Arm backend: Cherry-pick tosa patch to speed up large models

[ArmRyan]

• Only affects models >2gb

Change-Id: Id5ec79eecc4f1ae8d864e35ed13b34bbf761d30b

cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15592

• 📄 Preview Python docs built from this PR

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

❌ 1 Cancelled Job, 6 Unrelated Failures

As of commit 9650adc with merge base 964515c ():

CANCELLED JOB - The following job was cancelled. Please retry:

• pull / test-binary-size-linux-gcc / linux-job (gh)

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / test-setup-linux-gcc / linux-job (gh) (similar failure)
  ##[error]The operation was canceled.
• Test CUDA Builds / test-models-cuda (add) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

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

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

• pull / test-openvino-linux / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.
• pull / unittest / windows / windows-job (gh) (trunk failure)
  backends/xnnpack/test/ops/test_conv1d.py::TestConv1d::test_qs8_conv1d_batchnorm_seq
• pull / unittest-editable / windows / windows-job (gh) (trunk failure)
  backends/xnnpack/test/ops/test_conv1d.py::TestConv1d::test_qs8_conv1d_batchnorm_seq
• trunk / unittest-release / windows / windows-job (gh) (trunk failure)
  backends/xnnpack/test/ops/test_conv1d.py::TestConv1d::test_qs8_conv1d_batchnorm_seq

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

[zingo]

@SS-JIA here is a change bumping TOSA slightly that you might also want to test, we think it will work if you keep the old version internally if you prefer, this change should only be needed on larger models that are not yet in the tests flow.

[ArmRyan]

@SS-JIA here is a change bumping TOSA slighly that you might also wen to test, we also think to will work if you keep the old version internally, this change should only be needed on larger models that are not yet in the tests flow.

This is just a 2line change in an existing function it doesnt add anything new or change any behaviour, it just resereves memory correctly so nothing changes functionally and no new files or structure changes - wont affect anything

[YufengShi-dudu]

It has been tested internally and shows a significant speedup on large models.