[libtorch] Update to 2.13

ports/libtorch/fix-disable-xnnpack.patch

@@ -0,0 +1,9 @@
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -610,5 +610,5 @@ if(USE_XNNPACK AND NOT USE_SYSTEM_XNNPACK)
+   include_directories(SYSTEM ${XNNPACK_INCLUDE_DIR})
+   list(APPEND Caffe2_DEPENDENCY_LIBS XNNPACK microkernels-prod)
+-elseif(NOT TARGET XNNPACK AND USE_SYSTEM_XNNPACK)
++elseif(USE_XNNPACK AND NOT TARGET XNNPACK AND USE_SYSTEM_XNNPACK)
+   add_library(XNNPACK SHARED IMPORTED)
+   add_library(microkernels-prod SHARED IMPORTED)

ports/libtorch/fix-fbgemm-arch-check.patch

@@ -0,0 +1,11 @@
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -913,7 +913,7 @@ include(ExternalProject)
+
+ # ---[ Dependencies ---[ FBGEMM doesn't work on x86 32bit and
+ # CMAKE_SYSTEM_PROCESSOR thinks its 64bit
+-if(USE_FBGEMM AND NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|aarch64)$")
++if(USE_FBGEMM AND NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64|aarch64|ARM64)$")
+   message(WARNING
+     "x86_64 or aarch64 required for FBGEMM. "
+     "Not compiling with FBGEMM. "

ports/libtorch/fix-glog.patch

@@ -45,3 +45,33 @@
  }
  } // namespace c10
 
+--- a/c10/util/Exception.cpp
++++ b/c10/util/Exception.cpp
+@@ -9,18 +9,6 @@
+ #include <string>
+ #include <utility>
+
+-// Google glog's api does not have an external function that allows one to check
+-// if glog is initialized or not. It does have an internal function - so we are
+-// declaring it here. This is a hack but has been used by a bunch of others too
+-// (e.g. Torch, common/init). See also Logging.cpp in this directory.
+-#ifdef C10_USE_GLOG
+-namespace google {
+-namespace glog_internal_namespace_ {
+-bool IsGoogleLoggingInitialized();
+-} // namespace glog_internal_namespace_
+-} // namespace google
+-#endif
+-
+ namespace c10 {
+
+ Error::Error(std::string msg, Backtrace backtrace, const void* caller)
+@@ -265,7 +253,7 @@
+   // During static initialization (before InitGoogleLogging), glog's global
+   // flags may not be constructed yet. Accessing them causes SIOF crashes
+   // (T253115013, D96553733). Fall back to stderr in that case.
+-  if (!::google::glog_internal_namespace_::IsGoogleLoggingInitialized()) {
++  if (!::google::IsGoogleLoggingInitialized()) {
+     std::cerr << warning.source_location().file << ':'
+               << warning.source_location().line
+               << ": Warning: " << warning.msg() << " (function "

ports/libtorch/fix-headeronly-glob.patch

@@ -0,0 +1,19 @@
+--- a/torch/headeronly/CMakeLists.txt
++++ b/torch/headeronly/CMakeLists.txt
+@@ -19,13 +19,15 @@ configure_file(
+     ${CMAKE_BINARY_DIR}/torch/headeronly/macros/cmake_macros.h)
+
+ file(GLOB HEADERONLY_HEADERS CONFIGURE_DEPENDS
+-    *.h
+     core/**/*.h
+     cpu/**/*.h
+     macros/*.h
+     util/*.h
+ )
+
++set_source_files_properties(version.h PROPERTIES GENERATED TRUE)
++list(APPEND HEADERONLY_HEADERS ${CMAKE_CURRENT_LIST_DIR}/version.h)
++
+ add_library(headeronly INTERFACE ${HEADERONLY_HEADERS})
+
+ install(FILES ${CMAKE_BINARY_DIR}/torch/headeronly/macros/cmake_macros.h

