diff options
author | 2024-08-09 18:12:10 +0200 | |
---|---|---|
committer | 2024-08-09 18:13:52 +0200 | |
commit | e3b7cde9d825c40eba0c1ee43a71d9455aa77c2c (patch) | |
tree | a9366002f93faf50038b0ccdb1c126b338944cc6 /sci-libs | |
parent | kernel-install.eclass: run depmod after compressing modules (diff) | |
download | gentoo-e3b7cde9d825c40eba0c1ee43a71d9455aa77c2c.tar.gz gentoo-e3b7cde9d825c40eba0c1ee43a71d9455aa77c2c.tar.bz2 gentoo-e3b7cde9d825c40eba0c1ee43a71d9455aa77c2c.zip |
sci-libs/caffe2: update SRC_URI
Signed-off-by: Alfredo Tupone <tupone@gentoo.org>
Diffstat (limited to 'sci-libs')
23 files changed, 48 insertions, 980 deletions
diff --git a/sci-libs/caffe2/Manifest b/sci-libs/caffe2/Manifest index 0ab8e6d1b824..d190ba229493 100644 --- a/sci-libs/caffe2/Manifest +++ b/sci-libs/caffe2/Manifest @@ -1,4 +1,4 @@ -DIST caffe2-20240809.tar.gz 15242 BLAKE2B 77503c61487e7d85cca5afcab9a6e638f9833a70861845638cf1b62bc492d7b6650e6db81d53ebb2f39c6313509250d339f725f04d03ec6dd23dd0cf70843d8c SHA512 74b3b0b6671b655ecac93f7436c4ed7cb0157a83aafbf6afcc0811e11cef341cd8f638db1a111bcbb01e1a6dd4daf3a36b96d7a8ce90f04c2fa091bd6e3a142b +DIST caffe2-patches-20240809.tar.gz 15242 BLAKE2B 77503c61487e7d85cca5afcab9a6e638f9833a70861845638cf1b62bc492d7b6650e6db81d53ebb2f39c6313509250d339f725f04d03ec6dd23dd0cf70843d8c SHA512 74b3b0b6671b655ecac93f7436c4ed7cb0157a83aafbf6afcc0811e11cef341cd8f638db1a111bcbb01e1a6dd4daf3a36b96d7a8ce90f04c2fa091bd6e3a142b DIST pytorch-2.2.2.tar.gz 116367503 BLAKE2B 0be22f2ec4b9aac6f5e976664cae01facf07929a32565cd57d7cc5b2d9888e9ae71ca301853752fe8f31d174d04c9974eb9ed2f3d452360a50ccf024f200726a SHA512 7990e0f9484038c3458c0bda2c863bf2b19e56edab81fc5938c6e0f08b17558287f853bb67350e8cca8f42bec0f1d4ba0e94e50a145db8da44bdd4bd703d91d0 DIST pytorch-2.3.0.tar.gz 117029829 BLAKE2B 8f9c0d71ee0a9219b495eddccdcc65107f7ad537c43c68100b229f3d27b0e6c01ccb1659c7fffc356a48d80f2adc0a10361305dc8f1df20446de837d380f89f6 SHA512 67f7e9a096c3ffb952206ebf9105bedebb68c24ad82456083adf1d1d210437fcaa9dd52b68484cfc97d408c9eebc9541c76868c34a7c9982494dc3f424cfb07c DIST pytorch-2.3.1.tar.gz 117035696 BLAKE2B d419d7fa1342f1fb317ffce09ec9dc1447414627cc83d36578fe60f68c283c620b2b4d49f414cd206d537b90b16432a06cd1941662720db05d5e2b6c493325f5 SHA512 e1bcae44f9939fc7ccb1360a9b1970d92426f25e5de73e36964df3dd15ad5d8d9f5bd2f9a7dda6b8f64e2bba3674005bd869f542489cc442ad0125a02676f587 diff --git a/sci-libs/caffe2/caffe2-2.2.2-r1.ebuild b/sci-libs/caffe2/caffe2-2.2.2-r1.ebuild index 6649975ddf2d..773808bc4f76 100644 --- a/sci-libs/caffe2/caffe2-2.2.2-r1.ebuild +++ b/sci-libs/caffe2/caffe2-2.2.2-r1.ebuild @@ -13,7 +13,8 @@ MYP=${MYPN}-${PV} DESCRIPTION="A deep learning framework" HOMEPAGE="https://pytorch.org/" SRC_URI="https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz - -> ${MYP}.tar.gz" + -> ${MYP}.tar.gz + https://dev.gentoo.org/~tupone/distfiles/${PN}-patches-20240809.tar.gz" S="${WORKDIR}"/${MYP} @@ -97,16 +98,16 @@ DEPEND=" " PATCHES=( - "${FILESDIR}"/${PN}-2.2.1-gentoo.patch - "${FILESDIR}"/${PN}-1.13.0-install-dirs.patch - "${FILESDIR}"/${PN}-1.12.0-glog-0.6.0.patch - "${FILESDIR}"/${PN}-1.13.1-tensorpipe.patch - "${FILESDIR}"/${PN}-2.0.0-gcc13.patch - "${FILESDIR}"/${PN}-2.0.0-cudnn_include_fix.patch - "${FILESDIR}"/${PN}-2.1.2-fix-rpath.patch - "${FILESDIR}"/${PN}-2.1.2-fix-openmp-link.patch - "${FILESDIR}"/${PN}-2.1.2-rocm-fix-std-cpp17.patch - "${FILESDIR}"/${P}-musl.patch + ../patches/${PN}-2.2.1-gentoo.patch + ../patches/${PN}-1.13.0-install-dirs.patch + ../patches/${PN}-1.12.0-glog-0.6.0.patch + ../patches/${PN}-1.13.1-tensorpipe.patch + ../patches/${PN}-2.0.0-gcc13.patch + ../patches/${PN}-2.0.0-cudnn_include_fix.patch + ../patches/${PN}-2.1.2-fix-rpath.patch + ../patches/${PN}-2.1.2-fix-openmp-link.patch + ../patches/${PN}-2.1.2-rocm-fix-std-cpp17.patch + ../patches/${P}-musl.patch ) src_prepare() { diff --git a/sci-libs/caffe2/caffe2-2.3.0-r3.ebuild b/sci-libs/caffe2/caffe2-2.3.0-r3.ebuild index 666800d8f4b6..7fe4818311cb 100644 --- a/sci-libs/caffe2/caffe2-2.3.0-r3.ebuild +++ b/sci-libs/caffe2/caffe2-2.3.0-r3.ebuild @@ -13,7 +13,8 @@ MYP=${MYPN}-${PV} DESCRIPTION="A deep learning framework" HOMEPAGE="https://pytorch.org/" SRC_URI="https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz - -> ${MYP}.tar.gz" + -> ${MYP}.tar.gz + https://dev.gentoo.org/~tupone/distfiles/${PN}-patches-20240809.tar.gz" S="${WORKDIR}"/${MYP} @@ -106,21 +107,21 @@ DEPEND=" " PATCHES=( - "${FILESDIR}"/${PN}-2.2.1-gentoo.patch - "${FILESDIR}"/${PN}-1.13.0-install-dirs.patch - "${FILESDIR}"/${PN}-1.12.0-glog-0.6.0.patch - "${FILESDIR}"/${PN}-1.13.1-tensorpipe.patch - "${FILESDIR}"/${P}-cudnn_include_fix.patch - "${FILESDIR}"/${PN}-2.1.2-fix-rpath.patch - "${FILESDIR}"/${PN}-2.1.2-fix-openmp-link.patch - "${FILESDIR}"/${P}-rocm-fix-std-cpp17.patch - "${FILESDIR}"/${PN}-2.2.2-musl.patch - "${FILESDIR}"/${P}-CMakeFix.patch - "${FILESDIR}"/${PN}-2.3.0-exclude-aotriton.patch - "${FILESDIR}"/${PN}-2.3.0-fix-rocm-gcc14-clamp.patch - "${FILESDIR}"/${PN}-2.3.0-optional-hipblaslt.patch - "${FILESDIR}"/${PN}-2.3.0-fix-libcpp.patch - "${FILESDIR}"/${PN}-2.3.0-fix-gcc-clang-abi-compat.patch + ../patches/${PN}-2.2.1-gentoo.patch + ../patches/${PN}-1.13.0-install-dirs.patch + ../patches/${PN}-1.12.0-glog-0.6.0.patch + ../patches/${PN}-1.13.1-tensorpipe.patch + ../patches/${P}-cudnn_include_fix.patch + ../patches/${PN}-2.1.2-fix-rpath.patch + ../patches/${PN}-2.1.2-fix-openmp-link.patch + ../patches/${P}-rocm-fix-std-cpp17.patch + ../patches/${PN}-2.2.2-musl.patch + ../patches/${P}-CMakeFix.patch + ../patches/${PN}-2.3.0-exclude-aotriton.patch + ../patches/${PN}-2.3.0-fix-rocm-gcc14-clamp.patch + ../patches/${PN}-2.3.0-optional-hipblaslt.patch + ../patches/${PN}-2.3.0-fix-libcpp.patch + ../patches/${PN}-2.3.0-fix-gcc-clang-abi-compat.patch ) src_prepare() { diff --git a/sci-libs/caffe2/caffe2-2.3.1.ebuild b/sci-libs/caffe2/caffe2-2.3.1.ebuild index ee1da28aa12f..ff2a9caebd59 100644 --- a/sci-libs/caffe2/caffe2-2.3.1.ebuild +++ b/sci-libs/caffe2/caffe2-2.3.1.ebuild @@ -13,7 +13,8 @@ MYP=${MYPN}-${PV} DESCRIPTION="A deep learning framework" HOMEPAGE="https://pytorch.org/" SRC_URI="https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz - -> ${MYP}.tar.gz" + -> ${MYP}.tar.gz + https://dev.gentoo.org/~tupone/distfiles/${PN}-patches-20240809.tar.gz" S="${WORKDIR}"/${MYP} @@ -106,21 +107,21 @@ DEPEND=" " PATCHES=( - "${FILESDIR}"/${PN}-2.2.1-gentoo.patch - "${FILESDIR}"/${PN}-1.13.0-install-dirs.patch - "${FILESDIR}"/${PN}-1.12.0-glog-0.6.0.patch - "${FILESDIR}"/${PN}-1.13.1-tensorpipe.patch - "${FILESDIR}"/${PN}-2.3.0-cudnn_include_fix.patch - "${FILESDIR}"/${PN}-2.1.2-fix-rpath.patch - "${FILESDIR}"/${PN}-2.1.2-fix-openmp-link.patch - "${FILESDIR}"/${PN}-2.3.0-rocm-fix-std-cpp17.patch - "${FILESDIR}"/${PN}-2.2.2-musl.patch - "${FILESDIR}"/${PN}-2.3.0-CMakeFix.patch - "${FILESDIR}"/${PN}-2.3.0-exclude-aotriton.patch - "${FILESDIR}"/${PN}-2.3.0-fix-rocm-gcc14-clamp.patch - "${FILESDIR}"/${PN}-2.3.0-optional-hipblaslt.patch - "${FILESDIR}"/${PN}-2.3.0-fix-libcpp.patch - "${FILESDIR}"/${PN}-2.3.0-fix-gcc-clang-abi-compat.patch + ../patches/${PN}-2.2.1-gentoo.patch + ../patches/${PN}-1.13.0-install-dirs.patch + ../patches/${PN}-1.12.0-glog-0.6.0.patch + ../patches/${PN}-1.13.1-tensorpipe.patch + ../patches/${PN}-2.3.0-cudnn_include_fix.patch + ../patches/${PN}-2.1.2-fix-rpath.patch + ../patches/${PN}-2.1.2-fix-openmp-link.patch + ../patches/${PN}-2.3.0-rocm-fix-std-cpp17.patch + ../patches/${PN}-2.2.2-musl.patch + ../patches/${PN}-2.3.0-CMakeFix.patch + ../patches/${PN}-2.3.0-exclude-aotriton.patch + ../patches/${PN}-2.3.0-fix-rocm-gcc14-clamp.patch + ../patches/${PN}-2.3.0-optional-hipblaslt.patch + ../patches/${PN}-2.3.0-fix-libcpp.patch + ../patches/${PN}-2.3.0-fix-gcc-clang-abi-compat.patch ) src_prepare() { diff --git a/sci-libs/caffe2/caffe2-2.4.0.ebuild b/sci-libs/caffe2/caffe2-2.4.0.ebuild index 730ea4d365f2..524dafcaacef 100644 --- a/sci-libs/caffe2/caffe2-2.4.0.ebuild +++ b/sci-libs/caffe2/caffe2-2.4.0.ebuild @@ -14,7 +14,7 @@ DESCRIPTION="A deep learning framework" HOMEPAGE="https://pytorch.org/" SRC_URI="https://github.com/pytorch/${MYPN}/archive/refs/tags/v${PV}.tar.gz -> ${MYP}.tar.gz - https://dev.gentoo.org/~tupone/distfiles/caffe2-20240809.tar.gz" + https://dev.gentoo.org/~tupone/distfiles/${PN}-patches-20240809.tar.gz" S="${WORKDIR}"/${MYP} diff --git a/sci-libs/caffe2/files/caffe2-1.12.0-glog-0.6.0.patch b/sci-libs/caffe2/files/caffe2-1.12.0-glog-0.6.0.patch deleted file mode 100644 index 6c06d2cca654..000000000000 --- a/sci-libs/caffe2/files/caffe2-1.12.0-glog-0.6.0.patch +++ /dev/null @@ -1,29 +0,0 @@ -https://github.com/pytorch/pytorch/issues/58054 - ---- a/c10/util/Logging.cpp -+++ b/c10/util/Logging.cpp -@@ -192,23 +192,13 @@ - google::GLOG_WARNING, - "The minimum log level that caffe2 will output."); - --// 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). --namespace google { --namespace glog_internal_namespace_ { --bool IsGoogleLoggingInitialized(); --} // namespace glog_internal_namespace_ --} // namespace google -- - namespace c10 { - namespace { - - void initGoogleLogging(char const* name) { - #if !defined(_MSC_VER) - // This trick can only be used on UNIX platforms -- if (!::google::glog_internal_namespace_::IsGoogleLoggingInitialized()) -+ if (!::google::IsGoogleLoggingInitialized()) - #endif - { - ::google::InitGoogleLogging(name); diff --git a/sci-libs/caffe2/files/caffe2-1.13.0-install-dirs.patch b/sci-libs/caffe2/files/caffe2-1.13.0-install-dirs.patch deleted file mode 100644 index 299c9f88a173..000000000000 --- a/sci-libs/caffe2/files/caffe2-1.13.0-install-dirs.patch +++ /dev/null @@ -1,121 +0,0 @@ ---- a/c10/CMakeLists.txt -+++ b/c10/CMakeLists.txt -@@ -112,7 +112,7 @@ - # 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. --install(TARGETS c10 EXPORT Caffe2Targets DESTINATION lib) -+install(TARGETS c10 EXPORT Caffe2Targets DESTINATION ${CMAKE_INSTALL_LIBDIR}) - install(DIRECTORY ${CMAKE_CURRENT_LIST_DIR} - DESTINATION include - FILES_MATCHING PATTERN "*.h") ---- a/c10/cuda/CMakeLists.txt -+++ b/c10/cuda/CMakeLists.txt -@@ -64,7 +64,7 @@ add_subdirectory(test) - # 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. --install(TARGETS c10_cuda EXPORT Caffe2Targets DESTINATION lib) -+install(TARGETS c10_cuda EXPORT Caffe2Targets DESTINATION ${CMAKE_INSTALL_LIBDIR}) - foreach(file ${C10_CUDA_HEADERS}) - get_filename_component( dir ${file} DIRECTORY ) - install( FILES ${file} DESTINATION include/c10/cuda/${dir} ) ---- a/c10/hip/CMakeLists.txt -+++ b/c10/hip/CMakeLists.txt -@@ -55,7 +55,7 @@ target_include_directories( - add_subdirectory(test) - - # ---[ Installation --install(TARGETS c10_hip EXPORT Caffe2Targets DESTINATION lib) -+install(TARGETS c10_hip EXPORT Caffe2Targets DESTINATION ${CMAKE_INSTALL_LIBDIR}) - install(DIRECTORY ${CMAKE_CURRENT_LIST_DIR} - DESTINATION include - FILES_MATCHING PATTERN "*.h") ---- a/modules/detectron/CMakeLists.txt -+++ b/modules/detectron/CMakeLists.txt -@@ -20,7 +20,7 @@ - if(USE_MKLDNN) - target_link_libraries(caffe2_detectron_ops_gpu PRIVATE caffe2::mkldnn) - endif() -- install(TARGETS caffe2_detectron_ops_gpu DESTINATION lib) -+ install(TARGETS caffe2_detectron_ops_gpu DESTINATION ${CMAKE_INSTALL_LIBDIR}) - if(MSVC) - install(FILES $<TARGET_PDB_FILE:caffe2_detectron_ops_gpu> DESTINATION lib OPTIONAL) - endif() -@@ -37,7 +37,7 @@ - target_link_libraries(caffe2_detectron_ops_hip PRIVATE caffe2::mkldnn) - endif() - target_link_libraries(caffe2_detectron_ops_hip PRIVATE torch) -- install(TARGETS caffe2_detectron_ops_hip DESTINATION lib) -+ install(TARGETS caffe2_detectron_ops_hip DESTINATION ${CMAKE_INSTALL_LIBDIR}) - elseif(NOT IOS_PLATFORM) - add_library(caffe2_detectron_ops SHARED ${Detectron_CPU_SRCS}) - if(HAVE_SOVERSION) -@@ -49,7 +49,7 @@ - if(USE_MKLDNN) - target_link_libraries(caffe2_detectron_ops PRIVATE caffe2::mkldnn) - endif() -- install(TARGETS caffe2_detectron_ops DESTINATION lib) -+ install(TARGETS caffe2_detectron_ops DESTINATION ${CMAKE_INSTALL_LIBDIR}) - if(MSVC) - install(FILES $<TARGET_PDB_FILE:caffe2_detectron_ops> DESTINATION lib OPTIONAL) - endif() ---- a/modules/module_test/CMakeLists.txt -+++ b/modules/module_test/CMakeLists.txt -@@ -16,7 +16,7 @@ if(BUILD_TEST AND NOT BUILD_LITE_INTERPRETER) - VERSION ${TORCH_VERSION} SOVERSION ${TORCH_SOVERSION}) - endif() - target_link_libraries(caffe2_module_test_dynamic torch_library) -- install(TARGETS caffe2_module_test_dynamic DESTINATION lib) -+ install(TARGETS caffe2_module_test_dynamic DESTINATION ${CMAKE_INSTALL_LIBDIR}) - if(MSVC AND BUILD_SHARED_LIBS) - install(FILES $<TARGET_PDB_FILE:caffe2_module_test_dynamic> DESTINATION lib OPTIONAL) - endif() ---- a/modules/observers/CMakeLists.txt -+++ b/modules/observers/CMakeLists.txt -@@ -21,7 +21,7 @@ endif() - target_link_libraries(caffe2_observers PUBLIC torch_library) - target_include_directories(caffe2_observers PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/..) - target_compile_options(caffe2_observers PRIVATE "-DCAFFE2_BUILD_OBSERVER_LIB") --install(TARGETS caffe2_observers DESTINATION lib) -+install(TARGETS caffe2_observers DESTINATION ${CMAKE_INSTALL_LIBDIR}) - caffe2_interface_library(caffe2_observers caffe2_observers_library) - if(MSVC AND BUILD_SHARED_LIBS) - install(FILES $<TARGET_PDB_FILE:caffe2_observers> DESTINATION lib OPTIONAL) ---- a/modules/rocksdb/CMakeLists.txt -+++ b/modules/rocksdb/CMakeLists.txt -@@ -63,7 +63,7 @@ add_library(caffe2_rocksdb ${CMAKE_CURRENT_SOURCE_DIR}/rocksdb.cc) - target_link_libraries(caffe2_rocksdb PUBLIC torch_library) - target_link_libraries(caffe2_rocksdb PRIVATE ${RocksDB_LIBRARIES}) - target_include_directories(caffe2_rocksdb PRIVATE ${RocksDB_INCLUDE_DIR}) --install(TARGETS caffe2_rocksdb DESTINATION lib) -+install(TARGETS caffe2_rocksdb DESTINATION ${CMAKE_INSTALL_LIBDIR}) - - # ---[ Last, Append the library to Caffe2_MODULES, if we are building with - # the main repo. ---- a/test/cpp/c10d/CMakeLists.txt -+++ b/test/cpp/c10d/CMakeLists.txt -@@ -51,7 +51,7 @@ if(USE_CUDA) - if(INSTALL_TEST) - install(TARGETS ProcessGroupNCCLTest DESTINATION bin) - install(TARGETS ProcessGroupNCCLErrorsTest DESTINATION bin) -- install(TARGETS c10d_cuda_test DESTINATION lib) -+ install(TARGETS c10d_cuda_test DESTINATION ${CMAKE_INSTALL_LIBDIR}) - endif() - endif() - if(USE_UCC AND USE_C10D_UCC) ---- a/test/cpp/jit/CMakeLists.txt -+++ b/test/cpp/jit/CMakeLists.txt -@@ -32,9 +32,9 @@ endif() - target_link_libraries(backend_with_compiler torch) - - if(INSTALL_TEST) -- install(TARGETS torchbind_test DESTINATION lib) -- install(TARGETS jitbackend_test DESTINATION lib) -- install(TARGETS backend_with_compiler DESTINATION lib) -+ install(TARGETS torchbind_test DESTINATION ${CMAKE_INSTALL_LIBDIR}) -+ install(TARGETS jitbackend_test DESTINATION ${CMAKE_INSTALL_LIBDIR}) -+ install(TARGETS backend_with_compiler DESTINATION ${CMAKE_INSTALL_LIBDIR}) - endif() - - # Build the cpp gtest binary containing the cpp-only tests. diff --git a/sci-libs/caffe2/files/caffe2-1.13.1-tensorpipe.patch b/sci-libs/caffe2/files/caffe2-1.13.1-tensorpipe.patch deleted file mode 100644 index ae0cac9fb947..000000000000 --- a/sci-libs/caffe2/files/caffe2-1.13.1-tensorpipe.patch +++ /dev/null @@ -1,10 +0,0 @@ ---- a/cmake/Dependencies.cmake 2023-02-28 14:14:49.099057348 +0100 -+++ b/cmake/Dependencies.cmake 2023-02-28 14:15:05.326790806 +0100 -@@ -1404,7 +1404,6 @@ - - # Tensorpipe uses cuda_add_library - torch_update_find_cuda_flags() -- add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/tensorpipe) - - list(APPEND Caffe2_DEPENDENCY_LIBS tensorpipe) - if(USE_CUDA) diff --git a/sci-libs/caffe2/files/caffe2-2.0.0-cudnn_include_fix.patch b/sci-libs/caffe2/files/caffe2-2.0.0-cudnn_include_fix.patch deleted file mode 100644 index ff64e4108087..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.0.0-cudnn_include_fix.patch +++ /dev/null @@ -1,12 +0,0 @@ -diff -uar pytorch-2.0.0/cmake/Dependencies.cmake pytorch-2.0.0orig/cmake/Dependencies.cmake ---- a/cmake/Dependencies.cmake 2023-04-23 09:43:20.767566047 -0400 -+++ b/cmake/Dependencies.cmake 2023-03-09 17:42:00.000000000 -0500 -@@ -1235,7 +1235,7 @@ - - # ---[ cuDNN - if(USE_CUDNN) -- set(CUDNN_FRONTEND_INCLUDE_DIR ${CMAKE_CURRENT_LIST_DIR}/../third_party/cudnn_frontend/include) -+ set(CUDNN_FRONTEND_INCLUDE_DIR /opt/cuda/include) - target_include_directories(torch::cudnn INTERFACE ${CUDNN_FRONTEND_INCLUDE_DIR}) - endif() - diff --git a/sci-libs/caffe2/files/caffe2-2.0.0-gcc13.patch b/sci-libs/caffe2/files/caffe2-2.0.0-gcc13.patch deleted file mode 100644 index acbcebad0a5d..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.0.0-gcc13.patch +++ /dev/null @@ -1,41 +0,0 @@ ---- a/c10/util/Registry.h 2023-03-09 17:42:00.000000000 -0500 -+++ b/c10/util/Registry.h 2023-04-09 20:38:33.108135511 -0400 -@@ -16,6 +16,7 @@ - #include <memory> - #include <mutex> - #include <string> -+#include <stdexcept> - #include <unordered_map> - #include <vector> - ---- a/torch/csrc/jit/passes/quantization/quantization_type.h 2023-03-09 17:42:00.000000000 -0500 -+++ b/torch/csrc/jit/passes/quantization/quantization_type.h 2023-04-09 20:43:43.124806308 -0400 -@@ -1,5 +1,6 @@ - #pragma once - #include <ostream> -+#include <cstdint> - - namespace torch { - namespace jit { - ---- a/torch/csrc/jit/runtime/logging.cpp 2023-03-09 17:42:00.000000000 -0500 -+++ b/torch/csrc/jit/runtime/logging.cpp 2023-04-09 20:47:49.758142941 -0400 -@@ -1,6 +1,7 @@ - #include <torch/csrc/jit/runtime/logging.h> - - #include <atomic> -+#include <stdexcept> - #include <mutex> - #include <unordered_map> - - ---- a/torch/csrc/lazy/core/multi_wait.cpp 2023-03-09 17:42:00.000000000 -0500 -+++ b/torch/csrc/lazy/core/multi_wait.cpp 2023-04-09 20:50:36.608145172 -0400 -@@ -1,6 +1,7 @@ - #include <torch/csrc/lazy/core/multi_wait.h> - - #include <chrono> -+#include <stdexcept> - #include <exception> - - namespace torch { diff --git a/sci-libs/caffe2/files/caffe2-2.1.2-fix-openmp-link.patch b/sci-libs/caffe2/files/caffe2-2.1.2-fix-openmp-link.patch deleted file mode 100644 index 3f2d0ae3c30a..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.1.2-fix-openmp-link.patch +++ /dev/null @@ -1,15 +0,0 @@ -Fix "undefined symbol: omp_get_max_active_levels" in mkl + <nothing else> builds -https://github.com/pytorch/pytorch/issues/116576 ---- a/caffe2/CMakeLists.txt -+++ b/caffe2/CMakeLists.txt -@@ -1575,6 +1575,10 @@ if(BUILD_SHARED_LIBS) - target_link_libraries(torch_global_deps TBB::tbb) - endif() - -+ if(USE_OPENMP) -+ target_link_libraries(torch_global_deps OpenMP::OpenMP_CXX) -+ endif() -+ - install(TARGETS torch_global_deps DESTINATION "${TORCH_INSTALL_LIB_DIR}") - endif() - diff --git a/sci-libs/caffe2/files/caffe2-2.1.2-fix-rpath.patch b/sci-libs/caffe2/files/caffe2-2.1.2-fix-rpath.patch deleted file mode 100644 index 731227fa25ee..