-128-256-32-2-128]
6.42s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-128-128-False-False-int8-int8-int32-128-128-64-2-128]
6.33s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-128-128-True-False-int8-int8-int32-128-128-64-2-128]
6.30s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[512-1024-768-True-True-float16-float16-float32-128-256-32-2-128]
6.29s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[512-1024-768-False-True-bfloat16-bfloat16-float32-128-256-32-2-128]
6.20s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[512-1024-768-False-False-float16-float16-float32-128-256-32-2-128]
6.18s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-128-128-True-True-int8-int8-int32-128-128-64-2-128]
6.16s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-128-True-True-float32-float32-float32-128-128-32-2-128]
6.14s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-128-128-False-True-int8-int8-int32-128-128-64-2-128]
6.13s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[512-1024-768-True-False-float16-float16-float32-128-256-32-2-128]
6.12s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-128-False-False-float32-float32-float32-128-128-32-2-128]
6.12s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-8-128-False-True-int8-int8-int32-128-8-64-2-128]
6.11s call     testing/python/autotune/test_tilelang_autotune_with_inputs.py::test_autotune_matmul_symbolic_m
6.09s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[512-1024-768-False-False-float16-float16-float32-128-256-32-2-128]
6.07s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[512-1024-768-False-True-float16-float16-float32-128-256-32-2-128]
6.03s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-False-True-float32-float32-float32-128-128-32-2-128]
6.00s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-8-128-False-True-float16-float16-float32-128-8-32-2-128]
5.97s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[512-1024-768-True-True-float16-float16-float16-128-256-32-2-128]
5.94s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[512-1024-768-True-True-float16-float16-float32-128-256-32-2-128]
5.94s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[512-1024-768-False-True-bfloat16-bfloat16-float32-128-256-32-2-128]
5.93s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-True-False-int8-int8-int32-128-128-64-2-128]
5.91s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-128-128-True-False-int8-int8-int32-128-128-64-2-128]
5.90s call     testing/python/profiler/test_tilelang_profiler.py::test_profiler
5.88s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[512-1024-768-False-False-float16-float16-float32-128-256-32-2-128]
5.88s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_pad_f16f16f16_nn
5.86s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[512-1024-768-False-True-float16-float16-float16-128-256-32-2-128]
5.82s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[512-1024-768-False-True-float16-float16-float16-128-256-32-2-128]
5.80s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-False-True-int8-int32-int32-128-128-64-2-128]
5.78s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-128-128-False-True-int8-int8-int32-128-128-128-2-128]
5.78s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rr[128-128-128-True-True-int8-int8-int32-128-128-64-2-128]
5.71s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-False-True-int8-int8-int32-128-128-32-2-128]
5.69s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-False-False-int8-int8-int32-128-128-64-2-128]
5.67s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-False-False-float32-float32-float32-128-128-32-2-128]
5.66s call     testing/python/transform/test_tilelang_transform_config_index_bitwidth.py::test_sta_attention
5.63s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-128-128-True-True-int8-int8-int32-128-128-64-2-128]
5.62s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-32-2-128]
5.62s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[512-1024-768-True-False-float16-float16-float32-128-128-32-2-128]
5.60s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[512-1024-768-True-True-float16-float16-float32-128-128-32-2-128]
5.50s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-False-False-int8-int8-int32-128-128-32-2-128]
5.50s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_pad_aligned_f16f16f16_nn
5.50s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-True-False-int8-int8-int32-128-128-32-2-128]
5.47s call     testing/python/language/test_tilelang_language_parallel.py::test_parallel_dynamic_extent
5.46s call     testing/python/language/test_tilelang_language_any_of.py::test_block_sparse_matmul_local
5.45s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[512-1024-768-False-False-float16-float16-float32-128-128-32-2-128]
5.45s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-True-True-float8_e5m2-float8_e5m2-float32-128-128-32-2-128]
5.41s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-True-False-float32-float32-float32-128-128-32-2-128]
5.40s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[512-1024-768-True-True-float16-float16-float16-128-256-32-2-128]
5.39s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_mixed_pp
5.38s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[512-1024-768-True-False-float16-float16-float16-128-256-32-2-128]
5.35s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-128-False-True-float32-float32-float32-128-128-32-2-128]
5.33s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-True-True-float32-float32-float32-128-128-32-2-128]
5.31s call     testing/python/language/test_tilelang_language_all_of.py::test_block_sparse_matmul_global
5.31s call     testing/python/analysis/test_tilelang_nested_loop_checker.py::test_nested_pipelines
5.30s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-128-True-False-float32-float32-float32-128-128-32-2-128]
5.28s call     testing/python/language/test_tilelang_language_any_of.py::test_block_sparse_matmul_shared
5.25s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[512-1024-768-False-False-float16-float16-float16-128-256-32-2-128]
5.21s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-32-True-True-int8-int8-int32-128-128-32-2-128]
5.19s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[512-1024-768-False-False-float16-float16-float16-128-256-32-2-128]
5.19s call     testing/python/language/test_tilelang_language_all_of.py::test_block_sparse_matmul_local
5.19s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[512-1024-768-False-True-float16-float16-float32-128-128-32-2-128]
5.13s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[512-1024-768-True-False-float16-float16-float16-128-256-32-2-128]
5.11s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-128-128-True-True-int8-int8-int32-128-128-32-2-128]
