Revert "Arm backend: Rerun duplicate-user fusion after TOSA lowering" (#21106)

Reverts #21041

This change breaks executorch/backends/arm/test:addmm
test_addmm_u55_INT[basic] on Corstone-300: the model output is all zeros
instead of the reference [[2,4],[6,8]].

Root cause: running FuseDuplicateUsersPass after InsertRescalePass. In
the
[basic] case x1 and x3 are value-identical, so FuseEqualPlaceholdersPass
merges them into a single placeholder that feeds both the matmul operand
and the bias/add path. After rescale insertion, that shared producer has
multiple tosa.RESCALE users, and fusing them corrupts the quantized
addmm
data flow, zeroing the output. Bisected to the late
FuseDuplicateUsersPass
insertion specifically (reordering EnsureUniqueOutputNodesPass is not
the
cause).

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils
@Sebastian-Larsson @robell @rascani

Author: metascroy

backends/arm/_passes/arm_pass_manager.py

@@ -656,14 +656,9 @@ def _tosa_pipeline(
                 SymbolicToTosaShapesPass(),
                 InsertDynamicPaddingPass(),
                 FuseConsecutiveConcatShapesPass(),
-                # No-op removal can expose duplicate users and outputs, so run
-                # FuseDuplicateUsersPass and EnsureUniqueOutputNodesPass afterward.
+                EnsureUniqueOutputNodesPass(),
                 RemoveNoopPass(),
                 InsertRescalePass(),
-                # Late TOSA transformations can introduce duplicate users after
-                # the first FuseDuplicateUsersPass invocation.
-                FuseDuplicateUsersPass(),
-                EnsureUniqueOutputNodesPass(),
             ]
         )
 

backends/arm/test/passes/test_fuse_duplicate_users_pass.py

@@ -8,17 +8,13 @@
 import executorch.backends.arm.tosa.dialect  # noqa: F401
 import torch
 from executorch.backends.arm._passes import FuseDuplicateUsersPass
-from executorch.backends.arm._passes.arm_pass_manager import ArmPassManager
 from executorch.backends.arm.test import common
 from executorch.backends.arm.test.tester.test_pipeline import PassPipeline
-from executorch.backends.arm.tosa.compile_spec import TosaCompileSpec
 from executorch.backends.arm.tosa.specification import (
     TosaLoweringContext,
     TosaSpecification,
 )
-from executorch.exir import EdgeCompileConfig, to_edge
 from executorch.exir.dialects._ops import ops as exir_ops
-from torch.export import export
 from torch.fx import Graph, GraphModule
 
 input_t = Tuple[torch.Tensor]  # Input x
@@ -171,45 +167,3 @@ def test_fuse_duplicate_users_removes_identical_rescale_users():
     assert len(rescale_nodes) == 1
     output_node = result.graph_module.graph.output_node()
     assert output_node.args[0] == (rescale_nodes[0], rescale_nodes[0])
-
-
-class LateDuplicateUsers(torch.nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.register_buffer("first", torch.ones(2, 3))
-        self.register_buffer("second", torch.ones(2, 3))
-
-    def forward(self, x):
-        return x + self.first, x + self.second
-
-
-def test_fuse_duplicate_users_runs_after_tosa_transformations():
-    exported_program = export(LateDuplicateUsers(), (torch.ones(2, 3),), strict=True)
-    edge_program = to_edge(
-        exported_program,
-        compile_config=EdgeCompileConfig(_check_ir_validity=False),
-    )
-    edge_exported_program = edge_program.exported_program()
-
-    graph_module = ArmPassManager(
-        TosaCompileSpec("TOSA-1.0+FP")
-    ).transform_to_backend_pipeline(
-        edge_exported_program, edge_exported_program.graph_module
-    )
-
-    add_nodes = [
-        node
-        for node in graph_module.graph.nodes
-        if node.target == exir_ops.backend.tosa.ADD.default
-    ]
-    identity_nodes = [
-        node
-        for node in graph_module.graph.nodes
-        if node.target == exir_ops.backend.tosa.IDENTITY.default
-    ]
-
-    graph_module.graph.lint()
-    assert len(add_nodes) == 1
-    assert len(identity_nodes) == 2
-    assert all(node.args[0] is add_nodes[0] for node in identity_nodes)
-    assert graph_module.graph.output_node().args[0] == tuple(identity_nodes)