Move xnnpack.passes to xnnpack._passes (#6088)

Move xnnpack.passes to xnnpack._passes (#5917)

Summary:
Changing `xnnpack.passes` to `xnnpack._passes` to indicate that these passes are not covered under the API stability guarantee.

Pull Request resolved: #5917

Reviewed By: Olivia-liu, helunwencser

Differential Revision: D63925008

fbshipit-source-id: 3d9f13c0a3bd61c66d07cebd62047a3e24f8af1d
(cherry picked from commit 59cc817)

Co-authored-by: Tarun Karuturi <tkaruturi@meta.com>

backends/xnnpack/TARGETS

@@ -16,8 +16,8 @@ runtime.python_library(
     ],
     deps = [
         "//executorch/backends/transforms:lib",
+        "//executorch/backends/xnnpack/_passes:xnnpack_passes",
         "//executorch/backends/xnnpack/operators:operators",
-        "//executorch/backends/xnnpack/passes:xnnpack_passes",
         "//executorch/backends/xnnpack/serialization:xnnpack_serializer",
         "//executorch/exir:graph_module",
         "//executorch/exir/backend:backend_details",

backends/xnnpack/passes/__init__.py → backends/xnnpack/_passes/__init__.py

@@ -6,23 +6,27 @@
 
 from typing import List, Optional, Type
 
-from executorch.backends.xnnpack.passes.channels_last_tagged_reshape_pass import (
+from executorch.backends.xnnpack._passes.channels_last_tagged_reshape_pass import (
     ChannelsLastTaggedReshapePass,
 )
-from executorch.backends.xnnpack.passes.conv1d_unsqueeze_pass import Conv1dUnsqueezePass
-from executorch.backends.xnnpack.passes.convert_to_linear import ConvertToLinearPass
-from executorch.backends.xnnpack.passes.convert_to_sdpa import ConvertToSDPAPass
-from executorch.backends.xnnpack.passes.convert_to_upsample_bilinear2d import (
+from executorch.backends.xnnpack._passes.conv1d_unsqueeze_pass import (
+    Conv1dUnsqueezePass,
+)
+from executorch.backends.xnnpack._passes.convert_to_linear import ConvertToLinearPass
+from executorch.backends.xnnpack._passes.convert_to_sdpa import ConvertToSDPAPass
+from executorch.backends.xnnpack._passes.convert_to_upsample_bilinear2d import (
     ConvertToUpsampleBilinear2d,
 )
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
-from executorch.backends.xnnpack.passes.fuse_batch_norm_with_conv import (
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
+from executorch.backends.xnnpack._passes.fuse_batch_norm_with_conv import (
     FuseBatchNormWithConvPass,
 )
-from executorch.backends.xnnpack.passes.prelu_reshape_pass import PReLUReshapePass
-from executorch.backends.xnnpack.passes.remove_getitem_op import RemoveGetItemPass
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.prelu_reshape_pass import PReLUReshapePass
+from executorch.backends.xnnpack._passes.remove_getitem_op import RemoveGetItemPass
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 
 from executorch.exir.pass_base import ExportPass
 

backends/xnnpack/passes/channels_last_tagged_reshape_pass.py → backends/xnnpack/_passes/channels_last_tagged_reshape_pass.py

@@ -7,7 +7,7 @@
 from typing import Optional, Tuple
 
 import torch
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.utils.utils import is_param_node
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import PassResult

backends/xnnpack/passes/conv1d_unsqueeze_pass.py → backends/xnnpack/_passes/conv1d_unsqueeze_pass.py

@@ -7,7 +7,7 @@
 from typing import Optional
 
 import torch
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.utils.quant_utils import is_dequant, is_quant
 from executorch.backends.xnnpack.utils.utils import get_param_tensor, is_param_node
 from executorch.exir.dialects._ops import ops as exir_ops

backends/xnnpack/passes/convert_to_sdpa.py → backends/xnnpack/_passes/convert_to_sdpa.py

@@ -9,9 +9,9 @@
 
 import torch
 from executorch.backends.transforms import get_shape
-from executorch.backends.xnnpack.partition.graphs import sdpa
 