5.11s call     testing/python/jit/test_tilelang_jit_callback.py::test_gemm_jit_kernel
5.08s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f16f16f32_nn
5.07s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_rs[128-8-64-False-True-float16-float16-float32-128-8-32-0-128]
5.03s call     testing/python/language/test_tilelang_language_all_of.py::test_block_sparse_matmul_shared
4.99s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f16f16f16_nn
4.98s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_max
4.96s call     testing/python/language/test_tilelang_language_alias.py::test_matmul
4.95s call     testing/python/jit/test_tilelang_jit_tvm_ffi.py::test_tvm_ffi_kernel_do_bench
4.93s call     testing/python/issue/test_tilelang_issue_96.py::test_pipeline_small_matrix
4.92s call     testing/python/language/test_tilelang_language_warp_reduce.py::test_warp_reduce_bitand
4.92s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_max_shared
4.90s call     testing/python/language/test_tilelang_language_let_layout.py::test_blocksparse_copy_cp_async
4.87s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f16f16f16_tn
4.83s call     testing/python/transform/test_tilelang_transform_simplify.py::test_matmul
4.82s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_sr[128-8-64-False-True-float16-float16-float32-128-8-32-0-128]
4.81s call     testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_gemm_annot
4.77s call     testing/python/debug/test_tilelang_debug_print.py::test_debug_print_buffer_conditional
4.76s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-8-128-False-True-float16-float16-float16-128-8-32-2-128]
4.76s call     testing/python/jit/test_tilelang_jit_gemm.py::test_gemm_f16f16f16_nn_kernel_jit
4.75s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_memory_order
4.74s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[tan-tan]
4.71s call     testing/python/fastmath/test_mathops_fastmath.py::test_abs_maps_to_fabs
4.71s call     testing/python/jit/test_tilelang_jit_nullptr.py::test_nullptr
4.70s call     testing/python/issue/test_tilelang_issue_96.py::test_pipeline_large_matrix
4.69s call     testing/python/language/test_tilelang_language_any_of.py::test_block_sparse_matmul_global
4.67s call     testing/python/language/test_tilelang_language_annotate_safe_value.py::test_tilelang_copy
4.67s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-32-True-False-int8-int8-int32-128-128-32-2-128]
4.66s call     testing/python/issue/test_tilelang_issue_1008.py::test_fill_with_dynamic_region_kernel
4.64s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[sin-sin]
4.64s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f32f32f32_nt
4.64s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__exp-__exp]
4.63s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py::test_gemm_ss[128-8-64-False-True-float16-float16-float32-128-8-32-0-128]
4.63s call     testing/python/language/test_tilelang_language_lazy_jit.py::test_jit2_gemm_ptr
4.63s call     testing/python/layout/test_tilelang_layout_fused_replicate.py::test_layout_infer_compiles_and_runs
4.61s call     testing/python/language/test_tilelang_language_clear.py::test_matmul
4.61s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_shuffle_elect_block_leader
4.59s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_pad_f16f16f32_nn
4.58s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[sinh-sinh]
4.57s call     testing/python/language/test_tilelang_language_let_layout.py::test_blocksparse_copy_tma
4.57s call     testing/python/language/test_tilelang_language_parallel.py::test_parallel_static_extent
4.56s call     testing/python/issue/test_tilelang_issue_830.py::test_empty_kernel_lowering
4.56s call     testing/python/language/test_tilelang_language_if_range.py::test_tilelang_if_range
4.55s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_i8i8i32_tn
4.55s call     testing/python/debug/test_tilelang_debug_print.py::test_debug_print_value_conditional
4.55s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_bf16bf16f32_nn
4.55s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f32f32f32_nn
4.54s call     testing/python/language/test_tilelang_language_alloc.py::test_alloc_var
4.54s call     testing/python/kernel/test_tilelang_kernel_gemm_mma_intrinsic.py::test_assert_tl_matmul_bfloat16
4.54s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_min
4.53s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[log2-log2]
4.53s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[exp2-exp2]
4.51s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[sqrt-sqrt]
4.51s call     testing/python/language/test_tilelang_language_annot.py::test_tensor_annot_mul_add
4.50s call     testing/python/language/test_tilelang_language_mask_op.py::test_tilelang_copy_mask_copy_range
4.49s call     testing/python/language/test_tilelang_language_warp_reduce.py::test_warp_reduce_min
4.49s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[ceil-ceil]
4.48s call     testing/python/debug/test_tilelang_debug_print.py::test_debug_print_msg
4.48s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-32-False-False-int8-int8-int32-128-128-32-2-128]
4.47s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[tanh-tanh]
4.47s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_absmax_shared
4.47s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f16f16f16_nt
4.47s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f32f32f32_tn
4.45s call     testing/python/language/test_tilelang_language_warp_reduce.py::test_warp_reduce_bitor
4.45s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-128-32-False-True-int8-int8-int32-128-128-32-2-128]
4.44s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[nearbyint-nearbyint]
4.43s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[erf-erf]
4.43s call     testing/python/jit/test_tilelang_jit_tvm_ffi.py::test_tvm_ffi_im2col_tma_desc
4.42s call     testing/python/debug/test_tilelang_debug_print.py::test_debug_print_register_files
4.42s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_min_shared
4.41s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_rr[128-8-128-False-True-int8-int8-int32-128-8-32-2-128]
4.41s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_group_idx_custom
4.39s call     testing/python/kernel/test_tilelang_kernel_gemm_with_stride.py::test_tilelang_kernel_gemm_with_stride
4.39s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_i8i8i32_nt
4.34s call     testing/python/language/test_tilelang_language_ternary.py::test_tilelang_ternary
4.32s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[atan-atan]
4.30s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[log10-log10]
4.29s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[log-log]
4.29s call     testing/python/kernel/test_tilelang_kernel_gemm.py::test_gemm_f64f64f64_nt
