Qualcomm AI Engine Direct - Deprecate convert_bmm_to_matmul pass (pytorch#11392)

### Summary
- delete convert_bmm_to_matmul pass
- add torch.ops.aten.matmul.default in skip_decomp_table

### Test plan
General CI

Author: DannyYuyang-quic

backends/qualcomm/_passes/__init__.py

@@ -8,7 +8,6 @@
 from .annotate_quant_attrs import AnnotateQuantAttrs
 from .annotate_stack import AnnotateStack
 from .annotate_unbind import AnnotateUnbind
-from .convert_bmm_to_matmul import ConvertBmmToMatmul
 from .convert_conv1d_to_conv2d import ConvertConv1dToConv2d
 from .convert_square_to_pow import ConvertSquareToPow
 from .decompose_any import DecomposeAny
@@ -46,7 +45,6 @@
     AnnotateQuantAttrs,
     AnnotateStack,
     AnnotateUnbind,
-    ConvertBmmToMatmul,
     ConvertConv1dToConv2d,
     ConvertSquareToPow,
     DecomposeAny,

backends/qualcomm/_passes/convert_bmm_to_matmul.py

@@ -13,7 +13,6 @@
     AnnotateQuantAttrs,
     AnnotateStack,
     AnnotateUnbind,
-    ConvertBmmToMatmul,
     ConvertConv1dToConv2d,
     ConvertSquareToPow,
     DecomposeAny,
@@ -81,7 +80,6 @@ def get_capture_program_passes():
         (AnnotateQuantAttrs, True),
         (AnnotateStack, True),
         (AnnotateUnbind, True),
-        (ConvertBmmToMatmul, True),
         (ConvertConv1dToConv2d, True),
         (DecomposeAny, True),
         (DecomposeColIm, True),

backends/qualcomm/_passes/qnn_pass_manager.py

@@ -64,7 +64,6 @@ def get_passes_dependency_for_capture_program():
         AnnotateQuantAttrs,
         AnnotateStack,
         AnnotateUnbind,
-        ConvertBmmToMatmul,
         ConvertConv1dToConv2d,
         DecomposeAny,
         DecomposeColIm,
@@ -85,12 +84,10 @@ def get_passes_dependency_for_capture_program():
         AnnotateAdaptiveAvgPool1D: [RemoveRedundancy],
         AnnotateQuantAttrs: [
             RecomposePixelUnshuffle,
-            ConvertBmmToMatmul,
             RemoveRedundancy,
         ],
         AnnotateStack: [RemoveRedundancy],
         AnnotateUnbind: [RemoveRedundancy],
-        ConvertBmmToMatmul: [RecomposePixelUnshuffle],
         DecomposeAny: [RemoveRedundancy],
         DecomposeColIm: [FoldQDQ],
         DecomposeLinalgVectorNorm: [RemoveRedundancy],

backends/qualcomm/_passes/utils.py

@@ -50,6 +50,7 @@ def get_skip_decomp_table() -> List[torch._ops.OperatorBase]:
         torch.ops.aten.instance_norm.default,
         torch.ops.aten.leaky_relu.default,
         torch.ops.aten.linear.default,
+        torch.ops.aten.matmul.default,
         torch.ops.aten.pixel_shuffle.default,
         torch.ops.aten.pixel_unshuffle.default,
         torch.ops.aten.prelu.default,