Merge branch 'main' into check-tensor-dim-in-IR-verifier

Author: patelvndn

.ci/docker/ci_commit_pins/buck2.txt

@@ -1 +1 @@
-2024-12-16
+2025-05-06

.ci/scripts/build-qnn-sdk.sh

@@ -33,6 +33,7 @@ set_up_aot() {
   cmake .. \
       -DCMAKE_INSTALL_PREFIX=$PWD \
       -DEXECUTORCH_BUILD_QNN=ON \
+      -DANDROID_NATIVE_API_LEVEL=30 \
       -DQNN_SDK_ROOT=${QNN_SDK_ROOT} \
       -DEXECUTORCH_BUILD_DEVTOOLS=ON \
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \

.github/workflows/_link_check.yml

@@ -7,35 +7,51 @@ on:
 
 jobs:
   lint-urls:
+    if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-url-lint') }}
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-linter
-      submodules: 'none'
+      submodules: false
       fetch-depth: 0
       ref: ${{ inputs.ref }}
-      timeout: 90
+      timeout: 120
       script: |
         ./scripts/lint_urls.sh $(
-          [ "${{ github.event_name }}" = "pull_request" ] \
-            && git diff --name-only ${{ github.event.pull_request.base.sha }} ${{ github.event.pull_request.head.sha }} \
-          || [ "${{ github.event_name }}" = "push" ] \
-            && git diff --name-only ${{ github.event.before }} ${{ github.sha }}
-        )
+          { [ "${{ github.event_name }}" = "pull_request" ] \
+              && git diff --name-only "${{ github.event.pull_request.base.sha }}...${{ github.event.pull_request.head.sha }}"; } \
+          || \
+          { [ "${{ github.event_name }}" = "push" ] \
+              && git diff --name-only "${{ github.event.before }}...${{ github.sha }}"; }
+        ) || {
+          echo
+          echo "URL lint failed."
+          echo "If this is a transient outage, you can bypass it by adding the \`skip-url-lint\` label to your PR."
+          echo "Or add \`@lint-ignore\` somewhere on the same line as the URL you want to skip checking."
+          exit 1
+        }
 
   lint-xrefs:
+    if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-xref-lint') }}
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-linter
-      submodules: 'none'
+      submodules: false
       fetch-depth: 0
       ref: ${{ inputs.ref }}
-      timeout: 90
+      timeout: 60
       script: |
         ./scripts/lint_xrefs.sh $(
-          [ "${{ github.event_name }}" = "pull_request" ] \
-            && git diff --name-only ${{ github.event.pull_request.base.sha }} ${{ github.event.pull_request.head.sha }} \
-          || [ "${{ github.event_name }}" = "push" ] \
-            && git diff --name-only ${{ github.event.before }} ${{ github.sha }}
-        )
+          { [ "${{ github.event_name }}" = "pull_request" ] \
+              && git diff --name-only "${{ github.event.pull_request.base.sha }}...${{ github.event.pull_request.head.sha }}"; } \
+          || \
+          { [ "${{ github.event_name }}" = "push" ] \
+              && git diff --name-only "${{ github.event.before }}...${{ github.sha }}"; }
+        ) || {
+          echo
+          echo "Xref lint failed."
+          echo "If this is a transient outage, you can bypass it by adding the \`skip-xref-lint\` label to your PR."
+          echo "Or add \`@lint-ignore\` somewhere on the same line as the reference you want to skip checking."
+          exit 1
+        }

.github/workflows/apple.yml

@@ -5,6 +5,8 @@ on:
     branches:
       - main
       - release/*
+    tags:
+      - ciflow/trunk/*
   pull_request:
     paths:
       - .ci/scripts/setup-ios.sh

.github/workflows/build-presets.yml

@@ -0,0 +1,13 @@
+name: Build Presets
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+      - release/*
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
+  cancel-in-progress: true

CMakeLists.txt

@@ -44,6 +44,14 @@
 
 cmake_minimum_required(VERSION 3.24)
 project(executorch)
+
+# MARK: - Start EXECUTORCH_H12025_BUILD_MIGRATION --------------------------------------------------
+
+include(${PROJECT_SOURCE_DIR}/tools/cmake/common/preset.cmake)
+include(${PROJECT_SOURCE_DIR}/tools/cmake/preset/default.cmake)
+
+# MARK: - End EXECUTORCH_H12025_BUILD_MIGRATION ----------------------------------------------------
+
 include(tools/cmake/Utils.cmake)
 include(CMakeDependentOption)
 
@@ -96,9 +104,6 @@ set(EXECUTORCH_PAL_DEFAULT
           "Which PAL default implementation to use: one of {posix, minimal}"
 )
 
-option(EXECUTORCH_ENABLE_LOGGING "Build with ET_LOG_ENABLED"
-       ${_default_release_disabled_options}
-)
 if(NOT EXECUTORCH_ENABLE_LOGGING)
   # Avoid pulling in the logging strings, which can be large. Note that this
   # will set the compiler flag for all targets in this directory, and for all
@@ -608,7 +613,7 @@ endif()
 # any backends.
 #
 add_library(executorch ${_executorch__srcs})
-target_link_libraries(executorch PUBLIC executorch_core)
+target_link_libraries(executorch PRIVATE executorch_core)
 target_include_directories(executorch PUBLIC ${_common_include_directories})
 target_compile_definitions(executorch PUBLIC C10_USING_CUSTOM_GENERATED_MACROS)
 target_compile_options(executorch PUBLIC ${_common_compile_options})

backends/apple/mps/setup.md

@@ -76,12 +76,12 @@ cd executorch
 ## Run the mv3 generated model using the mps_executor_runner
 
 ```bash
-./cmake-out/examples/apple/mps/mps_executor_runner --model_path mv3_mps_bundled_fp16.pte --bundled_program
+./cmake-out/examples/apple/mps/mps_executor_runner --model_path mv3_mps_float16_bundled.pte --bundled_program
 ```
 