ports/libtorch/fix-python-package-data.patch

@@ -0,0 +1,22 @@
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -1466,13 +1466,17 @@
+ # inductor templates, and symlink-replacement copies.  Previously handled
+ # by setup.py; now consolidated here for both scikit-build-core and
+ # setuptools builds.
+-include(cmake/FileMirroring.cmake)
++if(BUILD_PYTHON)
++  include(cmake/FileMirroring.cmake)
++endif()
+
+ # Package data: install non-Python data files (type stubs, templates,
+ # headers, etc.) into the torch package.  Previously declared via
+ # setup.py package_data; now consolidated here for both scikit-build-core
+ # and setuptools builds.
+-include(cmake/PackageData.cmake)
++if(BUILD_PYTHON)
++  include(cmake/PackageData.cmake)
++endif()
+
+ include(cmake/PostBuildSteps.cmake)
+

ports/libtorch/fix-system-concurrentqueue.patch

@@ -0,0 +1,10 @@
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -1774,3 +1774,5 @@
+ # Include moodycamel
+-add_library(moodycamel INTERFACE IMPORTED)
+-include_directories(moodycamel SYSTEM INTERFACE ${PROJECT_SOURCE_DIR}/third_party/concurrentqueue)
++find_package(concurrentqueue CONFIG REQUIRED)
++add_library(moodycamel ALIAS concurrentqueue::concurrentqueue)
++get_target_property(moodycamel_INCLUDE_DIRS concurrentqueue::concurrentqueue INTERFACE_INCLUDE_DIRECTORIES)
++include_directories(SYSTEM ${moodycamel_INCLUDE_DIRS})

ports/libtorch/fix-system-fmt.patch

@@ -1,28 +1,23 @@
+diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
+index 2baa187..65f9e01 100644
 --- a/cmake/Dependencies.cmake
 +++ b/cmake/Dependencies.cmake
-@@ -1597,22 +1597,24 @@
- # Install `fmtlib` header.
+@@ -1633,6 +1633,7 @@ endif()
  # This was the default behavior before version 12.0.0.
  # Since PyTorch C API depends on it, make it available for projects that
  # depend on PyTorch.
 +if(FALSE)
  set(FMT_INSTALL ON)
  set(TEMP_BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS})
  set(BUILD_SHARED_LIBS OFF CACHE BOOL "Build shared libs" FORCE)
- add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/fmt)
-
- # Disable compiler feature checks for `fmt`.
- #
- # CMake compiles a little program to check compiler features. Some of our build
- # configurations (notably the mobile build analyzer) will populate
- # CMAKE_CXX_FLAGS in ways that break feature checks. Since we already know
- # `fmt` is compatible with a superset of the compilers that PyTorch is, it
- # shouldn't be too bad to just disable the checks.
- set_target_properties(fmt-header-only PROPERTIES INTERFACE_COMPILE_FEATURES "")
+@@ -1658,8 +1659,9 @@ target_compile_definitions(fmt PUBLIC "FMT_NO_UNIQUE_ADDRESS=${_fmt_no_unique_ad
+ target_compile_definitions(fmt-header-only INTERFACE "FMT_NO_UNIQUE_ADDRESS=${_fmt_no_unique_address}")
+ unset(_fmt_no_unique_address)
 
 +endif()
 +find_package(fmt REQUIRED)
  list(APPEND Caffe2_DEPENDENCY_LIBS fmt::fmt-header-only)
 -set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS} CACHE BOOL "Build shared libs" FORCE)
 
