Merge branch 'main' into dev1/danny/support_qnn_ir_backend

Author: DannyYuyang-quic

.github/scripts/extract_benchmark_results.py

@@ -349,7 +349,10 @@ def transform(
     # Overwrite the device name here with the job name as it has more information about
     # the device, i.e. Samsung Galaxy S22 5G instead of just Samsung
     for r in benchmark_results:
-        r["deviceInfo"]["device"] = job_name
+        is_private_device = job_report.get("is_private_instance", False)
+        r["deviceInfo"]["device"] = (
+            f"{job_name} (private)" if is_private_device else job_name
+        )
 
     # From https://github.com/pytorch/pytorch/wiki/How-to-integrate-with-PyTorch-OSS-benchmark-database
     return [
@@ -363,6 +366,7 @@ def transform(
                     "benchmark_config": json.dumps(benchmark_config),
                     "job_conclusion": "SUCCESS",
                     "job_arn": job_report.get("arn", ""),
+                    "instance_arn": job_report.get("instance_arn", ""),
                 },
             },
             "model": {

.github/workflows/apple-perf-private-device-experiment.yml

@@ -1,18 +1,16 @@
 name: apple-perf (private devices)
 
 on:
-  # TODO (huydhn): Disable the schedule run until we land the change to add device pool and device name
-  # to separate between public and private iOS devices
-  # schedule:
-  # - cron: 0 0,4,8,12,16,20 * * *
+  schedule:
+   - cron: 0 0,4,8,12,16,20 * * *
   pull_request:
     paths:
       - .github/workflows/apple-perf-private-device-experiment.yml
-  # push:
-  #   branches:
-  #     - main
-  #   paths:
-  #     - .github/workflows/apple-perf-private-device-experiment.yml
+  push:
+    branches:
+      - main
+    paths:
+      - .github/workflows/apple-perf-private-device-experiment.yml
   # Note: GitHub has an upper limit of 10 inputs
   workflow_dispatch:
     inputs:

.lintrunner.toml

@@ -10,7 +10,7 @@ exclude_patterns = [
     'exir/serde/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -19,7 +19,7 @@ command = [
     '@{{PATHSFILE}}'
 ]
 init_command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -41,7 +41,7 @@ exclude_patterns = [
     'exir/serde/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -50,7 +50,7 @@ command = [
     '@{{PATHSFILE}}'
 ]
 init_command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -83,7 +83,7 @@ exclude_patterns = [
     'runtime/core/portable_type/c10/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -94,7 +94,7 @@ command = [
     '@{{PATHSFILE}}'
 ]
 init_command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -116,7 +116,7 @@ exclude_patterns = [
     '**/third-party/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -126,7 +126,7 @@ command = [
     '@{{PATHSFILE}}',
 ]
 init_command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -150,7 +150,7 @@ exclude_patterns = [
     '**/third-party/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -191,7 +191,7 @@ exclude_patterns = [
     'extension/llm/custom_ops/spinquant/test/fast_hadamard_transform_special_unstrided_cpu.h',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -226,7 +226,7 @@ exclude_patterns = [
     'util/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -275,7 +275,7 @@ exclude_patterns = [
     'util/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -325,7 +325,7 @@ exclude_patterns = [
     'backends/arm/test/**',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -337,7 +337,7 @@ command = [
     '@{{PATHSFILE}}'
 ]
 init_command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',
@@ -356,7 +356,7 @@ exclude_patterns = [
     '.lintrunner.toml',
 ]
 command = [
-    'python',
+    'python3',
     '-m',
     'lintrunner_adapters',
     'run',

README.md

@@ -49,8 +49,8 @@ Key value propositions of ExecuTorch are:
 ## Getting Started
 To get started you can:
 
-- Visit the [Step by Step Tutorial](https://pytorch.org/executorch/main/index) to get things running locally and deploy a model to a device
-- Use this [Colab Notebook](https://pytorch.org/executorch/main/getting-started-setup#quick-setup-colab-jupyter-notebook-prototype) to start playing around right away
+- Visit the [Step by Step Tutorial](https://pytorch.org/executorch/stable/getting-started.html) to get things running locally and deploy a model to a device
+- Use this [Colab Notebook](https://colab.research.google.com/drive/1qpxrXC3YdJQzly3mRg-4ayYiOjC6rue3?usp=sharing) to start playing around right away
 - Jump straight into LLM use cases by following specific instructions for [Llama](examples/models/llama/README.md) and [Llava](examples/models/llava/README.md)
 
 ## Feedback and Engagement

backends/arm/operators/op_avg_pool2d.py

@@ -85,8 +85,12 @@ def define_node(
     ) -> None:
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
-        input_tensor = inputs[0]
-        assert input_tensor.dtype == ts.DType.INT8
+        supported_dtypes = [ts.DType.INT8]
+        if inputs[0].dtype not in supported_dtypes:
+            raise TypeError(
+                f"IO data type needs to be one of {supported_dtypes}, got "
+                f'"{inputs[0].dtype}"'
+            )
 
         accumulator_type = ts.DType.INT32
 
@@ -118,9 +122,12 @@ def define_node(
     ) -> None:
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
-        assert (
-            inputs[0].dtype == ts.DType.INT8 or inputs[0].dtype == ts.DType.FP32
-        ), "Only FP32 and INT8 supported"
+        supported_dtypes = [ts.DType.INT8, ts.DType.FP32]
+        if inputs[0].dtype not in supported_dtypes:
+            raise TypeError(
+                f"IO data type needs to be one of {supported_dtypes}, got "
+                f'"{inputs[0].dtype}"'
+            )
 
         if inputs[0].dtype == ts.DType.INT8:
             super().define_node(node, tosa_graph, inputs, output)
@@ -205,8 +212,12 @@ def define_node(
     ) -> None:
         import serializer.tosa_serializer as ts  # type: ignore
 
