2024-10-30 nightly release (db38bcc)

Author: pytorchbot

.github/workflows/pull.yml

@@ -35,9 +35,6 @@ jobs:
     name: test-setup-linux-gcc
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -50,7 +47,7 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
 
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         # Build and test ExecuTorch with the add model on portable backend.
@@ -89,20 +86,11 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        build-tool: [buck2, cmake]
         mode: [portable, xnnpack+custom, xnnpack+custom+qe]
         include:
           - dtype: bf16
-            build-tool: cmake
-            mode: portable
-          - dtype: bf16
-            build-tool: buck2
             mode: portable
           - dtype: bf16
-            build-tool: cmake
-            mode: custom
-          - dtype: bf16
-            build-tool: buck2
             mode: custom
       fail-fast: false
     with:
@@ -117,11 +105,11 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         DTYPE=${{ matrix.dtype }}
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         MODE=${{ matrix.mode }}
 
         # Setup executorch
-        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh buck2
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         # Install requirements for export_llama
         PYTHON_EXECUTABLE=python bash examples/models/llama/install_requirements.sh
         # Test llama2
@@ -131,9 +119,6 @@ jobs:
     name: test-llama-runner-linux-android
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -146,18 +131,14 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python \
         bash .ci/scripts/build_llama_android.sh  "${BUILD_TOOL}"
 
   test-custom-ops-linux:
     name: test-custom-ops-linux
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: buck2
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -170,7 +151,7 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         # Test custom ops
         PYTHON_EXECUTABLE=python bash examples/portable/custom_ops/test_custom_ops.sh "${BUILD_TOOL}"
@@ -179,10 +160,6 @@ jobs:
     name: test-selective-build-linux
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: buck2
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -195,7 +172,7 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         # Test selective build
         PYTHON_EXECUTABLE=python bash examples/selective_build/test_selective_build.sh "${BUILD_TOOL}"
@@ -235,9 +212,6 @@ jobs:
     name: test-quantized-aot-lib-linux
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -250,17 +224,14 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         PYTHON_EXECUTABLE=python bash examples/xnnpack/quantization/test_quantize.sh "${BUILD_TOOL}" mv2
 
   test-pybind-build-linux:
     name: test-pybind-build-linux
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
     strategy:
-      matrix:
-        include:
-          - build-tool: cmake
       fail-fast: false
     with:
       runner: linux.2xlarge
@@ -274,7 +245,7 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         # build module for executorch.extension.pybindings.portable_lib
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         PYTHON_EXECUTABLE=python \
         EXECUTORCH_BUILD_XNNPACK=ON \
         EXECUTORCH_BUILD_PYBIND=ON \
@@ -357,10 +328,6 @@ jobs:
 
   unittest-arm:
     uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
-    strategy:
-      matrix:
-        include:
-          - build-tool: buck2
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-arm-sdk
@@ -374,7 +341,7 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
 
         # Setup MacOS dependencies as there is no Docker support on MacOS atm
         PYTHON_EXECUTABLE=python \
@@ -396,7 +363,6 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        build-tool: [cmake]
         mode: [qnn]
       fail-fast: false
     with:
@@ -411,14 +377,14 @@ jobs:
         conda activate "${CONDA_ENV}"
 
         DTYPE=${{ matrix.dtype }}
-        BUILD_TOOL=${{ matrix.build-tool }}
+        BUILD_TOOL="cmake"
         MODE=${{ matrix.mode }}
 
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
         PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
 
         # Setup executorch
-        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh buck2
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "${BUILD_TOOL}"
         # Install requirements for export_llama
         PYTHON_EXECUTABLE=python bash examples/models/llama/install_requirements.sh
         # Test llama2

backends/arm/arm_partitioner.py

@@ -62,6 +62,7 @@ def is_node_supported(self, submodules, node: torch.fx.Node) -> bool:
             exir_ops.edge.aten.relu.default,
             exir_ops.edge.aten.rsqrt.default,
             exir_ops.edge.aten._softmax.default,
+            exir_ops.edge.aten.select_copy.int,
             exir_ops.edge.aten._log_softmax.default,
             exir_ops.edge.aten.slice_copy.Tensor,
             exir_ops.edge.aten.sub.Tensor,

backends/arm/operators/__init__.py

@@ -28,6 +28,7 @@
     op_relu,
     op_repeat,
     op_rsqrt,
+    op_select,
     op_sigmoid,
     op_slice,
     op_squeeze,

backends/arm/operators/op_select.py

@@ -0,0 +1,69 @@
+# Copyright 2024 Arm Limited and/or its affiliates.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from typing import List
+
+import serializer.tosa_serializer as ts
+from executorch.backends.arm.operators.node_visitor import (
+    NodeVisitor,
+    register_node_visitor,
+)
+
+from executorch.backends.arm.tosa_mapping import TosaArg
+
+from executorch.backends.arm.tosa_utils import build_reshape, tosa_shape
+from serializer.tosa_serializer import TosaOp
+from torch.fx import Node
+
+
+@register_node_visitor
+class SelectVisitor(NodeVisitor):
+    target = "aten.select_copy.int"
+
+    def __init__(self, *args):
+        super().__init__(*args)
+
+    def define_node(
+        self,
+        node: Node,
+        tosa_graph: ts.TosaSerializer,
+        inputs: List[TosaArg],
+        output: TosaArg,
+        is_quant_node: bool,
+    ) -> None:
+
+        assert len(inputs) == 3
+        input_node, dim, index = inputs
+        shape = input_node.shape
+        rank = len(shape)
+
+        dim = dim.number % rank if dim.number < 0 else dim.number
+        index = index.number % rank if index.number < 0 else index.number
+
+        # For aten.select_copy, the output will be rank[input_shape - 1]
+        # For TOSA rank(in) == rank(out).
+        # Add an intermediate with the same rank
+        expanded_shape = tuple(1 if i == dim else shape[i] for i in range(rank))
+        expanded_shape = tosa_shape(expanded_shape, input_node.dim_order)
+
+        output_reshaped = tosa_graph.addIntermediate(
+            expanded_shape, ts.DType.INT8 if is_quant_node else output.dtype
+        )
+
+        attr_slice = ts.TosaSerializerAttribute()
+
+        start_attr = [index if i == dim else 0 for i in input_node.dim_order]
+        size_attr = [
+            1 if i == dim else input_node.shape[i] for i in input_node.dim_order
+        ]
+
+        attr_slice.SliceAttribute(start_attr, size_attr)
+
+        tosa_graph.addOperator(
+            TosaOp.Op().SLICE, [input_node.name], [output_reshaped.name], attr_slice
+        )
+
+        # Reshape back to original rank of output.
+        build_reshape(tosa_graph, output_reshaped.name, output.shape, output.name)

backends/arm/quantizer/quantization_annotation/generic_annotator.py

@@ -34,6 +34,8 @@
     # torch.ops.aten.view_as_real.default,
     # torch.ops.aten.view_as_real_copy.default,
     torch.ops.aten.view_copy.default,
+    torch.ops.aten.select.int,
+    torch.ops.aten.select_copy.int,
     torch.ops.aten.slice.Tensor,
     torch.ops.aten.slice_copy.Tensor,
     # 'concat' should be handled separately as it has a sequence of inputs and