[CI] Compile on M1 natively (pytorch#95719)

We have plenty of runners now, let's use them for compilation as well.
To achieve that, remove `xcode-version: "13.3.1"` property and tweak Metal framework detection logic to work with command line tools(which are installed in `/Library/Developer/CommandLineTools`) and SDK is in `/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk`) rather than full Xcode installation.

TODO: Fix/enable OpenMP accelerated native builds (which are currently broken with `OMP: Error pytorch#15: Initializing libomp.dylib, but found libomp.dylib already initialized.`), but this matches existing behavior as cross-builds are compiled  with OpenMP disabled.
Pull Request resolved: pytorch#95719
Approved by: https://github.com/huydhn

Author: malfet

.ci/pytorch/macos-build.sh

@@ -40,6 +40,16 @@ cross_compile_arm64() {
   USE_DISTRIBUTED=0 CMAKE_OSX_ARCHITECTURES=arm64 MACOSX_DEPLOYMENT_TARGET=11.0 USE_MKLDNN=OFF USE_QNNPACK=OFF WERROR=1 BUILD_TEST=OFF USE_PYTORCH_METAL=1 python setup.py bdist_wheel
 }
 
+compile_arm64() {
+  # Compilation for arm64
+  # TODO: Compile with OpenMP support (but this causes CI regressions as cross-compilation were done with OpenMP disabled)
+  USE_DISTRIBUTED=0 USE_OPENMP=0 MACOSX_DEPLOYMENT_TARGET=11.0 WERROR=1 BUILD_TEST=OFF USE_PYTORCH_METAL=1 python setup.py bdist_wheel
+}
+
+compile_x86_64() {
+  USE_DISTRIBUTED=0 WERROR=1 python setup.py bdist_wheel
+}
+
 compile_x86_64() {
   USE_DISTRIBUTED=0 WERROR=1 python setup.py bdist_wheel
 }
@@ -63,7 +73,11 @@ build_lite_interpreter() {
 }
 
 if [[ ${BUILD_ENVIRONMENT} = *arm64* ]]; then
-  cross_compile_arm64
+  if [[ $(uname -m) == "arm64" ]]; then
+    compile_arm64
+  else
+    cross_compile_arm64
+  fi
 elif [[ ${BUILD_ENVIRONMENT} = *lite-interpreter* ]]; then
   export BUILD_LITE_INTERPRETER=1
   build_lite_interpreter

.github/workflows/mac-mps.yml

@@ -17,8 +17,7 @@ jobs:
     with:
       sync-tag: macos-12-py3-arm64-build
       build-environment: macos-12-py3-arm64
-      xcode-version: "13.3.1"
-      runner-type: macos-12-xl
+      runner-type: macos-m1-12
       build-generates-artifacts: true
       # To match the one pre-installed in the m1 runners
       python_version: 3.9.12

.github/workflows/trunk.yml

@@ -200,8 +200,7 @@ jobs:
     with:
       sync-tag: macos-12-py3-arm64-build
       build-environment: macos-12-py3-arm64
-      xcode-version: "13.3.1"
-      runner-type: macos-12-xl
+      runner-type: macos-m1-12
       build-generates-artifacts: true
       # To match the one pre-installed in the m1 runners
       python_version: 3.9.12

cmake/Metal.cmake

@@ -19,7 +19,9 @@ if(NOT DEFINED CMAKE_OSX_DEVELOPER_ROOT)
         set(CMAKE_OSX_DEVELOPER_ROOT ${XCODE_POST_43_ROOT})
     elseif(EXISTS ${XCODE_PRE_43_ROOT})
         set(CMAKE_OSX_DEVELOPER_ROOT ${XCODE_PRE_43_ROOT})
-    endif(EXISTS ${XCODE_POST_43_ROOT})
+    elseif(EXISTS ${CMAKE_XCODE_DEVELOPER_DIR} AND ${CMAKE_XCODE_DEVELOPER_DIR} STREQUAL "/Library/Developer/CommandLineTools")
+            set(CMAKE_OSX_DEVELOPER_ROOT ${CMAKE_XCODE_DEVELOPER_DIR})
+    endif()
 endif(NOT DEFINED CMAKE_OSX_DEVELOPER_ROOT)
 set(CMAKE_OSX_DEVELOPER_ROOT ${CMAKE_OSX_DEVELOPER_ROOT} CACHE PATH "Location of OSX SDKs root directory")