CudnnNormConvolution is no longer supported on NVIDIA Hopper GPUs #48203
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
Xreki
approved these changes
Nov 22, 2022
| @@ -45,6 +45,14 @@ struct NormConvolutionArgs { | |||
| int stride, | |||
| int dilation, | |||
| int group) { | |||
| PADDLE_ENFORCE_LT( | |||
There was a problem hiding this comment.
有个小疑问，当初为了进一步优化ResNet50性能，使用cudnnFusedOpsPlan_t相关接口实现了多个融合算子，代码分别在cudnn_bn_stats_finalize.cu.h、cudnn_norm_conv.cu.h和cudnn_scale_bias_add_relu.cu.h，公共类实现在cudnn_fusion_helper.h，请问只有cudnn_norm_conv.cu.h不再支持了吗？
There was a problem hiding this comment.
目前在H100上只看到cudnn_norm_conv.cu.h相关的test挂掉，别的暂时没问题。
zlsh80826
pushed a commit
to zlsh80826/Paddle
that referenced
this pull request
Nov 24, 2022
…ddlePaddle#48203) * Skip tests that use fused_ops on H100 * Add error message to FusedOps on H100
qingqing01
pushed a commit
that referenced
this pull request
Nov 28, 2022
* Reduce squeeze2_matmul_fuse_pass, flattent tests time (#47098) * Add missing fp32 config and reduce the testing combination * Reduce trt matmul pass test max examples * Loose TRT fp16 tests tolerance (#47100) * Loose TRT half test tolerance to 1e-3 (#47101) * Loose TRT half test tolerance to 1e-3 (#47106) * Update distributed_strategy.proto (#46531) * Close popen pipe after used (#47053) * Add launch_bounds (#47285) * Fix TRT UT failures (#47488) * Format cherry-picked commits * CudnnNormConvolution is no longer supported on NVIDIA Hopper GPUs (#48203) * Skip tests that use fused_ops on H100 * Add error message to FusedOps on H100 Co-authored-by: Shijie <505749828@qq.com> Co-authored-by: Leo Chen <39020268+leo0519@users.noreply.github.com> Co-authored-by: Tian Zheng <tizheng@nvidia.com>
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
Bug fixes
PR changes
OPs
Describe
CudnnNormConvolution relies on CUDNN fused_ops, which is deprecated and is no longer supported on Hopper GPUs (H100 and later). We add necessary prompts when the user tries to use this OP on Hopper devices.
Meanwhile, we fix the related unit tests on the new hardware.
Note: This is a duplicate of #47089, which was accidentally closed and could not be reopen.