You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by "Lunderberg (via GitHub)" <gi...@apache.org> on 2024/03/11 16:11:01 UTC
[PR] [Cutlass][Build] Clean up temporary files after build [tvm]
Lunderberg opened a new pull request, #16700:
URL: https://github.com/apache/tvm/pull/16700
Prior to this commit, compiling cutlass kernels left temporary files in the working directory. These temporary files are then left on disk after building. This commit updates the default behavior when building cutlass to produce files in a temporary directory, which is cleaned up on completion.
Since this change on its own would make it difficult to view and debug the generated cutlass code, this commit also maintains a copy of the generated source in the cutlass module. This can be inspected with `for ext in mod.attrs['external_modules']: print(ext.get_source())`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] [Cutlass][Build] Clean up temporary files after build [tvm]
Posted by "Lunderberg (via GitHub)" <gi...@apache.org>.
Lunderberg commented on PR #16700:
URL: https://github.com/apache/tvm/pull/16700#issuecomment-2004966514
Good point. In that case, we should make it explicit that these are intended as cached directories. Currently, they are named `tmp` on disk and are named `tmp_dir` in the code. All indication is that they are a temporary directory that wasn't cleaned up correctly, not a cache directory which may be used next time.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] [Cutlass][Build] Clean up temporary files after build [tvm]
Posted by "Lunderberg (via GitHub)" <gi...@apache.org>.
Lunderberg commented on PR #16700:
URL: https://github.com/apache/tvm/pull/16700#issuecomment-2102511489
Closing to clean up my open PRs, may revisit in the future.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] [Cutlass][Build] Clean up temporary files after build [tvm]
Posted by "Lunderberg (via GitHub)" <gi...@apache.org>.
Lunderberg closed pull request #16700: [Cutlass][Build] Clean up temporary files after build
URL: https://github.com/apache/tvm/pull/16700
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] [Cutlass][Build] Clean up temporary files after build [tvm]
Posted by "masahi (via GitHub)" <gi...@apache.org>.
masahi commented on PR #16700:
URL: https://github.com/apache/tvm/pull/16700#issuecomment-1989358360
@Lunderberg I think it is better to keep the build dir by default. When we offload matmul or conv2d to cutlass, the build directly contains a cache and many executables for kernel profiling (like below). We want to keep them to make build faster.
```
$ ls tmp
cutlass.o cutlass_tensorop_h1688gemm_128x128_32x2_tn_align8
cutlass_conv2d_cache.pickle cutlass_tensorop_h1688gemm_128x128_32x2_tt_align1
cutlass_gemm_cache.pickle cutlass_tensorop_h1688gemm_128x128_32x2_tt_align2
cutlass_tensorop_h16816fprop_optimized_128x128_32x3_nhwc_align4 cutlass_tensorop_h1688gemm_128x128_32x2_tt_align8
cutlass_tensorop_h16816fprop_optimized_128x128_32x3_nhwc_align8 cutlass_tensorop_h1688gemm_128x256_32x2_tn_align1
cutlass_tensorop_h16816fprop_optimized_128x128_32x4_nhwc_align4 cutlass_tensorop_h1688gemm_128x256_32x2_tn_align2
cutlass_tensorop_h16816fprop_optimized_128x128_32x4_nhwc_align8 cutlass_tensorop_h1688gemm_128x256_32x2_tn_align8
cutlass_tensorop_h16816fprop_optimized_128x128_32x5_nhwc_align4 cutlass_tensorop_h1688gemm_128x256_32x2_tt_align1
cutlass_tensorop_h16816fprop_optimized_128x128_32x5_nhwc_align8 cutlass_tensorop_h1688gemm_128x256_32x2_tt_align2
cutlass_tensorop_h16816fprop_optimized_128x128_64x3_nhwc_align4 cutlass_tensorop_h1688gemm_128x256_32x2_tt_align8
cutlass_tensorop_h16816fprop_optimized_128x128_64x3_nhwc_align8 cutlass_tensorop_h1688gemm_128x64_32x2_tn_align1
cutlass_tensorop_h16816fprop_optimized_128x128_64x4_nhwc_align4 cutlass_tensorop_h1688gemm_128x64_32x2_tn_align2
cutlass_tensorop_h16816fprop_optimized_128x128_64x4_nhwc_align8 cutlass_tensorop_h1688gemm_128x64_32x2_tn_align8
cutlass_tensorop_h16816fprop_optimized_128x256_32x3_nhwc_align4 cutlass_tensorop_h1688gemm_128x64_32x2_tt_align1
cutlass_tensorop_h16816fprop_optimized_128x256_32x3_nhwc_align8 cutlass_tensorop_h1688gemm_128x64_32x2_tt_align2
cutlass_tensorop_h16816fprop_optimized_128x256_64x3_nhwc_align4 cutlass_tensorop_h1688gemm_128x64_32x2_tt_align8
cutlass_tensorop_h16816fprop_optimized_128x256_64x3_nhwc_align8 cutlass_tensorop_h1688gemm_256x128_32x2_tn_align1
cutlass_tensorop_h16816fprop_optimized_128x64_32x6_nhwc_align4 cutlass_tensorop_h1688gemm_256x128_32x2_tn_align2
cutlass_tensorop_h16816fprop_optimized_128x64_32x6_nhwc_align8 cutlass_tensorop_h1688gemm_256x128_32x2_tn_align8
cutlass_tensorop_h16816fprop_optimized_128x64_64x3_nhwc_align4 cutlass_tensorop_h1688gemm_256x128_32x2_tt_align1
cutlass_tensorop_h16816fprop_optimized_128x64_64x3_nhwc_align8 cutlass_tensorop_h1688gemm_256x128_32x2_tt_align2
cutlass_tensorop_h16816fprop_optimized_256x128_32x3_nhwc_align4 cutlass_tensorop_h1688gemm_256x128_32x2_tt_align8
...
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org