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack.partition.graphs import sdpa
 from executorch.exir.dialects._ops import ops as exir_ops
 
 from torch.fx.passes.infra.pass_base import PassResult

backends/xnnpack/passes/convert_to_upsample_bilinear2d.py → backends/xnnpack/_passes/convert_to_upsample_bilinear2d.py

@@ -5,8 +5,8 @@
 # LICENSE file in the root directory of this source tree.
 
 import torch
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.partition.graphs import bilinear_2d
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.utils.utils import check_or_raise
 from executorch.exir.dialects._ops import ops as exir_ops
 from torch.fx.passes.infra.pass_base import PassResult

backends/xnnpack/passes/fuse_activation_pass.py → backends/xnnpack/_passes/fuse_activation_pass.py

@@ -7,7 +7,7 @@
 
 import torch
 
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import OutputMinMax
 
 from executorch.backends.xnnpack.utils.utils import check_or_raise

backends/xnnpack/passes/fuse_batch_norm_with_conv.py → backends/xnnpack/_passes/fuse_batch_norm_with_conv.py

@@ -8,7 +8,7 @@
 
 import torch
 
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 
 from executorch.backends.xnnpack.utils.utils import get_param_tensor, is_param_node
 from executorch.exir import ExportedProgram

backends/xnnpack/passes/prelu_reshape_pass.py → backends/xnnpack/_passes/prelu_reshape_pass.py

@@ -5,7 +5,7 @@
 # LICENSE file in the root directory of this source tree.
 
 import torch
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.utils.utils import (
     check_or_raise,
     get_param_tensor,

backends/xnnpack/passes/tag_implicit_q_dq_pass.py → backends/xnnpack/_passes/tag_implicit_q_dq_pass.py

@@ -7,11 +7,11 @@
 from typing import cast, List, Optional
 
 import torch
+from executorch.backends.xnnpack._passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.partition.configs import (
     SUPPORTED_IMPLICIT_Q_DQ_MODULES_SET,
     SUPPORTED_IMPLICIT_Q_DQ_OP_NAMES_SET,
 )
-from executorch.backends.xnnpack.passes.xnnpack_pass import XNNPACKPass
 from executorch.backends.xnnpack.utils.quant_utils import (
     is_dequant,
     is_dynamic_qdq,

backends/xnnpack/operators/node_visitor.py

@@ -11,12 +11,12 @@
 import torch
 from executorch.backends.transforms import get_shape
 
-from executorch.backends.xnnpack.operators.quant_params import QuantParams
-
-from executorch.backends.xnnpack.passes.channels_last_tagged_reshape_pass import (
+from executorch.backends.xnnpack._passes.channels_last_tagged_reshape_pass import (
     ChannelsLastTaggedReshapePass,
 )
 
+from executorch.backends.xnnpack.operators.quant_params import QuantParams
+
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     ConstantDataOffset,
     PerChannelGroupQuant,

backends/xnnpack/operators/op_add.py

@@ -7,12 +7,12 @@
 from typing import Dict
 
 import torch
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNAdd,
     XNNGraph,

backends/xnnpack/operators/op_conv2d.py

@@ -8,12 +8,12 @@
 
 import torch
 from executorch.backends.transforms import get_shape
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNConv2d,
     XNNDepthwiseConv2d,

backends/xnnpack/operators/op_dequantize_per_tensor.py

@@ -7,12 +7,14 @@
 from typing import Dict
 
 import torch
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNConvert,
     XNNGraph,

backends/xnnpack/operators/op_linear.py

@@ -7,13 +7,13 @@
 from typing import Dict
 
 import torch
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.operators.node_visitor import (
     get_input_node,
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNFullyConnected,
     XNNGraph,

backends/xnnpack/operators/op_permute.py

@@ -7,13 +7,13 @@
 from typing import cast, Dict, List
 
 import torch
+from executorch.backends.xnnpack._passes.channels_last_tagged_reshape_pass import (
+    ChannelsLastTaggedReshapePass,
+)
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
-from executorch.backends.xnnpack.passes.channels_last_tagged_reshape_pass import (
-    ChannelsLastTaggedReshapePass,
-)
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNGraph,
     XNNStaticTranspose,