  # ---[ Kineto
+ # edge profiler depends on KinetoProfiler but it only does cpu

ports/libtorch/fix-windows-install-dirs.patch

@@ -1,6 +1,8 @@
+diff --git a/c10/CMakeLists.txt b/c10/CMakeLists.txt
+index bd138d2..8e63fe8 100644
 --- a/c10/CMakeLists.txt
 +++ b/c10/CMakeLists.txt
-@@ -161,7 +161,10 @@
+@@ -162,7 +162,10 @@ if(NOT BUILD_LIBTORCHLESS)
    # Note: for now, we will put all export path into one single Caffe2Targets group
    # to deal with the cmake deployment need. Inside the Caffe2Targets set, the
    # individual libraries like libc10.so and libcaffe2.so are still self-contained.
@@ -12,9 +14,11 @@
  endif()
 
  install(DIRECTORY ${CMAKE_CURRENT_LIST_DIR}
+diff --git a/c10/cuda/CMakeLists.txt b/c10/cuda/CMakeLists.txt
+index a337614..425287e 100644
 --- a/c10/cuda/CMakeLists.txt
 +++ b/c10/cuda/CMakeLists.txt
-@@ -86,7 +86,10 @@
+@@ -86,7 +86,10 @@ if(NOT BUILD_LIBTORCHLESS)
  # Note: for now, we will put all export path into one single Caffe2Targets group
  # to deal with the cmake deployment need. Inside the Caffe2Targets set, the
  # individual libraries like libc10.so and libcaffe2.so are still self-contained.
@@ -26,9 +30,11 @@
 
  endif()
 
+diff --git a/c10/hip/CMakeLists.txt b/c10/hip/CMakeLists.txt
+index 0f86f03..64e0bb3 100644
 --- a/c10/hip/CMakeLists.txt
 +++ b/c10/hip/CMakeLists.txt
-@@ -56,7 +56,10 @@
+@@ -52,7 +52,10 @@ if(NOT BUILD_LIBTORCHLESS)
        $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/../..>
        $<BUILD_INTERFACE:${CMAKE_BINARY_DIR}>
        $<INSTALL_INTERFACE:include>)
@@ -40,9 +46,11 @@
    set(C10_HIP_LIB c10_hip)
  endif()
 
+diff --git a/c10/xpu/CMakeLists.txt b/c10/xpu/CMakeLists.txt
+index 349b209..b1ba10e 100644
 --- a/c10/xpu/CMakeLists.txt
 +++ b/c10/xpu/CMakeLists.txt
-@@ -49,7 +49,10 @@
+@@ -49,7 +49,10 @@ if(NOT BUILD_LIBTORCHLESS)
        $<BUILD_INTERFACE:${CMAKE_BINARY_DIR}>
        $<INSTALL_INTERFACE:include>
        )
@@ -54,9 +62,11 @@
    set(C10_XPU_LIB c10_xpu)
    add_subdirectory(test)
  endif()
+diff --git a/caffe2/CMakeLists.txt b/caffe2/CMakeLists.txt
+index e4a0454..2901fc4 100644
 --- a/caffe2/CMakeLists.txt
 +++ b/caffe2/CMakeLists.txt
-@@ -573,7 +573,10 @@
+@@ -561,7 +561,10 @@ if(USE_CUDA)
    endif()
 
    target_link_libraries(caffe2_nvrtc PRIVATE caffe2::nvrtc ${DELAY_LOAD_FLAGS})
@@ -68,7 +78,7 @@
    if(USE_NCCL)
      list(APPEND Caffe2_GPU_SRCS
        ${TORCH_SRC_DIR}/csrc/cuda/nccl.cpp)
-@@ -656,7 +659,10 @@
+@@ -645,7 +648,10 @@ if(USE_ROCM)
    target_link_libraries(caffe2_nvrtc hip::amdhip64 hiprtc::hiprtc)
    target_include_directories(caffe2_nvrtc PRIVATE ${CMAKE_BINARY_DIR})
    target_compile_definitions(caffe2_nvrtc PRIVATE USE_ROCM __HIP_PLATFORM_AMD__)
@@ -80,7 +90,7 @@
  endif()
 
