sci-libs/caffe2: add 2.0.0

Signed-off-by: Alfredo Tupone <tupone@gentoo.org>

3 files changed, 374 insertions, 0 deletions

diff --git a/sci-libs/caffe2/Manifest b/sci-libs/caffe2/Manifest
index 616b75a49763..1424256938e9 100644
--- a/sci-libs/caffe2/Manifest
+++ b/sci-libs/caffe2/Manifest
@@ -1 +1,2 @@
 DIST pytorch-1.13.1.tar.gz 108279745 BLAKE2B 75de03b74dfdaf8d8fb5ea743fcc0c1b0e408a714ad4160c487921220a7b1755e5fa6e587e6bbc8c9f34dd75e096d2e6dd69c80d24821835fff6c833314434d3 SHA512 f16f89d027efade11d057245cad5b69a390e88b458398310ae30de2dbff7c8fd7f1165be7b8da7ea989c81ac3f5a66c5cb9050610e441a97c83fb8aa28c0bd62
+DIST pytorch-2.0.0.tar.gz 111327292 BLAKE2B 6d593a975c0ade714f0b189f7e3c4ff704b9a9a2377b5e441a9cefc202fa22779966d08948e63671912c6ea5a0eee124042155f4f57a654db34e19e42f013cc9 SHA512 4dd76160711c0d87f3026c8b7fa3ed149dd86b8ac0ee9ecea0eaf80d2e6ce8c29368392e77b9466d90b60634087b462b782495997a5d33367cc8ca9fe14c8a14
diff --git a/sci-libs/caffe2/caffe2-2.0.0.ebuild b/sci-libs/caffe2/caffe2-2.0.0.ebuild
new file mode 100644
index 000000000000..d06eaa5b375f
--- /dev/null
+++ b/sci-libs/caffe2/caffe2-2.0.0.ebuild
@@ -0,0 +1,184 @@
+# Copyright 2022-2023 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{9..11} )
+inherit python-single-r1 cmake cuda flag-o-matic
+
+MYPN=pytorch
+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"
+
+LICENSE="BSD"
+SLOT="0"
+KEYWORDS="~amd64"
+IUSE="cuda distributed ffmpeg mpi nnpack +numpy opencl opencv openmp qnnpack tensorpipe xnnpack"
+RESTRICT="test"
+REQUIRED_USE="
+	${PYTHON_REQUIRED_USE}
+	ffmpeg? ( opencv )
+	mpi? ( distributed )
+	tensorpipe? ( distributed )
+" # ?? ( cuda rocm )
+
+# CUDA 12 not supported yet: https://github.com/pytorch/pytorch/issues/91122
+RDEPEND="
+	${PYTHON_DEPS}
+	dev-cpp/gflags:=
+	>=dev-cpp/glog-0.5.0
+	dev-libs/cpuinfo
+	dev-libs/libfmt
+	dev-libs/protobuf:=
+	dev-libs/pthreadpool
+	dev-libs/sleef
+	sci-libs/lapack
+	>=sci-libs/onnx-1.12.0
+	sci-libs/foxi
+	cuda? (
+		=dev-libs/cudnn-8*
+		dev-libs/cudnn-frontend:0/8
+		<dev-util/nvidia-cuda-toolkit-12:=[profiler]
+	)
+	ffmpeg? ( media-video/ffmpeg:= )
+	mpi? ( sys-cluster/openmpi )
+	nnpack? ( sci-libs/NNPACK )
+	numpy? ( $(python_gen_cond_dep '
+		dev-python/numpy[${PYTHON_USEDEP}]
+		') )
+	opencl? ( virtual/opencl )
+	opencv? ( media-libs/opencv:= )
+	qnnpack? ( sci-libs/QNNPACK )
+	tensorpipe? ( sci-libs/tensorpipe )
+	xnnpack? ( sci-libs/XNNPACK )
+"
+DEPEND="
+	${RDEPEND}
+	dev-cpp/eigen
+	cuda? ( dev-libs/cutlass )
+	dev-libs/psimd
+	dev-libs/FP16
+	dev-libs/FXdiv
+	dev-libs/pocketfft
+	dev-libs/flatbuffers
+	sci-libs/kineto
+	$(python_gen_cond_dep '
+		dev-python/pyyaml[${PYTHON_USEDEP}]
+		dev-python/pybind11[${PYTHON_USEDEP}]
+	')
+"
+
+S="${WORKDIR}"/${MYP}
+
+PATCHES=(
+	"${FILESDIR}"/${P}-gentoo.patch
+	"${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
+)
+
+src_prepare() {
+	filter-lto #bug 862672
+	cmake_src_prepare
+	pushd torch/csrc/jit/serialization || die
+	flatc --cpp --gen-mutable --scoped-enums mobile_bytecode.fbs || die
+	popd
+}
+
+src_configure() {
+	if use cuda && [[ -z ${TORCH_CUDA_ARCH_LIST} ]]; then
+		ewarn "WARNING: caffe2 is being built with its default CUDA compute capabilities: 3.5 and 7.0."