4.28s call     testing/python/kernel/test_tilelang_kernel_element_wise_add.py::test_elementwise_add_i32
4.27s call     testing/python/issue/test_tilelang_issue_830.py::test_empty_with_dead_code_kernel
4.27s call     testing/python/kernel/test_tilelang_kernel_element_wise_add.py::test_elementwise_add_f32f16
4.25s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[cos-cos]
4.24s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_sum_shared
4.24s call     testing/python/language/test_tilelang_language_unroll.py::test_unroll_with_step
4.24s call     testing/python/kernel/test_tilelang_kernel_element_wise_add.py::test_elementwise_add_f32
4.23s call     testing/python/language/test_tilelang_language_reduce.py::test_reduce_abssum_shared
4.23s call     testing/python/language/test_tilelang_language_warp_reduce.py::test_warp_reduce_max
4.22s call     testing/python/language/test_tilelang_language_int64.py::test_fill_symbolic
4.20s call     testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py::test_gemm_sr[128-8-32-False-True-float16-float16-float16-128-8-32-0-128]
4.20s call     testing/python/debug/test_device_assert.py::test_device_assert_no_trigger
4.20s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__exp10-__exp10]
4.17s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_load_store
4.17s call     testing/python/language/test_tilelang_language_warp_reduce.py::test_warp_reduce_sum
4.16s call     testing/python/language/test_tilelang_language_copy.py::test_tilelang_copy_buffer_load_with_parallel
4.16s call     testing/python/language/test_tilelang_language_mask_op.py::test_tilelang_copy_mask_copy
4.14s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[rsqrt-rsqrt]
4.13s call     testing/python/issue/test_tilelang_issue_814.py::test_issue_814
4.12s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_addx4
4.12s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__cos-__cos]
4.12s call     testing/python/language/test_tilelang_language_mask_op.py::test_tilelang_copy_mask_parallel
4.12s call     testing/python/jit/test_tilelang_jit_callback.py::test_gemm_f16f16f16_nn
4.12s call     testing/python/language/test_tilelang_language_annot.py::test_tensor_annot_mul
4.12s call     testing/python/issue/test_tilelang_issue_1008.py::test_fill_with_static_region_kernel
4.11s call     testing/python/kernel/test_tilelang_kernel_element_wise_add.py::test_elementwise_add_f16
4.08s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__tan-__tan]
4.06s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__log-__log]
4.05s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__log2-__log2]
4.05s call     testing/python/language/test_tilelang_language_annot.py::test_tensor_annot_add
4.04s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_lane_idx_default
4.04s call     testing/python/issue/test_tilelang_issue_1001.py::test_cumsum_view_infer_layout
4.04s call     testing/python/runtime/test_tilelang_runtime_dynamic_shared_memory.py::test_dynamic_shared_memory_varies_across_calls
4.02s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_group_idx_default
4.02s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_return_prev
4.01s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[cosh-cosh]
4.01s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[trunc-trunc]
3.99s call     testing/python/transform/test_nullable_buffer_params.py::test_nullable_shared_shape
3.98s call     testing/python/math/test_math_ieee_math.py::test_ieee_frsqrt_rn_only
3.98s call     testing/python/language/test_tilelang_language_mask_op.py::test_tilelang_copy_mask_parallel_range
3.98s call     testing/python/language/test_tilelang_language_ceildiv.py::test_ceildiv_dyn
3.97s call     testing/python/language/test_tilelang_language_atomic_add.py::test_atomic_addx2
3.93s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__sin-__sin]
3.93s call     testing/python/jit/test_tilelang_jit_tvm_ffi.py::test_tvm_ffi_l2_persistent_map
3.93s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_lane_idx_custom
3.92s call     testing/python/language/test_tilelang_language_alloc.py::test_alloc_var_add
3.91s call     testing/python/fastmath/test_mathops_fastmath.py::test_fastmath_versions[__log10-__log10]
3.89s call     testing/python/language/test_tilelang_language_assume.py::test_assume_enable_vectorization
3.89s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[round-round]
3.86s call     testing/python/language/test_tilelang_capture.py::test_tilelang_capture
3.85s call     testing/python/language/test_tilelang_language_var_init.py::test_var_assign
3.85s call     testing/python/language/test_tilelang_language_chain_equal.py::test_chain_equal
3.85s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[exp10-exp10]
3.83s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_idx_sync_default
3.82s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_shuffle_elect_default
3.81s call     testing/python/language/test_tilelang_language_unroll.py::test_unroll_with_unroll_factor
3.80s call     testing/python/language/test_tilelang_language_alloc.py::test_alloc_var_with_initializer
3.80s call     testing/python/language/test_tilelang_language_copy.py::test_tilelang_copy_bufferload
3.80s call     testing/python/language/test_tilelang_language_int64.py::test_fill_static
3.79s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_idx_custom
3.78s call     testing/python/language/test_tilelang_language_frontend_v2.py::test_while_loop
3.77s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[exp-exp]
3.77s call     testing/python/language/test_tilelang_language_assume.py::test_assume_complex_indexing
3.76s call     testing/python/language/test_tilelang_language_frontend_v2.py::test_var_assign
3.76s call     testing/python/jit/test_tilelang_jit_nvrtc.py::test_gemm_f16f16f16_nn
3.76s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_idx_default
3.76s call     testing/python/language/test_tilelang_language_get_warp_info.py::test_get_warp_idx_sync_custom
3.75s call     testing/python/issue/test_tilelang_issue_1210.py::test_make_packed_api_no_free_loop_var
3.69s call     testing/python/fastmath/test_mathops_fastmath.py::test_mathops_generate_no_fastmath[floor-floor]
3.67s call     testing/python/language/test_tilelang_language_alloc.py::test_alloc_multi_vars_with_initializer
3.67s call     testing/python/language/test_tilelang_language_assume.py::test_assume_remove_boundary_check
3.62s call     testing/python/language/test_tilelang_language_frontend_v2.py::test_frame_inside_macro
3.61s call     testing/python/language/test_tilelang_language_intrinsics_codegen.py::test_language_ldg_codegen
3.27s call     testing/python/jit/test_tilelang_jit_nvrtc.py::test_nvrtc_kernel_do_bench
3.04s call     testing/python/jit/test_tilelang_jit_nvrtc.py::test_nvrtc_im2col_tma_desc
2.75s call     testing/python/jit/test_tilelang_jit_nvrtc.py::test_nvrtc_l2_persistent_map

