[inductor] Fix triton compiler error in multilayer any #109325
Conversation
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109325
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit 86f8837 with merge base 5f6216b ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. ghstack-source-id: d3294c05eb8f46e10eebbcec05299819b4c8dad1 Pull Request resolved: #109325
| other = V.kernel.cse.generate( | ||
| V.kernel.compute, | ||
| f"tl.full({result}.shape, {triton_constant(other)}, {result}.dtype)", | ||
| ) |
There was a problem hiding this comment.
Simply using triton_constant(other) should work, as when both inputs are in the same category, no broadcasting happens.
There was a problem hiding this comment.
I'm not sure what you mean? The previous code already used triton_constant.
There was a problem hiding this comment.
Ah, sorry, what I meant is that it would be enough to give it the right type. It could be simply by casting the 0 to int, float, bool depending on the type of result. But we don't have a way to know the dtype of result during compilation time so I guess this is alright.
We could also always use False, but this would be really odd, so nvm.
|
@pytorchbot merge |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge -r |
|
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx ipiszy ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
|
Successfully rebased |
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. ghstack-source-id: bfea77d766b16a4ddace24d366a501d56c8dc1a3 Pull Request resolved: #109325
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: 1 jobs have failed, first few of them are: inductor-periodic / cuda12.1-py3.10-gcc9-sm86-periodic-dynamo-benchmarks / test (aot_eager_torchbench, 1, 1, linux.g5.4xlarge.nvidia.gpu) Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot rebase |
|
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx ipiszy ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
|
Successfully rebased |
Fixes #109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. ghstack-source-id: 59266820ba8a88b2a7538a489f2e3e242ed3f252 Pull Request resolved: #109325
|
@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: 1 mandatory check(s) failed. The first few are: Dig deeper by viewing the failures on hud |
|
@pytorchbot merge -i |
Merge startedYour change will be merged while ignoring the following 1 checks: pull / linux-focal-py3.8-clang10 / test (dynamo, 1, 2, linux.2xlarge) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
@pytorchbot drci (Please ignore this, I's testing Dr.CI) |
Fixes pytorch#109196 When we have a split reduction and the tensor is not an even multiple of the split size, we use `ops.masked` to pad to an even multiple. In the case here we generated: ```python tmp5 = tl.where(mask, tmp4, 0) ``` which implicitly promotes our boolean value to `int32`. The fix is to give the default value the same dtype as `result`. Pull Request resolved: pytorch#109325 Approved by: https://github.com/lezcano
Stack from ghstack (oldest at bottom):
Fixes #109196
When we have a split reduction and the tensor is not an even multiple of the split size,
we use
ops.maskedto pad to an even multiple. In the case here we generated:which implicitly promotes our boolean value to
int32. The fix is to give the defaultvalue the same dtype as
result.cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @ipiszy @ngimel @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov