Skip to content

[inductor][cutlass backend] Add 2 stage autotuning aka prescreening #153335

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
wants to merge 11 commits into from
Closed

[inductor][cutlass backend] Add 2 stage autotuning aka prescreening #153335

wants to merge 11 commits into from

Conversation

henrylhtsang
Copy link
Contributor

@henrylhtsang henrylhtsang commented May 10, 2025 •
edited
Loading

Stack from ghstack (oldest at bottom):

Motivation:
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2

AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1

with prescreening:

AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov

@pytorch-bot PyTorch Bot
Copy link

pytorch-bot bot commented May 10, 2025 •
edited
Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/153335

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 2e7b8ff with merge base 59c5fff (image):
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

henrylhtsang added a commit that referenced this pull request May 10, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
7c993ab 
ghstack-source-id: 44042cf
Pull Request resolved: #153335
@henrylhtsang henrylhtsang added the topic: not user facing topic category label May 10, 2025
@henrylhtsang henrylhtsang requested review from eellison and mlazos May 12, 2025 20:28
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
ea671d3 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 13, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
cc9d696 
ghstack-source-id: 427de0b
Pull Request resolved: #153335
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
cc700d4 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 13, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
1167610 
ghstack-source-id: 3ff0c76
Pull Request resolved: #153335
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
638efae 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
72775c9 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
f483135 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 15, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
9620d2f 
ghstack-source-id: c426f6f
Pull Request resolved: #153335
eellison
eellison approved these changes May 15, 2025
View reviewed changes
Copy link
Contributor

@eellison eellison left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

looks good, maybe @ColinPeppler @mlazos as well

torch/_inductor/select_algorithm.py

log.debug("Before pruning using prescreening timings, %d choices", len(choices))
sorted_choices = sorted(timings.keys(), key=lambda choice: timings[choice])
num_to_keep = max(int(math.sqrt(len(choices)) / 4), 8)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should this be config-able ?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should this be config-able ?

I was worried about adding too many configs, so want to hold on a bit.

torch/_inductor/select_algorithm.py Outdated
# reopen DLL
for choice in choices_to_keep:
if isinstance(choice, CUDATemplateCaller):
choice.bmreq.ensure_dll_loaded()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this needed ? the dll_loaded will happen elsewhere as well right ?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this needed ? the dll_loaded will happen elsewhere as well right ?

This is needed otherwise we will lose out the one we just tuned. I can double check again.

torch/_inductor/select_algorithm.py Outdated
choices_to_prune = OrderedSet(
choice.hash_key()
for pos, choice in enumerate(sorted_choices)
if pos > num_to_keep or timings[choice] == float("inf")
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do we need this if pos > num_to_keep ? won't that already be achieved with sorted_choices[:num_to_keep] ?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

do we need this if pos > num_to_keep ? won't that already be achieved with sorted_choices[:num_to_keep] ?

I want to make sure we didn't add the ones with timings=inf to choices.

torch/_inductor/select_algorithm.py Outdated
if pos > num_to_keep or timings[choice] == float("inf")
)
choices = [
choice for choice in choices if choice.hash_key() not in choices_to_prune
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a reason we need choices_to_prune and choices_to_keep ? won't those be mutually exclusive sets ?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a reason we need choices_to_prune and choices_to_keep ? won't those be mutually exclusive sets ?

they are mutually exclusive. I wrote it this way because I didn't want to check if choice in choices_to_prune many times, which is inefficient. Let me re-check if there is a better way

torch/_inductor/select_algorithm.py Outdated
@staticmethod
def prescreen_choices(
choices: list[ChoiceCaller],
):
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: return type

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: return type

mypy keep telling me to return list[CUDATemplateCaller] instead of list[ChoiceCaller]...

torch/_inductor/select_algorithm.py
Comment on lines +2388 to +2389
if len(candidates) < 10:
return []
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So, I guess the logic is, if we have a small enough number of candidates, we don't need to prune ? add comment ?

