Remove quantize_io from compile_spec (#7647)

quantize_io was only used in arm_partitioner
and is not needed there anymore when
running the delegate in the graph.

Signed-off-by: Erik Lundell <[email protected]>

backends/arm/arm_backend.py

@@ -49,8 +49,7 @@ def __init__(self):
         self.compiler_flags = []
         self.output_format = None
         self.path_for_intermediates = None
-        self.quantize_io = False
-        self.tosa_spec = None
+        self.tosa_version = None
         self.input_order = None
 
     def ethosu_compile_spec(
@@ -123,14 +122,6 @@ def dump_intermediate_artifacts_to(
         self.path_for_intermediates = output_path
         return self
 
-    def set_quantize_io(self, quantize_io: bool = False) -> "ArmCompileSpecBuilder":
-        """
-        Quantization of inputs and dequantization of outputs for cases where
-        whole graph is quantized and method signature is not of quantized type.
-        """
-        self.quantize_io = quantize_io
-        return self
-
     def set_input_order(
         self, input_order: Optional[str] = None
     ) -> "ArmCompileSpecBuilder":
@@ -170,9 +161,6 @@ def build(self) -> List[CompileSpec]:
                 )
             )
 
-        if self.quantize_io:
-            self.compile_spec.append(CompileSpec("quantize_io", "True".encode()))
-
         return self.compile_spec
 
 
@@ -183,13 +171,6 @@ def is_tosa(compile_spec: List[CompileSpec]) -> bool:
     return False
 
 
-def is_quantize_io(compile_specs: List[CompileSpec]) -> bool:
-    for spec in compile_specs:
-        if spec.key == "quantize_io" and spec.value.decode() == "True":
-            return True
-    return False
-
-
 def get_tosa_version(compile_spec: List[CompileSpec]) -> TosaSpecification:
     for spec in compile_spec:
         if spec.key == "tosa_version":

backends/arm/arm_partitioner.py

@@ -12,7 +12,6 @@
 import torch
 from executorch.backends.arm.arm_backend import (
     ArmBackend,
-    is_quantize_io,
 )  # usort: skip
 from executorch.backends.arm.operator_support.tosa_supported_operators import (
     TOSASupportedOperators,
@@ -89,9 +88,6 @@ def is_partitioned(node: torch.fx.Node, tag=tag) -> bool:
                 node.meta["delegation_tag"] = tag
                 partition_tags[tag] = self.delegation_spec
 
-            if not is_quantize_io(self.delegation_spec.compile_specs):
-                continue
-
             # De-tag outmost q-nodes upwards and dq-nodes downwards.
             # De-tag if at least one input/ output is not part of partition.
             for node in partition.nodes:

backends/arm/test/common.py

@@ -78,44 +78,38 @@ def get_tosa_compile_spec_unbuilt(
         ArmCompileSpecBuilder()
         .tosa_compile_spec(tosa_spec)
         .dump_intermediate_artifacts_to(custom_path)
-        .set_quantize_io(True)
     )
 
     return compile_spec_builder
 
 
 def get_u55_compile_spec(
-    quantize_io=True,
     custom_path=None,
     reorder_inputs=None,
 ) -> list[CompileSpec]:
     """
     Default compile spec for Ethos-U55 tests.
     """
     return get_u55_compile_spec_unbuilt(
-        quantize_io=quantize_io,
         custom_path=custom_path,
         reorder_inputs=reorder_inputs,
     ).build()
 
 
 def get_u85_compile_spec(
-    quantize_io=True,
     custom_path=None,
     reorder_inputs=None,
 ) -> list[CompileSpec]:
     """
     Default compile spec for Ethos-U85 tests.
     """
     return get_u85_compile_spec_unbuilt(
-        quantize_io=quantize_io,
         custom_path=custom_path,
         reorder_inputs=reorder_inputs,
     ).build()
 
 
 def get_u55_compile_spec_unbuilt(
-    quantize_io=True,
     custom_path=None,
     reorder_inputs=None,
 ) -> ArmCompileSpecBuilder:
@@ -133,15 +127,13 @@ def get_u55_compile_spec_unbuilt(
             memory_mode="Shared_Sram",
             extra_flags="--debug-force-regor --output-format=raw",
         )
-        .set_quantize_io(quantize_io)
         .dump_intermediate_artifacts_to(artifact_path)
         .set_input_order(reorder_inputs)
     )
     return compile_spec
 
 
 def get_u85_compile_spec_unbuilt(
-    quantize_io=True,
     custom_path=None,
     reorder_inputs=None,
 ) -> list[CompileSpec]:
@@ -157,7 +149,6 @@ def get_u85_compile_spec_unbuilt(
             memory_mode="Shared_Sram",
             extra_flags="--output-format=raw",
         )
-        .set_quantize_io(quantize_io)
         .dump_intermediate_artifacts_to(artifact_path)
         .set_input_order(reorder_inputs)
     )

backends/arm/test/ops/test_depthwise_conv.py

@@ -259,58 +259,48 @@ def test_dw_conv_tosa_BI(self, test_name: str, model: torch.nn.Module):
 
