Merge branch 'main' into export-D70903186

Author: zonglinpeng

.buckconfig

@@ -39,3 +39,6 @@
 
 [buck2]
 restarter=true
+
+[oss]
+folly_cxx_tests = False

.ci/scripts/gather_test_models.py

@@ -30,6 +30,7 @@
         "dl3": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
+        "phi-4-mini": "linux.4xlarge.memory",
     }
 }
 

.ci/scripts/unittest-buck2.sh

@@ -8,13 +8,17 @@ set -eux
 
 # TODO: expand this to //...
 # TODO: can't query cadence & vulkan backends
+# TODO: can't query //kernels/prim_ops because of a cpp_unittest and
+# broken code in shim to read oss.folly_cxx_tests. Sending fix but it
+# needs to propagate and we need a submodule update.
 buck2 query "//backends/apple/... + //backends/example/... + \
 //backends/mediatek/... + //backends/test/... + //backends/transforms/... + \
-//backends/xnnpack/... + //configurations/... + //kernels/portable/cpu/... + \
-//runtime/... + //schema/... + //test/... + //util/..."
+//backends/xnnpack/... + //configurations/... + //kernels/aten/... + \
+//kernels/optimized/... + //kernels/portable/... + //kernels/quantized/... + \
+//kernels/test/... + //runtime/... + //schema/... + //test/... + //util/..."
 
 # TODO: expand the covered scope of Buck targets.
 # //runtime/kernel/... is failing because //third-party:torchgen_files's shell script can't find python on PATH.
 # //runtime/test/... requires Python torch, which we don't have in our OSS buck setup.
-buck2 build //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...
-buck2 test //runtime/backend/... //runtime/core/... //runtime/executor: //runtime/kernel/... //runtime/platform/...
+buck2 test //kernels/portable/... //runtime/backend/... //runtime/core/... \
+      //runtime/executor: //runtime/kernel/... //runtime/platform/...

.github/workflows/doc-build.yml

@@ -83,6 +83,7 @@ jobs:
         cp -rf docs/_build/html/* "${RUNNER_DOCS_DIR}"
 
         mv docs/_build/html "${RUNNER_ARTIFACT_DIR}"
+        cp -rf "${RUNNER_DOCS_DIR}"/javadoc "${RUNNER_ARTIFACT_DIR}"/html
 
         ls -R "${RUNNER_ARTIFACT_DIR}"/*/*.html
 

backends/arm/test/passes/test_fuse_constant_ops_pass.py

@@ -4,6 +4,7 @@
 # LICENSE file in the root directory of this source tree.
 
 import operator
+import unittest
 from typing import Tuple
 
 import torch
@@ -107,6 +108,7 @@ def test_fuse_batchnorm_tosa_MI(module):
     pipeline.run()
 
 
+@unittest.skip("Test failing on internal CI")
 @common.parametrize("module", modules)
 def test_fuse_batchnorm_tosa_BI(module):
     pipeline = TosaPipelineBI[input_t](

backends/cadence/fusion_g3/operators/op_mean.cpp

@@ -118,7 +118,7 @@ Tensor& mean_out(
     for (int i = 0; i < kNnlibMaxDim; i++) {
       out_shape[i] = 1;
       inp_shape[i] = 1;
-      p_axis[i] = 1;
+      p_axis[i] = -1;
     }
 
     int num_axis_dims = prepare_data(
@@ -135,20 +135,10 @@ Tensor& mean_out(
       num_out_dims = 1;
     }
 
-    int inp_shape_max = inp_shape[p_axis[0]];
-    for (int i = 1; i < num_axis_dims; i++) {
-      if (inp_shape[p_axis[i]] > inp_shape_max) {
-        inp_shape_max = inp_shape[p_axis[i]];
-      }
+    if ((out.dim() == 0) && (out.numel())) {
+      num_out_dims = 1;
     }
 
