[quant][pt2e][be] Rename qnnpack quantizer to xnnpack quantizer (#105551

)

Summary: att

Test Plan: sandcastle CI and OSS CI

Reviewed By: andrewor14

Differential Revision: D47422894

Pull Request resolved: #105551
Approved by: https://github.com/andrewor14

test/quantization/pt2e/test_quantize_pt2e.py

@@ -21,7 +21,7 @@
     EmbeddingQuantizer,
     FixedQParamsQuantizationSpec,
     OperatorConfig,
-    QNNPackQuantizer,
+    XNNPACKQuantizer,
     QuantizationAnnotation,
     QuantizationSpec,
     Quantizer,
@@ -30,7 +30,7 @@
 from torch.ao.quantization.pt2e.quantizer.composable_quantizer import (  # noqa: F811
     ComposableQuantizer,
 )
-from torch.ao.quantization.pt2e.quantizer.qnnpack_quantizer import (
+from torch.ao.quantization.pt2e.quantizer.xnnpack_quantizer import (
     get_symmetric_quantization_config,
 )
 from torch.ao.quantization.quantize_pt2e import (
@@ -291,7 +291,7 @@ def _verify_symmetric_qnnpack_qat_numerics(
         # PT2 export
 
         model_pt2e = copy.deepcopy(model)
-        quantizer = QNNPackQuantizer()
+        quantizer = XNNPACKQuantizer()
         quantizer.set_global(
             get_symmetric_quantization_config(
                 is_per_channel=is_per_channel, is_qat=True
@@ -354,7 +354,7 @@ def _verify_symmetric_qnnpack_qat_graph(
         with fake quantizes inserted into the correct places.
         # TODO: also verify that metadata is copied over to the new nodes.
         """
-        quantizer = QNNPackQuantizer()
+        quantizer = XNNPACKQuantizer()
         quantizer.set_global(
             get_symmetric_quantization_config(is_per_channel, is_qat=True)
         )
@@ -991,8 +991,8 @@ def get_supported_operators(cls) -> List[OperatorConfig]:
             m, expected_node_list=node_list, expected_node_occurrence=node_occurrence
         )
 
-    def test_qnnpack_quantizer_conv(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_conv(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         example_inputs = (torch.randn(1, 3, 5, 5),)
@@ -1017,8 +1017,8 @@ def test_qnnpack_quantizer_conv(self):
             node_list,
         )
 
-    def test_qnnpack_quantizer_linear(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_linear(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m_eager = TestHelperModules.TwoLinearModule().eval()
@@ -1047,8 +1047,8 @@ def test_qnnpack_quantizer_linear(self):
                 qconfig_mapping,
             )
 
-    def test_qnnpack_quantizer_conv_linear_no_permute(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_conv_linear_no_permute(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         node_occurrence = {
@@ -1072,8 +1072,8 @@ def test_qnnpack_quantizer_conv_linear_no_permute(self):
             qconfig_mapping,
         )
 
-    def test_qnnpack_quantizer_conv_linear(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_conv_linear(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
 
@@ -1097,8 +1097,8 @@ def test_qnnpack_quantizer_conv_linear(self):
             qconfig_mapping,
         )
 
-    def test_qnnpack_quantizer_linear_with_dynamic_shape(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_linear_with_dynamic_shape(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m_eager = TestHelperModules.TwoLinearModule().eval()
@@ -1125,8 +1125,8 @@ def test_qnnpack_quantizer_linear_with_dynamic_shape(self):
             export_with_dynamic_shape=True,
         )
 
-    def test_qnnpack_quantizer_obs_sharing_ops(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_obs_sharing_ops(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m = TestHelperModules.Conv2dWithObsSharingOps().eval()
@@ -1157,7 +1157,7 @@ def test_qnnpack_quantizer_obs_sharing_ops(self):
         self._test_quantizer(m, example_inputs, quantizer, node_occurrence, node_list)
 
     def test_propagate_annotation(self):
-        quantizer = QNNPackQuantizer()
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m = TestHelperModules.Conv2dPropAnnotaton().eval()
@@ -1196,8 +1196,8 @@ def test_propagate_annotation(self):
         }
         self.checkGraphModuleNodes(m, expected_node_occurrence=node_occurrence)
 
-    def test_qnnpack_quantizer_dynamic_linear(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_dynamic_linear(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(
             is_per_channel=True, is_dynamic=True
         )
@@ -1239,8 +1239,8 @@ def test_qnnpack_quantizer_dynamic_linear(self):
                 qconfig_mapping,
             )
 
