Use libjpeg-turbo

[NicolasHug]

libjpeg-turbo is now available on the pytorch conda channel so we should be able to switch (at least for cuda builds).
Probably superseeds #5951

Addresses #7660

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/7672

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 11 New Failures, 5 Unrelated Failures

As of commit 69e4261:

NEW FAILURES - The following jobs have failed:

• macos (macos-12) / macos-job (gh)
• python-types / linux-job (gh)
• unittests-macos (3.10, macos-12) / macos-job (gh)
• unittests-macos (3.11, macos-12) / macos-job (gh)
• unittests-macos (3.8, macos-12) / macos-job (gh)
• unittests-macos (3.9, macos-12) / macos-job (gh)
• unittests-windows (3.10, windows.4xlarge, cpu) / windows-job (gh)
• unittests-windows (3.11, windows.4xlarge, cpu) / windows-job (gh)
• unittests-windows (3.8, windows.4xlarge, cpu) / windows-job (gh)
• unittests-windows (3.9, windows.4xlarge, cpu) / windows-job (gh)
• windows (windows.4xlarge, cpu) / windows-job (gh)

BROKEN TRUNK - The following jobs failed but were present on the merge base 641fdd9:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• unittests-linux (3.10, linux.12xlarge, cpu) / linux-job (gh)
• unittests-linux (3.11, linux.12xlarge, cpu) / linux-job (gh)
• unittests-linux (3.8, linux.12xlarge, cpu) / linux-job (gh)
• unittests-linux (3.9, linux.12xlarge, cpu) / linux-job (gh)
• unittests-macos (3.8, macos-m1-12) / macos-job (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[pmeier]

libjpeg-turbo is now available on the pytorch conda channel

That is not true and the reason why we are seeing all the errors above. It is available from the pytorch-nightly channel though: https://anaconda.org/pytorch-nightly/libjpeg-turbo

[pmeier]

Meaning, we can't just do conda install -c pytorch-nightly libjpeg-turbo, since that won't work for Windows and macOS: https://anaconda.org/pytorch-nightly/libjpeg-turbo/files

We could either set the channel based on the OS, or let conda figure it out automatically by proving both channels to the initial env creation.

I've opted for the latter for now. Let's see if CI is happy with it.

[pmeier]

We have reverted the fixes I made since libjpeg-turbo should be accessible on the stable pytorch channel soon for all OSs and archs. For now we only have it on pytorch-nightly and only for Linux / x86: https://anaconda.org/pytorch-nightly/libjpeg-turbo/files

Still, note that CI is green for fd39e13 minus a few flaky tests. Meaning, it should work out smoothly.

[pmeier]

LGTM if CI is green. Thanks Nicolas!

[NicolasHug]

CC @malfet @atalman , just making sure this is OK with you before merging. TL;DR we now enforce that build jobs (and tests job) are built with libjpeg-turbo. This will prevent compatibility issues as noticed in #7660

[atalman]

lgtm, we may need to build libjpeg-turbo for aarch64

[github-actions]

Hey @NicolasHug!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py

[NicolasHug]

(labelling as bug, not strictly the case but I want to make sure we advertize this properly in our release notes)