Add op_amax support

[cccclai]

Summary: As title, add op_amax to support an internal model, add unit test in test_qnn_delegate.py

Differential Revision: D72613814

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9955

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure

As of commit 553aa3a with merge base a238d6a ():

NEW FAILURE - The following job has failed:

• pull / test-llava-runner-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 57d57793e6952825b38e905cace16af300279bef17b64d93114464590402b513 /exec failed with exit code 139

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

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D72613814

[shewu-quic]

Thanks for your effort.

There is one small thing need to add.

diff --git a/backends/qualcomm/_passes/layout_transform.py b/backends/qualcomm/_passes/layout_transform.py
index 17960a602..4d47c38bc 100644
--- a/backends/qualcomm/_passes/layout_transform.py
+++ b/backends/qualcomm/_passes/layout_transform.py
@@ -47,6 +47,7 @@ class LayoutTransform(ExportPass):
     layout_agnostic_ops = {
         exir_ops.edge.aten.abs.default,
         exir_ops.edge.aten.add.Tensor,
+        exir_ops.edge.aten.amax.default,
         exir_ops.edge.aten.bitwise_or.Tensor,
         exir_ops.edge.aten.bmm.default,
         exir_ops.edge.aten.bitwise_and.Tensor,

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D72613814

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D72613814

[cccclai]

Thanks for your effort.

There is one small thing need to add.

diff --git a/backends/qualcomm/_passes/layout_transform.py b/backends/qualcomm/_passes/layout_transform.py
index 17960a602..4d47c38bc 100644
--- a/backends/qualcomm/_passes/layout_transform.py
+++ b/backends/qualcomm/_passes/layout_transform.py
@@ -47,6 +47,7 @@ class LayoutTransform(ExportPass):
     layout_agnostic_ops = {
         exir_ops.edge.aten.abs.default,
         exir_ops.edge.aten.add.Tensor,
+        exir_ops.edge.aten.amax.default,
         exir_ops.edge.aten.bitwise_or.Tensor,
         exir_ops.edge.aten.bmm.default,
         exir_ops.edge.aten.bitwise_and.Tensor,

thanks! updated

[facebook-github-bot]

This pull request was exported from Phabricator. Differential Revision: D72613814

[cccclai]

Quick question, what is the reason that insert_permute defined in LayoutTransform defaults to False? I currently manually set it to True to continue lowering, but it seems required anyway, correct?

[haowhsu-quic]

Quick question, what is the reason that insert_permute defined in LayoutTransform defaults to False? I currently manually set it to True to continue lowering, but it seems required anyway, correct?

This will actually insert the permute node according to layout information but will break graph if you're currently in partitioner stage (the new layout will not match pytorch cpu convention).
We only turn insert_permute to true in preprocess stage for QNN to build graph correctly.