Skip to content

Conversation

@ekuznetsov139
Copy link
Contributor

This fixes a few problems that surfaced in ROCm after the switch to hip-clang with version 3.5.

  • Several kernels crash or execute incorrectly without these fixes
  • impl_rsqrt requires changes because hip-clang does FP rounding differently, and code that used to work correctly now results in NaN's for inputs close to zero

@google-ml-butler google-ml-butler bot added the size:S CL Change Size: Small label Aug 12, 2020
Comment on lines 282 to 285
#if TENSORFLOW_COMPILER_IS_HIP_CLANG
// HIP-CLANG has some kind of problem here with 32 threads (possibly because
// the warpsize is 64). Reenable when working properly
} else if (true) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ekuznetsov139
this part is already upstreamed....see lines 251-254 above

@gbaned gbaned self-assigned this Aug 13, 2020
@gbaned gbaned added the comp:gpu GPU related issues label Aug 13, 2020
@gbaned gbaned requested a review from chsigg August 13, 2020 03:23
@gbaned gbaned added the awaiting review Pull request awaiting review label Aug 18, 2020
chsigg
chsigg previously approved these changes Aug 19, 2020
@google-ml-butler google-ml-butler bot added kokoro:force-run Tests on submitted change ready to pull PR ready for merge process labels Aug 19, 2020
@kokoro-team kokoro-team removed the kokoro:force-run Tests on submitted change label Aug 19, 2020
@gbaned gbaned removed the awaiting review Pull request awaiting review label Aug 20, 2020
deven-amd added a commit to ROCm/tensorflow-upstream that referenced this pull request Aug 26, 2020
@google-ml-butler google-ml-butler bot removed the ready to pull PR ready for merge process label Aug 26, 2020
@gbaned gbaned requested a review from chsigg August 27, 2020 06:10
@gbaned gbaned added the awaiting review Pull request awaiting review label Aug 27, 2020
@ekuznetsov139 ekuznetsov139 added the kokoro:force-run Tests on submitted change label Aug 28, 2020
@kokoro-team kokoro-team removed the kokoro:force-run Tests on submitted change label Aug 28, 2020
@ekuznetsov139 ekuznetsov139 force-pushed the google-upstream-rocm35-0812 branch from 599dd70 to 0940c65 Compare September 1, 2020 22:37
@gbaned gbaned requested review from chsigg and removed request for chsigg September 2, 2020 18:46
@ekuznetsov139 ekuznetsov139 force-pushed the google-upstream-rocm35-0812 branch from 0940c65 to c6df176 Compare September 6, 2020 23:28
@gbaned gbaned requested review from chsigg and removed request for chsigg September 11, 2020 18:08
@gbaned gbaned requested a review from cheshire September 23, 2020 11:08
deven-amd added a commit to ROCm/tensorflow-upstream that referenced this pull request Sep 24, 2020
tensorflow-copybara pushed a commit that referenced this pull request Sep 25, 2020
…ocm35-0812

PiperOrigin-RevId: 333738541
Change-Id: Ic848afd875bb7dee980bd7c47a637ae6189f90e3
@gbaned
Copy link
Contributor

gbaned commented Sep 25, 2020

Changes have been merged into master by commit eb0f1ed. So closing the PR.

@gbaned gbaned closed this Sep 25, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

awaiting review Pull request awaiting review cla: yes comp:gpu GPU related issues size:S CL Change Size: Small

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants