Update on "[ET-VK] Removing descriptor pool intialization from DescriptorPool ctor."

Descriptor pool is initialized with proper capacity while ComputeGraph is build, in ComputeGraph::prepare function.
Thus, initializing descriptor pool in constructor just add throwaway work.

Differential Revision: [D74347030](https://our.internmc.facebook.com/intern/diff/D74347030/)

[ghstack-poisoned]

Author: trivedivivek

.github/workflows/build-presets.yml

@@ -11,3 +11,20 @@ on:
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
   cancel-in-progress: true
+
+jobs:
+  apple:
+    uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
+    strategy:
+      matrix:
+        preset: [macos-arm64]
+    with:
+      job-name: build
+      runner: macos-latest-xlarge
+      python-version: 3.12
+      submodules: recursive
+      script: |
+        set -eux
+        ${CONDA_RUN} ./install_requirements.sh > /dev/null
+        ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
+        ${CONDA_RUN} cmake --build cmake-out --parallel

CMakeLists.txt

@@ -48,8 +48,13 @@ project(executorch)
 # MARK: - Start EXECUTORCH_H12025_BUILD_MIGRATION --------------------------------------------------
 
 include(${PROJECT_SOURCE_DIR}/tools/cmake/common/preset.cmake)
+
+load_build_preset()
 include(${PROJECT_SOURCE_DIR}/tools/cmake/preset/default.cmake)
 
+# Print all the configs that were called with announce_configured_options.
+print_configured_options()
+
 # MARK: - End EXECUTORCH_H12025_BUILD_MIGRATION ----------------------------------------------------
 
 include(tools/cmake/Utils.cmake)
@@ -175,8 +180,6 @@ option(EXECUTORCH_BUILD_ARM_BAREMETAL
        "Build the Arm Baremetal flow for Cortex-M and Ethos-U" OFF
 )
 
-option(EXECUTORCH_BUILD_COREML "Build the Core ML backend" OFF)
-
 option(EXECUTORCH_BUILD_KERNELS_CUSTOM "Build the custom kernels" OFF)
 
 option(EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT "Build the custom ops lib for AOT"

CMakePresets.json

@@ -0,0 +1,33 @@
+{
+  "version": 10,
+  "cmakeMinimumRequired": {
+    "major": 3,
+    "minor": 31,
+    "patch": 0
+  },
+  "$comment": "On-device AI across mobile, embedded and edge for PyTorch.",
+  "configurePresets": [
+    {
+      "name": "common",
+      "hidden": true,
+      "binaryDir": "${sourceDir}/cmake-out",
+      "generator": "Unix Makefiles"
+    },
+    {
+      "name": "macos-arm64",
+      "inherits": ["common"],
+      "generator": "Xcode",
+      "cacheVariables": {
+        "CMAKE_TOOLCHAIN_FILE": "${sourceDir}/third-party/ios-cmake/ios.toolchain.cmake",
+        "EXECUTORCH_BUILD_PRESET_FILE": "${sourceDir}/tools/cmake/preset/macos-arm64.cmake",
+        "PLATFORM": "MAC_ARM64",
+        "DEPLOYMENT_TARGET": "10.15"
+      },
+      "condition": {
+        "lhs": "${hostSystemName}",
+        "type": "equals",
+        "rhs": "Darwin"
+      }
+    }
+  ]
+}

backends/arm/operators/op_permute.py

@@ -46,24 +46,26 @@ def permutation_matrix_to_vector(permutation_matrix: torch.Tensor) -> list[int]:
     (1,0,2)
     """
     N = len(permutation_matrix)
-    assert N == len(
-        permutation_matrix[0]
-    ), f"A permutation matrix must be square, got shape {permutation_matrix.shape}"
+    if N != len(permutation_matrix[0]):
+        raise ValueError(
+            f"A permutation matrix must be square, got shape {permutation_matrix.shape}"
+        )
 
     p = [0] * N
     for row_index, row in enumerate(permutation_matrix):
         saw_one = False
         for col_index, value in enumerate(row):
             if value == 1:
-                assert (
-                    not saw_one
-                ), f"A permutation matrix can only have one 1 per row, got row {row}."
+                if saw_one:
+                    raise ValueError(
+                        f"A permutation matrix can only have one 1 per row, got {row=}"
+                    )
                 p[row_index] = col_index
                 saw_one = True
-            else:
-                assert (
-                    value == 0
-                ), f"A permutation matrix only contains 1's and 0's, got value {value}."
+            elif value != 0:
+                raise ValueError(
+                    f"A permutation matrix only contains 1's and 0's, got {value=}"
+                )
     return p
 
 

backends/arm/operators/op_where.py

@@ -69,8 +69,6 @@ def define_node(
     ) -> None:
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
-        validate_num_inputs(self.target, inputs, 3)
-
         bi_supported_dtypes = [
             ts.DType.INT8,
             ts.DType.INT16,
@@ -99,8 +97,6 @@ def define_node(
     ) -> None:
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
-        validate_num_inputs(self.target, inputs, 3)
-
         mi_supported_dtypes = [
             ts.DType.FP16,
             ts.DType.FP32,
@@ -163,8 +159,6 @@ def define_node(
     ) -> None:
         import serializer.tosa_serializer as ts
 
-        validate_num_inputs(self.target, inputs, 3)
-
         bi_supported_dtypes = [
             ts.DType.INT8,
             ts.DType.INT16,
@@ -193,8 +187,6 @@ def define_node(
     ) -> None:
         import serializer.tosa_serializer as ts
 
-        validate_num_inputs(self.target, inputs, 3)
-
         mi_supported_dtypes = [
             ts.DType.FP16,
             ts.DType.FP32,

backends/arm/scripts/parse_test_names.py

@@ -5,7 +5,15 @@
 from executorch.exir.dialects.edge.spec.utils import SAMPLE_INPUT
 
 # Add edge ops which we lower but which are not included in exir/dialects/edge/edge.yaml here.
-CUSTOM_EDGE_OPS = ["linspace.default", "eye.default"]
+CUSTOM_EDGE_OPS = [
+    "linspace.default",
+    "eye.default",
+    "hardsigmoid.default",
+    "hardswish.default",
+    "linear.default",
+    "maximum.default",
+    "adaptive_avg_pool2d.default",
+]
 ALL_EDGE_OPS = SAMPLE_INPUT.keys() | CUSTOM_EDGE_OPS
 
 # Add all targets and TOSA profiles we support here.

backends/arm/test/common.py

@@ -259,17 +259,15 @@ def decorator_func(func):
                     raise RuntimeError(
                         "xfail info needs to be str, or tuple[str, type[Exception]]"
                     )
-                pytest_param = pytest.param(
-                    test_parameters,
-                    id=id,
-                    marks=pytest.mark.xfail(
-                        reason=reason, raises=raises, strict=strict
-                    ),
+                # Set up our fail marker
+                marker = (
+                    pytest.mark.xfail(reason=reason, raises=raises, strict=strict),
                 )
             else:
-                pytest_param = pytest.param(test_parameters, id=id)
-            pytest_testsuite.append(pytest_param)
+                marker = ()
 
+            pytest_param = pytest.param(test_parameters, id=id, marks=marker)
+            pytest_testsuite.append(pytest_param)
         return pytest.mark.parametrize(arg_name, pytest_testsuite)(func)
 
     return decorator_func

backends/arm/test/conftest.py

@@ -12,12 +12,6 @@
 
 import pytest
 
-try:
-    import tosa_tools.v0_80.tosa_reference_model as tosa_reference_model
-except ImportError:
-    logging.warning("tosa_reference_model not found, can't run reference model tests")
-    tosa_reference_model = None
-
 """
 This file contains the pytest hooks, fixtures etc. for the Arm test suite.
 """
@@ -50,10 +44,11 @@ def pytest_configure(config):
     if getattr(config.option, "fast_fvp", False):
         pytest._test_options["fast_fvp"] = config.option.fast_fvp  # type: ignore[attr-defined]
 
-    # TODO: remove this flag once we have a way to run the reference model tests with Buck
-    pytest._test_options["tosa_ref_model"] = False  # type: ignore[attr-defined]
-    if tosa_reference_model is not None:
-        pytest._test_options["tosa_ref_model"] = True  # type: ignore[attr-defined]
+    if config.option.arm_run_tosa_version:
+        pytest._test_options["tosa_version"] = config.option.arm_run_tosa_version
+
+    pytest._test_options["tosa_ref_model"] = True  # type: ignore[attr-defined]
+
     logging.basicConfig(level=logging.INFO, stream=sys.stdout)
 
 
@@ -76,6 +71,7 @@ def try_addoption(*args, **kwargs):
         nargs="+",
         help="List of two files. Firstly .pt file. Secondly .json",
     )
+    try_addoption("--arm_run_tosa_version", action="store", default="0.80")
 
 
 def pytest_sessionstart(session):

backends/arm/test/models/test_llama.py

@@ -33,27 +33,35 @@
 class TestLlama(unittest.TestCase):
     """
     Test class of Llama models. Type of Llama model depends on command line parameters:
-    --llama_inputs <path to .pt file> <path to json file>
-    Example: --llama_inputs stories110M/stories110M.pt stories110M/params.json
+    --llama_inputs <path to .pt file> <path to json file> <name of model variant>
+    Example: --llama_inputs stories110M/stories110M.pt stories110M/params.json stories110m
+    For more examples and info see examples/models/llama/README.md.
     """
 
     def prepare_model(self):
 
         checkpoint = None
         params_file = None
+        usage = "To run use --llama_inputs <.pt/.pth> <.json> <name>"
+
         if conftest.is_option_enabled("llama_inputs"):
             param_list = conftest.get_option("llama_inputs")
-            assert (
-                isinstance(param_list, list) and len(param_list) == 2
-            ), "invalid number of inputs for --llama_inputs"
+
+            if not isinstance(param_list, list) or len(param_list) != 3:
+                raise RuntimeError(
+                    f"Invalid number of inputs for --llama_inputs. {usage}"
+                )
+            if not all(isinstance(param, str) for param in param_list):
+                raise RuntimeError(
+                    f"All --llama_inputs are expected to be strings. {usage}"
+                )
+
             checkpoint = param_list[0]
             params_file = param_list[1]
-            assert isinstance(checkpoint, str) and isinstance(
-                params_file, str
-            ), "invalid input for --llama_inputs"
+            model_name = param_list[2]
         else:
             logger.warning(
-                "Skipping Llama test because of lack of input. To run use --llama_inputs <.pt> <.json>"
+                "Skipping Llama tests because of missing --llama_inputs. {usage}"
             )
             return None, None, None
 
@@ -71,7 +79,7 @@ def prepare_model(self):
             "-p",
             params_file,
             "--model",
-            "stories110m",
+            model_name,
         ]
         parser = build_args_parser()
         args = parser.parse_args(args)
@@ -122,6 +130,7 @@ def test_llama_tosa_BI(self):
                 .quantize()
                 .export()
                 .to_edge_transform_and_lower()
+                .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
                 .to_executorch()
                 .run_method_and_compare_outputs(
                     inputs=llama_inputs,