Skip to content

Improve memory viewer and tf-op-profile tool. #2525

Merged
qiuminxu merged 4 commits into
tensorflow:masterfrom
qiuminxu:fix_demo
Aug 10, 2019
Merged

Improve memory viewer and tf-op-profile tool. #2525
qiuminxu merged 4 commits into
tensorflow:masterfrom
qiuminxu:fix_demo

Conversation

@qiuminxu
Copy link
Copy Markdown
Contributor

@qiuminxu qiuminxu commented Aug 8, 2019

  • Motivation for features / changes
    Add bug fixes and feature requests that have been made internally but are not in Github repository.
  • Technical description of changes
    Memory viewer:
  1. Identify hbm heapSimulatorTrace from logical buffer color.
  2. Fix a bug that XLA made changes to heap simulator proto to have must alias support. 'SHARE_WITH' type allocation now needs to be taken into account in heap simulator.
    Op Profile:
  3. Add wasted time and sort by wasted time in the table.
  4. Add memory bandwidth utilization to the header and make its color higher is red (bad) and lower is green (good).
  5. Move unavailable info upper to where performance info is in the op details card.
  • Screenshots of UI changes
    op_profile_demo

(No change to memory viewer UI, only bug fixes.)

  • Detailed steps to verify changes work correctly (as executed by you)
    tensorboard --logdir=gs://cloud-tpu-tools
    Run: 2019-08-02_23:38:56 op_profile and memory viewer tool.

  • Alternate designs / implementations considered

@qiuminxu
Copy link
Copy Markdown
Contributor Author

qiuminxu commented Aug 8, 2019

Error message:
The following URLs are not properly mirrored:

@wchargin
Copy link
Copy Markdown
Contributor

wchargin commented Aug 8, 2019 •
edited
Loading

The following URLs are not properly mirrored:

This is likely a spurious error (network failure); that file has been
mirrored since 2019-05-18T05:51:02Z. You’ve already triggered another
build, which should hopefully succeed. If it persists, let me know and
I’ll investigate.

@qiuminxu
Copy link
Copy Markdown
Contributor Author

qiuminxu commented Aug 8, 2019

The following URLs are not properly mirrored:

This is likely a spurious error (network failure); that file has been
mirrored since 2019-05-18T05:51:02Z. You’ve already triggered another
build, which should hopefully succeed. If it persists, let me know and
I’ll investigate.

Thanks, yes, the test passed after rebuild.

@qiuminxu qiuminxu removed the request for review from tensorboard-gardener August 8, 2019 23:17
stephanwlee
stephanwlee reviewed Aug 8, 2019
View reviewed changes
Comment thread tensorboard/plugins/profile/tf_op_profile/utils.ts Outdated
Comment thread tensorboard/plugins/profile/tf_op_profile/tf-op-table.html Outdated
Comment thread tensorboard/plugins/profile/memory_viewer/memory_usage/memory-usage.ts
Comment thread tensorboard/plugins/profile/tf_op_profile/tf-op-profile.html Outdated
stephanwlee
stephanwlee approved these changes Aug 9, 2019
View reviewed changes
Copy link
Copy Markdown
Contributor

@stephanwlee stephanwlee left a comment

Choose a reason for hiding this comment

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

Of course, make sure to fix the lint error :)

@qiuminxu
Copy link
Copy Markdown
Contributor Author

qiuminxu commented Aug 9, 2019

The first travis build stuck in queued for 8 hours, but when I click into the details and click on the build, it shows all tests passed. Any idea on what's happening and how to fix this? @wchargin @stephanwlee

@qiuminxu qiuminxu merged commit 9759e27 into tensorflow:master Aug 10, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@stephanwlee stephanwlee stephanwlee approved these changes

Assignees

No one assigned

Labels

cla: yes

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@qiuminxu @wchargin @stephanwlee @googlebot