 - You should see the following results. Note that no output file will be generated in this example:
 ```
-I 00:00:00.003290 executorch:mps_executor_runner.mm:286] Model file mv3_mps_bundled_fp16.pte is loaded.
+I 00:00:00.003290 executorch:mps_executor_runner.mm:286] Model file mv3_mps_float16_bundled.pte is loaded.
 I 00:00:00.003306 executorch:mps_executor_runner.mm:292] Program methods: 1
 I 00:00:00.003308 executorch:mps_executor_runner.mm:294] Running method forward
 I 00:00:00.003311 executorch:mps_executor_runner.mm:349] Setting up non-const buffer 1, size 606112.
@@ -118,7 +118,7 @@ python3 -m examples.apple.mps.scripts.mps_example --model_name="mv3" --generate_
 ```
 2. Run your Program on the ExecuTorch runtime and generate an [ETDump](../../../docs/source/etdump.md).
 ```
-./cmake-out/examples/apple/mps/mps_executor_runner --model_path mv3_mps_bundled_fp16.pte --bundled_program --dump-outputs
+./cmake-out/examples/apple/mps/mps_executor_runner --model_path mv3_mps_float16_bundled.pte --bundled_program --dump-outputs
 ```
 3. Create an instance of the Inspector API by passing in the ETDump you have sourced from the runtime along with the optionally generated ETRecord from step 1.
 ```bash

backends/arm/_passes/TARGETS

@@ -11,5 +11,6 @@ python_library(
         "//executorch/backends/xnnpack/_passes:xnnpack_passes",
         "//executorch/exir:lib",
         "//executorch/backends/transforms:utils",
+        "//executorch/backends/transforms:decompose_sdpa",
     ],
 )

backends/arm/_passes/__init__.py

@@ -19,6 +19,7 @@
 from .convert_squeezes_to_view import ConvertSqueezesToViewPass  # noqa
 from .convert_to_clamp import ConvertToClampPass  # noqa
 from .decompose_batchnorm_pass import DecomposeBatchNormPass  # noqa
+from .decompose_cosine_similarity_pass import DecomposeCosineSimilarityPass  # noqa
 from .decompose_div_pass import DecomposeDivPass  # noqa
 from .decompose_gelu_pass import DecomposeGeluPass  # noqa
 from .decompose_layernorm_pass import DecomposeLayerNormPass  # noqa
@@ -57,4 +58,5 @@
 from .size_adjust_conv2d_pass import SizeAdjustConv2DPass  # noqa
 from .unsqueeze_before_repeat_pass import UnsqueezeBeforeRepeatPass  # noqa
 from .unsqueeze_scalar_placeholders_pass import UnsqueezeScalarPlaceholdersPass  # noqa
+from .replace_inf_values_pass import ReplaceInfValues  # noqa  # usort: skip
 from .arm_pass_manager import ArmPassManager  # noqa  # usort: skip

backends/arm/_passes/annotate_decomposed_matmul.py

@@ -1,13 +1,12 @@
 # Copyright 2024-2025 Arm Limited and/or its affiliates.
-# All rights reserved.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
 # pyre-unsafe
 
 import itertools
-
+import operator
 from typing import List
 
 import torch
@@ -22,7 +21,7 @@
 
 class AnnotateDecomposedMatmulPass(ExportPass):
     """
-    torch.matmul can be decomposed in many ways, for instance:
+    torch.matmul and it's equivalent operator @ can be decomposed in many ways, for instance:
     dq -> matmul -> q can become
     dq -> repeat -> view -> bmm -> view -> dq which makes quantization folding
     difficult. This helper function find all matmul partitions and annotate its
@@ -50,6 +49,7 @@ def call(self, graph_module: GraphModule) -> PassResult:
             graph_module.graph,
             [
                 torch.matmul,
+                operator.matmul,
             ],
             None,
         )
@@ -70,17 +70,14 @@ def call(self, graph_module: GraphModule) -> PassResult:
             if quantized_input:
                 matmul_args = matmul_node.all_input_nodes
                 for node in matmul_args:
+                    # Find the dq-node connected to this mm/bmm arg
                     input_node = self._match_partition_to_node(
                         node, partition.input_nodes
                     )
-
-                    # Remove partition input dq-node
-                    input_node.replace_all_uses_with(input_node.all_input_nodes[0])
-                    graph_module.graph.erase_node(input_node)
                     input_node_qargs = QuantArgs.from_operator(
                         input_node.target, input_node.args
                     )
-
+                    # Insert new dq-node just before the mm/bmm with input_node's qparams
                     with graph_module.graph.inserting_before(matmul_node):
                         # Create new dq-node before matmul
                         dq_node = create_node(
@@ -90,6 +87,13 @@ def call(self, graph_module: GraphModule) -> PassResult:
                         dq_node.args = (node, *input_node_qargs)
                         matmul_node.replace_input_with(node, dq_node)
 
+                for partition_input in partition.input_nodes:
+                    # Remove partition input dq-node
+                    partition_input.replace_all_uses_with(
+                        partition_input.all_input_nodes[0]
+                    )
+                    graph_module.graph.erase_node(partition_input)
+
             partition_output = list(partition.output_nodes[0].users)[0]
             quantized_output = partition_output.target == q_op
             if quantized_output: