Skip to content

Make the requant pass call the per_tensor overload #11789

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 19, 2025
Merged

Make the requant pass call the per_tensor overload #11789

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 5 additions & 23 deletions backends/cadence/aot/fuse_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -712,32 +712,14 @@ def _create_requantize_node(
out_dtype: torch.dtype,
graph: torch.fx.Graph,
) -> torch.fx.Node:
in_scale_tensor = graph.call_function(
exir_ops.edge.aten.full.default, args=((1,), in_scale)
)
in_zero_point_tensor = graph.call_function(
exir_ops.edge.aten.full.default,
args=((1,), in_zero_point),
kwargs={"dtype": torch.int32},
)
out_scale_tensor = graph.call_function(
exir_ops.edge.aten.full.default, args=((1,), out_scale)
)
out_zero_point_tensor = graph.call_function(
exir_ops.edge.aten.full.default,
args=((1,), out_zero_point),
kwargs={"dtype": torch.int32},
)
# cadence::requantize(Tensor input, Tensor in_scale, Tensor in_zero_point, Tensor out_scale, Tensor out_zero_point, ScalarType out_dtype) -> Tensor Y
# TODO(hardiksharma): Add support for per-tensor requantize.
return graph.call_function(
exir_ops.edge.cadence.requantize.default,
exir_ops.edge.cadence.requantize.per_tensor,
args=(
in_tensor,
in_scale_tensor,
in_zero_point_tensor,
out_scale_tensor,
out_zero_point_tensor,
in_scale,
in_zero_point,
out_scale,
out_zero_point,
out_dtype,
),
)
Expand Down
2 changes: 1 addition & 1 deletion backends/cadence/aot/remove_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -447,7 +447,7 @@ def call_operator(
kwargs: dict[str, Argument],
meta: NodeMetadata,
) -> ProxyValue:
if op != exir_ops.edge.cadence.requantize.default:
if op != exir_ops.edge.cadence.requantize.per_tensor:
return super().call_operator(op, args, kwargs, meta)

# Parse the args
Expand Down
10 changes: 5 additions & 5 deletions backends/cadence/aot/tests/test_fusion_ops_passes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -298,7 +298,7 @@ def test_force_quant_dequant_fusion(self):
# Verify that dequant/quant pair was replaced with requantize.
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 0,
exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
exir_ops.edge.cadence.requantize.default: 1,
exir_ops.edge.cadence.requantize.per_tensor: 1,
},
)

Expand Down Expand Up @@ -328,7 +328,7 @@ def test_no_replace_quant_permute_dequant_with_requantize(self):
# quantize -> permute -> dequantize should not be replaced with requantize.
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 1,
exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 1,
exir_ops.edge.cadence.requantize.default: 0,
exir_ops.edge.cadence.requantize.per_tensor: 0,
},
)

Expand Down Expand Up @@ -357,7 +357,7 @@ def test_replace_quant_view_dequant_with_requantize(self):
# Verify that dequant/quant pair was replaced with requantize.
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 0,
exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
exir_ops.edge.cadence.requantize.default: 1,
exir_ops.edge.cadence.requantize.per_tensor: 1,
},
)

Expand All @@ -383,7 +383,7 @@ def test_replace_dequant_quant_with_requantize(self):
# Verify that dequant -> quant was replaced with requantize.
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 0,
exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
exir_ops.edge.cadence.requantize.default: 1,
exir_ops.edge.cadence.requantize.per_tensor: 1,
},
)

Expand Down Expand Up @@ -413,7 +413,7 @@ def test_replace_dequant_permute_quant_with_requantize(self):
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 0,
exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
exir_ops.edge.aten.permute_copy.default: 1,
exir_ops.edge.cadence.requantize.default: 1,
exir_ops.edge.cadence.requantize.per_tensor: 1,
},
)

Expand Down
2 changes: 1 addition & 1 deletion backends/cadence/aot/tests/test_reorder_ops_passes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,7 +214,7 @@ def test_advance_branched_quantize(self):
self.assertEqual(
count_node(
graph_module,
exir_ops.edge.cadence.requantize.default,
exir_ops.edge.cadence.requantize.per_tensor,
),
1,
)
Expand Down
Loading