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
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153335
Approved by: https://github.com/eellison
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
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153335
Approved by: https://github.com/eellison
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
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153335
Approved by: https://github.com/eellison
When using Bazel, it’s common to encounter issues like [this](https://github.com/bazelbuild/bazel/issues/14640) and [this](https://github.com/bazel-contrib/rules_python/issues/792) where the `PYTHONPATH` environment variable becomes too long and results in an error such as: `OSError: [Errno 7] Argument list too long` . To work around this, users often resort to custom logic to manipulate PYTHONPATH.
Currently, PyTorch Inductor constructs the PYTHONPATH for a subprocess using sys.path, which can lead to this issue in certain environments.
This PR introduces support for a new environment variable, `TORCH_CUSTOM_PYTHONPATH`, allowing users to override the default `PYTHONPATH` passed to the subprocess. This provides a clean way to avoid an exception when using PyTorch in Bazel.
Please let me know if I need to add some documentation to support this PR. I haven't found an open issue specific to this change but I'm confident that this change (or a similar one) would be appreciated by few.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152832
Approved by: https://github.com/masnesral
Summary: This change is to support remote autotuning. I want to use all the same benchmarking utilities in select_algorithm.py. For remote autotuning, I'll reuse the TritonBenchmarkRequest class used for subprocess autotuning because it's already serializable. That class is also used in standard, in-process autotuning, but via TritonTemplateCaller.benchmark() which sets the output_tensor param when calling the underlying TritonBenchmarkRequest. For remote, I'll be using the TritonBenchmarkRequest request directly so I want the parameter to be named 'out' to avoid "got an unexpected keyword argument 'out'".
Test Plan: Existing unit tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153169
Approved by: https://github.com/aorenste, https://github.com/eellison
Summary:
This PR fixes a bug in the Triton kernel invocation path where the `workspace_tensor` was inserted before the unpacked `extra_args` list in the final kernel argument list. This broke the expected ordering of arguments when dynamic shape size hints are emitted.
When dynamic shapes are used, `extra_args` contains both size hint arguments and grid arguments. The kernel expects the argument list to follow the order: **size hints → workspace tensor → grid args**. But previously, the `workspace_tensor` was inserted before unpacking `extra_args`, resulting in: **workspace tensor → size hints → grid args**, which is incorrect.
This fix constructs the workspace tensor earlier, allowing it to be slotted in after the size hints and before the grid arguments, restoring the expected argument layout.
Test Plan:
contbuild and OSS CI
Reviewers: paulzhan
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152660
Approved by: https://github.com/PaulZhang12, https://github.com/drisspg
Summary:
Currently only `num_warps` and `num_stages` are supported as one of the kernel options for inductor auto-tuning using `TritonTemplate`.
In order to allow warp-specialization kernel options should allow specifying `num_consumer_groups` and `num_buffers_warp_spec` as well.
NOTE: Currently gating changes to FBCODE using HAS_WARP_SPEC which is only available on triton/release-3.3.x
Test Plan:
## Unit test
Added tests for `test_triton_template_warp_specialization` to verify generated kenrnel contains configs for `num_consumer_groups` and `num_buffers_warp_spec`.
## Functional Testing
Specific to flexattention.
```
import torch
from torch.nn.attention.flex_attention import flex_attention
from triton.testing import do_bench
make_tensor = lambda: torch.rand(8, 16, 8192, 128, device="cuda", dtype=torch.bfloat16)
q, k, v = make_tensor(), make_tensor(), make_tensor()
flex_compiled = torch.compile(flex_attention, fullgraph=True)
print(do_bench(lambda: flex_compiled(q, k, v, kernel_options={"num_warps": 4})))
```
triton do_bench results:
- default compile: 15.176783561706543
- with warp-spec: 9.452800750732422
## Extra notes
- generated triton kernel using `TORCH_LOGS=output_code`: P1740612877
- TTGIR for fused kernel: P1740614685
Differential Revision: D71982587
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150122
Approved by: https://github.com/eellison, https://github.com/zou3519, https://github.com/jansel
Summary: A couple follow-ups noted in review from https://github.com/pytorch/pytorch/pull/149700:
1. Make sure we correctly signal _all_ subproces to shutdown, even in the case where some processes are currently benchmarking.
2. Change how the pool singleton is created. That also allows us to fully initialize the object in the ctor and remove a bunch of asserts.
Test Plan: existing unit tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149890
Approved by: https://github.com/aorenste
ghstack dependencies: #149700
Summary: The primary change is to update the autotune-in-a-subproc implementation to avoid using multiprocessing spawn. Spawn (re)executes the toplevel script in the subproc, which can be problematic. The approach here is similar to Triton parallel compile: we Popen a subproc on a controlled entry point and communicate over pipes. That change drove a lot of refactoring in the TuningProcess class, so I took the opportunity to simplify some things, rename some methods, etc.