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.1.2-fix-rpath.patch +++ /dev/null @@ -1,12 +0,0 @@ -Unset rpath to support blas-lapack-switch -Bug: https://bugs.gentoo.org/921129 ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -10,7 +10,6 @@ endif(APPLE) - set(CMAKE_SKIP_BUILD_RPATH FALSE) - # Don't use the install-rpath during the build phase - set(CMAKE_BUILD_WITH_INSTALL_RPATH FALSE) --set(CMAKE_INSTALL_RPATH "${_rpath_portable_origin}") - # Automatically add all linked folders that are NOT in the build directory to - # the rpath (per library?) - set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE) diff --git a/sci-libs/caffe2/files/caffe2-2.1.2-rocm-fix-std-cpp17.patch b/sci-libs/caffe2/files/caffe2-2.1.2-rocm-fix-std-cpp17.patch deleted file mode 100644 index cb0fa0c48e80..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.1.2-rocm-fix-std-cpp17.patch +++ /dev/null @@ -1,68 +0,0 @@ -Fix for error: invalid argument '-std=c++17' not allowed with 'C' -https://github.com/pytorch/pytorch/issues/103222 ---- a/c10/hip/CMakeLists.txt -+++ b/c10/hip/CMakeLists.txt -@@ -30,6 +30,7 @@ hip_add_library(c10_hip ${C10_HIP_SRCS} ${C10_HIP_HEADERS}) - - # Propagate HIP_CXX_FLAGS that were set from Dependencies.cmake - target_compile_options(c10_hip PRIVATE ${HIP_CXX_FLAGS}) -+set_target_properties(c10_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - - # caffe2_hip adds a bunch of dependencies like rocsparse, but c10/hip is supposed to be - # minimal. I'm not sure if we need hip_hcc or not; for now leave it out ---- a/caffe2/CMakeLists.txt -+++ b/caffe2/CMakeLists.txt -@@ -1598,6 +1598,7 @@ if(USE_ROCM) - - # Since PyTorch files contain HIP headers, these flags are required for the necessary definitions to be added. - target_compile_options(torch_hip PUBLIC ${HIP_CXX_FLAGS}) # experiment -+ set_target_properties(torch_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - target_link_libraries(torch_hip PUBLIC c10_hip) - - if(NOT INTERN_BUILD_MOBILE) -@@ -1774,6 +1775,7 @@ if(BUILD_TEST) - target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>) - target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE} ${Caffe2_HIP_INCLUDE}) - target_compile_options(${test_name} PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(${test_name} PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>) - if(INSTALL_TEST) - install(TARGETS ${test_name} DESTINATION test) -@@ -1955,6 +1957,7 @@ if(BUILD_PYTHON) - endif() - if(NOT MSVC) - target_compile_options(caffe2_pybind11_state_hip PRIVATE ${HIP_CXX_FLAGS} -fvisibility=hidden) -+ set_target_properties(caffe2_pybind11_state_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - endif() - set_target_properties(caffe2_pybind11_state_hip PROPERTIES PREFIX "") - set_target_properties(caffe2_pybind11_state_hip PROPERTIES SUFFIX ${PY_EXT_SUFFIX}) ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -1287,7 +1287,6 @@ if(USE_ROCM) - list(APPEND HIP_CXX_FLAGS -Wno-duplicate-decl-specifier) - list(APPEND HIP_CXX_FLAGS -DCAFFE2_USE_MIOPEN) - list(APPEND HIP_CXX_FLAGS -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_HIP) -- list(APPEND HIP_CXX_FLAGS -std=c++17) - add_definitions(-DROCM_VERSION=${ROCM_VERSION_DEV_INT}) - add_definitions(-DTORCH_HIP_VERSION=${TORCH_HIP_VERSION}) - message("TORCH_HIP_VERSION=${TORCH_HIP_VERSION} is added as a compiler defines") ---- a/cmake/public/utils.cmake -+++ b/cmake/public/utils.cmake -@@ -335,6 +335,7 @@ function(caffe2_hip_binary_target target_name_or_src) - caffe2_binary_target(${target_name_or_src}) - - target_compile_options(${__target} PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(${__target} PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - target_include_directories(${__target} PRIVATE ${Caffe2_HIP_INCLUDE}) - endfunction() - ---- a/modules/detectron/CMakeLists.txt -+++ b/modules/detectron/CMakeLists.txt -@@ -31,6 +31,7 @@ if(BUILD_CAFFE2_OPS) - ${Detectron_CPU_SRCS} - ${Detectron_HIP_SRCS}) - target_compile_options(caffe2_detectron_ops_hip PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(caffe2_detectron_ops_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - if(USE_MKLDNN) - target_link_libraries(caffe2_detectron_ops_hip PRIVATE caffe2::mkldnn) - endif() diff --git a/sci-libs/caffe2/files/caffe2-2.2.1-gentoo.patch b/sci-libs/caffe2/files/caffe2-2.2.1-gentoo.patch deleted file mode 100644 index 5472a2c41836..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.2.1-gentoo.patch +++ /dev/null @@ -1,195 +0,0 @@ ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -474,7 +474,7 @@ - endif() - - # ---[ QNNPACK --if(USE_QNNPACK) -+if(FALSE) - set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party") - - if(NOT DEFINED QNNPACK_SOURCE_DIR) -@@ -530,7 +530,7 @@ - endif() - - # ---[ Caffe2 Int8 operators (enabled by USE_QNNPACK) depend on gemmlowp and neon2sse headers --if(USE_QNNPACK) -+if(FALSE) - set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party") - include_directories(SYSTEM "${CAFFE2_THIRD_PARTY_ROOT}/gemmlowp") - include_directories(SYSTEM "${CAFFE2_THIRD_PARTY_ROOT}/neon2sse") -@@ -780,7 +780,7 @@ - endif() - - # ---[ FBGEMM --if(USE_FBGEMM) -+if(FALSE) - set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party") - if(NOT DEFINED FBGEMM_SOURCE_DIR) - set(FBGEMM_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/fbgemm" CACHE STRING "FBGEMM source directory") -@@ -828,6 +828,7 @@ - endif() - - if(USE_FBGEMM) -+ list(APPEND Caffe2_DEPENDENCY_LIBS fbgemm) - caffe2_update_option(USE_FBGEMM ON) - else() - caffe2_update_option(USE_FBGEMM OFF) -@@ -1529,7 +1530,6 @@ - set_target_properties(onnx_proto PROPERTIES CXX_STANDARD 17) - endif() - endif() -- add_subdirectory(${CMAKE_CURRENT_LIST_DIR}/../third_party/foxi EXCLUDE_FROM_ALL) - - add_definitions(-DONNX_NAMESPACE=${ONNX_NAMESPACE}) - if(NOT USE_SYSTEM_ONNX) -@@ -1796,7 +1796,6 @@ - # - 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`. - # -@@ -1805,9 +1804,7 @@ - # 