ROCm ❤ TensorExpr

[t-vi]

This might be an alternative to reverting #45396 .
The obvious rough edge is that I'm not really seeing the work group limits that TensorExpr produces.

[dr-ci]

💊 CI failures summary and remediations

As of commit 71bf92a (more details on the Dr. CI page):

---

• 2/3 failures possibly* introduced in this PR
  • 1/2 non-CircleCI failure(s)
• 1/3 broken upstream at merge base b66ac1e on Sep 29 from 12:16pm to 12:31pm PDT (1 commit; b66ac1e - 147c88e)

---

XLA failure

Job pytorch_xla_linux_bionic_py3_6_clang9_build is failing. Please create an issue with title prefixed by [PT_BREAK] in pytorch/xla and link to to this PR. If you have questions, please reach out to @ailzhang / @dlibenzi / @JackCaoG.

---

🚧 1 fixed upstream failure:

These were probably caused by upstream breakages that were already fixed.

Please rebase on the viable/strict branch (expand for instructions)

Since your merge base is older than viable/strict, run these commands:

git fetch https://github.com/pytorch/pytorch viable/strict
git rebase FETCH_HEAD

Check out the recency history of this "viable master" tracking branch.

• pytorch_ios_11_2_1_x86_64_build on Sep 29 from 12:16pm to 12:31pm PDT (1 commit; b66ac1e - 147c88e)
  • 🔁 rerun

---

ci.pytorch.org: 1 failed

• Failed: pr/pytorch-linux-bionic-rocm3.7-py3.6

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 20 times.

[t-vi]

@jeffdaily @ZolotukhinM @mruberry @walterddr

[Krovatkin]

[facebook-github-bot]

@Krovatkin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[nickgg]

@t-vi yeah, uh we don't limit the blockDim in the Cuda backend, we just know that (currently) we limit the thread loop to 512 elements. A definite TODO, but for now might make sense to disable it for ROCm and come back to it.

[t-vi]

Doesn't Cuda have bounds, too? I thought 1024 was the block size limit there.

[zheng-xq]

LGTM in general. Some minor comments.

[zheng-xq]

Please give the constant 128 and 1024 a name here. So people know its meaning.

Also if possible, please include a reference/link where they come from, so future developers know how to update them. Thanks!

[t-vi]

Would describing them in the comment more explicitly work or do you prefer a #define or somesuch?

[zheng-xq]

I would prefer a "static const int kBlcokSizeLimit..." or something like that. I think having both comments and variable names will be even more helpful for future developers. But it is up to you. Thanks!

[t-vi]

I think with the new commentary, it should work. It's a much better comment now, thank you for suggesting that it needed improvement.
What do you think?

[facebook-github-bot]

@Krovatkin has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@Krovatkin merged this pull request in 22a34bc.

[facebook-github-bot]

@Krovatkin merged this pull request in 22a34bc.