     @parameterized.expand(testsuite_conv2d[:4], skip_on_empty=True)
     @pytest.mark.corstone_fvp
-    def test_dw_conv2d_u55_BI(
-        self, test_name: str, model: torch.nn.Module, set_quantize_io: bool = True
-    ):
+    def test_dw_conv2d_u55_BI(self, test_name: str, model: torch.nn.Module):
         self._test_dw_conv_ethos_BI_pipeline(
             model,
-            common.get_u55_compile_spec(quantize_io=set_quantize_io),
+            common.get_u55_compile_spec(),
             model.get_inputs(),
         )
 
     @parameterized.expand(testsuite_conv2d[4:], skip_on_empty=True)
     @pytest.mark.corstone_fvp
     @conftest.expectedFailureOnFVP  # TODO: MLETORCH-516
-    def test_dw_conv2d_u55_BI_xfails(
-        self, test_name: str, model: torch.nn.Module, set_quantize_io: bool = False
-    ):
+    def test_dw_conv2d_u55_BI_xfails(self, test_name: str, model: torch.nn.Module):
         self._test_dw_conv_ethos_BI_pipeline(
             model,
-            common.get_u55_compile_spec(quantize_io=set_quantize_io),
+            common.get_u55_compile_spec(),
             model.get_inputs(),
         )
 
     @parameterized.expand(testsuite_conv1d, skip_on_empty=True)
     @pytest.mark.corstone_fvp
-    def test_dw_conv1d_u55_BI(
-        self, test_name: str, model: torch.nn.Module, set_quantize_io: bool = True
-    ):
+    def test_dw_conv1d_u55_BI(self, test_name: str, model: torch.nn.Module):
         self._test_dw_conv_ethos_BI_pipeline(
             model,
-            common.get_u55_compile_spec(quantize_io=set_quantize_io),
+            common.get_u55_compile_spec(),
             model.get_inputs(),
         )
 
     @parameterized.expand(testsuite_conv1d + testsuite_conv2d_u85)
     @pytest.mark.corstone_fvp
-    def test_dw_conv_u85_BI(
-        self, test_name: str, model: torch.nn.Module, set_quantize_io: bool = True
-    ):
+    def test_dw_conv_u85_BI(self, test_name: str, model: torch.nn.Module):
         self._test_dw_conv_ethos_BI_pipeline(
             model,
-            common.get_u85_compile_spec(quantize_io=set_quantize_io),
+            common.get_u85_compile_spec(),
             model.get_inputs(),
         )
 
     # All test cases except 3x3_1x3x256x256_gp3_st1 have numerical issues on FVP. MLETORCH-520
     @parameterized.expand(testsuite_conv2d_u85_xfails)
     @pytest.mark.corstone_fvp
     @conftest.expectedFailureOnFVP
-    def test_dw_conv_u85_BI_xfails(
-        self, test_name: str, model: torch.nn.Module, set_quantize_io: bool = True
-    ):
+    def test_dw_conv_u85_BI_xfails(self, test_name: str, model: torch.nn.Module):
         self._test_dw_conv_ethos_BI_pipeline(
             model,
-            common.get_u85_compile_spec(quantize_io=set_quantize_io),
+            common.get_u85_compile_spec(),
             model.get_inputs(),
         )

examples/arm/aot_arm_compiler.py

@@ -275,7 +275,6 @@ def get_compile_spec(
                 memory_mode=memory_mode,
                 extra_flags="--debug-force-regor --output-format=raw --verbose-operators --verbose-cycle-estimate",
             )
-            .set_quantize_io(True)
             .set_input_order(reorder_inputs)
         )
     elif "ethos-u85" in target:
@@ -287,7 +286,6 @@ def get_compile_spec(
                 memory_mode=memory_mode,
                 extra_flags="--output-format=raw --verbose-operators --verbose-cycle-estimate",
             )
-            .set_quantize_io(True)
             .set_input_order(reorder_inputs)
         )