torch/_inductor/select_algorithm.py Outdated
@staticmethod
def prune_choices_postscreen(
choices: list[ChoiceCaller],
timings: dict[ChoiceCaller, float],
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: more explicit timings ? the timings here are a subset of choices, which isnt necessarily clear

@henrylhtsang henrylhtsang requested a review from ColinPeppler May 15, 2025 20:35
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
1584191 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

with prescreening:
```
AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 16, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
1db74d0 
ghstack-source-id: e72601f
Pull Request resolved: #153335
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
c6d1174 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

with prescreening:
```
AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
@henrylhtsang
Copy link
Contributor Author

@pytorchbot merge

@pytorch-bot pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label May 19, 2025
@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

pytorchmergebot added a commit that referenced this pull request May 20, 2025
@pytorchmergebot
Revert "[inductor][cutlass backend] Add 2 stage autotuning aka prescr…
7b7604f 
…eening (#153335)"

This reverts commit 0c04492.

Reverted #153335 on behalf of https://github.com/malfet due to Breaks lint, see https://hud.pytorch.org/hud/pytorch/pytorch/3742b7fb3ab677cdfd4294a0b683044b67f4c358/1?per_page=50&name_filter=lintrunner&mergeEphemeralLF=true ([comment](#153335 (comment)))
@pytorchmergebot pytorchmergebot added Reverted ci-no-td Do not run TD on this PR labels May 20, 2025
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
8205097 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

with prescreening:
```
AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 20, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
c463258 
ghstack-source-id: d75134a
Pull Request resolved: #153335
@henrylhtsang
Copy link
Contributor Author

@pytorchbot merge

@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

@yangw-dev
Copy link
Contributor

@pytorchbot revert -m "sorry the pr is failed internally D75155648" -c ghfirst

@pytorchmergebot
Copy link
Collaborator

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

pytorchmergebot added a commit that referenced this pull request May 22, 2025
@pytorchmergebot
Revert "[inductor][cutlass backend] Add 2 stage autotuning aka prescr…
025c5cc 
…eening (#153335)"

This reverts commit d237629.

Reverted #153335 on behalf of https://github.com/yangw-dev due to sorry the pr is failed internally [D75155648](https://www.internalfb.com/diff/D75155648) ([comment](#153335 (comment)))
@pytorchmergebot
Copy link
Collaborator

@henrylhtsang your PR has been successfully reverted.

@henrylhtsang
Copy link
Contributor Author

What I did wrong was changing the hash key, which can change the precompilation seen choices logic.

@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
f3e9972 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

with prescreening:
```
AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
@henrylhtsang
Update on "[inductor][cutlass backend] Add 2 stage autotuning aka pre…
2e7b8ff 
…screening"


Motivation: 
By default, we are tuning the cutlass backend kernels on 3 swizzles. There are runtime params, so they share the same underlying kernel, which saves a lot of compilation time. However, autotuning all combinations of {configs} x {swizzles} is still expensive.

Observations:
Winner of the {configs} x {swizzles} autotuning is the same as if we do a greedy search: first find the top X winners of {configs} with swizzle 2 (hardcoded), then autotune on the {top X winner configs} x {swizzles}. In other words, we can use a Greedy algorithm to reduce autotuning time.

I attach the logs below. This somewhat depends on what X is, but a number like 5-10 works pretty well from empirical observations.

