[Runtime] Flush L2 cache in time eval #15305
Merged
Conversation
This file contains 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
|
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment. Generated by tvm-bot |
tqchen
reviewed
Jul 12, 2023
|
Also CC: @yzh119 |
|
I suppose we already have a and you can reuse the |
I don't want it to be cuda only |
|
Yes the major concern is that L2 cache size is device specific, and later architectures may have L2 cache greater than 256mb |
|
To make it generalized. how about we instead introduce a l2_cache_flush_bytes, which default to 0, and use that as a parameter to indicate what array to allocate. This way it would generalize across GPUs as long as we set this argument right |
|
The implementation per se is not specific to L2 either. We could say it’s cache_flush_bytes |
|
cache_flush_bytes sounds good to me |
junrushao
approved these changes
Jul 14, 2023
tqchen
approved these changes
Jul 14, 2023
junrushao
changed the title
|
@tvm-bot rerun |
This PR introduces an optional cache flush functionality to `time_evaluator`. It is implemented by allocating two large empty NDArrays on the device so that the L2 cache are flushed. This gives us more accurate evaluation on the performance of a runtime function.
MasterJH5574
pushed a commit
that referenced
this pull request
Jul 18, 2023
Followup of #15305 , this PR creates API to query device L2 cache size in bytes. Currently, the API-supported devices includes CUDA, OpenCL, and ROCM. Note that OpenCL's API does not return the accurate device L2 cache size. I cannot find a Vulkan API that returns L2 texture cache size, but the `vkCmdPipelineBarrier` call will flush the L2 texture cache automatically(https://zeux.io/2020/02/27/writing-an-efficient-vulkan-renderer/), thus we return 0 by default.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 24, 2023
Followup of apache#15305 , this PR creates API to query device L2 cache size in bytes. Currently, the API-supported devices includes CUDA, OpenCL, and ROCM. Note that OpenCL's API does not return the accurate device L2 cache size. I cannot find a Vulkan API that returns L2 texture cache size, but the `vkCmdPipelineBarrier` call will flush the L2 texture cache automatically(https://zeux.io/2020/02/27/writing-an-efficient-vulkan-renderer/), thus we return 0 by default.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 24, 2023
This PR introduces an optional cache flush functionality to `time_evaluator`. It is implemented by allocating two large empty NDArrays on the device so that the L2 cache are flushed. This gives us more accurate evaluation on the performance of a runtime function.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 27, 2023
Followup of apache#15305 , this PR creates API to query device L2 cache size in bytes. Currently, the API-supported devices includes CUDA, OpenCL, and ROCM. Note that OpenCL's API does not return the accurate device L2 cache size. I cannot find a Vulkan API that returns L2 texture cache size, but the `vkCmdPipelineBarrier` call will flush the L2 texture cache automatically(https://zeux.io/2020/02/27/writing-an-efficient-vulkan-renderer/), thus we return 0 by default.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 27, 2023
This PR introduces an optional cache flush functionality to `time_evaluator`. It is implemented by allocating two large empty NDArrays on the device so that the L2 cache are flushed. This gives us more accurate evaluation on the performance of a runtime function.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 30, 2023
Followup of apache#15305 , this PR creates API to query device L2 cache size in bytes. Currently, the API-supported devices includes CUDA, OpenCL, and ROCM. Note that OpenCL's API does not return the accurate device L2 cache size. I cannot find a Vulkan API that returns L2 texture cache size, but the `vkCmdPipelineBarrier` call will flush the L2 texture cache automatically(https://zeux.io/2020/02/27/writing-an-efficient-vulkan-renderer/), thus we return 0 by default.
junrushao
pushed a commit
to junrushao/tvm
that referenced
this pull request
Jul 30, 2023
This PR introduces an optional cache flush functionality to `time_evaluator`. It is implemented by allocating two large empty NDArrays on the device so that the L2 cache are flushed. This gives us more accurate evaluation on the performance of a runtime function.
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.
This PR introduces an optional cache flush functionality to
time_evaluator. It is implemented by allocating two large empty NDArrays on the device so that the L2 cache are flushed. This gives us more accurate evaluation on the performance of a runtime function.