[profiler] support cuLaunchKernel (for triton kernel launches) & update kineto submodule #99571
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…te kineto submodule **Background**: Prior to this PR, traces for PT2 w/ inductor don't contain connections between CUDA kernels and the CPU launch site. This PR adds those connections. **Details**: Triton kernels launched by inductor use cuLaunchKernel instead of cudaLaunchKernel. cuLaunchKernel is part of the driver API, while cudaLaunchKernel is part of the runtime API. In order to support cuLaunchKernel, we added support in kineto (pytorch/kineto#752) to also start listening to driver events; hence why we need to update the kineto submodule. After the change in kineto, we just need to turn this on in the PyTorch repo by adding CUDA_DRIVER activity type to the CPU and CUDA activity type lists; then **Testing**: Added test/inductor/test_profiler.py to check for `cuLaunchKernel` in json trace files. Also, I ran this test: ```python import torch x = torch.rand((2, 2), device='cuda') def fn(x): return x.relu() fn_c = torch.compile(fn) fn_c(x) with torch.profiler.profile(with_stack=True) as prof: fn_c(x) prof.export_chrome_trace("relu_profile.json") ``` [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/99571
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 26d50ae: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
davidberard98
added a commit
that referenced
this pull request
Apr 19, 2023
…te kineto submodule **Background**: Prior to this PR, traces for PT2 w/ inductor don't contain connections between CUDA kernels and the CPU launch site. This PR adds those connections. **Details**: Triton kernels launched by inductor use cuLaunchKernel instead of cudaLaunchKernel. cuLaunchKernel is part of the driver API, while cudaLaunchKernel is part of the runtime API. In order to support cuLaunchKernel, we added support in kineto (pytorch/kineto#752) to also start listening to driver events; hence why we need to update the kineto submodule. After the change in kineto, we just need to turn this on in the PyTorch repo by adding CUDA_DRIVER activity type to the CPU and CUDA activity type lists; then **Testing**: Added test/inductor/test_profiler.py to check for `cuLaunchKernel` in json trace files. Also, I ran this test: ```python import torch x = torch.rand((2, 2), device='cuda') def fn(x): return x.relu() fn_c = torch.compile(fn) fn_c(x) with torch.profiler.profile(with_stack=True) as prof: fn_c(x) prof.export_chrome_trace("relu_profile.json") ``` ghstack-source-id: 85fe6b4 Pull Request resolved: #99571
ngimel
approved these changes
Apr 19, 2023
aaronenyeshi
approved these changes
Apr 20, 2023
|
@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 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack:
Background: Prior to this PR, traces for PT2 w/ inductor don't contain connections between CUDA kernels and the CPU launch site. This PR adds those connections.
Details: Triton kernels launched by inductor use cuLaunchKernel instead of cudaLaunchKernel. cuLaunchKernel is part of the driver API, while cudaLaunchKernel is part of the runtime API. In order to support cuLaunchKernel, we added support in kineto (pytorch/kineto#752) to also start listening to driver events; hence why we need to update the kineto submodule.
After the change in kineto, we just need to turn this on in the PyTorch repo by adding CUDA_DRIVER activity type to the CPU and CUDA activity type lists; then
Testing: Added test/inductor/test_profiler.py to check for
cuLaunchKernelin json trace files.Also, I ran this test:
which generated this chrometrace:
in which you can see flows between a
cuLaunchKernelon the CPU side, and the triton kernel on the GPU.Kineto Updates: To get the kineto-side changes required for cupti driver events, this PR updates the kineto pin. In that updated kineto submodule, we also have:
autograd.profile.profilerfails to export valid chrome trace files on Windows #98432)cc @robieta @chaekit @aaronenyeshi @ngimel @nbcsm @guotuofeng @guyang3532 @gaoteng-git @tiffzhaofb @dzhulgakov @soumith @voznesenskym @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @desertfire