[libtorch] regenerate patches, remove some features

ports/libtorch/fix-cmake.patch

@@ -26,7 +26,7 @@ index 0087dd9..257c0bd 100644
    endif()
    list(APPEND ATen_CPU_DEPENDENCY_LIBS sleef)
 diff --git a/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt b/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt
-index fd6b7ff..12007a5 100644
+index fd6b7ff..ba6b7a3 100644
 --- a/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt
 +++ b/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt
 @@ -337,13 +337,8 @@ target_link_libraries(pytorch_qnnpack PUBLIC clog)
@@ -45,6 +45,25 @@ index fd6b7ff..12007a5 100644
  elseif(NOT TARGET cpuinfo)
    set(CPUINFO_BUILD_TOOLS OFF CACHE BOOL "")
    set(CPUINFO_BUILD_UNIT_TESTS OFF CACHE BOOL "")
+@@ -364,14 +359,11 @@ if(NOT TARGET pthreadpool AND NOT USE_SYSTEM_PTHREADPOOL)
+     "${CONFU_DEPENDENCIES_BINARY_DIR}/pthreadpool")
+ elseif(NOT TARGET pthreadpool AND USE_SYSTEM_PTHREADPOOL)
+   add_library(pthreadpool SHARED IMPORTED)
+-  find_library(PTHREADPOOL_LIBRARY pthreadpool)
+-  if(NOT PTHREADPOOL_LIBRARY)
+-    message(FATAL_ERROR "Cannot find pthreadpool")
+-  endif()
+-  message("-- Found pthreadpool: ${PTHREADPOOL_LIBRARY}")
+-  set_target_properties(pthreadpool PROPERTIES
+-    IMPORTED_LOCATION "${PTHREADPOOL_LIBRARY}")
++  find_library(PTHREADPOOL_LIBRARY pthreadpool REQUIRED)
++  set_target_properties(pthreadpool PROPERTIES IMPORTED_LOCATION "${PTHREADPOOL_LIBRARY}")
+   add_library(pthreadpool_interface INTERFACE)
++  find_path(PTHREADPOOL_INCLUDE_DIR "pthreadpool.h" REQUIRED)
++  set_target_properties(pthreadpool_interface PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${PTHREADPOOL_INCLUDE_DIR}")
+ endif()
+ if(PYTORCH_QNNPACK_CUSTOM_THREADPOOL)
+   # Depend on pthreadpool interface, but not on implementation.
 diff --git a/caffe2/CMakeLists.txt b/caffe2/CMakeLists.txt
 index 89c31fa..c08ab68 100644
 --- a/caffe2/CMakeLists.txt
@@ -74,26 +93,35 @@ index 89c31fa..c08ab68 100644
    install(TARGETS torch_global_deps DESTINATION "${TORCH_INSTALL_LIB_DIR}")
  endif()
 diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake
-index c4661e3..9d50b55 100644
+index c4661e3..6e51b7d 100644
 --- a/cmake/Dependencies.cmake
 +++ b/cmake/Dependencies.cmake
-@@ -370,13 +370,8 @@ if(INTERN_BUILD_MOBILE OR NOT DISABLE_NNPACK_AND_FAMILY)
-
+@@ -371,12 +371,9 @@ if(INTERN_BUILD_MOBILE OR NOT DISABLE_NNPACK_AND_FAMILY)
    if(NOT TARGET pthreadpool)
      if(USE_SYSTEM_PTHREADPOOL)
--      add_library(pthreadpool SHARED IMPORTED)
+       add_library(pthreadpool SHARED IMPORTED)
 -      find_library(PTHREADPOOL_LIBRARY pthreadpool)
 -      set_property(TARGET pthreadpool PROPERTY IMPORTED_LOCATION "${PTHREADPOOL_LIBRARY}")
 -      if(NOT PTHREADPOOL_LIBRARY)
 -        message(FATAL_ERROR "Cannot find pthreadpool")
 -      endif()
 -      message("-- Found pthreadpool: ${PTHREADPOOL_LIBRARY}")
-+      find_package(unofficial-pthreadpool CONFIG REQUIRED)
-+      add_library(pthreadpool ALIAS unofficial::pthreadpool)
++      find_library(PTHREADPOOL_LIBRARY pthreadpool REQUIRED)
++      set_target_properties(pthreadpool PROPERTIES IMPORTED_LOCATION "${PTHREADPOOL_LIBRARY}")
++      message(STATUS "Found pthreadpool: ${PTHREADPOOL_LIBRARY}")
      elseif(NOT USE_INTERNAL_PTHREADPOOL_IMPL)
        if(NOT DEFINED PTHREADPOOL_SOURCE_DIR)
          set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party")
-@@ -407,13 +402,8 @@ if(NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(s390x|ppc64le)$")
+@@ -396,7 +393,7 @@ if(INTERN_BUILD_MOBILE OR NOT DISABLE_NNPACK_AND_FAMILY)
+     if(USE_INTERNAL_PTHREADPOOL_IMPL)
+       set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DUSE_INTERNAL_PTHREADPOOL_IMPL")
+     else()
+-      list(APPEND Caffe2_DEPENDENCY_LIBS pthreadpool)
++      list(APPEND Caffe2_DEPENDENCY_LIBS "${PTHREADPOOL_LIBRARY}")
+     endif()
+   endif()
+ else()
+@@ -407,13 +404,8 @@ if(NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(s390x|ppc64le)$")
    # ---[ Caffe2 uses cpuinfo library in the thread pool
    # ---[ But it doesn't support s390x/powerpc and thus not used on s390x/powerpc
    if(NOT TARGET cpuinfo AND USE_SYSTEM_CPUINFO)
@@ -109,7 +137,7 @@ index c4661e3..9d50b55 100644
    elseif(NOT TARGET cpuinfo)
      if(NOT DEFINED CPUINFO_SOURCE_DIR)
        set(CPUINFO_SOURCE_DIR "${CMAKE_CURRENT_LIST_DIR}/../third_party/cpuinfo" CACHE STRING "cpuinfo source directory")
-@@ -569,8 +559,8 @@ elseif(NOT TARGET XNNPACK AND USE_SYSTEM_XNNPACK)
+@@ -569,8 +561,8 @@ elseif(NOT TARGET XNNPACK AND USE_SYSTEM_XNNPACK)
    if(NOT XNNPACK_LIBRARY)
      message(FATAL_ERROR "Cannot find XNNPACK")
    endif()
@@ -120,7 +148,7 @@ index c4661e3..9d50b55 100644
  endif()
 
