View Raw Ebuild
# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
PYTHON_COMPAT=( python3_{9..14} )
ROCM_VERSION=5.1.3
DISTUTILS_EXT=1
DISTUTILS_USE_PEP517=setuptools
inherit distutils-r1 prefix pypi rocm cuda
DESCRIPTION="CuPy: A NumPy-compatible array library accelerated by CUDA"
HOMEPAGE="https://cupy.dev/"
SRC_URI="$(pypi_sdist_url "${PN^}" "${PV}")"
LICENSE="MIT"
SLOT="0"
KEYWORDS=""
IUSE="rocm +cuda cudnn"
REQUIRED_USE="
^^ ( cuda rocm )
cudnn? ( cuda )
rocm? ( ${ROCM_REQUIRED_USE} )
"
DEPEND="
>=dev-python/cython-3.1.0[${PYTHON_USEDEP}]
>=dev-python/numpy-1.18.0[${PYTHON_USEDEP}]
cuda? ( dev-util/nvidia-cuda-toolkit[profiler]
dev-libs/cudnn )
cudnn? ( dev-libs/cudnn )
rocm? ( >=dev-util/hip-${ROCM_VERSION}
>=dev-util/roctracer-${ROCM_VERSION}
>=sci-libs/hipBLAS-${ROCM_VERSION}[${ROCM_USEDEP}]
>=sci-libs/hipCUB-${ROCM_VERSION}[${ROCM_USEDEP}]
>=sci-libs/hipFFT-${ROCM_VERSION}[${ROCM_USEDEP}]
>=sci-libs/hipRAND-${ROCM_VERSION}[${ROCM_USEDEP}]
>=sci-libs/rocThrust-${ROCM_VERSION}[${ROCM_USEDEP}]
>=sci-libs/hipSPARSE-${ROCM_VERSION}[${ROCM_USEDEP}] )
"
#dev-libs/cusparselt
RDEPEND=">=dev-python/fastrlock-0.8.1
${DEPEND}"
distutils_enable_tests pytest
#PATCHES=( "${FILESDIR}"/${PN}-11.6-add_dir.patch )
src_prepare ()
{
#sed -i -e 's:_from_dict(CUDA_nccl,:#_from_dict(CUDA_nccl,:' install/cupy_builder/_features.py || die
#sed -i -e 's:_from_dict(CUDA_cutensor,:#_from_dict(CUDA_cutensor,:' install/cupy_builder/_features.py || die
#sed -i -e 's:_from_dict(CUDA_cusparselt:#_from_dict(CUDA_cusparselt:' install/cupy_builder/_features.py || die
#sed -i -e 's:cuda/cupy_cutensor.h:#stub/cupy_cutensor.h:' cupy_backends/cupy_cutensor.h || die
default
eprefixify cupy/cuda/compiler.py
use cuda && cuda_src_prepare
}
src_compile() {
if use rocm; then
addpredict /dev/kfd
addpredict /dev/dri/
export CUPY_INSTALL_USE_HIP=1
export ROCM_HOME="${EPREFIX}/usr"
local AMDGPU_FLAGS=$(get_amdgpu_flags)
export HCC_AMDGPU_TARGET=${AMDGPU_FLAGS//;/,}
elif use cuda; then
local supported=() codegen=()
local s target ok
while IFS= read -r s; do
supported+=( "${s}" )
done < <(nvcc --list-gpu-code) || die
for target in ${NVPTX_TARGETS}; do
ok=
for s in "${supported[@]}"; do
[[ ${s} == "${target}" ]] && ok=1 && break
done
if [[ -n ${ok} ]]; then
codegen+=( "arch=${target/sm_/compute_},code=${target}" )
else
ewarn "Skipping unsupported CUDA target ${target} for toolkit $(cuda_toolkit_version)"
fi
done
[[ ${#codegen[@]} -gt 0 ]] || die "No supported NVPTX_TARGETS selected"
local IFS=';'
export CUPY_NVCC_GENERATE_CODE="${codegen[*]}"
export NVCC="nvcc ${NVCCFLAGS}"
fi
distutils-r1_src_compile
}
##compare with caffee2