+		ewarn "These may not be optimal for your GPU."
+		ewarn ""
+		ewarn "To configure caffe2 with the CUDA compute capability that is optimal for your GPU,"
+		ewarn "set TORCH_CUDA_ARCH_LIST in your make.conf, and re-emerge caffe2."
+		ewarn "For example, to use CUDA capability 7.5 & 3.5, add: TORCH_CUDA_ARCH_LIST=7.5,3.5"
+		ewarn "For a Maxwell model GPU, an example value would be: TORCH_CUDA_ARCH_LIST=Maxwell"
+		ewarn ""
+		ewarn "You can look up your GPU's CUDA compute capability at https://developer.nvidia.com/cuda-gpus"
+		ewarn "or by running /opt/cuda/extras/demo_suite/deviceQuery | grep 'CUDA Capability'"
+	fi
+
+	local mycmakeargs=(
+		-DBUILD_CUSTOM_PROTOBUF=OFF
+		-DBUILD_SHARED_LIBS=ON
+
+		-DUSE_CCACHE=OFF
+		-DUSE_CUDA=$(usex cuda)
+		-DUSE_CUDNN=$(usex cuda)
+		-DUSE_FAST_NVCC=$(usex cuda)
+		-DTORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5 7.0}"
+		-DUSE_DISTRIBUTED=$(usex distributed)
+		-DUSE_MPI=$(usex mpi)
+		-DUSE_FAKELOWP=OFF
+		-DUSE_FBGEMM=OFF # TODO
+		-DUSE_FFMPEG=$(usex ffmpeg)
+		-DUSE_GFLAGS=ON
+		-DUSE_GLOG=ON
+		-DUSE_GLOO=OFF
+		-DUSE_KINETO=OFF # TODO
+		-DUSE_LEVELDB=OFF
+		-DUSE_MAGMA=OFF # TODO: In GURU as sci-libs/magma
+		-DUSE_MKLDNN=OFF
+		-DUSE_NCCL=OFF # TODO: NVIDIA Collective Communication Library
+		-DUSE_NNPACK=$(usex nnpack)
+		-DUSE_QNNPACK=$(usex qnnpack)
+		-DUSE_XNNPACK=$(usex xnnpack)
+		-DUSE_SYSTEM_XNNPACK=$(usex xnnpack)
+		-DUSE_TENSORPIPE=$(usex tensorpipe)
+		-DUSE_PYTORCH_QNNPACK=OFF
+		-DUSE_NUMPY=$(usex numpy)
+		-DUSE_OPENCL=$(usex opencl)
+		-DUSE_OPENCV=$(usex opencv)
+		-DUSE_OPENMP=$(usex openmp)
+		-DUSE_ROCM=OFF # TODO
+		-DUSE_SYSTEM_CPUINFO=ON
+		-DUSE_SYSTEM_PYBIND11=ON
+		-DUSE_UCC=OFF
+		-DUSE_VALGRIND=OFF
+		-DPYBIND11_PYTHON_VERSION="${EPYTHON#python}"
+		-DPYTHON_EXECUTABLE="${PYTHON}"
+		-DUSE_ITT=OFF
+		-DBLAS=Eigen # avoid the use of MKL, if found on the system
+		-DUSE_SYSTEM_EIGEN_INSTALL=ON
+		-DUSE_SYSTEM_PTHREADPOOL=ON
+		-DUSE_SYSTEM_FXDIV=ON
+		-DUSE_SYSTEM_FP16=ON
+		-DUSE_SYSTEM_GLOO=ON
+		-DUSE_SYSTEM_ONNX=ON
+		-DUSE_SYSTEM_SLEEF=ON
+
+		-Wno-dev
+		-DTORCH_INSTALL_LIB_DIR="${EPREFIX}"/usr/$(get_libdir)
+		-DLIBSHM_INSTALL_LIB_SUBDIR="${EPREFIX}"/usr/$(get_libdir)
+	)
+
+	if use cuda; then
+		addpredict "/dev/nvidiactl" # bug 867706
+
+		mycmakeargs+=(
+			-DCMAKE_CUDA_FLAGS="$(cuda_gccdir -f | tr -d \")"
+		)
+	fi
+	cmake_src_configure
+}
+
+src_install() {
+	cmake_src_install
+
+	insinto "/var/lib/${PN}"
+	doins "${BUILD_DIR}"/CMakeCache.txt
+
+	rm -rf python
+	mkdir -p python/torch/include || die
+	mv "${ED}"/usr/lib/python*/site-packages/caffe2 python/ || die
+	mv "${ED}"/usr/include/torch python/torch/include || die
+	cp torch/version.py python/torch/ || die
+	rm -rf "${ED}"/var/tmp || die
+	python_domodule python/caffe2
+	python_domodule python/torch
+}
diff --git a/sci-libs/caffe2/files/caffe2-2.0.0-gentoo.patch b/sci-libs/caffe2/files/caffe2-2.0.0-gentoo.patch
new file mode 100644