-    def test_qnnpack_quantizer_dynamic_linear_with_conv(self):
-        quantizer = QNNPackQuantizer()
+    def test_xnnpack_quantizer_dynamic_linear_with_conv(self):
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(
             is_per_channel=False, is_dynamic=True
         )
@@ -1280,12 +1280,12 @@ def test_qnnpack_quantizer_dynamic_linear_with_conv(self):
         )
 
     def test_composable_quantizer_linear_conv(self):
-        dynamic_quantizer = QNNPackQuantizer()
+        dynamic_quantizer = XNNPACKQuantizer()
         operator_config_dynamic = get_symmetric_quantization_config(
             is_per_channel=False, is_dynamic=True
         )
         dynamic_quantizer.set_global(operator_config_dynamic)
-        static_quantizer = QNNPackQuantizer()
+        static_quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         static_quantizer.set_global(operator_config)
         # Note that dynamic quantization must be applied first here.
@@ -1348,7 +1348,7 @@ def validate(self, model: torch.fx.GraphModule) -> None:
             def get_supported_operators(cls) -> List[OperatorConfig]:
                 pass
 
-        quantizer = QNNPackQuantizer()
+        quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         bad_quantizer = BadQuantizer()
@@ -1472,12 +1472,12 @@ def test_embedding_conv_linear_quantization(self):
         example_inputs = (indices,)
 
         embedding_quantizer = EmbeddingQuantizer()
-        dynamic_quantizer = QNNPackQuantizer()
+        dynamic_quantizer = XNNPACKQuantizer()
         operator_config_dynamic = get_symmetric_quantization_config(
             is_per_channel=True, is_dynamic=True
         )
         dynamic_quantizer.set_global(operator_config_dynamic)
-        static_quantizer = QNNPackQuantizer()
+        static_quantizer = XNNPACKQuantizer()
         operator_config = get_symmetric_quantization_config(is_per_channel=True)
         static_quantizer.set_global(operator_config)
         composed_quantizer = ComposableQuantizer(
@@ -1705,7 +1705,7 @@ def _get_getitem_nodes(m: torch.fx.GraphModule):
         (_, original_conv_bn_getitem_node) = _get_getitem_nodes(m)
 
         # Prepare QAT
-        quantizer = QNNPackQuantizer()
+        quantizer = XNNPACKQuantizer()
         quantizer.set_global(
             get_symmetric_quantization_config(is_per_channel=False, is_qat=True)
         )
@@ -1774,10 +1774,8 @@ def __init__(self):
             def forward(self, x, y):
                 return x + y
 
-        import torch.ao.quantization.pt2e.quantizer.qnnpack_quantizer as qq
-
-        quantizer = QNNPackQuantizer()
-        operator_config = qq.get_symmetric_quantization_config(is_per_channel=True)
+        quantizer = XNNPACKQuantizer()
+        operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m_eager = M().eval()
 
@@ -1799,10 +1797,8 @@ def __init__(self):
             def forward(self, x, y):
                 return x + y
 
-        import torch.ao.quantization.pt2e.quantizer.qnnpack_quantizer as qq
-
-        quantizer = QNNPackQuantizer()
-        operator_config = qq.get_symmetric_quantization_config(is_per_channel=True)
+        quantizer = XNNPACKQuantizer()
+        operator_config = get_symmetric_quantization_config(is_per_channel=True)
         quantizer.set_global(operator_config)
         m_eager = M().eval()
 
@@ -1880,7 +1876,7 @@ def forward(self, input_tensor, hidden_tensor):
                 aten_graph=True,
                 tracing_mode="real",
             )
-            quantizer = QNNPackQuantizer()
+            quantizer = XNNPACKQuantizer()
             operator_config = get_symmetric_quantization_config(
                 is_per_channel=False, is_dynamic=False
             )
@@ -1944,7 +1940,7 @@ def forward(self, input_tensor, hidden_tensor):
                 aten_graph=True,
                 tracing_mode="real",
             )
-            quantizer = QNNPackQuantizer()
+            quantizer = XNNPACKQuantizer()
             operator_config = get_symmetric_quantization_config(
                 is_per_channel=False, is_dynamic=False
             )
@@ -1972,7 +1968,7 @@ def test_resnet18_with_quantizer_api(self):
                 aten_graph=True,
             )
 
-            quantizer = QNNPackQuantizer()
+            quantizer = XNNPACKQuantizer()
             operator_config = get_symmetric_quantization_config(is_per_channel=True)
             quantizer.set_global(operator_config)
             m = prepare_pt2e(m, quantizer)

torch/_dynamo/skipfiles.py

@@ -140,7 +140,7 @@ def _module_dir(m: types.ModuleType):
 # `torch.ao.quantization._pt2e`, which interferes with memory profiling
 FILENAME_ALLOWLIST |= {
     _module_dir(torch) + "ao/quantization/pt2e/qat_utils.py",
-    _module_dir(torch) + "ao/quantization/pt2e/quantizer/qnnpack_quantizer.py",
+    _module_dir(torch) + "ao/quantization/pt2e/quantizer/xnnpack_quantizer.py",
     _module_dir(torch) + "ao/quantization/pt2e/representation/rewrite.py",
     _module_dir(torch) + "ao/quantization/pt2e/utils.py",
 }