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

✅ Successfully generated implementation plan.

💡 Use @coderabbitai <feedback> to replan

Examples

• Example 1
• Example 2

---

🔗 Similar Issues

Related Issues

• [Bug] PyTest Failures on H100 #506
• [Question] Downgraded performance of flashattention on A100 #1488
• [Bug] Symbolic shape with dynamic alignment config results in incorrect results #708
• Auto-tuning failed #778
• [Bug] Autotuner failed to compile the program which can be sucessfully compiled by tilelang.compile #349
• [Questions] Compilation Error -- Check failed: (load->indices.size() == ndim) is false: #340
• [Question] l20 run example_warp_specialize_gemm_copy_1_gemm_0 fial #1394

🔗 Related PRs

#761 - [Feature] Add 1D TMA support [merged]
#875 - [FastMath] Disable default TVM fastmath intrinsic dispatch and add explicit fastmath op to invoke [merged]
#882 - [Precision] Introduce T.ieee_rsqrt and related high precision op [merged]
#901 - [Example] Add topk into sparse mla example and append some docs [merged]
#977 - [Refactor] Refactor Pass InjectFenceProxy and expose some warp group primitives in frontend [merged]
#1174 - [Language] Add Correctness and performance check scripts for V2 [merged]
#1421 - feat(cutedsl): add CuTeDSL backend [merged]
#1444 - [Refactor] Reduce direct dependency on PyTorch due to its limited type support [merged]

👤 Suggested Assignees

• LeiWang1999
• xxw-keju
• Xiang-cd
• Rachmanino
• lqf0624

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord or schedule a call!

[coderabbitai]

Coding Plan

Approach

Ticket Summary

Reduce CI test execution time from 60+ minutes to ~20-30 minutes by implementing CI-specific test configurations that reduce test shapes, optimize autotune parameters, and enable selective test execution through pytest markers.

Observations

Codebase Summary

The TileLanguage project contains two test suites: examples/ with non-parametrized integration tests using hardcoded large matrix dimensions, and testing/python/ with heavily parametrized unit tests (14-16 cases per function). Tests use pytest with basic configuration in conftest.py for seeding, but lack CI-specific markers or environment-based test selection. The AutoTuner evaluates 27 configurations with 30-100 second timeouts per configuration.

Assumptions

Assumption 1: Test coverage strategy in CI vs local development

Options Considered:

• Option A: Run all tests but with reduced shapes (128×128 instead of 512×1024)
• Option B: Run subset of tests marked as critical with full shapes

Chosen Option: Option A

Rationale: Run all tests with reduced shapes. This maintains full functional coverage while reducing execution time through smaller problem sizes. This approach is more aligned with CI best practices where breadth of coverage is prioritized over exhaustive stress testing. The codebase already shows tests are primarily validating correctness rather than performance, making smaller shapes acceptable for CI.

Assumption 2: Autotune configuration reduction approach

Options Considered:

• Option A: Reduce timeout and configuration count globally for CI
• Option B: Skip autotune tests entirely in CI
• Option C: Create separate fast-autotune tests with minimal configs

Chosen Option: Option A

Rationale: Reduce timeout to 10s and configuration count to 8-10 representative configs when running in CI. This preserves autotune functionality testing while cutting execution time by 70%. Complete autotune validation can be reserved for nightly or pre-release testing.

💡 User Tips

Regenerate the plan with different choices with @coderabbitai <feedback>.

Implementation Steps

Phase 1: Establish CI Environment Detection and Test Markers

Create the foundational infrastructure for CI-aware test execution by adding environment detection, pytest markers, and configuration helpers.

Task 1: Add CI environment detection utilities

Create a new module tilelang/testing/ci_config.py that provides environment detection and CI-specific configuration helpers.

• Implement is_ci_environment() function that checks common CI environment variables (CI, GITHUB_ACTIONS, GITLAB_CI, etc.)
• Implement get_ci_shape_scale() function that returns a scaling factor for test shapes (0.5 for CI, 1.0 for local)
• Implement get_ci_param_count() function that returns maximum parametrization count for CI (default 6 for CI, unlimited for local)
• Export these utilities from tilelang/testing/__init__.py alongside existing test helpers

Task 2: Register custom pytest markers

Extend pyproject.toml to register custom pytest markers for test categorization.

• Add markers section under [tool.pytest.ini_options] with definitions for @pytest.mark.slow (tests >10s), @pytest.mark.large_shape (tests requiring >512×512 matrices), and @pytest.mark.autotune (tests involving autotuning)
• Add pytest command-line option configuration to allow marker-based test selection (e.g., -m "not slow")

Task 3: Create pytest plugin for automatic marker detection

Add a new file testing/pytest_plugins/ci_markers.py that implements pytest hooks for automatic test marking based on execution time.