-        input_tensor = inputs[0]
-        assert input_tensor.dtype == ts.DType.INT8
+        supported_dtypes = [ts.DType.INT8]
+        if inputs[0].dtype not in supported_dtypes:
+            raise TypeError(
+                f"IO data type needs to be one of {supported_dtypes}, got "
+                f'"{inputs[0].dtype}"'
+            )
 
         accumulator_type = ts.DType.INT32
 
@@ -241,9 +252,12 @@ def define_node(
     ) -> None:
         import serializer.tosa_serializer as ts  # type: ignore
 
-        assert (
-            inputs[0].dtype == ts.DType.INT8 or inputs[0].dtype == ts.DType.FP32
-        ), "Only FP32 and INT8 supported"
+        supported_dtypes = [ts.DType.INT8, ts.DType.FP32]
+        if inputs[0].dtype not in supported_dtypes:
+            raise TypeError(
+                f"IO data type needs to be one of {supported_dtypes}, got "
+                f'"{inputs[0].dtype}"'
+            )
 
         if inputs[0].dtype == ts.DType.INT8:
             super().define_node(node, tosa_graph, inputs, output)

backends/arm/quantizer/__init__.py

@@ -13,3 +13,16 @@
 
 # Used in tests
 from .arm_quantizer_utils import is_annotated  # noqa
+
+# Load quantized ops library.
+try:
+    import executorch.extension.pybindings.portable_lib
+    import executorch.kernels.quantized  # noqa
+except:
+    import logging
+
+    logging.info(
+        "Failed to load portable_lib and quantized_aot_lib. To run quantized kernels AOT, either build "
+        "Executorch with pybindings, or load your own custom built op library using torch.ops.load_library."
+    )
+    del logging

backends/arm/scripts/build_executorch_runner.sh renamed to backends/arm/scripts/build_executor_runner.sh

@@ -27,7 +27,7 @@ VERBS="Add|Fix|Update|Refactor|Improve|Remove|Change|Implement|Create|Modify|"\
 "Handle|Ignore|Interpret|Instantiate|Invoke|Limit|Load|Modify|Permit|Print|"\
 "Profile|Recalculate|Reconstruct|Redefine|Redesign|Reevaluate|Relocate|Remap|"\
 "Render|Reposition|Request|Revert|Sanitize|Specify|Strengthen|Stub|Substitute|"\
-"Tag|Tweak|Unify|Unlock|Unset|Use|Validate|Verify"
+"Tag|Tweak|Unify|Unlock|Unset|Use|Validate|Verify|Rename"
 
 # Remote branch
 REMOTE=$(git rev-parse --abbrev-ref --symbolic-full-name @{u} 2>/dev/null)

backends/arm/scripts/build_quantized_ops_aot_lib.sh

@@ -5,10 +5,8 @@
 
 import logging
 import os
-import platform
 import random
 import shutil
-import subprocess
 import sys
 from typing import Any
 
@@ -81,8 +79,7 @@ def try_addoption(*args, **kwargs):
 
 
 def pytest_sessionstart(session):
-    if not session.config.option.collectonly:
-        _load_libquantized_ops_aot_lib()
+    pass
 
 
 def pytest_sessionfinish(session, exitstatus):
@@ -172,32 +169,3 @@ def get_option(option: str) -> Any | None:
     if option in pytest._test_options:  # type: ignore[attr-defined]
         return pytest._test_options[option]  # type: ignore[attr-defined]
     return None
-
-
-def _load_libquantized_ops_aot_lib():
-    """
-    Find and load the libquantized_ops_aot_lib shared library.
-    """
-    so_ext = {
-        "Darwin": "dylib",
-        "Linux": "so",
-        "Windows": "dll",
-    }.get(platform.system(), None)
-
-    find_lib_cmd = [
-        "find",
-        "cmake-out-aot-lib",
-        "-name",
-        f"libquantized_ops_aot_lib.{so_ext}",
-    ]
-
-    res = subprocess.run(find_lib_cmd, capture_output=True)
-    if res.returncode == 0:
-        library_path = res.stdout.decode().strip()
-        import torch
-
-        torch.ops.load_library(library_path)
-    else:
-        raise RuntimeError(
-            f"Did not find libquantized_ops_aot_lib.{so_ext} in cmake-out-aot-lib. Did you build it?"
-        )

backends/arm/scripts/pre-push

@@ -52,8 +52,6 @@ function build_semihosting_executorch_runner() {
     find ${build_test_dir} -name "arm_executor_runner"
 }
 
-cd $et_root_dir && backends/arm/scripts/build_quantized_ops_aot_lib.sh
-
 # Use most optimal system_configs for testing
 build_semihosting_executorch_runner corstone-300 Ethos_U55_High_End_Embedded