--- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -987,7 +987,7 @@ set(CMAKE_COLOR_DIAGNOSTICS ON) endif() if(NOT MSVC) - string(APPEND CMAKE_CXX_FLAGS " -O2 -fPIC") + string(APPEND CMAKE_CXX_FLAGS " -O2") # This prevents use of `c10::optional`, `c10::nullopt` etc within the codebase string(APPEND CMAKE_CXX_FLAGS " -DC10_NODEPRECATED") @@ -998,7 +998,6 @@ # Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459 string(APPEND CMAKE_CXX_FLAGS " -Wall") string(APPEND CMAKE_CXX_FLAGS " -Wextra") - append_cxx_flag_if_supported("-Werror=return-type" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-Werror=non-virtual-dtor" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-Werror=braced-scalar-init" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-Werror=range-loop-construct" CMAKE_CXX_FLAGS) @@ -1083,7 +1082,6 @@ endif() append_cxx_flag_if_supported("-fno-math-errno" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-fno-trapping-math" CMAKE_CXX_FLAGS) - append_cxx_flag_if_supported("-Werror=format" CMAKE_CXX_FLAGS) if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 13) append_cxx_flag_if_supported("-Wno-dangling-reference" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-Wno-error=dangling-reference" CMAKE_CXX_FLAGS) append_cxx_flag_if_supported("-Wno-error=redundant-move" CMAKE_CXX_FLAGS) --- a/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt +++ b/aten/src/ATen/native/quantized/cpu/qnnpack/CMakeLists.txt @@ -323,7 +323,7 @@ set_target_properties(pytorch_qnnpack PROPERTIES PUBLIC_HEADER include/qnnpack_func.h) # ---[ Configure clog -if(NOT TARGET clog) +if(FALSE) set(CLOG_BUILD_TESTS OFF CACHE BOOL "") set(CLOG_RUNTIME_TYPE "${CPUINFO_RUNTIME_TYPE}" CACHE STRING "") add_subdirectory( @@ -335,7 +335,8 @@ target_compile_options(clog PRIVATE "-Wno-unused-result") endif() endif() -target_link_libraries(pytorch_qnnpack PUBLIC clog) +find_library(CLOG_LIBRARY NAMES clog REQUIRED) +target_link_libraries(pytorch_qnnpack PUBLIC ${CLOG_LIBRARY}) # ---[ Configure cpuinfo if(NOT TARGET cpuinfo AND USE_SYSTEM_CPUINFO) --- a/caffe2/CMakeLists.txt +++ b/caffe2/CMakeLists.txt @@ -87,7 +87,7 @@ endif() # Note: the folders that are being commented out have not been properly # addressed yet. -if(NOT MSVC AND USE_XNNPACK) +if(FALSE) if(NOT TARGET fxdiv) set(FXDIV_BUILD_TESTS OFF CACHE BOOL "") set(FXDIV_BUILD_BENCHMARKS OFF CACHE BOOL "") @@ -1195,7 +1195,6 @@ if(USE_XPU) endif() if(NOT MSVC AND USE_XNNPACK) - TARGET_LINK_LIBRARIES(torch_cpu PRIVATE fxdiv) endif() # ========================================================== @@ -1307,17 +1306,6 @@ target_include_directories(torch_cpu PRIVATE "/usr/include/kineto") -if(USE_KINETO) - target_include_directories(torch_cpu PRIVATE - ${TORCH_ROOT}/third_party/kineto/libkineto/src) -endif() - -target_include_directories(torch_cpu PRIVATE - ${TORCH_ROOT}/third_party/cpp-httplib) - -target_include_directories(torch_cpu PRIVATE - ${TORCH_ROOT}/third_party/nlohmann/include) - install(DIRECTORY "${TORCH_SRC_DIR}/csrc" "${TORCH_SRC_DIR}/headeronly" --- a/cmake/Codegen.cmake +++ b/cmake/Codegen.cmake @@ -64,7 +64,7 @@ if(INTERN_BUILD_ATEN_OPS) if(MSVC) set(OPT_FLAG "/fp:strict ") else(MSVC) - set(OPT_FLAG "-O3 ") + set(OPT_FLAG " ") if("${CMAKE_BUILD_TYPE}" MATCHES "Debug") set(OPT_FLAG " ") endif() --- a/cmake/Dependencies.cmake +++ b/cmake/Dependencies.cmake @@ -461,7 +461,9 @@ set_property(TARGET pytorch_qnnpack PROPERTY POSITION_INDEPENDENT_CODE ON) set_property(TARGET cpuinfo PROPERTY POSITION_INDEPENDENT_CODE ON) # QNNPACK depends on gemmlowp headers - target_include_directories(pytorch_qnnpack PRIVATE "${CAFFE2_THIRD_PARTY_ROOT}/gemmlowp") + find_package(gemmlowp REQUIRED) + get_target_property(GEMMLOWP_INCLUDE_DIRS gemmlowp::gemmlowp INTERFACE_INCLUDE_DIRECTORIES) + target_include_directories(pytorch_qnnpack PRIVATE ${GEMMLOWP_INCLUDE_DIRS}) endif() list(APPEND Caffe2_DEPENDENCY_LIBS pytorch_qnnpack) @@ -556,7 +558,7 @@ find_library(microkernels-prod_LIBRARY microkernels-prod) set_property(TARGET XNNPACK PROPERTY IMPORTED_LOCATION "${XNNPACK_LIBRARY}") set_property(TARGET