• Implement pytest_configure() hook to register the plugin
• Implement pytest_runtest_makereport() hook to track test execution times
• Add logic to automatically apply @pytest.mark.slow to tests exceeding 10-second threshold on first run
• Create mechanism to persist slow test information to .pytest_cache/ for subsequent runs

Task 4: Update conftest.py files for CI awareness

Modify both examples/conftest.py and testing/conftest.py to enable CI-specific behavior.

• Import and integrate the new CI detection utilities from tilelang/testing/ci_config.py
• Add pytest_collection_modifyitems() hook to automatically skip or modify tests based on CI environment
• Preserve existing seeding and validation logic while adding conditional behavior for CI

🤖 Prompt for AI agents

Set up CI-aware test infrastructure to enable environment detection and
selective test execution.

Create `tilelang/testing/ci_config.py` with:
- `is_ci_environment()` checking CI, GITHUB_ACTIONS, GITLAB_CI environment
variables
- `get_ci_shape_scale()` returning 0.5 for CI, 1.0 for local
- `get_ci_param_count()` returning 6 for CI, unlimited for local
- Export these utilities from `tilelang/testing/__init__.py`

Update `pyproject.toml`:
- Add markers section under `[tool.pytest.ini_options]` with `slow`,
`large_shape`, and `autotune` markers
- Configure marker-based test selection

Create `testing/pytest_plugins/ci_markers.py`:
- Implement `pytest_configure()` and `pytest_runtest_makereport()` hooks
- Track test execution times and auto-apply `@pytest.mark.slow` to tests >10s
- Persist slow test information to `.pytest_cache/`

Update `examples/conftest.py` and `testing/conftest.py`:
- Import CI detection utilities
- Add `pytest_collection_modifyitems()` hook for CI-specific test behavior
- Preserve existing seeding and validation logic

Phase 2: Reduce Test Shapes in Example Tests

Modify example test files to use CI-aware shape configurations, targeting the slowest tests first to maximize CI time savings.

Task 1: Implement dynamic shape configuration for examples

Create a new module examples/test_utils.py with shape configuration helpers.

• Implement get_gemm_shapes() function that returns (M, N, K, block_M, block_N, block_K) tuples scaled for CI (e.g., 128×128×128 instead of 512×1024×768)
• Implement get_attention_shapes() function for attention tests returning (batch, seq_len, head_dim) scaled appropriately
• Provide scale_shape() utility that accepts base dimensions and applies CI scaling factor
• Import CI detection from tilelang/testing/ci_config.py to drive scaling decisions

Task 2: Refactor slowest example tests to use dynamic shapes

Modify the four slowest example tests to accept shape parameters from the new configuration helpers.

• Update examples/gemm_sp/example_gemm_sp.py main() function to accept optional shape parameters and default to get_gemm_shapes() when not provided
• Update examples/sparse_tensorcore/example_sparse_tensorcore.py similarly to use configurable shapes
• Update examples/dequantize_gemm/example_dequantize_gemm.py to use reduced block sizes (block_M=64, block_N=64, block_K=32) in CI
• Update examples/gemm/example_gemm.py to reduce from 1024×1024×1024 to 256×256×256 in CI
• Ensure all test wrappers in corresponding test_example_*.py files call main() without arguments to get CI-aware defaults

Task 3: Apply slow markers to long-running example tests

Add @pytest.mark.slow decorators to tests identified in the ticket as taking >10 seconds.

• Mark tests in gemm_sp, sparse_tensorcore, dequantize_gemm, gemm, attention_sink, and flash_attention directories
• Mark autotune variants separately with both @pytest.mark.slow and @pytest.mark.autotune
• Add @pytest.mark.large_shape to tests that use shapes >512 even after CI reduction

🤖 Prompt for AI agents

Implement CI-aware shape configurations for example tests to reduce execution
time while maintaining correctness validation.

Create `examples/test_utils.py`:
- `get_gemm_shapes()` returning scaled tuples (M, N, K, block_M, block_N,
block_K) - 128×128×128 for CI vs 512×1024×768 for local
- `get_attention_shapes()` returning scaled (batch, seq_len, head_dim)
- `scale_shape()` utility applying CI scaling factor
- Import and use CI detection from `tilelang/testing/ci_config.py`

Update slowest example tests:
- Modify `examples/gemm_sp/example_gemm_sp.py` main() to accept optional shape
parameters, defaulting to `get_gemm_shapes()`
- Update `examples/sparse_tensorcore/example_sparse_tensorcore.py` similarly
- Update `examples/dequantize_gemm/example_dequantize_gemm.py` to use
block_M=64, block_N=64, block_K=32 in CI
- Update `examples/gemm/example_gemm.py` to reduce from 1024³ to 256³ in CI
- Ensure test wrappers call main() without arguments

Add pytest markers:
- `@pytest.mark.slow` to tests in gemm_sp, sparse_tensorcore, dequantize_gemm,
gemm, attention_sink, flash_attention
- Both `@pytest.mark.slow` and `@pytest.mark.autotune` to autotune variants
- `@pytest.mark.large_shape` to tests with shapes >512

Phase 3: Optimize Parametrized Testing/Python Tests

Reduce parametrization breadth in the heavily parametrized tilelibrary GEMM tests while maintaining representative coverage.

Task 1: Create CI-aware parametrization helpers

Add testing/python/test_helpers.py module with parametrization generation utilities.