Logs:
Baseline:
https://gist.github.com/henrylhtsang/9a604f150a270dc19524f72a5d4dfac2
```
AUTOTUNE mm(2048x2048, 2048x2048)
strides: [2048, 1], [1, 2048]
dtypes: torch.bfloat16, torch.bfloat16
  cuda_cutlass_gemm_1776 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1777 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1778 0.0291 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1800 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1801 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1802 0.0293 ms 99.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9012 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9013 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9014 0.0294 ms 98.9% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8940 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8941 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8942 0.0296 ms 98.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8934 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8935 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8936 0.0297 ms 98.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_2001 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_2002 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_2003 0.0297 ms 97.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1848 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1849 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1850 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8964 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8965 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8966 0.0298 ms 97.6% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_8958 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_8959 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_8960 0.0298 ms 97.5% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1929 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1930 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1931 0.0302 ms 96.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1770 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1771 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1772 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1953 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1954 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1955 0.0302 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_tnn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1995 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1996 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1997 0.0303 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1794 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1795 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1796 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1842 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_1843 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_1844 0.0303 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_9006 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cuda_cutlass_gemm_9007 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cuda_cutlass_gemm_9008 0.0304 ms 95.7% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cuda_cutlass_gemm_1923 0.0306 ms 95.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x1x1_0_tnn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
```

with prescreening:
```
AUTOTUNE mm(147456x6144, 6144x2048)
strides: [6144, 1], [2048, 1]
dtypes: torch.bfloat16, torch.bfloat16
  cutlass_1a5e81af 4.5469 ms 100.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6328 ms 98.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.6836 ms 97.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_161b8b81 4.7224 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_161b8b81 4.7234 ms 96.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7274 ms 96.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_853b6347 4.7369 ms 96.0% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_aa6f899c 4.7404 ms 95.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_161b8b81 4.7711 ms 95.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8148 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_8bc6fbda 4.8159 ms 94.4% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_8bc6fbda 4.8214 ms 94.3% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_8bc6fbda 4.8302 ms 94.1% cutlass3x_sm90_tensorop_s64x256x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_0a1c55af 4.8487 ms 93.8% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_0a1c55af 4.8527 ms 93.7% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_02780d72 4.8617 ms 93.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_0a1c55af 4.8737 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_0a1c55af 4.8738 ms 93.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_02780d72 4.9348 ms 92.1% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_02780d72 4.9763 ms 91.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_853b6347 4.9805 ms 91.3% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.0225 ms 90.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.0271 ms 90.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_02780d72 5.0595 ms 89.9% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_853b6347 5.1434 ms 88.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.1574 ms 88.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=8
  cutlass_1a5e81af 5.1916 ms 87.6% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2018 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=4
  cutlass_c1ffa14b 5.2019 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=1
  cutlass_c1ffa14b 5.2037 ms 87.4% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_256x128x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_1a5e81af 5.5329 ms 82.2% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_2x1x1_0_ttn_align8_stream_k_warpspecialized_cooperative_epi_tma swizzle=2
  cutlass_aa6f899c 11.5046 ms 39.5% cutlass3x_sm90_tensorop_s64x128x16gemm_bf16_bf16_f32_void_bf16_128x256x64_1x2x1_0_ttn_align8_warpspecialized_cooperative_epi_tma swizzle=8
SingleProcess AUTOTUNE benchmarking takes 1.9526 seconds and 0.0352 seconds precompiling for 32 choices
```

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov

[ghstack-poisoned]
henrylhtsang added a commit that referenced this pull request May 23, 2025
@henrylhtsang
[inductor][cutlass backend] Add 2 stage autotuning aka prescreening
0bc6498 
ghstack-source-id: 0c3d53a
Pull Request resolved: #153335
@henrylhtsang
Copy link
Contributor Author

@pytorchbot merge

@pytorchmergebot
Copy link
Collaborator

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging
Check the merge workflow status
here

@henrylhtsang
Copy link
Contributor Author

fyi I think I didn't do the hashing correctly...

@github-actions github-actions bot deleted the gh/henrylhtsang/78/head branch June 27, 2025 02:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@eellison eellison eellison approved these changes

@mlazos mlazos Awaiting requested review from mlazos

@ColinPeppler ColinPeppler Awaiting requested review from ColinPeppler

Assignees
No one assigned
Labels
ci-no-td Do not run TD on this PR ciflow/inductor ciflow/trunk Trigger trunk jobs on your pull request Merged module: inductor Reverted topic: not user facing topic category
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@henrylhtsang @pytorchmergebot @malfet @yangw-dev @eellison