One other notable change is around the timeout / kill approach. After a timeout, we were previously attempting to stop the subproc in three steps (graceful shutdown, sigkill if graceful fails, sigterm if sigkill fails). I'm gonna argue think that's not useful: 1) The graceful shutdown is never going to work unless the subproc happens to have just completed its task and is ready to receive the next command. 2) If we're going to kill the subproc, let's just take the most aggressive approach and move on as quickly as possible to restarting it rather than waiting to see if previous shutdown attempts succeeded. The only downside that I can find find is maybe a little log spew?, e.g., ` ResourceWarning: subprocess 2987680 is still running`
List of changes:
* Use Popen instead of spawn for the autotuning subprocess.
* Introduced a new entry point `__autotune_main__.py`
* Renamed some TuningProcess methods. For example `shutdown` makes more sense than `terminate` because the latter implies a forced kill.
* Simplified the implementation around benchmarking timeout and how we kill the subproc after a timeout.
* Deprecated the unused timeout configs in `_inductor/config.py`
* Moved `get_ld_library_path` helper to a common utils file.
* Added more unit tests for subproc crashes / timeouts / exceptions, etc.
Test plan:
* New unit tests
* Also ran internally with all combinations of: build mode `opt` and `dev-nosan`, and `buck run` vs. executing the `.par` file directly.
* Made sure the functionality to parallelize autotuning across different GPUs is working (it wasn't clear to me this was behaving the way we wanted it to).
Differential Revision: [D71976971](https://our.internmc.facebook.com/intern/diff/D71976971)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149700
Approved by: https://github.com/aorenste, https://github.com/jansel, https://github.com/eellison
Summary:
Currently only `num_warps` and `num_stages` are supported as one of the kernel options for inductor auto-tuning using `TritonTemplate`. In order to allow warp-specialization kernel options should allow specifying `num_consumer_groups` and `num_buffers_warp_spec` as well.
Test Plan:
## Unit test
Added tests for `test_triton_template_warp_specialization` to verify generated kenrnel contains configs for `num_consumer_groups` and `num_buffers_warp_spec`.
## Functional Testing
Specific to flexattention.
```
import torch
from torch.nn.attention.flex_attention import flex_attention
from triton.testing import do_bench
make_tensor = lambda: torch.rand(8, 16, 8192, 128, device="cuda", dtype=torch.bfloat16)
q, k, v = make_tensor(), make_tensor(), make_tensor()
flex_compiled = torch.compile(flex_attention, fullgraph=True)
print(do_bench(lambda: flex_compiled(q, k, v, kernel_options={"num_warps": 4})))
```
triton do_bench results:
- default compile: 15.176783561706543
- with warp-spec: 9.452800750732422
## Extra notes
- generated triton kernel using `TORCH_LOGS=output_code`: P1740612877
- TTGIR for fused kernel: P1740614685
Differential Revision: D70212243
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148503
Approved by: https://github.com/eellison
Summary: The primary change is to update the autotune-in-a-subproc implementation to avoid using multiprocessing spawn. Spawn (re)executes the toplevel script in the subproc, which can be problematic. The approach here is similar to Triton parallel compile: we Popen a subproc on a controlled entry point and communicate over pipes. That change drove a lot of refactoring in the TuningProcess class, so I took the opportunity to simplify some things, rename some methods, etc.
One other notable change is around the timeout / kill approach. After a timeout, we were previously attempting to stop the subproc in three steps (graceful shutdown, sigkill if graceful fails, sigterm if sigkill fails). I'm gonna argue think that's not useful: 1) The graceful shutdown is never going to work unless the subproc happens to have just completed its task and is ready to receive the next command. 2) If we're going to kill the subproc, let's just take the most aggressive approach and move on as quickly as possible to restarting it rather than waiting to see if previous shutdown attempts succeeded. The only downside that I can find find is maybe a little log spew?, e.g., ` ResourceWarning: subprocess 2987680 is still running`
List of changes:
* Use Popen instead of spawn for the autotuning subprocess.
* Introduced a new entry point `__autotune_main__.py`
* Renamed some TuningProcess methods. For example `shutdown` makes more sense than `terminate` because the latter implies a forced kill.
* Simplified the implementation around benchmarking timeout and how we kill the subproc after a timeout.
* Deprecated the unused timeout configs in `_inductor/config.py`
* Moved `get_ld_library_path` helper to a common utils file.
* Added more unit tests for subproc crashes / timeouts / exceptions, etc.
Test plan:
* New unit tests
* Also ran internally with all combinations of: build mode `opt` and `dev-nosan`, and `buck run` vs. executing the `.par` file directly.
