Add a pass to remove certain redundant branched quant/dequant nodes

backends/cadence/aot/pass_utils.py

@@ -104,6 +104,16 @@ def count_node(graph_module: torch.fx.GraphModule, target: torch.fx.node.Target)
     return total
 
 
+def op_counts_match(
+    graph_module: torch.fx.GraphModule,
+    expected_op_counts: dict[EdgeOpOverload, int],
+) -> bool:
+    for op, count in expected_op_counts.items():
+        if count_node(graph_module, op) != count:
+            return False
+    return True
+
+
 # Testing utils
 # Return the compute/function nodes in the graph
 def get_compute_nodes_in_gm(graph_module: torch.fx.GraphModule) -> List[torch.fx.Node]:

backends/cadence/aot/remove_ops.py

@@ -33,7 +33,7 @@
 from executorch.backends.cadence.aot.utils import get_edge_overload_packet
 from executorch.backends.transforms.remove_clone_ops import RemoveCloneOpsTransform
 from executorch.exir.dialects._ops import ops as exir_ops
-from executorch.exir.dialects.edge._ops import EdgeOpOverload
+from executorch.exir.dialects.edge._ops import EdgeOpOverload, EdgeOpOverloadPacket
 from executorch.exir.pass_base import ExportPass, NodeMetadata, PassResult, ProxyValue
 from executorch.exir.pass_manager import PassManager, PassType
 from executorch.exir.passes import dead_code_elimination_pass
@@ -745,6 +745,68 @@ def permute_shape(
         return [shape[p] for p in permute_dims]
 
 
+@register_cadence_pass(CadencePassAttribute(opt_level=1))
+class RemoveBranchedQuantDequant(ExportPass):
+    """
+    This pass looks for adjacent quant and dequant nodes with identical
+    parameters, where the quant node has other users in addition to the
+    dequant. The quant and dequant pair would be removed by the
+    FuseQuantDequantToRequantizePass if not for the multiple users. This pass
+    removes just the dequant node by connecting it to the quant's parent node
+    """
+
+    quantize_op_packets: set[EdgeOpOverloadPacket] = {
+        exir_ops.edge.cadence.quantize_per_tensor,
+        exir_ops.edge.quantized_decomposed.quantize_per_tensor,
+    }
+    dequantize_op_packets: set[EdgeOpOverloadPacket] = {
+        exir_ops.edge.cadence.dequantize_per_tensor,
+        exir_ops.edge.quantized_decomposed.dequantize_per_tensor,
+    }
+
+    def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
+        self.remove_branched(
+            graph_module, self.quantize_op_packets, self.dequantize_op_packets
+        )
+        self.remove_branched(
+            graph_module, self.dequantize_op_packets, self.quantize_op_packets
+        )
+
+        graph_module.graph.eliminate_dead_code()
+        result = super().call(graph_module)
+        return result
+
+    def remove_branched(
+        self,
+        graph_module: torch.fx.GraphModule,
+        producer_pkts: set[EdgeOpOverloadPacket],
+        consumer_pkts: set[EdgeOpOverloadPacket],
+    ) -> None:
+        for node in graph_module.graph.nodes:
+            if (
+                node.op != "call_function"
+                or not isinstance(node.target, EdgeOpOverload)
+                or get_edge_overload_packet(node.target) not in producer_pkts
+            ):
+                continue
+
+            if len(node.users) < 2:
+                continue
+
+            for user in node.users:
+                if (
+                    not isinstance(user.target, EdgeOpOverload)
+                    or get_edge_overload_packet(user.target) not in consumer_pkts
+                ):
+                    continue
+
+                # check qparams match
+                if node.args[1:] != user.args[1:]:
+                    continue
+
+                user.replace_all_uses_with(node.args[0])
+
+
 # The following class consolidates functions to remove ops that are redundant
 # in Jarvis. Currently, each function in this class iterates over each node of
 # the graph module once. In future, we could consolidate them into a monolithic
@@ -765,4 +827,5 @@ class CadenceRemoveNops:
         RemoveNopMulOpPass,
         RemoveNopAddOpPass,
         RemoveNopLinalgVectorNormOpPass,
+        RemoveBranchedQuantDequant,
     ]

backends/cadence/aot/tests/test_fusion_ops_passes.py

@@ -20,7 +20,7 @@
     FuseTransposeOpPairsPass,
 )
 from executorch.backends.cadence.aot.graph_builder import GraphBuilder
-from executorch.backends.cadence.aot.pass_utils import count_node
+from executorch.backends.cadence.aot.pass_utils import count_node, op_counts_match
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.dialects.edge._ops import EdgeOpOverload
 from torch import nn
@@ -32,8 +32,7 @@ def check_op_counts(
         graph_module: torch.fx.GraphModule,
         expected_op_counts: dict[EdgeOpOverload, int],
     ) -> None:
-        for op, count in expected_op_counts.items():
-            self.assertEqual(count_node(graph_module, op), count)
+        self.assertTrue(op_counts_match(graph_module, expected_op_counts))
 
 
 class TestFusionPasses(TestFusionPassesBase):

backends/cadence/aot/tests/test_remove_ops_passes.py

@@ -17,10 +17,11 @@
 from executorch.backends.cadence.aot import compiler
 from executorch.backends.cadence.aot.compiler import export_to_edge
 
-from executorch.backends.cadence.aot.pass_utils import count_node
+from executorch.backends.cadence.aot.pass_utils import count_node, op_counts_match
 from executorch.backends.cadence.aot.quantizer.quantizer import CadenceDefaultQuantizer
 from executorch.backends.cadence.aot.remove_ops import (
     RemoveAliasCopyOpPass,
+    RemoveBranchedQuantDequant,
     RemoveCloneOpPass,
     RemoveContiguousOpPass,
     RemoveDetachCopyPass,
@@ -709,3 +710,34 @@ def forward(self, x):
         self.assertEqual(
             count_node(graph_module, exir_ops.edge.aten.permute_copy.default), 2
         )
+
+    def test_remove_dequant_on_branch(self):
+        class M(torch.nn.Module):
+            def forward(self, x):
+                x = torch.abs(x)
+                x0 = torch.ops.quantized_decomposed.quantize_per_tensor(
+                    x, 1.2, 3, 0, 127, torch.int8
+                )
+                x1 = torch.abs(x0)
+                y0 = torch.ops.quantized_decomposed.dequantize_per_tensor(
+                    x0, 1.2, 3, 0, 127, torch.int8
+                )
+                y1 = y0.view(-1)
+                return x1, y1
+
+        inputs = torch.rand(1, 8, 4, 6)
+        model = M()
+        graph_module = export_to_edge(model, (inputs,)).exported_program().graph_module
+
+        graph_module = RemoveBranchedQuantDequant()(graph_module).graph_module
+        self.assertTrue(
+            op_counts_match(
+                graph_module,
+                expected_op_counts={
+                    exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 1,
+                    # we expect the pass to remove the dequantize node
+                    exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
+                    exir_ops.edge.aten.abs.default: 2,
+                },
+            )
+        )