Add CastInt64ToInt32Pass (#5842)

Summary:
Integer placeholders, for example in this examples x = torch.Tensor([1.]) + 1
Are by default lowered as int64 tensors.

As int64 is not a valid TOSA dtype, we cast these
tensors to int32.

Reference:
https://www.mlplatform.org/tosa/tosa_spec.html#_supported_number_formats

Change-Id: Ice59625b7fd68ff3a544ee4648fe05f69b9333e6

Pull Request resolved: #5842

Reviewed By: mergennachin

Differential Revision: D64047286

Pulled By: digantdesai

fbshipit-source-id: 66cce1c5e360082cbe168d45ef2ba9ee492e8787

Author: Erik-Lundell

backends/arm/passes/arm_pass_manager.py

@@ -11,6 +11,7 @@
 from executorch.backends.arm.passes.annotate_channels_last_dim_order_pass import (
     AnnotateChannelsLastDimOrder,
 )
+from executorch.backends.arm.passes.cast_int64_pass import CastInt64ToInt32Pass
 from executorch.backends.arm.passes.convert_expand_copy_to_repeat import (
     ConvertExpandCopyToRepeatPass,
 )
@@ -40,6 +41,7 @@ def transform_to_backend_pipeline(
         self, exported_program: ExportedProgram, compile_spec: list[CompileSpec]
     ):
         """Apply passes before transforming program to backend"""
+        self.add_pass(CastInt64ToInt32Pass(exported_program))
         self.add_pass(SizeAdjustConv2DPass())
         self.add_pass(RemoveClonePass())
         self.add_pass(ConvertExpandCopyToRepeatPass())

backends/arm/passes/cast_int64_pass.py

@@ -0,0 +1,35 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+from executorch.exir.pass_base import ExportPass, PassResult
+
+
+class CastInt64ToInt32Pass(ExportPass):
+    def __init__(self, exported_program: torch.export.ExportedProgram):
+        super(CastInt64ToInt32Pass, self).__init__()
+        self.exported_program = exported_program
+
+    def _to_int32(self, graph_module: torch.fx.GraphModule):
+        for node in graph_module.graph.nodes:
+            fake_tensor = node.meta["val"]
+            if isinstance(fake_tensor, torch._subclasses.fake_tensor.FakeTensor):
+                if node.meta["val"].dtype == torch.int64:
+                    node.meta["val"] = node.meta["val"].to(torch.int32)
+                    buffer_name = (
+                        self.exported_program.graph_signature.inputs_to_buffers[
+                            node.name
+                        ]
+                    )
+                    new_tensor = self.exported_program.state_dict[buffer_name].to(
+                        torch.int32
+                    )
+                    self.exported_program.state_dict[buffer_name] = new_tensor
+
+    def call(self, graph_module: torch.fx.GraphModule):
+        self._to_int32(graph_module)
+        graph_module.recompile()
+        graph_module = super().call(graph_module).graph_module
+        return PassResult(graph_module, True)