index 000000000000..021a06e343e4
--- /dev/null
+++ b/sci-libs/caffe2/files/caffe2-2.0.0-gentoo.patch
@@ -0,0 +1,189 @@
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -1554,7 +1554,6 @@
+       set_target_properties(onnx_proto PROPERTIES CXX_STANDARD 17)
+     endif()
+   endif()
+-  add_subdirectory(${CMAKE_CURRENT_LIST_DIR}/../third_party/foxi EXCLUDE_FROM_ALL)
+ 
+   add_definitions(-DONNX_NAMESPACE=${ONNX_NAMESPACE})
+   if(NOT USE_SYSTEM_ONNX)
+@@ -1819,7 +1818,6 @@
+ #
+ 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`.
+ #
+@@ -1828,9 +1826,7 @@
+ # 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 ${TEMP_BUILD_SHARED_LIBS} CACHE BOOL "Build shared libs" FORCE)
+ 
+ # ---[ Kineto
+--- a/c10/CMakeLists.txt
++++ b/c10/CMakeLists.txt
+@@ -65,7 +65,7 @@
+ if(${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)
+ 
+ find_package(Backtrace)
+ if(Backtrace_FOUND)
+--- a/torch/CMakeLists.txt
++++ b/torch/CMakeLists.txt
+@@ -66,15 +66,9 @@
+     ${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
+     ${TORCH_SRC_DIR}/csrc/api/include
+@@ -87,7 +81,6 @@
+     python::python
+     pybind::pybind11
+     shm
+-    fmt::fmt-header-only
+     ATEN_CPU_FILES_GEN_LIB)
+ 
+ set(TORCH_PYTHON_COMPILE_DEFINITIONS)
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -799,12 +799,11 @@
+ 
+ # ---[ Build flags
+ if(NOT MSVC)
+-  string(APPEND CMAKE_CXX_FLAGS " -O2 -fPIC")
++  string(APPEND CMAKE_CXX_FLAGS " -fPIC")
+   # Eigen fails to build with some versions, so convert this to a warning
+   # Details at http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1459
+   string(APPEND CMAKE_CXX_FLAGS " -Wall")
+   string(APPEND CMAKE_CXX_FLAGS " -Wextra")
+-  append_cxx_flag_if_supported("-Werror=return-type" CMAKE_CXX_FLAGS)
+   append_cxx_flag_if_supported("-Werror=non-virtual-dtor" CMAKE_CXX_FLAGS)
+   append_cxx_flag_if_supported("-Werror=braced-scalar-init" CMAKE_CXX_FLAGS)
+   append_cxx_flag_if_supported("-Werror=range-loop-construct" CMAKE_CXX_FLAGS)
+@@ -912,8 +911,6 @@
+   string(APPEND CMAKE_LINKER_FLAGS_DEBUG " -fno-omit-frame-pointer -O0")
+   append_cxx_flag_if_supported("-fno-math-errno" CMAKE_CXX_FLAGS)
+   append_cxx_flag_if_supported("-fno-trapping-math" CMAKE_CXX_FLAGS)
+-  append_cxx_flag_if_supported("-Werror=format" CMAKE_CXX_FLAGS)
+-  append_cxx_flag_if_supported("-Werror=cast-function-type" CMAKE_CXX_FLAGS)
+ endif()
+ 
+ if(USE_ASAN)
+--- a/cmake/public/utils.cmake
++++ b/cmake/public/utils.cmake
+@@ -507,8 +507,6 @@
+   endif()
+ 
+   # Use -O2 for release builds (-O3 doesn't improve perf, and -Os results in perf regression)
+-  target_compile_options(${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("${CMAKE_BUILD_TYPE}" MATCHES "Debug")
+       set(OPT_FLAG " ")
+     endif()
+--- a/caffe2/CMakeLists.txt
++++ b/caffe2/CMakeLists.txt
+@@ -106,7 +106,7 @@
+ # Note: the folders that are being commented out have not been properly
+ # addressed yet.