-    int scratch_size = in.numel() / inp_shape_max;
-
-    executorch::runtime::Result<void*> temp_mem =
-        ctx.allocate_temp(scratch_size * sizeof(float));
-
-    void* __restrict__ p_scratch_in = (void* __restrict__)(temp_mem.get());
-
     XT_KERNEL_CHECK(
         ctx,
         out,
@@ -160,8 +150,7 @@ Tensor& mean_out(
         inp_shape,
         num_inp_dims,
         p_axis,
-        num_axis_dims,
-        p_scratch_in);
+        num_axis_dims);
   } else {
     ET_KERNEL_CHECK(
         ctx,

backends/transforms/test/test_create_delete_constant_placeholder.py

@@ -15,7 +15,6 @@
 
 
 class EmptyNetwork(torch.nn.Module):
-
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         return x
 
@@ -30,7 +29,7 @@ def _test_create_delete(kind: InputKind, persistent_buffer: bool = None):
     # Toy network with two nodes, input and output
     # The result should be 0 = 0
     module = EmptyNetwork()
-    exported_program = export(module, args=module.test_data)
+    exported_program = export(module, args=module.test_data, strict=True)
     exported_program = to_edge(exported_program).exported_program()
     graph = exported_program.graph_module.graph
     assert len(graph.nodes) == 2

backends/xnnpack/CMakeLists.txt

@@ -10,7 +10,7 @@
 # cmake-format -i CMakeLists.txt
 # ~~~
 
-cmake_minimum_required(VERSION 3.19)
+cmake_minimum_required(VERSION 3.24)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
@@ -124,6 +124,10 @@ else()
   list(APPEND xnn_executor_runner_libs portable_ops_lib)
 endif()
 
+if(EXECUTORCH_BUILD_KERNELS_CUSTOM)
+  list(APPEND xnn_executor_runner_libs $<LINK_LIBRARY:WHOLE_ARCHIVE,custom_ops>)
+endif()
+
 list(APPEND xnn_executor_runner_libs xnnpack_backend executorch)
 
 # ios can only build library but not binary

build/build_android_library.sh

@@ -40,6 +40,8 @@ build_android_native_library() {
     EXECUTORCH_BUILD_NEURON=OFF
   fi
 
+  EXECUTORCH_BUILD_VULKAN="${EXECUTORCH_BUILD_VULKAN:-OFF}"
+
   cmake . -DCMAKE_INSTALL_PREFIX="${CMAKE_OUT}" \
     -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" \
     -DANDROID_ABI="${ANDROID_ABI}" \
@@ -60,6 +62,7 @@ build_android_native_library() {
     -DNEURON_BUFFER_ALLOCATOR_LIB="${NEURON_BUFFER_ALLOCATOR_LIB}" \
     -DEXECUTORCH_BUILD_QNN="${EXECUTORCH_BUILD_QNN}" \
     -DQNN_SDK_ROOT="${QNN_SDK_ROOT}" \
+    -DEXECUTORCH_BUILD_VULKAN="${EXECUTORCH_BUILD_VULKAN}" \
     -DCMAKE_BUILD_TYPE="${EXECUTORCH_CMAKE_BUILD_TYPE}" \
     -B"${CMAKE_OUT}"
 
@@ -168,6 +171,7 @@ collect_artifacts_to_be_uploaded() {
   MINIBENCH_APP_DIR="${ARTIFACTS_DIR_NAME}/minibench"
   mkdir -p "${MINIBENCH_APP_DIR}"
   cp extension/benchmark/android/benchmark/app/build/outputs/apk/debug/*.apk "${MINIBENCH_APP_DIR}"
+  cp extension/benchmark/android/benchmark/app/build/outputs/apk/androidTest/debug/*.apk "${MINIBENCH_APP_DIR}"
   # Collect Java library test
   JAVA_LIBRARY_TEST_DIR="${ARTIFACTS_DIR_NAME}/library_test_dir"
   mkdir -p "${JAVA_LIBRARY_TEST_DIR}"