sci-libs/caffe2: add 2.1.2

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

2 files changed, 220 insertions, 0 deletions

diff --git a/sci-libs/caffe2/Manifest b/sci-libs/caffe2/Manifest
index d51cccdc3c37..2fc8f8663e91 100644
--- a/sci-libs/caffe2/Manifest
+++ b/sci-libs/caffe2/Manifest
@@ -2,3 +2,4 @@ DIST pytorch-1.13.1.tar.gz 108279745 BLAKE2B 75de03b74dfdaf8d8fb5ea743fcc0c1b0e4
 DIST pytorch-2.0.0.tar.gz 111327292 BLAKE2B 6d593a975c0ade714f0b189f7e3c4ff704b9a9a2377b5e441a9cefc202fa22779966d08948e63671912c6ea5a0eee124042155f4f57a654db34e19e42f013cc9 SHA512 4dd76160711c0d87f3026c8b7fa3ed149dd86b8ac0ee9ecea0eaf80d2e6ce8c29368392e77b9466d90b60634087b462b782495997a5d33367cc8ca9fe14c8a14
 DIST pytorch-2.0.1.tar.gz 111335778 BLAKE2B 7a10cc2b2d5e2422aef7e060a0c3a62ca5c7460c6e0b9becade9b98939501975c74ed5a175a653731f43ca824d2c9bd31f41d1f633c2b139779ab23d5331e9ce SHA512 2309a22b3be3ccdb36d8d9781a59a7bdcc2fdb8d95ada205702ec77862480f0cbb12cd5d6b8cd3114d01a6e33b7743d0fe9de93debf37138ca5c14403cdb0c43
 DIST pytorch-2.1.1.tar.gz 116317162 BLAKE2B d9819256cba0b9951aabe95d86fb135e97d8bafa2c010d13162cd9b3373ca75f20d218e31279ace41981f3f76308721c522f9e53745a1ff9e6386fa10634f9ad SHA512 31b36e7732ee086ae7565a3811ab2d1b2869e79057bea7a4ffc4a3c95c544757e656a6d2289ee11fe7508828aca144e4220ef1e9ab1878e075e1259cf6ff9ca4
+DIST pytorch-2.1.2.tar.gz 116316469 BLAKE2B c5a55ee264bc3477d3556ba6376b5591117e992e56e0dd0c9ba93d12526e2727f7840f6f1e0730a38223b6492c9556840c4ebf22ffd220e97225c2abff303747 SHA512 a8961d78ad785b13c959a0612563a60e0de17a7c8bb9822ddea9a24072796354d07e81c47b6cc8761b21a6448845b088cf80e1661d9e889b0ed5474d3dc76756
diff --git a/sci-libs/caffe2/caffe2-2.1.2.ebuild b/sci-libs/caffe2/caffe2-2.1.2.ebuild
new file mode 100644
index 000000000000..be2bdb80c02f
--- /dev/null
+++ b/sci-libs/caffe2/caffe2-2.1.2.ebuild
@@ -0,0 +1,219 @@
+# 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 prefix
+
+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 fbgemm ffmpeg gloo mpi nnpack +numpy opencl opencv openmp qnnpack tensorpipe xnnpack"
+RESTRICT="test"
+REQUIRED_USE="
+	${PYTHON_REQUIRED_USE}
+	ffmpeg? ( opencv )
+	mpi? ( distributed )
+	tensorpipe? ( distributed )
+	distributed? ( tensorpipe )
+	gloo? ( 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-0.9.2:0/8
+		dev-util/nvidia-cuda-toolkit:=[profiler]
+	)
+	fbgemm? ( >=dev-libs/FBGEMM-2023.11.02 )
+	ffmpeg? ( media-video/ffmpeg:= )
+	gloo? ( sci-libs/gloo[cuda?] )
+	mpi? ( virtual/mpi )
+	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[cuda?] )
+	xnnpack? ( >=sci-libs/XNNPACK-2022.12.22 )
+"
+DEPEND="
+	${RDEPEND}
+	dev-cpp/eigen
+	cuda? ( >=dev-libs/cutlass-3.1.0 )
+	dev-libs/psimd
+	dev-libs/FP16
+	dev-libs/FXdiv
+	dev-libs/pocketfft
+	dev-libs/flatbuffers
+	>=sci-libs/kineto-0.4.0_p20231031
+	$(python_gen_cond_dep '
+		dev-python/pyyaml[${PYTHON_USEDEP}]
+		dev-python/pybind11[${PYTHON_USEDEP}]
+	')
+"
+
+S="${WORKDIR}"/${MYP}
+
+PATCHES=(
+	"${FILESDIR}"/${PN}-2.1.1-gentoo.patch
+	"${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}-2.1.1-cudaExtra.patch
+)
+
+src_prepare() {
+	filter-lto #bug 862672
+	sed -i \
+		-e "/third_party\/gloo/d" \
+		cmake/Dependencies.cmake \
+		|| die
+	cmake_src_prepare
+	pushd torch/csrc/jit/serialization || die
+	flatc --cpp --gen-mutable --scoped-enums mobile_bytecode.fbs || die
+	popd
+	# prefixify the hardcoded paths, after all patches are applied
+	hprefixify \
+		aten/CMakeLists.txt \
+		caffe2/CMakeLists.txt \
+		cmake/Metal.cmake \
+		cmake/Modules/*.cmake \
+		cmake/Modules_CUDA_fix/FindCUDNN.cmake \
+		cmake/Modules_CUDA_fix/upstream/FindCUDA/make2cmake.cmake \
+		cmake/Modules_CUDA_fix/upstream/FindPackageHandleStandardArgs.cmake \
+		cmake/public/LoadHIP.cmake \
+		cmake/public/cuda.cmake \
+		cmake/Dependencies.cmake \
+		torch/CMakeLists.txt \
+		CMakeLists.txt
+}
+
+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=$(usex cuda)
+		-DTORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5 7.0}"
+		-DBUILD_NVFUSER=$(usex cuda)
+		-DUSE_DISTRIBUTED=$(usex distributed)
+		-DUSE_MPI=$(usex mpi)
+		-DUSE_FAKELOWP=OFF
+		-DUSE_FBGEMM=$(usex fbgemm)
+		-DUSE_FFMPEG=$(usex ffmpeg)
+		-DUSE_GFLAGS=ON
+		-DUSE_GLOG=ON
+		-DUSE_GLOO=$(usex gloo)
+		-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/$(get_libdir)
+	)
+
+	if use cuda; then
+		addpredict "/dev/nvidiactl" # bug 867706
+		addpredict "/dev/char"
+
+		mycmakeargs+=(
+			-DCMAKE_CUDA_FLAGS="$(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
+	if use cuda; then
+		mv "${ED}${S}"/nvfuser python/nvfuser || die
+		mv "${ED}"/usr/$(get_libdir)/nvfuser.so python/nvfuser/_C.so || die
+	fi
+	rm -rf "${ED}${S}"/test
+	rm -rf "${ED}${S}"/third_party
+	cp torch/version.py python/torch/ || die
+	python_domodule python/caffe2
+	python_domodule python/torch
+	if use cuda; then
+		python_domodule python/nvfuser
+	fi
+	find "${ED}" -empty -delete
+}