microkernels-prod PROPERTY IMPORTED_LOCATION "${microkernels-prod_LIBRARY}") - if(NOT XNNPACK_LIBRARY or NOT microkernels-prod_LIBRARY) + if(FALSE) message(FATAL_ERROR "Cannot find XNNPACK") endif() message("-- Found XNNPACK: ${XNNPACK_LIBRARY}") @@ -637,7 +639,7 @@ if(BUILD_TEST OR BUILD_MOBILE_BENCHMARK OR BUILD_MOBILE_TEST) endif() # ---[ FBGEMM -if(USE_FBGEMM) +if(FALSE) set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party") if(NOT DEFINED FBGEMM_SOURCE_DIR) set(FBGEMM_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/fbgemm" CACHE STRING "FBGEMM source directory") @@ -696,6 +698,7 @@ if(USE_FBGEMM) endif() if(USE_FBGEMM) + list(APPEND Caffe2_DEPENDENCY_LIBS fbgemm) caffe2_update_option(USE_FBGEMM ON) else() caffe2_update_option(USE_FBGEMM OFF) @@ -1140,7 +1140,6 @@ endif() set(TP_BUILD_LIBUV ON CACHE BOOL "" FORCE) add_compile_options(-DTORCH_USE_LIBUV) - include_directories(BEFORE SYSTEM ${CMAKE_CURRENT_LIST_DIR}/../third_party/tensorpipe/third_party/libuv/include) set(TP_STATIC_OR_SHARED STATIC CACHE STRING "" FORCE) # Tensorpipe uses cuda_add_library @@ -1712,11 +1712,9 @@ # Include cpp-httplib add_library(httplib INTERFACE IMPORTED) -target_include_directories(httplib SYSTEM INTERFACE ${PROJECT_SOURCE_DIR}/third_party/cpp-httplib) # Include nlohmann-json add_library(nlohmann INTERFACE IMPORTED) -include_directories(nlohmann SYSTEM INTERFACE ${PROJECT_SOURCE_DIR}/third_party/nlohmann/include) # Include moodycamel add_library(moodycamel INTERFACE IMPORTED) --- a/cmake/External/nnpack.cmake +++ b/cmake/External/nnpack.cmake @@ -56,7 +56,7 @@ set(PTHREADPOOL_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/pthreadpool" CACHE STRING "pthreadpool source directory") set(GOOGLETEST_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/googletest" CACHE STRING "Google Test source directory") - if(NOT TARGET nnpack) + if(FALSE) set(NNPACK_BUILD_TESTS OFF CACHE BOOL "") set(NNPACK_BUILD_BENCHMARKS OFF CACHE BOOL "") set(NNPACK_LIBRARY_TYPE "static" CACHE STRING "") --- a/cmake/public/utils.cmake +++ b/cmake/public/utils.cmake @@ -460,8 +460,6 @@ function(torch_compile_options libname) endif() # Use -O2 for release builds (-O3 doesn't improve perf, and -Os results in perf regression) - target_compile_options(${libname} PRIVATE - $<$,$,$>>:-O2>) endfunction() --- a/aten/src/ATen/CMakeLists.txt 2025-02-27 14:23:02.402742165 +0100 +++ b/aten/src/ATen/CMakeLists.txt 2025-02-27 14:23:40.445850718 +0100 @@ -326,8 +326,6 @@ if(USE_CUDA) list(APPEND ATen_CUDA_INCLUDE ${CMAKE_CURRENT_SOURCE_DIR}/cuda) # Next two lines are needed because TunableOp uses third-party/fmt - list(APPEND ATen_CUDA_INCLUDE $) - list(APPEND ATen_CUDA_DEPENDENCY_LIBS fmt::fmt-header-only) list(APPEND ATen_CUDA_CU_SRCS ${cuda_cu} ${native_cuda_cu} @@ -395,8 +393,6 @@ _pytorch_rocm_generate_ck_conf() # Next two lines are needed because TunableOp uses third-party/fmt - list(APPEND ATen_HIP_INCLUDE $) - list(APPEND ATen_HIP_DEPENDENCY_LIBS fmt::fmt-header-only) if(USE_FLASH_ATTENTION) list(APPEND ATen_HIP_INCLUDE ${CMAKE_CURRENT_SOURCE_DIR}/native/transformers/hip/flash_attn/ck) endif() --- a/torch/CMakeLists.txt +++ b/torch/CMakeLists.txt @@ -60,16 +60,10 @@ ${CMAKE_BINARY_DIR}/aten/src ${CMAKE_BINARY_DIR}/caffe2/aten/src ${CMAKE_BINARY_DIR}/third_party - ${CMAKE_BINARY_DIR}/third_party/onnx ${TORCH_ROOT}/third_party/valgrind-headers - ${TORCH_ROOT}/third_party/gloo - ${TORCH_ROOT}/third_party/onnx - ${TORCH_ROOT}/third_party/flatbuffers/include "/usr/include/kineto" - ${TORCH_ROOT}/third_party/cpp-httplib - ${TORCH_ROOT}/third_party/nlohmann/include ${TORCH_SRC_DIR}/csrc ${TORCH_SRC_DIR}/csrc/api/include --- a/cmake/FlatBuffers.cmake +++ b/cmake/FlatBuffers.cmake @@ -1,10 +1 @@ -set(FlatBuffers_Include ${PROJECT_SOURCE_DIR}/third_party/flatbuffers/include) -file(GLOB FlatBuffers_Library_SRCS - ${FlatBuffers_Include}/flatbuffers/*.h -) add_library(flatbuffers INTERFACE) -target_sources( - flatbuffers - INTERFACE ${FlatBuffers_Library_SRCS} -) -target_include_directories(flatbuffers INTERFACE ${FlatBuffers_Include})