  # ---[ Vulkan deps
-@@ -812,8 +802,9 @@ if(NOT TARGET fp16 AND NOT USE_SYSTEM_FP16)
+@@ -812,8 +804,9 @@ if(NOT TARGET fp16 AND NOT USE_SYSTEM_FP16)
      "${FP16_SOURCE_DIR}"
      "${CONFU_DEPENDENCIES_BINARY_DIR}/FP16")
  elseif(NOT TARGET fp16 AND USE_SYSTEM_FP16)
@@ -132,7 +160,7 @@ index c4661e3..9d50b55 100644
  endif()
  list(APPEND Caffe2_DEPENDENCY_LIBS fp16)
 
-@@ -996,7 +987,7 @@ if(USE_CUDNN)
+@@ -996,7 +989,7 @@ if(USE_CUDNN)
    if(CUDNN_VERSION VERSION_LESS 8.5)
      message(FATAL_ERROR "PyTorch needs CuDNN-8.5 or above, but found ${CUDNN_VERSION}. Builds are still possible with `USE_CUDNN=0`")
    endif()
@@ -141,7 +169,7 @@ index c4661e3..9d50b55 100644
    target_include_directories(torch::cudnn INTERFACE ${CUDNN_FRONTEND_INCLUDE_DIR})
  endif()
 
-@@ -1207,7 +1198,7 @@ if(USE_GLOO)
+@@ -1207,7 +1200,7 @@ if(USE_GLOO)
        if(NOT GLOO_LIBRARY)
          message(FATAL_ERROR "Cannot find gloo")
        endif()
@@ -150,7 +178,7 @@ index c4661e3..9d50b55 100644
        set_target_properties(gloo PROPERTIES IMPORTED_LOCATION ${GLOO_LIBRARY})
      endif()
      # Here is a little bit hacky. We have to put PROJECT_BINARY_DIR in front
-@@ -1288,7 +1279,7 @@ if(CAFFE2_CMAKE_BUILDING_WITH_MAIN_REPO AND NOT INTERN_DISABLE_ONNX)
+@@ -1288,7 +1281,7 @@ if(CAFFE2_CMAKE_BUILDING_WITH_MAIN_REPO AND NOT INTERN_DISABLE_ONNX)
        set_target_properties(onnx_proto PROPERTIES CXX_STANDARD 17)
      endif()
    endif()
@@ -159,7 +187,7 @@ index c4661e3..9d50b55 100644
 