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 "") - --list(APPEND Caffe2_DEPENDENCY_LIBS fmt::fmt-header-only) - set(BUILD_SHARED_LIBS ${TEMP_BUILD_SHARED_LIBS} CACHE BOOL "Build shared libs" FORCE) - - # ---[ Kineto ---- a/c10/CMakeLists.txt -+++ b/c10/CMakeLists.txt -@@ -89,7 +89,7 @@ - if(C10_USE_GLOG) - target_link_libraries(c10 PUBLIC glog::glog) - endif() --target_link_libraries(c10 PRIVATE fmt::fmt-header-only) -+target_link_libraries(c10 PRIVATE fmt) - - if(C10_USE_NUMA) - target_include_directories(c10 PRIVATE ${Numa_INCLUDE_DIR}) ---- a/torch/CMakeLists.txt -+++ b/torch/CMakeLists.txt -@@ -59,15 +59,9 @@ - ${CMAKE_BINARY_DIR} - ${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 -- ${TORCH_ROOT}/third_party/kineto/libkineto/include -+ /usr/include/kineto - - ${TORCH_SRC_DIR}/csrc - ${TORCH_SRC_DIR}/csrc/api/include -@@ -80,7 +74,6 @@ - python::python - pybind::pybind11 - shm -- fmt::fmt-header-only - ATEN_CPU_FILES_GEN_LIB) - - if(USE_ASAN AND TARGET Sanitizer::address) ---- a/CMakeLists.txt -+++ b/CMakeLists.txt -@@ -835,12 +835,11 @@ - # Re-include to override append_cxx_flag_if_supported from third_party/FBGEMM - include(cmake/public/utils.cmake) - if(NOT MSVC) -- string(APPEND CMAKE_CXX_FLAGS " -O2 -fPIC") -+ string(APPEND CMAKE_CXX_FLAGS " -O2") - # Eigen fails to build with some versions, so convert this to a warning - # Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459 - string(APPEND CMAKE_CXX_FLAGS " -Wall") - string(APPEND CMAKE_CXX_FLAGS " -Wextra") -- append_cxx_flag_if_supported("-Werror=return-type" CMAKE_CXX_FLAGS) - append_cxx_flag_if_supported("-Werror=non-virtual-dtor" CMAKE_CXX_FLAGS) - append_cxx_flag_if_supported("-Werror=braced-scalar-init" CMAKE_CXX_FLAGS) - append_cxx_flag_if_supported("-Werror=range-loop-construct" CMAKE_CXX_FLAGS) -@@ -930,7 +930,6 @@ - string(APPEND CMAKE_LINKER_FLAGS_DEBUG " -fno-omit-frame-pointer -O0") - append_cxx_flag_if_supported("-fno-math-errno" CMAKE_CXX_FLAGS) - append_cxx_flag_if_supported("-fno-trapping-math" CMAKE_CXX_FLAGS) -- append_cxx_flag_if_supported("-Werror=format" CMAKE_CXX_FLAGS) - else() - # skip unwanted includes from windows.h - add_compile_definitions(WIN32_LEAN_AND_MEAN) ---- a/cmake/public/utils.cmake -+++ b/cmake/public/utils.cmake -@@ -486,8 +486,6 @@ - endif() - - # Use -O2 for release builds (-O3 doesn't improve perf, and -Os results in perf regression) -- target_compile_options(${libname} PRIVATE -- $<$<AND:$<COMPILE_LANGUAGE:CXX>,$<OR:$<CONFIG:Release>,$<CONFIG:RelWithDebInfo>>>:-O2>) - - endfunction() - ---- a/cmake/Codegen.cmake -+++ b/cmake/Codegen.cmake -@@ -57,7 +57,7 @@ - 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/caffe2/CMakeLists.txt -+++ b/caffe2/CMakeLists.txt -@@ -107,7 +107,7 @@ - # 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 "") -@@ -1055,7 +1055,6 @@ - endif() - - if(NOT MSVC AND USE_XNNPACK) -- TARGET_LINK_LIBRARIES(torch_cpu PRIVATE fxdiv) - endif() - - # ========================================================== -@@ -1175,8 +1174,7 @@ - target_include_directories(torch_cpu PRIVATE - ${TORCH_ROOT}/third_party/miniz-2.1.0) - --target_include_directories(torch_cpu PRIVATE -- ${TORCH_ROOT}/third_party/kineto/libkineto/include) -+target_include_directories(torch_cpu PRIVATE /usr/include/kineto) - - if(USE_KINETO) - target_include_directories(torch_cpu PRIVATE ---- a/cmake/External/nnpack.cmake -+++ b/cmake/External/nnpack.cmake -@@ -56,7 +56,7 @@ - set(PTHREADPOOL_SOURCE_DIR "${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) - if(NOT USE_SYSTEM_PTHREADPOOL AND USE_INTERNAL_PTHREADPOOL_IMPL) - set(NNPACK_CUSTOM_THREADPOOL ON CACHE BOOL "") - endif() ---- a/functorch/CMakeLists.txt 2023-11-30 20:30:45.805209036 +0100 -+++ b/functorch/CMakeLists.txt 2023-11-30 20:31:13.284766157 +0100 -@@ -35,4 +35,4 @@ - if(NOT ${TORCH_PYTHON_LINK_FLAGS} STREQUAL "") - set_target_properties(${PROJECT_NAME} PROPERTIES LINK_FLAGS ${TORCH_PYTHON_LINK_FLAGS}) - endif() --install(TARGETS ${PROJECT_NAME} DESTINATION "${CMAKE_CURRENT_SOURCE_DIR}") -+install(TARGETS ${PROJECT_NAME} DESTINATION "${CMAKE_INSTALL_LIBDIR}") diff --git a/sci-libs/caffe2/files/caffe2-2.2.2-musl.patch b/sci-libs/caffe2/files/caffe2-2.2.2-musl.patch deleted file mode 100644 index f63e9f1df332..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.2.2-musl.patch +++ /dev/null @@ -1,13 +0,0 @@ ---- a/torch/csrc/profiler/unwind/unwind.cpp 2024-04-29 12:05:40.895667482 +0200 -+++ b/torch/csrc/profiler/unwind/unwind.cpp 2024-04-29 12:05:53.099524760 +0200 -@@ -112,8 +112,8 @@ - } - - struct Version { -- uint64_t adds_ = LONG_LONG_MAX; -- uint64_t subs_ = LONG_LONG_MAX; -+ uint64_t adds_ = LLONG_MAX; -+ uint64_t subs_ = LLONG_MAX; - }; - - struct UnwindCache { diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-CMakeFix.patch b/sci-libs/caffe2/files/caffe2-2.3.0-CMakeFix.patch deleted file mode 100644 index eba37d933cac..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-CMakeFix.patch +++ /dev/null @@ -1,11 +0,0 @@ ---- a/CMakeLists.txt 2024-04-29 20:32:26.259716769 +0200 -+++ b/CMakeLists.txt 2024-04-29 20:32:35.886384618 +0200 -@@ -50,7 +50,7 @@ - - # This define is needed to preserve behavior given anticpated changes to cccl/thrust - # https://nvidia.github.io/libcudacxx/standard_api/numerics_library/complex.html --string(APPEND CMAKE_CUDA_FLAGS "-DLIBCUDACXX_ENABLE_SIMPLIFIED_COMPLEX_OPERATIONS") -+string(APPEND CMAKE_CUDA_FLAGS " -DLIBCUDACXX_ENABLE_SIMPLIFIED_COMPLEX_OPERATIONS") - - if(LINUX) - include(cmake/CheckAbi.cmake) diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-cudnn_include_fix.patch b/sci-libs/caffe2/files/caffe2-2.3.0-cudnn_include_fix.patch deleted file mode 100644 index 77905dbd1ac8..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-cudnn_include_fix.patch +++ /dev/null @@ -1,11 +0,0 @@ ---- a/cmake/Dependencies.cmake 2024-04-29 18:37:34.005639858 +0200 -+++ b/cmake/Dependencies.cmake 2024-04-29 18:39:29.126587738 +0200 -@@ -1235,7 +1235,7 @@ - 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() -- set(CUDNN_FRONTEND_INCLUDE_DIR ${CMAKE_CURRENT_LIST_DIR}/../third_party/cudnn_frontend/include) -+ set(CUDNN_FRONTEND_INCLUDE_DIR /opt/cuda/include) - target_include_directories(torch::cudnn INTERFACE ${CUDNN_FRONTEND_INCLUDE_DIR}) - endif() - diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-exclude-aotriton.patch b/sci-libs/caffe2/files/caffe2-2.3.0-exclude-aotriton.patch deleted file mode 100644 index 2c65987acd85..