[dynamic shapes] skip fused linear path if not definitely contiguous #155051
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/155051
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 6eba051 with merge base ce9ba07 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…k/unbacked_contig_lin
pianpwk
changed the title
aten/src/ATen/native/Linear.cpp
Outdated
| // Also hit the fused path for contiguous 3D input, if not using xla | ||
| // backend. Reshaping/flattening has some performance implications on xla. | ||
| if (input.is_contiguous() && input_dim == 3) { | ||
| if (definitely_contiguous(input.sym_sizes(), input.sym_strides(), input.sym_numel()) && input_dim == 3) { |
There was a problem hiding this comment.
I see will have to change it again with my change but fine for now. I will handle it once i rebase.
aten/src/ATen/native/Linear.cpp
Outdated
| // Also hit the fused path for contiguous 3D input, if not using xla | ||
| // backend. Reshaping/flattening has some performance implications on xla. | ||
| if (input.is_contiguous() && input_dim == 3) { | ||
| if (definitely_contiguous(input.sym_sizes(), input.sym_strides(), input.sym_numel()) && input_dim == 3) { |
There was a problem hiding this comment.
there is no caching in the current form can you call
definitely_contiguous(input.sym_sizes(), input.sym_strides(), input.sym_numel()) only one time?
|
@pytorchbot merge |
Merge startedYour 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 |
Merge failedReason: 2 jobs have failed, first few of them are: linux-aarch64 / linux-jammy-aarch64-py3.10 / test (default, 2, 3, lf.linux.arm64.m7g.4xlarge), linux-aarch64 / linux-jammy-aarch64-py3.10 / test (default, 4, 4, lf.linux.arm64.2xlarge) Details for Dev Infra teamRaised by workflow job |
…k/unbacked_contig_lin
|
@pytorchbot merge |
Merge startedYour 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 |
3 participants
Falls back to non-fused linear -> add bias path for non-contiguous tensors with unbacked sizes