Skip to content

CUDNN v8 support#39577

Merged
tensorflow-copybara merged 4 commits intotensorflow:masterfrom
nluehr:CUDNN8
May 30, 2020
Merged

CUDNN v8 support#39577
tensorflow-copybara merged 4 commits intotensorflow:masterfrom
nluehr:CUDNN8

Conversation

@nluehr
Copy link
Copy Markdown
Contributor

@nluehr nluehr commented May 15, 2020

Enables building against the CUDNN v8 library.

Attn: @reedwm

Note this PR can be applied separately from the cuda 11 PR (39576), but both modify find_cuda_config.py. As a result, they will conflcit on find_cuda_config.py.gz.base64 and that file will need to be regenerated via compress_find_cuda_config.py when the second PR is merged.

@google-ml-butler google-ml-butler bot added the size:XL CL Change Size:Extra Large label May 15, 2020
@gbaned gbaned self-assigned this May 15, 2020
@gbaned gbaned requested a review from reedwm May 15, 2020 14:51
@gbaned gbaned added the awaiting review Pull request awaiting review label May 19, 2020
freedomtan
freedomtan reviewed May 20, 2020
View reviewed changes
@tensorflowbutler tensorflowbutler removed the awaiting review Pull request awaiting review label May 22, 2020
@chsigg
Copy link
Copy Markdown
Contributor

chsigg commented May 25, 2020

Hi Nathan, thank you very much for this PR. I'm going to temporarily mark it as 'ready to pull'. This will simplify testing it on our side and merging it.

@chsigg chsigg added ready to pull PR ready for merge process and removed ready to pull PR ready for merge process labels May 25, 2020
@pkanwar23 pkanwar23 self-requested a review May 29, 2020 17:51
@google-ml-butler google-ml-butler bot added kokoro:force-run Tests on submitted change ready to pull PR ready for merge process labels May 29, 2020
@kokoro-team kokoro-team removed the kokoro:force-run Tests on submitted change label May 29, 2020
@rthadur
Copy link
Copy Markdown
Contributor

rthadur commented May 30, 2020

@nluehr can you please resolve conflicts ?

@chsigg
Copy link
Copy Markdown
Contributor

chsigg commented May 30, 2020

@nluehr can you please resolve conflicts ?

Hi Nathan, you don't need to resolve, it's easier to merge this manually from our side and I'm actively working on that. The CUDA 11 changes have been merged already, and hopefully this one will land until Monday as well. There was some trouble with compiler commands going over the Windows character limit and unrelated breakages.

I'm removing the 'ready to pull' tag again so the automated tools don't try to submit it as well.

@chsigg chsigg removed the ready to pull PR ready for merge process label May 30, 2020
@tensorflow-copybara tensorflow-copybara merged commit 8b4dd17 into tensorflow:master May 30, 2020
tensorflow-copybara pushed a commit that referenced this pull request Jun 1, 2020
@tanzhenyu @tensorflower-gardener
Rollback PR #39577: CUDNN v8 support
5cf536f 
PiperOrigin-RevId: 314219496
Change-Id: I3e36453044e5b73274e05d84f8d7f2cdb5b144ae
ArmageddonKnight pushed a commit to UofT-EcoSystem/tensorflow that referenced this pull request Jun 2, 2020
@chsigg @tensorflower-gardener
Roll PR tensorflow#39577 (CUDNN v8 support) forward with fix:
255f590 
Handle '64_' prefix in the cuDNN version for Windows.

PiperOrigin-RevId: 314385732
Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
fo40225 pushed a commit to fo40225/tensorflow that referenced this pull request Jun 12, 2020
@chsigg @fo40225
Roll PR tensorflow#39577 (CUDNN v8 support) forward with fix:Handle '…
d3141e8 
…64_' prefix in the cuDNN version for Windows.PiperOrigin-RevId: 314385732Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
fo40225 pushed a commit to fo40225/tensorflow that referenced this pull request Jun 12, 2020
@chsigg @fo40225
Roll PR tensorflow#39577 (CUDNN v8 support) forward with fix:
21eee2d 
Handle '64_' prefix in the cuDNN version for Windows.

PiperOrigin-RevId: 314385732
Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
This was referenced Jun 26, 2020
Merged
Open
@YashasSamaga
Copy link
Copy Markdown

YashasSamaga commented Jun 28, 2020

What was the reason for not using WINOGRAD_NONFUSED algorithms?

fo40225 pushed a commit to fo40225/tensorflow that referenced this pull request Jun 30, 2020
@chsigg @fo40225
Roll PR tensorflow#39577 (CUDNN v8 support) forward with fix:
4cfabf1 
Handle '64_' prefix in the cuDNN version for Windows.

PiperOrigin-RevId: 314385732
Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
@YashasSamaga
Copy link
Copy Markdown

YashasSamaga commented Jul 23, 2020

What was the reason for not using WINOGRAD_NONFUSED algorithms?

Just leaving a comment here so that future visitors can get the answer to the above quesiton.

https://forums.developer.nvidia.com/t/variations-in-heuristics-for-selecting-conv-algorithm-in-different-apis/139730/2

@nluehr
Copy link
Copy Markdown
Contributor Author

nluehr commented Sep 4, 2020

The reason was just to match the behavior with cudnn v7.

@nluehr nluehr deleted the CUDNN8 branch January 13, 2021 18:41
gerasiov pushed a commit to gerasiov/tensorflow that referenced this pull request Oct 21, 2021
@chsigg @gerasiov
Roll PR tensorflow#39577 (CUDNN v8 support) forward with fix:
0f646e0 
Handle '64_' prefix in the cuDNN version for Windows.

PiperOrigin-RevId: 314385732
Change-Id: I9b36497dbe460e2e7cad3d755cf9ce8d41647ce3
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@pkanwar23 pkanwar23 pkanwar23 approved these changes

@reedwm reedwm Awaiting requested review from reedwm

+1 more reviewer

@freedomtan freedomtan freedomtan left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

@gbaned gbaned

Labels

cla: yes size:XL CL Change Size:Extra Large

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.