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-exclude-aotriton.patch +++ /dev/null @@ -1,35 +0,0 @@ -Disables aotriton download when both USE_FLASH_ATTENTION and USE_MEM_EFF_ATTENTION cmake flags are OFF -Backports upstream PR to 2.3.0: https://github.com/pytorch/pytorch/pull/130197 ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -1334,7 +1334,9 @@ if(USE_ROCM) - message(STATUS "Disabling Kernel Assert for ROCm") - endif() - -- include(${CMAKE_CURRENT_LIST_DIR}/External/aotriton.cmake) -+ if(USE_FLASH_ATTENTION) -+ include(${CMAKE_CURRENT_LIST_DIR}/External/aotriton.cmake) -+ endif() - if(USE_CUDA) - caffe2_update_option(USE_MEM_EFF_ATTENTION OFF) - endif() ---- a/aten/src/ATen/native/transformers/cuda/sdp_utils.cpp -+++ b/aten/src/ATen/native/transformers/cuda/sdp_utils.cpp -@@ -21,7 +21,7 @@ - #include <cmath> - #include <functional> - --#if USE_ROCM -+#if defined(USE_ROCM) && defined(USE_FLASH_ATTENTION) - #include <aotriton/flash.h> - #endif - -@@ -186,7 +186,7 @@ bool check_flash_attention_hardware_support(sdp_params const& params, bool debug - // Check that the gpu is capable of running flash attention - using sm80 = SMVersion<8, 0>; - using sm90 = SMVersion<9, 0>; --#if USE_ROCM -+#if defined(USE_ROCM) && defined(USE_FLASH_ATTENTION) - auto stream = at::cuda::getCurrentCUDAStream().stream(); - if (hipSuccess != aotriton::v2::flash::check_gpu(stream)) { - auto dprops = at::cuda::getCurrentDeviceProperties(); diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-fix-gcc-clang-abi-compat.patch b/sci-libs/caffe2/files/caffe2-2.3.0-fix-gcc-clang-abi-compat.patch deleted file mode 100644 index a6f981b7e054..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-fix-gcc-clang-abi-compat.patch +++ /dev/null @@ -1,17 +0,0 @@ - -When gcc builds libtorch_cpu.so and hipcc (clang-18) build libtorch_hip.so, -resulting binary fails in runtime due to different mangling. -Related issue in LLVM: https://github.com/llvm/llvm-project/issues/85656 -Fixed in pytorch-2.4.0 in https://github.com/pytorch/pytorch/commit/a89f442f0b103fa6f38103784a2dfedbd147f863 ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -1314,6 +1314,9 @@ if(USE_ROCM) - list(APPEND HIP_HIPCC_FLAGS -fdebug-info-for-profiling) - endif(CMAKE_BUILD_TYPE MATCHES Debug) - -+ # needed for compat with newer versions of hip-clang that introduced C++20 mangling rules -+ list(APPEND HIP_HIPCC_FLAGS -fclang-abi-compat=17) -+ - set(HIP_CLANG_FLAGS ${HIP_CXX_FLAGS}) - # Ask hcc to generate device code during compilation so we can use - # host linker to link. diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-fix-libcpp.patch b/sci-libs/caffe2/files/caffe2-2.3.0-fix-libcpp.patch deleted file mode 100644 index 75808fd7ec50..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-fix-libcpp.patch +++ /dev/null @@ -1,24 +0,0 @@ -Workaround for libc++ issue https://github.com/llvm/llvm-project/issues/100802 -"reference to __host__ function 'memcpy' in __device__ function" ---- a/c10/util/Half.h -+++ b/c10/util/Half.h -@@ -227,7 +227,7 @@ C10_HOST_DEVICE inline float fp16_ieee_to_fp32_value(uint16_t h) { - // const float exp_scale = 0x1.0p-112f; - constexpr uint32_t scale_bits = (uint32_t)15 << 23; - float exp_scale_val = 0; -- std::memcpy(&exp_scale_val, &scale_bits, sizeof(exp_scale_val)); -+ memcpy(&exp_scale_val, &scale_bits, sizeof(exp_scale_val)); - const float exp_scale = exp_scale_val; - const float normalized_value = - fp32_from_bits((two_w >> 4) + exp_offset) * exp_scale; -@@ -298,8 +298,8 @@ inline uint16_t fp16_ieee_from_fp32_value(float f) { - constexpr uint32_t scale_to_inf_bits = (uint32_t)239 << 23; - constexpr uint32_t scale_to_zero_bits = (uint32_t)17 << 23; - float scale_to_inf_val = 0, scale_to_zero_val = 0; -- std::memcpy(&scale_to_inf_val, &scale_to_inf_bits, sizeof(scale_to_inf_val)); -- std::memcpy( -+ memcpy(&scale_to_inf_val, &scale_to_inf_bits, sizeof(scale_to_inf_val)); -+ memcpy( - &scale_to_zero_val, &scale_to_zero_bits, sizeof(scale_to_zero_val)); - const float scale_to_inf = scale_to_inf_val; - const float scale_to_zero = scale_to_zero_val; diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-fix-rocm-gcc14-clamp.patch b/sci-libs/caffe2/files/caffe2-2.3.0-fix-rocm-gcc14-clamp.patch deleted file mode 100644 index 81ae075c67cc..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-fix-rocm-gcc14-clamp.patch +++ /dev/null @@ -1,18 +0,0 @@ -Fix hip compilation with gcc-14 -Upstream commit: https://github.com/pytorch/pytorch/commit/8c2c3a03fb87c3568a22362d83b00d82b9fb3db2 ---- a/aten/src/ATen/native/cuda/IndexKernel.cu -+++ b/aten/src/ATen/native/cuda/IndexKernel.cu -@@ -259,7 +259,13 @@ void index_put_kernel_quantized_cuda(TensorIterator& iter, const IntArrayRef ind - - gpu_index_kernel(iter, index_size, index_stride, [inv_scale, zero_point, qmin, qmax]C10_DEVICE(char* const out_data, const char* const in_data, const int64_t offset) { - int64_t qvalue = static_cast<int64_t>(zero_point + nearbyintf(*(float*)in_data * inv_scale)); -+ // See https://github.com/pytorch/pytorch/issues/127666 -+ // hip-clang std::clamp __glibcxx_assert_fail host function when building on Fedora40/gcc14 -+#ifndef USE_ROCM - qvalue = std::clamp(qvalue, qmin, qmax); -+#else -+ qvalue = (qvalue < qmin) ? qmin : (qmax < qvalue) ? qmax : qvalue; -+#endif - *(scalar_t*)(out_data + offset) = static_cast<scalar_t>(qvalue); - }); - }); diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-optional-hipblaslt.patch b/sci-libs/caffe2/files/caffe2-2.3.0-optional-hipblaslt.patch deleted file mode 100644 index dc544255c2bd..