  if(NOT NO_API AND NOT BUILD_LITE_INTERPRETER)
-@@ -1067,7 +1073,10 @@
+@@ -1065,7 +1071,10 @@ elseif(USE_CUDA)
      target_compile_definitions(torch_cuda PUBLIC USE_NVSHMEM)
      target_compile_definitions(torch_nvshmem PUBLIC USE_NVSHMEM)
      target_link_libraries(torch_cuda PRIVATE torch_nvshmem)
@@ -92,7 +102,7 @@
    else()
      message(STATUS "NVSHMEM not found, not building with NVSHMEM support.")
    endif()
-@@ -1130,7 +1139,10 @@
+@@ -1128,7 +1137,10 @@ elseif(USE_CUDA)
            CUDA::culibos ${CMAKE_DL_LIBS})
      endif()
      set_source_files_properties(${CMAKE_CURRENT_SOURCE_DIR}/../aten/src/ATen/native/cuda/LinearAlgebraStubs.cpp PROPERTIES COMPILE_FLAGS "-DBUILD_LAZY_CUDA_LINALG")
@@ -104,20 +114,20 @@
    endif()
 
    if(USE_PRECOMPILED_HEADERS)
-@@ -1528,8 +1540,10 @@
-       target_link_libraries(torch_rocshmem PRIVATE roc::rocshmem)
-       target_link_libraries(torch_hip PRIVATE torch_rocshmem)
+@@ -1520,8 +1532,10 @@ if(USE_ROCM)
+         target_link_libraries(torch_rocshmem PRIVATE roc::rocshmem)
+         target_link_libraries(torch_hip PRIVATE torch_rocshmem)
 
--      install(TARGETS torch_rocshmem EXPORT Caffe2Targets DESTINATION
--        "${TORCH_INSTALL_LIB_DIR}")
-+      install(TARGETS torch_rocshmem EXPORT Caffe2Targets
-+              RUNTIME DESTINATION "${TORCH_INSTALL_BIN_DIR}"
-+              LIBRARY DESTINATION "${TORCH_INSTALL_LIB_DIR}"
-+              ARCHIVE DESTINATION "${TORCH_INSTALL_LIB_DIR}")
-     elseif(rocshmem_FOUND)
-       message(STATUS
-         "rocSHMEM found but skipped: PYTORCH_ROCM_ARCH has no entry in "
-@@ -1676,17 +1690,32 @@
+-        install(TARGETS torch_rocshmem EXPORT Caffe2Targets DESTINATION
+-          "${TORCH_INSTALL_LIB_DIR}")
++        install(TARGETS torch_rocshmem EXPORT Caffe2Targets
++                RUNTIME DESTINATION "${TORCH_INSTALL_BIN_DIR}"
++                LIBRARY DESTINATION "${TORCH_INSTALL_LIB_DIR}"
++                ARCHIVE DESTINATION "${TORCH_INSTALL_LIB_DIR}")
+       else()
+         message(STATUS
+           "rocSHMEM found but skipped: PYTORCH_ROCM_ARCH has no entry in "
+@@ -1673,17 +1687,32 @@ endif()
 
  caffe2_interface_library(torch torch_library)
 
@@ -155,7 +165,7 @@
 
  target_link_libraries(torch PUBLIC torch_cpu_library)
 
-@@ -1829,7 +1858,10 @@
+@@ -1826,7 +1855,10 @@ if(BUILD_SHARED_LIBS)
        target_link_libraries(torch_global_deps torch::nvtoolsext)
      endif()
    endif()
@@ -167,23 +177,11 @@
  endif()
 