* Made sure the functionality to parallelize autotuning across different GPUs is working (it wasn't clear to me this was behaving the way we wanted it to).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149700
Approved by: https://github.com/aorenste, https://github.com/jansel, https://github.com/eellison
Summary:
Relands D69965761 / https://github.com/pytorch/pytorch/pull/147583
Before this PR, calling a triton kernel would look like:
```py
kernel.run(a, b, xnumel, grid=grid(xnumel), stream=stream0)
```
where the `grid=` was passed as a callable (function closure) arg. This PR removes the grid arg:
```py
kernel.run(a, b, xnumel, stream=stream0)
```
instead now the grid computation is included in the kernel launcher, with something like:
```py
def launcher(in_ptr0, out_ptr0, xnumel, stream):
grid_0 = ((xnumel + 1023) >> 10)
grid_1 = 1
grid_2 = 1
runner(grid_0, grid_1, grid_2, stream, function, metadata, None, launch_enter_hook, launch_exit_hook, in_ptr0, out_ptr0, xnumel)
```
This should be faster, since we remove multiple function/dict calls and are able to specialize the grid computation for each `triton.Config`.
It also allows us to unify the handling of grids between the Python and C++ wrapper code. Before this, C++ wrapper code didn't actually support dynamic grid sizes and instead burned in a static grid.
This unification allows this PR to be a net deletion of code.
Differential [disconnected] Revision: D70471332
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148305
Approved by: https://github.com/shunting314, https://github.com/eellison
Summary:
Relands D69965761 / https://github.com/pytorch/pytorch/pull/147583
Before this PR, calling a triton kernel would look like:
```py
kernel.run(a, b, xnumel, grid=grid(xnumel), stream=stream0)
```
where the `grid=` was passed as a callable (function closure) arg. This PR removes the grid arg:
```py
kernel.run(a, b, xnumel, stream=stream0)
```
instead now the grid computation is included in the kernel launcher, with something like:
```py
def launcher(in_ptr0, out_ptr0, xnumel, stream):
grid_0 = ((xnumel + 1023) >> 10)
grid_1 = 1
grid_2 = 1
runner(grid_0, grid_1, grid_2, stream, function, metadata, None, launch_enter_hook, launch_exit_hook, in_ptr0, out_ptr0, xnumel)
```
This should be faster, since we remove multiple function/dict calls and are able to specialize the grid computation for each `triton.Config`.
It also allows us to unify the handling of grids between the Python and C++ wrapper code. Before this, C++ wrapper code didn't actually support dynamic grid sizes and instead burned in a static grid.
This unification allows this PR to be a net deletion of code.
Differential Revision: D70471332
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148305
Approved by: https://github.com/shunting314, https://github.com/eellison
Before this PR, calling a triton kernel would look like:
```py
kernel.run(a, b, xnumel, grid=grid(xnumel), stream=stream0)
```
where the `grid=` was passed as a callable (function closure) arg. This PR removes the grid arg:
```py
kernel.run(a, b, xnumel, stream=stream0)
```
instead now the grid computation is included in the kernel launcher, with something like:
```py
def launcher(in_ptr0, out_ptr0, xnumel, stream):
grid_0 = ((xnumel + 1023) >> 10)
grid_1 = 1
grid_2 = 1
runner(grid_0, grid_1, grid_2, stream, function, metadata, None, launch_enter_hook, launch_exit_hook, in_ptr0, out_ptr0, xnumel)
```
This should be faster, since we remove multiple function/dict calls and are able to specialize the grid computation for each `triton.Config`.
It also allows us to unify the handling of grids between the Python and C++ wrapper code. Before this, C++ wrapper code didn't actually support dynamic grid sizes and instead burned in a static grid.
This unification allows this PR to be a net deletion of code.