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-optional-hipblaslt.patch +++ /dev/null @@ -1,235 +0,0 @@ -Makes hipblaslt optional to simplify build for non-datacenter GPUs. -Based on https://github.com/pytorch/pytorch/pull/120551 with added USE_HIPBLASLT cmake option. ---- a/CMakeLists.txt -+++ b/CMakeLists.txt -@@ -225,6 +225,9 @@ option(USE_FAKELOWP "Use FakeLowp operators" OFF) - option(USE_FFMPEG "Use ffmpeg" OFF) - option(USE_GFLAGS "Use GFLAGS" OFF) - option(USE_GLOG "Use GLOG" OFF) -+cmake_dependent_option( -+ USE_HIPBLASLT "Use hipBLASLt" ON -+ "USE_ROCM" OFF) - option(USE_LEVELDB "Use LEVELDB" OFF) - option(USE_LITE_PROTO "Use lite protobuf instead of full." OFF) - option(USE_LMDB "Use LMDB" OFF) ---- a/aten/src/ATen/cuda/CUDABlas.cpp -+++ b/aten/src/ATen/cuda/CUDABlas.cpp -@@ -14,7 +14,7 @@ - #include <c10/util/irange.h> - - #ifdef USE_ROCM --#if ROCM_VERSION >= 60000 -+#ifdef USE_HIPBLASLT - #include <hipblaslt/hipblaslt-ext.hpp> - #endif - // until hipblas has an API to accept flags, we must use rocblas here -@@ -781,7 +781,7 @@ void gemm<at::BFloat16>(CUDABLAS_GEMM_ARGTYPES(at::BFloat16)) { - } - } - --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - - #if defined(USE_ROCM) && ROCM_VERSION >= 50700 && ROCM_VERSION < 60000 - // only for rocm 5.7 where we first supported hipblaslt, it was difficult -@@ -912,6 +912,7 @@ class CuBlasLtMatmulPreference : public CuBlasLtDescriptor< - }; - } // namespace - -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - template <typename Dtype> - void gemm_and_bias( - bool transpose_mat1, -@@ -1124,7 +1125,7 @@ template void gemm_and_bias( - at::BFloat16* result_ptr, - int64_t result_ld, - GEMMAndBiasActivationEpilogue activation); -- -+#endif - void scaled_gemm( - char transa, - char transb, ---- a/aten/src/ATen/cuda/CUDABlas.h -+++ b/aten/src/ATen/cuda/CUDABlas.h -@@ -82,7 +82,7 @@ void gemm_internal<at::Half>(CUDABLAS_GEMM_ARGTYPES(at::Half)); - template <> - void gemm_internal<at::BFloat16>(CUDABLAS_GEMM_ARGTYPES(at::BFloat16)); - --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - enum GEMMAndBiasActivationEpilogue { - None, - RELU, ---- a/aten/src/ATen/cuda/CUDAContextLight.h -+++ b/aten/src/ATen/cuda/CUDAContextLight.h -@@ -9,7 +9,7 @@ - - // cublasLT was introduced in CUDA 10.1 but we enable only for 11.1 that also - // added bf16 support --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - #include <cublasLt.h> - #endif - -@@ -82,7 +82,7 @@ TORCH_CUDA_CPP_API c10::Allocator* getCUDADeviceAllocator(); - /* Handles */ - TORCH_CUDA_CPP_API cusparseHandle_t getCurrentCUDASparseHandle(); - TORCH_CUDA_CPP_API cublasHandle_t getCurrentCUDABlasHandle(); --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - TORCH_CUDA_CPP_API cublasLtHandle_t getCurrentCUDABlasLtHandle(); - #endif - ---- a/aten/src/ATen/cuda/CublasHandlePool.cpp -+++ b/aten/src/ATen/cuda/CublasHandlePool.cpp -@@ -29,7 +29,7 @@ namespace at::cuda { - - namespace { - --#if defined(USE_ROCM) && ROCM_VERSION >= 50700 -+#if defined(USE_ROCM) && defined(USE_HIPBLASLT) - void createCublasLtHandle(cublasLtHandle_t *handle) { - TORCH_CUDABLAS_CHECK(cublasLtCreate(handle)); - } -@@ -190,7 +190,7 @@ cublasHandle_t getCurrentCUDABlasHandle() { - return handle; - } - --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - cublasLtHandle_t getCurrentCUDABlasLtHandle() { - #ifdef USE_ROCM - c10::DeviceIndex device = 0; ---- a/aten/src/ATen/cuda/tunable/TunableGemm.h -+++ b/aten/src/ATen/cuda/tunable/TunableGemm.h -@@ -11,7 +11,7 @@ - - #include <ATen/cuda/tunable/GemmCommon.h> - #ifdef USE_ROCM --#if ROCM_VERSION >= 50700 -+#ifdef USE_HIPBLASLT - #include <ATen/cuda/tunable/GemmHipblaslt.h> - #endif - #include <ATen/cuda/tunable/GemmRocblas.h> -@@ -166,7 +166,7 @@ class GemmTunableOp : public TunableOp<GemmParams<T>, StreamTimer> { - } - #endif - --#if defined(USE_ROCM) && ROCM_VERSION >= 50700 -+#if defined(USE_ROCM) && defined(USE_HIPBLASLT) - static const char *env = std::getenv("PYTORCH_TUNABLEOP_HIPBLASLT_ENABLED"); - if (env == nullptr || strcmp(env, "1") == 0) { - // disallow tuning of hipblaslt with c10::complex -@@ -240,7 +240,7 @@ class GemmStridedBatchedTunableOp : public TunableOp<GemmStridedBatchedParams<T> - } - #endif - --#if defined(USE_ROCM) && ROCM_VERSION >= 50700 -+#if defined(USE_ROCM) && defined(USE_HIPBLASLT) - static const char *env = std::getenv("PYTORCH_TUNABLEOP_HIPBLASLT_ENABLED"); - if (env == nullptr || strcmp(env, "1") == 0) { - // disallow tuning of hipblaslt with c10::complex ---- a/aten/src/ATen/native/cuda/Blas.cpp -+++ b/aten/src/ATen/native/cuda/Blas.cpp -@@ -155,7 +155,7 @@ enum class Activation { - GELU, - }; - --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - cuda::blas::GEMMAndBiasActivationEpilogue activation_to_gemm_and_blas_arg(Activation a) { - switch (a) { - case Activation::None: -@@ -193,6 +193,7 @@ static bool getDisableAddmmCudaLt() { - - #ifdef USE_ROCM - static bool isSupportedHipLtROCmArch(int index) { -+#if defined(USE_HIPBLASLT) - hipDeviceProp_t* prop = at::cuda::getDeviceProperties(index); - std::string device_arch = prop->gcnArchName; - static const std::vector<std::string> archs = {"gfx90a", "gfx940", "gfx941", "gfx942"}; -@@ -203,6 +204,7 @@ static bool isSupportedHipLtROCmArch(int index) { - } - } - TORCH_CHECK(false, "Attempting to use hipBLASLt on a unsupported architecture!"); -+#endif - return false; - } - #endif -@@ -228,7 +230,7 @@ Tensor& addmm_out_cuda_impl(Tensor& result, const Tensor& self, const Tensor& ma - at::ScalarType scalar_type = self.scalar_type(); - c10::MaybeOwned<Tensor> self_; - if (&result != &self) { --#if (defined(CUDA_VERSION) && CUDA_VERSION >= 11040 && !defined(_MSC_VER)) || defined(USE_ROCM) && ROCM_VERSION >= 50700 -+#if (defined(CUDA_VERSION) && CUDA_VERSION >= 11040 && !defined(_MSC_VER)) || defined(USE_ROCM) && defined(USE_HIPBLASLT) - // Strangely, if mat2 has only 1 row or column, we get - // CUBLAS_STATUS_INVALID_VALUE error from cublasLtMatmulAlgoGetHeuristic. - // self.dim() == 1 && result.dim() == 2 && self.sizes()[0] == mat2_sizes[1] -@@ -271,7 +273,7 @@ Tensor& addmm_out_cuda_impl(Tensor& result, const Tensor& self, const Tensor& ma - } - self__sizes = self_->sizes(); - } else { --#if defined(USE_ROCM) && ROCM_VERSION >= 50700 -+#if defined(USE_ROCM) && defined(USE_HIPBLASLT) - useLtInterface = !disable_addmm_cuda_lt && - result.dim() == 2 && result.is_contiguous() && - isSupportedHipLtROCmArch(self.device().index()) && -@@ -322,7 +324,7 @@ Tensor& addmm_out_cuda_impl(Tensor& result, const