• Implement generate_gemm_params() function that creates parameter combinations for GEMM tests
• Accept ci_mode parameter that, when True, returns 4-6 representative cases instead of full 14-16 cases
• For CI mode, select representative combinations covering: small shapes (128×128×128), one medium shape (256×256×256), different data types (float16, int8, float8), key transpose combinations (NN, NT), and varied pipeline stages (0, 2)
• Ensure non-CI mode returns full parameter sets for comprehensive local testing

Task 2: Refactor tilelibrary GEMM test parametrization

Update the three slowest tilelibrary test files to use the new parametrization helpers.

• Modify testing/python/tilelibrary/test_tilelang_tilelibrary_gemm.py test_gemm_rr and test_gemm_sr to call generate_gemm_params(ci_mode=is_ci_environment())
• Update testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp.py test_gemm_sp_sm90 similarly
• Update testing/python/tilelibrary/test_tilelang_tilelibrary_gemm_sp_v2.py for test_gemm_rr, test_gemm_rs, test_gemm_ss, test_gemm_sr
• Replace hardcoded @pytest.mark.parametrize decorators with dynamic parameter generation from helpers
• Add @pytest.mark.slow and @pytest.mark.large_shape markers to these test functions

Task 3: Optimize language test shapes

Reduce matrix dimensions in the slowest language tests that don't require large shapes for correctness validation.

• Update testing/python/language/test_tilelang_language_vectorized_cast.py to use 128×128 tensors instead of defaults in CI
• Update cumsum, reduce, and vectorize tests in testing/python/language/ to use smaller buffer sizes (reduce from 1024 to 256 elements) when is_ci_environment() returns True
• Update pipeline and reshape tests to use 64×64 or 128×128 test matrices in CI
• Preserve existing shapes for local development by conditionally selecting shapes based on environment

Task 4: Mark and optimize slow language/debug tests

Apply markers and implement CI-specific optimizations for the slowest non-GEMM tests.

• Add @pytest.mark.slow to test_debug_print_buffer, test_cython_dynamic_shape, test_reduce_sum_clear, test_jit2_return, test_bitwise_reduce_ops
• For test_debug_print_buffer, reduce buffer size and iteration count in CI mode
• For JIT/Cython tests, reduce dynamic shape ranges to 64-256 instead of 64-1024 in CI
• For reduce tests, reduce tensor dimensions from 1024 to 256 in CI

🤖 Prompt for AI agents

Reduce parametrization breadth and test shapes in heavily parametrized unit
tests to cut execution time while maintaining representative coverage.

Create `testing/python/test_helpers.py`:
- `generate_gemm_params(ci_mode)` function generating parameter combinations
- Return 4-6 representative cases for CI (128×128×128, 256×256×256,
float16/int8/float8, NN/NT transposes, pipeline stages 0/2)
- Return full 14-16 cases for local testing

Update tilelibrary GEMM tests:
- Modify `test_tilelang_tilelibrary_gemm.py` test_gemm_rr and test_gemm_sr to
use `generate_gemm_params(ci_mode=is_ci_environment())`
- Update `test_tilelang_tilelibrary_gemm_sp.py` test_gemm_sp_sm90
- Update `test_tilelang_tilelibrary_gemm_sp_v2.py` test_gemm_rr, test_gemm_rs,
test_gemm_ss, test_gemm_sr
- Replace hardcoded `@pytest.mark.parametrize` with dynamic parameter generation
- Add `@pytest.mark.slow` and `@pytest.mark.large_shape` markers

Optimize language test shapes:
- Update `test_tilelang_language_vectorized_cast.py` to use 128×128 tensors in
CI
- Update cumsum, reduce, vectorize tests to use 256 elements instead of 1024 in
CI
- Update pipeline and reshape tests to use 64×64 or 128×128 matrices in CI
- Conditionally select shapes based on `is_ci_environment()`

Mark slow language/debug tests:
- Add `@pytest.mark.slow` to test_debug_print_buffer, test_cython_dynamic_shape,
test_reduce_sum_clear, test_jit2_return, test_bitwise_reduce_ops
- Reduce buffer sizes, iteration counts, and dynamic shape ranges in CI mode

Phase 4: Optimize Autotune Test Execution

Reduce autotune test execution time by limiting configuration search space and timeout values in CI environments.

Task 1: Create CI-aware autotune configuration generator

Add testing/python/autotune/autotune_test_helpers.py with configuration generation for autotune tests.

• Implement get_autotune_configs_for_ci() function that generates 8-10 representative configurations instead of full 27-config search space
• Select configs covering: 2-3 block size combinations (small: 64×64×32, medium: 128×128×64, large: 256×256×128), 2 thread counts (128, 256), representative data types
• Implement get_autotune_timeout() function returning 10s for CI, 30-100s for local development
• Provide these as replacement for get_configs() functions in example files

Task 2: Refactor example autotune tests

Update example files that use autotuning to leverage CI-aware configuration.

• Modify examples/elementwise/example_elementwise_add.py get_configs() to call CI-aware helper and return reduced config set in CI
• Update examples/gemm/example_gemm_autotune.py similarly to use 8-10 configs instead of full search space in CI
• Pass CI-aware timeout to AutoTuner initialization (modify calls to use timeout=get_autotune_timeout())
• Add @pytest.mark.autotune markers to autotune test functions in test files

Task 3: Optimize testing/python autotune tests

Apply similar optimizations to unit tests of the autotune system itself.

• Update testing/python/autotune/test_tilelang_autotune.py test_autotune_matmul to use reduced timeout (10s) and smaller matrix sizes (128×128 instead of 256×256) in CI
• Update testing/python/autotune/test_tilelang_autotune_with_inputs.py test_autotune_matmul_symbolic_m similarly
• Reduce number of trial configurations evaluated in these tests when in CI mode
• Add @pytest.mark.autotune and @pytest.mark.slow markers

🤖 Prompt for AI agents

Optimize autotune test execution by reducing configuration search space and
timeouts in CI to achieve ~70% time reduction.