+ 
+-if(NOT MSVC AND USE_XNNPACK)
++if(FALSE)
+   if(NOT TARGET fxdiv)
+     set(FXDIV_BUILD_TESTS OFF CACHE BOOL "")
+     set(FXDIV_BUILD_BENCHMARKS OFF CACHE BOOL "")
+@@ -533,7 +533,6 @@
+ if(NOT MSVC)
+   set_source_files_properties(${TORCH_SRC_DIR}/csrc/jit/tensorexpr/llvm_jit.cpp PROPERTIES COMPILE_FLAGS -Wno-noexcept-type)
+   # Force -Werror on several files
+-  set_source_files_properties(${CMAKE_CURRENT_LIST_DIR}/../aten/src/ATen/native/mkldnn/Pooling.cpp PROPERTIES COMPILE_FLAGS "-Werror")
+ endif()
+ # Disable certain warnings for GCC-9.X
+ if(CMAKE_COMPILER_IS_GNUCXX AND (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 9.0.0))
+@@ -811,7 +812,6 @@
+ torch_compile_options(torch_cpu)  # see cmake/public/utils.cmake
+ if(HAS_WERROR_SIGN_COMPARE AND WERROR)
+   # target_compile_options(torch_cpu PRIVATE "-Werror=sign-compare")
+-  set_property(SOURCE ${ATen_CORE_SRCS} ${ATen_CPU_SRCS} APPEND PROPERTY COMPILE_OPTIONS "-Werror=sign-compare")
+ endif()
+ 
+ set_property(SOURCE ${ATen_CORE_SRCS} APPEND
+@@ -975,7 +977,6 @@
+ endif()
+ 
+ if(NOT MSVC AND USE_XNNPACK)
+-  TARGET_LINK_LIBRARIES(torch_cpu PRIVATE fxdiv)
+ endif()
+ 
+ # ==========================================================
+@@ -1095,8 +1098,7 @@
+ target_include_directories(torch_cpu PRIVATE
+   ${TORCH_ROOT}/third_party/miniz-2.1.0)
+ 
+-target_include_directories(torch_cpu PRIVATE
+-  ${TORCH_ROOT}/third_party/kineto/libkineto/include)
++target_include_directories(torch_cpu PRIVATE /usr/include/kineto)
+ 
+ if(USE_KINETO)
+   target_include_directories(torch_cpu PRIVATE
+--- a/cmake/External/nnpack.cmake
++++ b/cmake/External/nnpack.cmake
+@@ -58,7 +58,7 @@
+   set(PTHREADPOOL_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/pthreadpool" CACHE STRING "pthreadpool source directory")
+   set(GOOGLETEST_SOURCE_DIR "${CAFFE2_THIRD_PARTY_ROOT}/googletest" CACHE STRING "Google Test source directory")
+ 
+-  if(NOT TARGET nnpack)
++  if(FALSE)
+     if(NOT USE_SYSTEM_PTHREADPOOL AND USE_INTERNAL_PTHREADPOOL_IMPL)
+       set(NNPACK_CUSTOM_THREADPOOL ON CACHE BOOL "")
+     endif()
+--- a/cmake/Dependencies.cmake
++++ b/cmake/Dependencies.cmake
+@@ -486,7 +486,7 @@ endif()
+ list(APPEND Caffe2_DEPENDENCY_LIBS cpuinfo)
+ 
+ # ---[ QNNPACK
+-if(USE_QNNPACK)
++if(FALSE)
+   set(CAFFE2_THIRD_PARTY_ROOT "${PROJECT_SOURCE_DIR}/third_party")
+ 
+   if(NOT DEFINED QNNPACK_SOURCE_DIR)
+@@ -542,7 +542,7 @@ if(USE_QNNPACK)
+ 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")