torch/ao/quantization/pt2e/_propagate_annotation.py

@@ -9,7 +9,7 @@
 
 
 def _is_share_obs_or_fq_op(op: Callable) -> bool:
-    # TODO: remove some of these ops in qnnpack_quantizer
+    # TODO: remove some of these ops in xnnpack_quantizer
     return op in [
         torch.ops.aten.hardtanh.default,
         torch.ops.aten.mean.default,

torch/ao/quantization/pt2e/quantizer/__init__.py

@@ -1,4 +1,4 @@
-from .qnnpack_quantizer import QNNPackQuantizer
+from .xnnpack_quantizer import XNNPACKQuantizer
 from .quantizer import (
     DerivedQuantizationSpec,
     EdgeOrNode,
@@ -25,7 +25,7 @@
     "QuantizationConfig",
     "EmbeddingQuantizer",
     "Quantizer",
-    "QNNPackQuantizer",
+    "XNNPACKQuantizer",
     "QuantizationSpecBase",
     "QuantizationSpec",
     "FixedQParamsQuantizationSpec",

torch/ao/quantization/pt2e/quantizer/x86_inductor_quantizer.py

@@ -19,7 +19,7 @@
     get_weight_qspec,
     get_bias_qspec,
 )
-from .qnnpack_quantizer import (
+from .xnnpack_quantizer import (
     _is_annotated,
 )
 from torch.ao.quantization.observer import (

torch/ao/quantization/pt2e/quantizer/qnnpack_quantizer.py → torch/ao/quantization/pt2e/quantizer/xnnpack_quantizer.py

@@ -10,6 +10,15 @@
 import torch
 import torch._dynamo as torchdynamo
 import torch.nn.functional as F
+from torch.ao.quantization.fake_quantize import FusedMovingAvgObsFakeQuantize
+from torch.ao.quantization.observer import (
+    HistogramObserver,
+    MinMaxObserver,
+    MovingAverageMinMaxObserver,
+    MovingAveragePerChannelMinMaxObserver,
+    PerChannelMinMaxObserver,
+    PlaceholderObserver,
+)
 
 from torch.ao.quantization.pt2e.graph_utils import find_sequential_partitions
 
@@ -23,15 +32,6 @@
     get_output_act_qspec,
     get_weight_qspec,
 )
-from torch.ao.quantization.fake_quantize import FusedMovingAvgObsFakeQuantize
-from torch.ao.quantization.observer import (
-    HistogramObserver,
-    MinMaxObserver,
-    MovingAverageMinMaxObserver,
-    MovingAveragePerChannelMinMaxObserver,
-    PerChannelMinMaxObserver,
-    PlaceholderObserver,
-)
 from torch.ao.quantization.qconfig import _ObserverOrFakeQuantizeConstructor
 
 from torch.fx import Node
@@ -50,7 +50,7 @@
 
 
 __all__ = [
-    "QNNPackQuantizer",
+    "XNNPACKQuantizer",
     "get_symmetric_quantization_config",
 ]
 
@@ -224,7 +224,7 @@ def _is_annotated(nodes: List[Node]):
     return annotated
 
 
-class QNNPackQuantizer(Quantizer):
+class XNNPACKQuantizer(Quantizer):
     supported_config_and_operators = _get_supported_config_and_operators()
 
     def __init__(self):
@@ -259,13 +259,13 @@ def get_supported_operator_for_quantization_config(
                 return ops
         return []
 
-    def set_global(self, quantization_config: QuantizationConfig) -> QNNPackQuantizer:
+    def set_global(self, quantization_config: QuantizationConfig) -> XNNPACKQuantizer:
         self.global_config = quantization_config
         return self
 
     def set_config_for_operator_type(
         self, operator_type: str, quantization_config: QuantizationConfig
-    ) -> QNNPackQuantizer:
+    ) -> XNNPACKQuantizer:
         self.operator_type_config[operator_type] = quantization_config
         return self
 

torch/ao/quantization/quantize_pt2e.py

@@ -28,7 +28,7 @@
     SharedQuantizationSpec,
     DerivedQuantizationSpec,
     QuantizationAnnotation,
-    QNNPackQuantizer,
+    XNNPACKQuantizer,
     EmbeddingQuantizer,
     ComposableQuantizer,
 )
@@ -38,7 +38,7 @@
     get_output_act_qspec,
     get_weight_qspec,
 )
-from torch.ao.quantization.pt2e.quantizer.qnnpack_quantizer import (  # noqa: F401
+from torch.ao.quantization.pt2e.quantizer.xnnpack_quantizer import (  # noqa: F401
     get_symmetric_quantization_config,
 )
 from torch.ao.quantization.backend_config import BackendConfig