backends/xnnpack/operators/op_quantize_per_tensor.py

@@ -7,12 +7,14 @@
 from typing import Dict
 
 import torch
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNConvert,
     XNNGraph,

backends/xnnpack/operators/op_sub.py

@@ -7,12 +7,12 @@
 from typing import Dict
 
 import torch
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.operators.node_visitor import (
     NodeVisitor,
     register_node_visitor,
 )
 from executorch.backends.xnnpack.operators.quant_params import QuantParams
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     XNNGraph,
     XNNSubtract,

backends/xnnpack/operators/quant_params.py

@@ -9,7 +9,9 @@
 from typing import cast, Optional, Union
 
 import torch
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
 from executorch.backends.xnnpack.utils.quant_utils import (
     extract_qdq_affine_op_args_for_decomposed_ops,
     is_affine_qdq,

backends/xnnpack/partition/config/node_configs.py

@@ -9,13 +9,13 @@
 from typing import List, Optional
 
 import torch
+from executorch.backends.xnnpack._passes.fuse_batch_norm_with_conv import (
+    FuseBatchNormWithConvPass,
+)
 from executorch.backends.xnnpack.partition.config.xnnpack_config import (
     ConfigPrecisionType,
     XNNPartitionerConfig,
 )
-from executorch.backends.xnnpack.passes.fuse_batch_norm_with_conv import (
-    FuseBatchNormWithConvPass,
-)
 from executorch.backends.xnnpack.utils.utils import is_param_node
 from executorch.exir.backend.canonical_partitioners.config_partitioner import (
     format_target_name,

backends/xnnpack/test/TARGETS

@@ -13,7 +13,7 @@ runtime.python_test(
         "test_xnnpack_utils_classes.py",
     ],
     deps = [
-        "//executorch/backends/xnnpack/passes:xnnpack_passes",
+        "//executorch/backends/xnnpack/_passes:xnnpack_passes",
         "//executorch/backends/xnnpack/test/tester:tester",
         "//executorch/backends/xnnpack/utils:xnnpack_utils",
         "//executorch/exir:lib",

backends/xnnpack/test/passes/test_activation_fusion.py

@@ -7,8 +7,8 @@
 import unittest
 
 import torch
-from executorch.backends.xnnpack.passes.convert_to_linear import ConvertToLinearPass
-from executorch.backends.xnnpack.passes.fuse_activation_pass import FuseActivationPass
+from executorch.backends.xnnpack._passes.convert_to_linear import ConvertToLinearPass
+from executorch.backends.xnnpack._passes.fuse_activation_pass import FuseActivationPass
 from executorch.backends.xnnpack.test.tester import RunPasses, Tester
 from executorch.exir.dialects._ops import ops as exir_ops
 

backends/xnnpack/test/passes/test_batch_norm_fusion.py

@@ -8,7 +8,7 @@
 from typing import Tuple
 
 import torch
-from executorch.backends.xnnpack.passes.fuse_batch_norm_with_conv import (
+from executorch.backends.xnnpack._passes.fuse_batch_norm_with_conv import (
     FuseBatchNormWithConvPass,
 )
 from executorch.backends.xnnpack.test.tester import RunPasses, Tester

backends/xnnpack/test/passes/test_channels_last_tagged_reshape.py

@@ -7,7 +7,7 @@
 import unittest
 
 import torch
-from executorch.backends.xnnpack.passes.channels_last_tagged_reshape_pass import (
+from executorch.backends.xnnpack._passes.channels_last_tagged_reshape_pass import (
     ChannelsLastTaggedReshapePass,
 )
 from executorch.backends.xnnpack.test.test_xnnpack_utils_classes import (

backends/xnnpack/test/passes/test_convert_to_linear.py

@@ -7,7 +7,7 @@
 import unittest
 
 import torch
-from executorch.backends.xnnpack.passes.convert_to_linear import ConvertToLinearPass
+from executorch.backends.xnnpack._passes.convert_to_linear import ConvertToLinearPass
 from executorch.backends.xnnpack.test.tester import RunPasses, Tester
 