  # ---[ Caffe2 HIP sources.
---- a/torch/CMakeLists.txt
-+++ b/torch/CMakeLists.txt
-@@ -459,7 +459,10 @@
-     set_target_properties(torch_python PROPERTIES LINK_FLAGS ${TORCH_PYTHON_LINK_FLAGS})
- endif()
-
--install(TARGETS torch_python DESTINATION "${TORCH_INSTALL_LIB_DIR}")
-+install(TARGETS torch_python
-+        RUNTIME DESTINATION "${TORCH_INSTALL_BIN_DIR}"
-+        LIBRARY DESTINATION "${TORCH_INSTALL_LIB_DIR}"
-+        ARCHIVE DESTINATION "${TORCH_INSTALL_LIB_DIR}")
-
- # Generate torch/version.py from the appropriate CMake cache variables.
- if(${CMAKE_BUILD_TYPE} STREQUAL "Debug")
+diff --git a/test/cpp/c10d/CMakeLists.txt b/test/cpp/c10d/CMakeLists.txt
+index aa98db5..616bba8 100644
 --- a/test/cpp/c10d/CMakeLists.txt
 +++ b/test/cpp/c10d/CMakeLists.txt
-@@ -79,7 +79,10 @@
+@@ -80,7 +80,10 @@ if(USE_CUDA)
        LINK_LIBRARIES torch_cpu c10d_cuda_test gtest_main __caffe2_nccl INSTALL_TEST ${INSTALL_TEST})
      target_compile_definitions(NCCLDevCommManagerTest PRIVATE USE_CUDA USE_NCCL)
      if(INSTALL_TEST)
@@ -195,7 +193,7 @@
      endif()
    endif()
    if(USE_UCC AND USE_C10D_UCC)
-@@ -91,7 +94,10 @@
+@@ -92,7 +95,10 @@ if(USE_CUDA)
        ProcessGroupUCCTest.cpp
        LINK_LIBRARIES torch_cpu c10d_cuda_test gtest_main __caffe2_ucc INSTALL_TEST ${INSTALL_TEST})
      if(INSTALL_TEST)
@@ -207,7 +205,7 @@
      endif()
    endif()
  elseif(USE_ROCM)
-@@ -107,7 +113,10 @@
+@@ -108,7 +114,10 @@ elseif(USE_ROCM)
        hip/ProcessGroupNCCLErrorsTest.cpp
        LINK_LIBRARIES torch_cpu c10d_hip_test gtest_main __caffe2_nccl INSTALL_TEST ${INSTALL_TEST})
      if(INSTALL_TEST)
@@ -219,7 +217,7 @@
      endif()
    endif()
    if(USE_UCC AND USE_C10D_UCC)
-@@ -115,7 +124,10 @@
+@@ -116,7 +125,10 @@ elseif(USE_ROCM)
        hip/ProcessGroupUCCTest.cpp
        LINK_LIBRARIES torch_cpu c10d_hip_test gtest_main __caffe2_ucc INSTALL_TEST ${INSTALL_TEST})
      if(INSTALL_TEST)
@@ -231,9 +229,11 @@
      endif()
    endif()
  else()
+diff --git a/test/cpp/jit/CMakeLists.txt b/test/cpp/jit/CMakeLists.txt
+index 552f8cf..2299940 100644
 --- a/test/cpp/jit/CMakeLists.txt
 +++ b/test/cpp/jit/CMakeLists.txt
-@@ -32,9 +32,18 @@
+@@ -32,9 +32,18 @@ endif()
  target_link_libraries(backend_with_compiler torch)
 
  if(INSTALL_TEST)
@@ -255,4 +255,19 @@
  endif()
 
  # Build the cpp gtest binary containing the cpp-only tests.
+diff --git a/torch/CMakeLists.txt b/torch/CMakeLists.txt
+index 23aa88d..2381304 100644
+--- a/torch/CMakeLists.txt
++++ b/torch/CMakeLists.txt
+@@ -439,7 +439,10 @@ if(NOT TORCH_PYTHON_LINK_FLAGS STREQUAL "")
+     set_target_properties(torch_python PROPERTIES LINK_FLAGS "${TORCH_PYTHON_LINK_FLAGS}")
+ endif()
+
+-install(TARGETS torch_python DESTINATION "${TORCH_INSTALL_LIB_DIR}")
++install(TARGETS torch_python
++        RUNTIME DESTINATION "${TORCH_INSTALL_BIN_DIR}"
++        LIBRARY DESTINATION "${TORCH_INSTALL_LIB_DIR}"
++        ARCHIVE DESTINATION "${TORCH_INSTALL_LIB_DIR}")
 
+ # Build the torch._C Python extension module from the thin stub that
+ # forwards to torch_python.  This was historically built by setuptools;