Note the attached diff contains some minor fbcode-only changes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147583
Approved by: https://github.com/eellison, https://github.com/shunting314
**Summary**
Enable the CPP Grouped GEMM Fusion, lowering and Grouped GEMM Template following the RFC: https://github.com/pytorch/pytorch/issues/144012
- Support flexible number of GEMMs
- Share activation across GEMMs
- The Grouped GEMM Template supports independent activations
- However, the pattern matcher requires an anchor node, which is as the shared activation across GEMMs
- Each GEMM can have a unique weight but same sizes
- Each GEMM can have a unique bias or None
- Current PR does not yet support biases; this will be addressed in a follow-up epilogue fusion PR
- Each GEMM have its own epilogues
- Epilogue fusion is not yet supported in this PR and will be enabled in an upcoming follow-up epilogue fusion PR
**Test Plan**
```
python -u -m pytest -s -v test/inductor/test_cpu_select_algorithm.py -k test_grouped_linear
python -u -m pytest -s -v test/inductor/test_cpu_select_algorithm.py -k test_grouped_linear_invalid
python -u -m pytest -s -v test/inductor/test_cpu_cpp_wrapper.py -k test_grouped_linear
```
**Example**
Here is the example and generated code
```
batch_size = 4
in_features = 512
out_features = 1024
dtype = torch.bfloat16
class M(torch.nn.Module):
def __init__(self, bias):
super().__init__()
self.linear0 = torch.nn.Linear(in_features, out_features, bias=False)
self.linear1 = torch.nn.Linear(in_features, out_features, bias=False)
def forward(self, x):
return self.linear0(x), self.linear1(x)
if __name__ == "__main__":
with torch.no_grad():
input = torch.randn(batch_size, in_features, dtype=dtype)
m = M(bias=bias).to(dtype=dtype).eval()
cm = torch.compile(m)
act_res = cm(input)
```
Generated Code: https://gist.github.com/leslie-fang-intel/ed2e8d23aeb3586eb504feeace692e16#file-grouped-gemm-generated-code-py
**Next Step**
- Support Epilogue fusion
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143796
Approved by: https://github.com/jgong5, https://github.com/jansel
This PR aims to add the functionality support of max-autotune for XPU. The current triton templates and configurations are not well optimized for XPU, so the performance is not ready yet. Also the `mm_plus_mm` template have accuracy issues in some cases. We will address these issues in the next PRs.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143266
Approved by: https://github.com/EikanWang, https://github.com/jansel
This PR aims to add the functionality support of max-autotune for XPU. The current triton templates and configurations are not well optimized for XPU, so the performance is not ready yet. Also the `mm_plus_mm` template have accuracy issues in some cases. We will address these issues in the next PRs.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143266
Approved by: https://github.com/EikanWang, https://github.com/jansel
Summary:
The recent tries on bandwidth profiler is not as expected. I have observed a few issues and tried to fix them in this diff:
1. The return of the DebugAutotuner class
2. Profiling results shows really large overhead.
DebugAutotuner.run() returns the benchmark time around 45ms while CachingAutotuner.run() returns the benchmark time around 0.45ms.
The `_find_names` and `re.match` takes 45ms: P1669186358
After we commenting out the above _find_names and re.match, the benchmark time become consistent with non-profiling mode: P1669185589
3. introduce a variable `bandwidth_info` to control the path in DebugAutotuner.run(). During benchmarking of configuration selection, we should turn off the `bandwidth_info`
After applying this diff, the profiling issues mentioned above are fixed: P1669273172
Test Plan:
```
TORCHINDUCTOR_FORCE_DISABLE_CACHES=1 TORCHINDUCTOR_PROFILE=1 TORCHINDUCTOR_PROFILE_OUTPUT=~/tmp/profile.txt TORCH_LOGS='+inductor,+schedule,output_code' TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 CUDA_VISIBLE_DEVICES=5 buck run mode/{opt,inplace} scripts/wwei6/triton_examples:test_mat 2>&1 | tee profiling-5.log
```
If we want to disable the Aten backend, just add TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS="TRITON"
Differential Revision: D64883079
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139607
Approved by: https://github.com/chenyang78
Here's a markdown summary for the PR:
# Add workspace buffer support for Triton templates
## Summary
Adds support for templates to allocate and use temporary workspace buffers
## Key Changes
- Add `WorkspaceArg` support in Triton template system
- Automatic workspace allocation/deallocation around kernel execution
- Zero-initialization support for workspace buffers
- Seamless integration with existing tensor management
## Example Usage
```python
def generate(self, ...):
workspace_arg = WorkspaceArg(
count=1024*1024, # 1MB workspace
zero_fill=True # Zero-initialized
)
return TritonTemplateCaller(..., workspace_arg=workspace_arg)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/138050
Approved by: https://github.com/Chillee, https://github.com/eellison
Summary:
We skip the save_gpu_kernel if kernel is being saved already.
This would give us a more accurate Triton profiling result. The
following trace shows before/after the change for a benchmarking of a
trivial addmm:
Before:
<img width="1255" alt="Screenshot 2024-09-23 at 10 26 53 AM" src="https://github.com/user-attachments/assets/5aea05ef-6ef0-464c-8da9-17b31c97b43a">
After:
<img width="910" alt="Screenshot 2024-09-23 at 10 27 03 AM" src="https://github.com/user-attachments/assets/488b7d4f-268f-41cf-8553-cb16ceeae118">
We can see that before the change, the benchmarking includes two parts,
(1) The overhead of our triton_heuristic call, which includes the
save/get, and the (expensive) hash computation.
(2) The exact computation of Triton kernel.
We see that (1) accounts >50% of time, which makes kernel selection
for profiling choosing aten kernels over Triton kernels.
Test Plan:
Existing OSS CI
python test/inductor/test_cuda_cpp_wrapper.py
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137073
Approved by: https://github.com/desertfire