 

backends/xnnpack/test/passes/test_remove_get_item_pass.py

@@ -7,7 +7,7 @@
 import unittest
 
 import torch
-from executorch.backends.xnnpack.passes.remove_getitem_op import RemoveGetItemPass
+from executorch.backends.xnnpack._passes.remove_getitem_op import RemoveGetItemPass
 from executorch.backends.xnnpack.test.tester import RunPasses, Tester
 
 

backends/xnnpack/test/passes/test_tag_implicit_q_dq_pass.py

@@ -7,7 +7,9 @@
 import unittest
 
 import torch
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
 from executorch.backends.xnnpack.test.tester import RunPasses, Tester
 from executorch.exir.backend.canonical_partitioners.duplicate_dequant_node_pass import (
     DuplicateDequantNodePass,

backends/xnnpack/test/tester/tester.py

@@ -14,8 +14,8 @@
 from typing import Any, Callable, Dict, List, Optional, Tuple, Type, Union
 
 import torch
+from executorch.backends.xnnpack._passes import XNNPACKPassManager
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
-from executorch.backends.xnnpack.passes import XNNPACKPassManager
 from executorch.backends.xnnpack.utils.configs import get_xnnpack_edge_compile_config
 from executorch.exir import (
     EdgeCompileConfig,

backends/xnnpack/xnnpack_preprocess.py

@@ -9,11 +9,13 @@
 from typing import Dict, final, List
 
 import torch
-from executorch.backends.xnnpack.operators.node_visitor import get_node_visitors
 
-from executorch.backends.xnnpack.passes import XNNPACKPassManager
-from executorch.backends.xnnpack.passes.convert_to_linear import ConvertToLinearPass
-from executorch.backends.xnnpack.passes.tag_implicit_q_dq_pass import TagImplicitQDqPass
+from executorch.backends.xnnpack._passes import XNNPACKPassManager
+from executorch.backends.xnnpack._passes.convert_to_linear import ConvertToLinearPass
+from executorch.backends.xnnpack._passes.tag_implicit_q_dq_pass import (
+    TagImplicitQDqPass,
+)
+from executorch.backends.xnnpack.operators.node_visitor import get_node_visitors
 
 from executorch.backends.xnnpack.serialization.xnnpack_graph_schema import (
     ConstantDataOffset,

docs/source/native-delegates-executorch-xnnpack-delegate.md

@@ -36,7 +36,7 @@ The `XnnpackPartitioner` also partitions using op targets. It traverses the grap
 
 ##### Passes
 
-Before any serialization, we apply passes on the subgraphs to prepare the graph. These passes are essentially graph transformations that help improve the performance of the delegate. We give an overview of the most significant passes and their function below. For a description of all passes see [here](https://github.com/pytorch/executorch/tree/main/backends/xnnpack/passes):
+Before any serialization, we apply passes on the subgraphs to prepare the graph. These passes are essentially graph transformations that help improve the performance of the delegate. We give an overview of the most significant passes and their function below. For a description of all passes see [here](https://github.com/pytorch/executorch/tree/main/backends/xnnpack/_passes):
 
 * Channels Last Reshape
     * ExecuTorch tensors tend to be contiguous before passing them into delegates, while XNNPACK only accepts channels-last memory layout. This pass minimizes the number of permutation operators inserted to pass in channels-last memory format.

extension/llm/export/builder.py

@@ -16,7 +16,7 @@
 from executorch.backends.transforms.duplicate_dynamic_quant_chain import (
     DuplicateDynamicQuantChainPass,
 )
-from executorch.backends.xnnpack.passes.convert_to_linear import ConvertToLinearPass
+from executorch.backends.xnnpack._passes.convert_to_linear import ConvertToLinearPass
 from executorch.exir import EdgeProgramManager
 from executorch.exir.backend.partitioner import Partitioner