Create `testing/python/autotune/autotune_test_helpers.py`:
- `get_autotune_configs_for_ci()` generating 8-10 representative configs instead
of 27
- Select configs covering block sizes (64×64×32, 128×128×64, 256×256×128),
thread counts (128, 256), representative data types
- `get_autotune_timeout()` returning 10s for CI, 30-100s for local

Update example autotune tests:
- Modify `examples/elementwise/example_elementwise_add.py` get_configs() to use
CI-aware helper
- Update `examples/gemm/example_gemm_autotune.py` to use 8-10 configs in CI
- Pass `timeout=get_autotune_timeout()` to AutoTuner initialization
- Add `@pytest.mark.autotune` markers to test functions

Optimize testing/python autotune tests:
- Update `test_tilelang_autotune.py` test_autotune_matmul to use 10s timeout and
128×128 matrices in CI
- Update `test_tilelang_autotune_with_inputs.py` test_autotune_matmul_symbolic_m
similarly
- Reduce trial configuration count in CI mode
- Add `@pytest.mark.autotune` and `@pytest.mark.slow` markers

Phase 5: CI Configuration and Documentation

Create CI pipeline configuration and documentation to enable the new CI-aware test execution strategy.

Task 1: Create GitHub Actions workflow configuration

Add .github/workflows/fast-ci-tests.yml workflow file that uses the new test selection capabilities.

• Configure pytest to run with -m "not slow or autotune" for fast CI feedback on pull requests
• Set CI=true environment variable explicitly
• Add separate workflow step for slow tests that runs on schedule or manual trigger
• Configure workflow to run full test suite (with CI shapes) on main branch merges

Task 2: Update pytest configuration files

Modify examples/pytest.ini and add testing/pytest.ini if needed to support marker-based execution.

• Add marker definitions matching those in pyproject.toml
• Configure default marker expression for CI (-m "not slow")
• Add addopts configuration for CI-specific pytest flags (e.g., -v --tb=short for concise output)

Task 3: Document CI test strategy

Create or update TESTING.md documentation file explaining the CI testing approach.

• Document the marker system (@pytest.mark.slow, @pytest.mark.large_shape, @pytest.mark.autotune)
• Explain how to run full test suite locally (pytest testing/ examples/ without markers)
• Explain how to run CI-equivalent tests locally (CI=true pytest -m "not slow" testing/ examples/)
• Provide guidance on when to add markers to new tests
• Document shape scaling expectations and how to override for local testing

Task 4: Add test execution metrics and reporting

Enhance pytest configuration to track and report test execution time improvements.

• Add --durations=30 flag to pytest configuration to report slowest tests
• Configure pytest-html or similar plugin to generate execution time reports
• Add comparison metrics to CI output showing time saved vs full test suite
• Consider adding pytest-xdist for parallel test execution to further reduce CI time

🤖 Prompt for AI agents

Set up CI pipeline configuration and comprehensive documentation to enable fast
CI feedback and full local testing capabilities.

Create `.github/workflows/fast-ci-tests.yml`:
- Configure pytest with `-m "not slow or autotune"` for PR fast feedback
- Set `CI=true` environment variable
- Add separate workflow step for slow tests (scheduled/manual trigger)
- Run full test suite with CI shapes on main branch merges

Update pytest configuration:
- Modify `examples/pytest.ini` and create `testing/pytest.ini` if needed
- Add marker definitions matching `pyproject.toml`
- Set default marker expression `-m "not slow"` for CI
- Add addopts for CI-specific flags (`-v --tb=short`)

Create/update `TESTING.md`:
- Document marker system (slow, large_shape, autotune)
- Explain local full test suite execution (`pytest testing/ examples/`)
- Explain CI-equivalent local execution (`CI=true pytest -m "not slow" testing/
examples/`)
- Provide guidance on adding markers to new tests
- Document shape scaling and override options

Add test execution metrics:
- Add `--durations=30` to pytest configuration
- Configure pytest-html for execution time reports
- Add CI output comparison metrics
- Consider pytest-xdist for parallel execution

---

🚀 Next Steps

🤖 All AI agent prompts combined

Task: 1

Set up CI-aware test infrastructure to enable environment detection and
selective test execution.

Create `tilelang/testing/ci_config.py` with:
- `is_ci_environment()` checking CI, GITHUB_ACTIONS, GITLAB_CI environment
variables
- `get_ci_shape_scale()` returning 0.5 for CI, 1.0 for local
- `get_ci_param_count()` returning 6 for CI, unlimited for local
- Export these utilities from `tilelang/testing/__init__.py`

Update `pyproject.toml`:
- Add markers section under `[tool.pytest.ini_options]` with `slow`,
`large_shape`, and `autotune` markers
- Configure marker-based test selection

Create `testing/pytest_plugins/ci_markers.py`:
- Implement `pytest_configure()` and `pytest_runtest_makereport()` hooks
- Track test execution times and auto-apply `@pytest.mark.slow` to tests >10s
- Persist slow test information to `.pytest_cache/`

Update `examples/conftest.py` and `testing/conftest.py`:
- Import CI detection utilities
- Add `pytest_collection_modifyitems()` hook for CI-specific test behavior
- Preserve existing seeding and validation logic
===============================================================================

Task: 2

Implement CI-aware shape configurations for example tests to reduce execution
time while maintaining correctness validation.