Tensor& self, const Tensor& ma - - TORCH_INTERNAL_ASSERT_DEBUG_ONLY(!args.result->is_conj()); - --#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && ROCM_VERSION >= 50700) -+#if (!defined(USE_ROCM) && !defined(_MSC_VER)) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - if (useLtInterface) { - AT_DISPATCH_FLOATING_TYPES_AND2( - at::ScalarType::Half, -@@ -876,7 +878,7 @@ _scaled_mm_out_cuda(const Tensor& mat1, const Tensor& mat2, - at::native::resize_output(out, {mat1_sizes[0], mat2_sizes[1]}); - at::native::resize_output(amax, {}); - --#if !defined(USE_ROCM) && !defined(_MSC_VER) || (defined(USE_ROCM) && ROCM_VERSION >= 60000) -+#if !defined(USE_ROCM) && !defined(_MSC_VER) || (defined(USE_ROCM) && defined(USE_HIPBLASLT)) - cublasCommonArgs args(mat1, mat2, out); - const auto out_dtype_ = args.result->scalar_type(); - TORCH_CHECK(args.transa == 't' && args.transb == 'n', "Only multiplication of row-major and column-major matrices is supported by cuBLASLt"); -@@ -906,7 +908,7 @@ _scaled_mm_out_cuda(const Tensor& mat1, const Tensor& mat2, - TORCH_CHECK(false, "_scaled_mm_out_cuda is not compiled for this platform."); - #endif - --#if defined(USE_ROCM) && ROCM_VERSION >= 60000 -+#if defined(USE_ROCM) && defined(USE_HIPBLASLT) - // rocm's hipblaslt does not yet support amax, so calculate separately - auto out_float32 = out.to(kFloat); - out_float32.abs_(); ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -1282,6 +1282,9 @@ if(USE_ROCM) - if(ROCM_VERSION_DEV VERSION_GREATER_EQUAL "6.0.0") - list(APPEND HIP_CXX_FLAGS -DHIPBLAS_V2) - endif() -+ if(hipblast_FOUND) -+ list(APPEND HIP_CXX_FLAGS -DHIPBLASLT) -+ endif() - if(HIPBLASLT_CUSTOM_DATA_TYPE) - list(APPEND HIP_CXX_FLAGS -DHIPBLASLT_CUSTOM_DATA_TYPE) - endif() ---- a/cmake/public/LoadHIP.cmake -+++ b/cmake/public/LoadHIP.cmake -@@ -155,7 +155,7 @@ if(HIP_FOUND) - find_package_and_print_version(hiprand REQUIRED) - find_package_and_print_version(rocblas REQUIRED) - find_package_and_print_version(hipblas REQUIRED) -- if(ROCM_VERSION_DEV VERSION_GREATER_EQUAL "5.7.0") -+ if(ROCM_VERSION_DEV VERSION_GREATER_EQUAL "5.7.0" AND USE_HIPBLASLT) - find_package_and_print_version(hipblaslt REQUIRED) - endif() - find_package_and_print_version(miopen REQUIRED) -@@ -191,7 +191,7 @@ if(HIP_FOUND) - # roctx is part of roctracer - find_library(ROCM_ROCTX_LIB roctx64 HINTS ${ROCM_PATH}/lib) - -- if(ROCM_VERSION_DEV VERSION_GREATER_EQUAL "5.7.0") -+ if(hipblastlt_FOUND) - # check whether hipblaslt is using its own datatype - set(file "${PROJECT_BINARY_DIR}/hipblaslt_test_data_type.cc") - file(WRITE ${file} "" diff --git a/sci-libs/caffe2/files/caffe2-2.3.0-rocm-fix-std-cpp17.patch b/sci-libs/caffe2/files/caffe2-2.3.0-rocm-fix-std-cpp17.patch deleted file mode 100644 index 127a31e4b225..000000000000 --- a/sci-libs/caffe2/files/caffe2-2.3.0-rocm-fix-std-cpp17.patch +++ /dev/null @@ -1,68 +0,0 @@ -Fix for error: invalid argument '-std=c++17' not allowed with 'C' -https://github.com/pytorch/pytorch/issues/103222 ---- a/c10/hip/CMakeLists.txt -+++ b/c10/hip/CMakeLists.txt -@@ -30,6 +30,7 @@ hip_add_library(c10_hip ${C10_HIP_SRCS} ${C10_HIP_HEADERS}) - - # Propagate HIP_CXX_FLAGS that were set from Dependencies.cmake - target_compile_options(c10_hip PRIVATE ${HIP_CXX_FLAGS}) -+set_target_properties(c10_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - - # caffe2_hip adds a bunch of dependencies like rocsparse, but c10/hip is supposed to be - # minimal. I'm not sure if we need hip_hcc or not; for now leave it out ---- a/caffe2/CMakeLists.txt -+++ b/caffe2/CMakeLists.txt -@@ -1712,6 +1712,7 @@ if(USE_ROCM) - - # Since PyTorch files contain HIP headers, these flags are required for the necessary definitions to be added. - target_compile_options(torch_hip PUBLIC ${HIP_CXX_FLAGS}) # experiment -+ set_target_properties(torch_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - target_link_libraries(torch_hip PUBLIC c10_hip) - - if(NOT INTERN_BUILD_MOBILE) -@@ -1908,6 +1909,7 @@ if(BUILD_TEST) - target_include_directories(${test_name} PRIVATE $<INSTALL_INTERFACE:include>) - target_include_directories(${test_name} PRIVATE ${Caffe2_CPU_INCLUDE} ${Caffe2_HIP_INCLUDE}) - target_compile_options(${test_name} PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(${test_name} PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - add_test(NAME ${test_name} COMMAND $<TARGET_FILE:${test_name}>) - if(INSTALL_TEST) - install(TARGETS ${test_name} DESTINATION test) -@@ -2092,6 +2094,7 @@ if(BUILD_PYTHON) - endif() - if(NOT MSVC) - target_compile_options(caffe2_pybind11_state_hip PRIVATE ${HIP_CXX_FLAGS} -fvisibility=hidden) -+ set_target_properties(caffe2_pybind11_state_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - endif() - set_target_properties(caffe2_pybind11_state_hip PROPERTIES PREFIX "") - set_target_properties(caffe2_pybind11_state_hip PROPERTIES SUFFIX ${PY_EXT_SUFFIX}) ---- a/cmake/Dependencies.cmake -+++ b/cmake/Dependencies.cmake -@@ -1278,7 +1278,6 @@ - list(APPEND HIP_CXX_FLAGS -Wno-duplicate-decl-specifier) - list(APPEND HIP_CXX_FLAGS -DCAFFE2_USE_MIOPEN) - list(APPEND HIP_CXX_FLAGS -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_HIP) -- list(APPEND HIP_CXX_FLAGS -std=c++17) - if(ROCM_VERSION_DEV VERSION_GREATER_EQUAL "6.0.0") - list(APPEND HIP_CXX_FLAGS -DHIPBLAS_V2) - endif() ---- a/cmake/public/utils.cmake -+++ b/cmake/public/utils.cmake -@@ -335,6 +335,7 @@ function(caffe2_hip_binary_target target_name_or_src) - caffe2_binary_target(${target_name_or_src}) - - target_compile_options(${__target} PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(${__target} PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - target_include_directories(${__target} PRIVATE ${Caffe2_HIP_INCLUDE}) - endfunction() - ---- a/modules/detectron/CMakeLists.txt -+++ b/modules/detectron/CMakeLists.txt -@@ -31,6 +31,7 @@ if(BUILD_CAFFE2_OPS) - ${Detectron_CPU_SRCS} - ${Detectron_HIP_SRCS}) - target_compile_options(caffe2_detectron_ops_hip PRIVATE ${HIP_CXX_FLAGS}) -+ set_target_properties(caffe2_detectron_ops_hip PROPERTIES CXX_STANDARD 17 CXX_EXTENSIONS OFF) - if(USE_MKLDNN) - target_link_libraries(caffe2_detectron_ops_hip PRIVATE caffe2::mkldnn) - endif() |