    add_definitions(-DONNX_NAMESPACE=${ONNX_NAMESPACE})
    if(NOT USE_SYSTEM_ONNX)
-@@ -1306,23 +1297,17 @@ if(CAFFE2_CMAKE_BUILDING_WITH_MAIN_REPO AND NOT INTERN_DISABLE_ONNX)
+@@ -1306,23 +1299,17 @@ if(CAFFE2_CMAKE_BUILDING_WITH_MAIN_REPO AND NOT INTERN_DISABLE_ONNX)
        target_compile_options(onnx PRIVATE -Wno-deprecated-declarations)
      endif()
    else()
@@ -192,7 +220,7 @@ index c4661e3..9d50b55 100644
    # Recover the build shared libs option.
    set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS})
  endif()
-@@ -1530,7 +1515,7 @@ endif()
+@@ -1530,7 +1517,7 @@ endif()
  #
  set(TEMP_BUILD_SHARED_LIBS ${BUILD_SHARED_LIBS})
  set(BUILD_SHARED_LIBS OFF CACHE BOOL "Build shared libs" FORCE)
@@ -201,7 +229,7 @@ index c4661e3..9d50b55 100644
 
  # Disable compiler feature checks for `fmt`.
  #
-@@ -1539,7 +1524,7 @@ add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/fmt)
+@@ -1539,7 +1526,7 @@ add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/fmt)
  # 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.
@@ -210,26 +238,27 @@ index c4661e3..9d50b55 100644
 
  list(APPEND Caffe2_DEPENDENCY_LIBS fmt::fmt-header-only)
  set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS} CACHE BOOL "Build shared libs" FORCE)
-@@ -1681,5 +1666,4 @@ endif()
+@@ -1681,5 +1668,5 @@ endif()
  include(${CMAKE_CURRENT_LIST_DIR}/FlatBuffers.cmake)
 
  # Include cpp-httplib
 -add_library(httplib INTERFACE IMPORTED)
 -target_include_directories(httplib SYSTEM INTERFACE ${PROJECT_SOURCE_DIR}/third_party/cpp-httplib)
 +find_package(httplib CONFIG REQUIRED) # httplib::httplib
++add_library(httplib ALIAS httplib::httplib)
 diff --git a/cmake/public/cuda.cmake b/cmake/public/cuda.cmake
-index b37036d..b988e1c 100644
+index b37036d..281b4ec 100644
 --- a/cmake/public/cuda.cmake
 +++ b/cmake/public/cuda.cmake
-@@ -66,7 +66,7 @@ if(NOT CMAKE_CUDA_COMPILER_VERSION VERSION_EQUAL CUDAToolkit_VERSION)
-                       "V${CUDAToolkit_VERSION} in '${CUDAToolkit_INCLUDE_DIRS}'")
+@@ -67,7 +67,7 @@ if(NOT CMAKE_CUDA_COMPILER_VERSION VERSION_EQUAL CUDAToolkit_VERSION)
  endif()
 
--if(NOT TARGET CUDA::nvToolsExt)
-+if(FALSE)
-   message(FATAL_ERROR "Failed to find nvToolsExt")
+ if(NOT TARGET CUDA::nvToolsExt)
+-  message(FATAL_ERROR "Failed to find nvToolsExt")
++  message(WARNING "Failed to find nvToolsExt")
  endif()
 
+ message(STATUS "Caffe2: CUDA detected: " ${CUDA_VERSION})
 @@ -183,10 +183,10 @@ else()
  endif()
 
@@ -246,17 +275,16 @@ index b37036d..b988e1c 100644
  # cublas
  add_library(caffe2::cublas INTERFACE IMPORTED)
 diff --git a/torch/CMakeLists.txt b/torch/CMakeLists.txt
-index 10a44af..3232f93 100644
+index 10a44af..5460ce7 100644
 --- a/torch/CMakeLists.txt
 +++ b/torch/CMakeLists.txt
-@@ -133,8 +133,9 @@ if(USE_CUDA)
-         list(APPEND TORCH_PYTHON_LINK_LIBRARIES torch::cudnn)
+@@ -134,7 +134,9 @@ if(USE_CUDA)
          list(APPEND TORCH_PYTHON_COMPILE_DEFINITIONS USE_CUDNN)
      endif()
--
+ 
 -    list(APPEND TORCH_PYTHON_LINK_LIBRARIES torch::nvtoolsext)
 +    if(TARGET torch::nvtoolsext)
-+      list(APPEND TORCH_PYTHON_LINK_LIBRARIES torch::nvtoolsext)
++      list(APPEND TORCH_PYTHON_LINK_LIBRARIES torch::nvtoolsext) # may not installed...
 +    endif()
  endif()
 

