[NPU] refine update_loss_scaling npu kernel #32580
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 your contribution! |
zhiqiu
reviewed
Apr 26, 2021
| auto g = out->mutable_data<T>(place); | ||
| platform::NPUMemsetAsync(static_cast<void*>(g), 0, | ||
| out->numel() * sizeof(T), stream); | ||
| auto runner_zeros = NpuOpRunner("ZerosLike", {*out}, {*out}); |
phlrain
approved these changes
May 8, 2021
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Outdated suggestions cannot be applied.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.
PR types
Performance optimization
PR changes
OPs
Describe
use
ZerosLikeandMemcpyinstead ofNPUMemsetAsync.As shown in the timeline, there is a blank correspondding to
update_loss_scaling_opcaused byNPUMemsetAsync.update_loss_scaling_opcost about 103 ms.If only use
ZerosLiketo replaceNPUMemsetAsync.update_loss_scaling_opwill launch manyZerosLikeNPU ops.update_loss_scaling_opcost about 22.2 ms.update_loss_scaling_opwill launch only 1ZerosLikeNPU op, and then useMemcpyto set tensor to 0.update_loss_scaling_opcost about 5.5 ms.Performance
Speed up: 19448 tokens/s -> 20679 tokens/s, +6.33 %