Run decompositions before the quantizer (#7111)

Summary:

In the current flow, decompositions run in `to_edge()`, long after the quantization process is done. This creates a lot of issues, since we cannot quantize any operations contained in the large operators that the graph tracer can give (e.g. aten.scaled_dot_product_attention, aten.rnn_<tanh, relu>.input, and a few others).
Any models using those will see many fp32 operators in the final graph. Running the decomps earlier solves the problem, but we need to retain a couple operators that we do rely on in the quantizer, like `aten.linear`, `aten.conv1d` and `aten.conv2d`.

Reviewed By: zonglinpeng

Differential Revision: D66461406

Author: mcremon-meta

backends/cadence/aot/compiler.py

@@ -28,6 +28,7 @@
     to_edge,
 )
 from executorch.exir.pass_base import PassResult
+from torch._inductor.decomposition import remove_decompositions
 from torch.ao.quantization.pt2e.export_utils import model_is_exported
 from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
 
@@ -59,15 +60,29 @@ def convert_pt2(
     """
 
     # Export with dynamo
-    model_gm = torch.export.export_for_training(model, inputs).module()
+    decomp_table = torch.export.default_decompositions()
+    ops_to_keep = [
+        torch.ops.aten.conv1d.default,
+        torch.ops.aten.conv2d.default,
+        torch.ops.aten.layer_norm.default,
+        torch.ops.aten.linear.default,
+        torch.ops.aten.matmul.default,
+    ]
+    # pyre-fixme[6]: For 1st argument expected `Dict[typing.Callable[..., typing.Any
+    remove_decompositions(decomp_table, ops_to_keep)
+    model_gm = (
+        torch.export.export_for_training(model, inputs)
+        .run_decompositions(decomp_table)
+        .module()
+    )
 
-    if model_gm_has_SDPA(model_gm):  # pyre-fixme[6]
+    if model_gm_has_SDPA(model_gm):
         # Decompose SDPA
-        DecomposeScaledDotProductAttention(False)(model_gm)  # pyre-fixme[6]
+        DecomposeScaledDotProductAttention(False)(model_gm)
 
         # Swap _safe_softmax with _softmax (see https://github.com/pytorch/pytorch/pull/133882
         # for details).
-        result = ReplaceSafeSoftmaxWithSoftmax()(model_gm)  # pyre-fixme[6]
+        result = ReplaceSafeSoftmaxWithSoftmax()(model_gm)
         assert result is not None
         model_gm = result.graph_module