ports/libtorch/portfile.cmake

@@ -55,15 +55,10 @@ get_filename_component(pybind11_DIR "${SITE_PACKAGES_DIR}/pybind11/share/cmake/p
 vcpkg_check_features(OUT_FEATURE_OPTIONS FEATURE_OPTIONS
   FEATURES
     dist    USE_DISTRIBUTED # MPI, Gloo, TensorPipe
-    zstd    USE_ZSTD
     fbgemm  USE_FBGEMM
-    opencv  USE_OPENCV
     # These are alternatives !
-    # tbb     USE_TBB
-    # tbb     AT_PARALLEL_NATIVE_TBB # AT_PARALLEL_ are alternatives
     # openmp  USE_OPENMP
     # openmp  AT_PARALLEL_OPENMP # AT_PARALLEL_ are alternatives
-    leveldb USE_LEVELDB
     opencl  USE_OPENCL
     cuda    USE_CUDA
     cuda    USE_CUDNN
@@ -74,15 +69,14 @@ vcpkg_check_features(OUT_FEATURE_OPTIONS FEATURE_OPTIONS
     cuda    AT_CUDNN_ENABLED
     cuda    USE_MAGMA
     vulkan  USE_VULKAN
-    #vulkan  USE_VULKAN_SHADERC_RUNTIME
     vulkan  USE_VULKAN_RELAXED_PRECISION
     rocm    USE_ROCM  # This is an alternative to cuda not a feature! (Not in vcpkg.json!) -> disabled
     llvm    USE_LLVM
     mpi     USE_MPI
     nnpack  USE_NNPACK  # todo: check use of `DISABLE_NNPACK_AND_FAMILY`
     nnpack  AT_NNPACK_ENABLED
-    qnnpack USE_QNNPACK # todo: check use of `USE_PYTORCH_QNNPACK`
-#   No feature in vcpkg yet so disabled. -> Requires numpy build by vcpkg itself
+    qnnpack USE_PYTORCH_QNNPACK
+    # No feature in vcpkg yet so disabled. -> Requires numpy build by vcpkg itself
     python  BUILD_PYTHON
     python  USE_NUMPY
 )
@@ -91,16 +85,16 @@ if("dist" IN_LIST FEATURES)
     if(VCPKG_TARGET_IS_LINUX OR VCPKG_TARGET_IS_OSX)
         list(APPEND FEATURE_OPTIONS -DUSE_TENSORPIPE=ON)
     endif()
-    if(VCPKG_TARGET_IS_WINDOWS OR VCPKG_TARGET_IS_OSX)
+    if(VCPKG_TARGET_IS_OSX)
         list(APPEND FEATURE_OPTIONS -DUSE_LIBUV=ON)
     endif()
     list(APPEND FEATURE_OPTIONS -DUSE_GLOO=${VCPKG_TARGET_IS_LINUX})
 endif()
 
 if(VCPKG_TARGET_IS_ANDROID OR VCPKG_TARGET_IS_IOS)
-    list(APPEND FEATURE_OPTIONS -DINTERN_BUILD_MOBILE=ON)
+    set(IS_MOBILE ON)
 else()
-    list(APPEND FEATURE_OPTIONS -DINTERN_BUILD_MOBILE=OFF)
+    set(IS_MOBILE OFF)
 endif()
 