Create `examples/test_utils.py`:
- `get_gemm_shapes()` returning scaled tuples (M, N, K, block_M, block_N,
block_K) - 128×128×128 for CI vs 512×1024×768 for local
- `get_attention_shapes()` returning scaled (batch, seq_len, head_dim)
- `scale_shape()` utility applying CI scaling factor
- Import and use CI detection from `tilelang/testing/ci_config.py`

Update slowest example tests:
- Modify `examples/gemm_sp/example_gemm_sp.py` main() to accept optional shape
parameters, defaulting to `get_gemm_shapes()`
- Update `examples/sparse_tensorcore/example_sparse_tensorcore.py` similarly
- Update `examples/dequantize_gemm/example_dequantize_gemm.py` to use
block_M=64, block_N=64, block_K=32 in CI
- Update `examples/gemm/example_gemm.py` to reduce from 1024³ to 256³ in CI
- Ensure test wrappers call main() without arguments

Add pytest markers:
- `@pytest.mark.slow` to tests in gemm_sp, sparse_tensorcore, dequantize_gemm,
gemm, attention_sink, flash_attention
- Both `@pytest.mark.slow` and `@pytest.mark.autotune` to autotune variants
- `@pytest.mark.large_shape` to tests with shapes >512
===============================================================================

Task: 3

Reduce parametrization breadth and test shapes in heavily parametrized unit
tests to cut execution time while maintaining representative coverage.

Create `testing/python/test_helpers.py`:
- `generate_gemm_params(ci_mode)` function generating parameter combinations
- Return 4-6 representative cases for CI (128×128×128, 256×256×256,
float16/int8/float8, NN/NT transposes, pipeline stages 0/2)
- Return full 14-16 cases for local testing

Update tilelibrary GEMM tests:
- Modify `test_tilelang_tilelibrary_gemm.py` test_gemm_rr and test_gemm_sr to
use `generate_gemm_params(ci_mode=is_ci_environment())`
- Update `test_tilelang_tilelibrary_gemm_sp.py` test_gemm_sp_sm90
- Update `test_tilelang_tilelibrary_gemm_sp_v2.py` test_gemm_rr, test_gemm_rs,
test_gemm_ss, test_gemm_sr
- Replace hardcoded `@pytest.mark.parametrize` with dynamic parameter generation
- Add `@pytest.mark.slow` and `@pytest.mark.large_shape` markers

Optimize language test shapes:
- Update `test_tilelang_language_vectorized_cast.py` to use 128×128 tensors in
CI
- Update cumsum, reduce, vectorize tests to use 256 elements instead of 1024 in
CI
- Update pipeline and reshape tests to use 64×64 or 128×128 matrices in CI
- Conditionally select shapes based on `is_ci_environment()`

Mark slow language/debug tests:
- Add `@pytest.mark.slow` to test_debug_print_buffer, test_cython_dynamic_shape,
test_reduce_sum_clear, test_jit2_return, test_bitwise_reduce_ops
- Reduce buffer sizes, iteration counts, and dynamic shape ranges in CI mode
===============================================================================

Task: 4

Optimize autotune test execution by reducing configuration search space and
timeouts in CI to achieve ~70% time reduction.

Create `testing/python/autotune/autotune_test_helpers.py`:
- `get_autotune_configs_for_ci()` generating 8-10 representative configs instead
of 27
- Select configs covering block sizes (64×64×32, 128×128×64, 256×256×128),
thread counts (128, 256), representative data types
- `get_autotune_timeout()` returning 10s for CI, 30-100s for local

Update example autotune tests:
- Modify `examples/elementwise/example_elementwise_add.py` get_configs() to use
CI-aware helper
- Update `examples/gemm/example_gemm_autotune.py` to use 8-10 configs in CI
- Pass `timeout=get_autotune_timeout()` to AutoTuner initialization
- Add `@pytest.mark.autotune` markers to test functions

Optimize testing/python autotune tests:
- Update `test_tilelang_autotune.py` test_autotune_matmul to use 10s timeout and
128×128 matrices in CI
- Update `test_tilelang_autotune_with_inputs.py` test_autotune_matmul_symbolic_m
similarly
- Reduce trial configuration count in CI mode
- Add `@pytest.mark.autotune` and `@pytest.mark.slow` markers
===============================================================================

Task: 5

Set up CI pipeline configuration and comprehensive documentation to enable fast
CI feedback and full local testing capabilities.

Create `.github/workflows/fast-ci-tests.yml`:
- Configure pytest with `-m "not slow or autotune"` for PR fast feedback
- Set `CI=true` environment variable
- Add separate workflow step for slow tests (scheduled/manual trigger)
- Run full test suite with CI shapes on main branch merges

Update pytest configuration:
- Modify `examples/pytest.ini` and create `testing/pytest.ini` if needed
- Add marker definitions matching `pyproject.toml`
- Set default marker expression `-m "not slow"` for CI
- Add addopts for CI-specific flags (`-v --tb=short`)

Create/update `TESTING.md`:
- Document marker system (slow, large_shape, autotune)
- Explain local full test suite execution (`pytest testing/ examples/`)
- Explain CI-equivalent local execution (`CI=true pytest -m "not slow" testing/
examples/`)
- Provide guidance on adding markers to new tests
- Document shape scaling and override options

Add test execution metrics:
- Add `--durations=30` to pytest configuration
- Configure pytest-html for execution time reports
- Add CI output comparison metrics
- Consider pytest-xdist for parallel execution

💡 Iterate on the plan with: @coderabbitai <feedback>

Example Feedback
- @coderabbitai You can skip phase 3. Add a simple unit test case for phase 2.
- @coderabbitai For assumption 1 go ahead with option 3 and replan.

---

💬 Have feedback or questions? Drop into our discord or schedule a call!