 if(VCPKG_TARGET_IS_OSX OR VCPKG_TARGET_IS_IOS)
@@ -114,61 +108,37 @@ string(COMPARE EQUAL "${VCPKG_CRT_LINKAGE}" "static" USE_STATIC_RUNTIME)
 vcpkg_cmake_configure(
     SOURCE_PATH "${SOURCE_PATH}"
     DISABLE_PARALLEL_CONFIGURE
-    # WINDOWS_USE_MSBUILD
     OPTIONS
         ${FEATURE_OPTIONS}
         -DProtobuf_PROTOC_EXECUTABLE:FILEPATH=${PROTOC}
         -DCAFFE2_CUSTOM_PROTOC_EXECUTABLE:FILEPATH=${PROTOC}
         -DPYTHON_EXECUTABLE:FILEPATH=${PYTHON3}
         -Dpybind11_DIR:PATH=${pybind11_DIR}
-        -DCAFFE2_STATIC_LINK_CUDA=ON
         -DCAFFE2_USE_MSVC_STATIC_RUNTIME=${USE_STATIC_RUNTIME}
         -DBUILD_CUSTOM_PROTOBUF=OFF
         -DUSE_LITE_PROTO=OFF
-        -DBUILD_TEST=OFF
-        -DATEN_NO_TEST=ON
         -DUSE_SYSTEM_LIBS=ON
-        -DUSE_METAL=${IS_APPLE}
         -DUSE_PYTORCH_METAL=${IS_APPLE}
         -DUSE_PYTORCH_METAL_EXPORT=${IS_APPLE}
         -DUSE_GFLAGS=ON
         -DUSE_GLOG=ON
-        -DUSE_LMDB=ON
+        -DUSE_XNNPACK=ON
         -DUSE_ITT=OFF
-        -DUSE_ROCKSDB=ON
         -DUSE_OBSERVERS=OFF
-        -DUSE_PYTORCH_QNNPACK=OFF
         -DUSE_KINETO=OFF
-        -DUSE_ROCM=OFF
-        -DUSE_NUMA=OFF
-        -DUSE_SYSTEM_ONNX=ON
-        -DUSE_SYSTEM_FP16=ON
-        -DUSE_SYSTEM_EIGEN_INSTALL=ON
-        -DUSE_SYSTEM_CPUINFO=ON
-        -DUSE_SYSTEM_PTHREADPOOL=ON
-        -DUSE_SYSTEM_XNNPACK=ON
-        -DUSE_SYSTEM_PYBIND11=ON
-        -DUSE_SYSTEM_ZSTD=ON
-        -DUSE_SYSTEM_GLOO=ON
-        -DUSE_SYSTEM_NCCL=ON
-        -DUSE_SYSTEM_LIBS=ON
-        -DUSE_SYSTEM_FXDIV=ON
-        -DUSE_SYSTEM_SLEEF=ON
+        -DUSE_LITE_INTERPRETER_PROFILER=OFF
+        -DUSE_NUMA=${VCPKG_TARGET_IS_LINUX}
+        -DUSE_MIMALLOC=${VCPKG_TARGET_IS_WINDOWS}
+        -DINTERN_BUILD_MOBILE=${IS_MOBILE}
         -DBUILD_JNI=${VCPKG_TARGET_IS_ANDROID}
         -DUSE_NNAPI=${VCPKG_TARGET_IS_ANDROID}
-        ${BLAS_OPTIONS}
         # BLAS=MKL not supported in this port
         -DUSE_MKLDNN=OFF
         -DUSE_MKLDNN_CBLAS=OFF
-        #-DCAFFE2_USE_MKL=ON
-        #-DAT_MKL_ENABLED=ON
         -DAT_MKLDNN_ENABLED=OFF
-        -DUSE_OPENCL=ON
-        -DUSE_NUMPY=ON
     MAYBE_UNUSED_VARIABLES
+        PYTHON_EXECUTABLE
         USE_NUMA
-        USE_SYSTEM_BIND11
-        MKLDNN_CPU_RUNTIME
 )
 
 vcpkg_cmake_install()

ports/libtorch/vcpkg.json

@@ -26,6 +26,10 @@
     "libkineto",
     "lmdb",
     "mimalloc",
+    {
+      "name": "mimalloc",
+      "platform": "windows"
+    },
     "onnx",
     "onnx-optimizer",
     "opencl",
@@ -55,10 +59,8 @@
   "default-features": [
     {
       "name": "cuda",
-      "platform": "windows & !x86"
-    },
-    "opencv",
-    "zstd"
+      "platform": "windows & x64"
+    }
   ],
   "features": {
     "cuda": {
@@ -97,13 +99,6 @@
         }
       ]
     },
-    "leveldb": {
-      "description": "Build with LevelDB",
-      "dependencies": [
-        "leveldb",
-        "snappy"
-      ]
-    },
     "llvm": {
       "description": "Build with LLVM",
       "dependencies": [
@@ -117,31 +112,19 @@
       ]
     },
     "nnpack": {
-      "description": "Build with NNPack",
+      "description": "Build with NNPACK",
       "supports": "linux | osx",
       "dependencies": [
         "nnpack"
       ]
     },
-    "opencv": {
-      "description": "Build with OpenCV 3.x",
-      "dependencies": [
-        "opencv"
-      ]
-    },
     "vulkan": {
       "description": "Build with Vulkan GPU backend",
       "dependencies": [
         "vulkan",
         "vulkan-loader",
         "vulkan-memory-allocator"
       ]
-    },
-    "zstd": {
-      "description": "Build with ZSTD",
-      "dependencies": [
-        "zstd"
-      ]
     }
   }
 }