Install this version:
emerge -a =dev-python/rmm-26.10.0
If this version is masked, you can unmask it using the autounmask tool or standard emerge options:
autounmask =dev-python/rmm-26.10.0
Or alternatively:
emerge --autounmask-write -a =dev-python/rmm-26.10.0
| Version | EAPI | Keywords | Slot |
|---|---|---|---|
| 26.10.0 | 8 | ~amd64 | 0 |
# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_USE_PEP517=scikit-build-core
DISTUTILS_EXT=1
PYTHON_COMPAT=( python3_{12..14} )
inherit distutils-r1
# GitHub and rmm/VERSION use zero-padded 26.10.00; PyPI normalizes to ${PV}.
MY_PV="26.10.00"
DESCRIPTION="RAPIDS Memory Manager — Python (Cython) bindings"
HOMEPAGE="
https://github.com/rapidsai/rmm
https://pypi.org/project/rmm/
"
SRC_URI="
https://github.com/rapidsai/rmm/archive/refs/tags/v${MY_PV}.tar.gz
-> rmm-${MY_PV}.gh.tar.gz
https://github.com/rapidsai/rapids-cmake/archive/refs/tags/v${MY_PV}.tar.gz
-> rapids-cmake-${MY_PV}.gh.tar.gz
"
S="${WORKDIR}/rmm-${MY_PV}/python/rmm"
LICENSE="Apache-2.0"
SLOT="0"
KEYWORDS="~amd64"
# Configure fetches CCCL through CPM, at the commit and SHA256 the pinned
# rapids-cmake lists. Installed librmm supplies rmm-config.cmake through its
# cmake.prefix entry point.
RESTRICT="network-sandbox test"
RDEPEND="
>=dev-python/cuda-bindings-13.0.1[${PYTHON_USEDEP}]
<dev-python/cuda-bindings-14[${PYTHON_USEDEP}]
~dev-python/librmm-26.10.0[${PYTHON_USEDEP}]
>=dev-python/numpy-2.0[${PYTHON_USEDEP}]
<dev-python/numpy-3[${PYTHON_USEDEP}]
"
DEPEND="
${RDEPEND}
dev-util/nvidia-cuda-toolkit:=
"
# Cython cimports cuda-bindings .pxd headers; without the build dependency,
# symbols degrade to Python objects that fail nogil blocks. # verified 2026-08-06
BDEPEND="
>=dev-build/cmake-4
dev-build/ninja
>=dev-python/cuda-bindings-13.0.1[${PYTHON_USEDEP}]
<dev-python/cuda-bindings-14[${PYTHON_USEDEP}]
>=dev-python/cython-3.2.2[${PYTHON_USEDEP}]
~dev-python/librmm-26.10.0[${PYTHON_USEDEP}]
>=dev-python/scikit-build-core-0.11.0[${PYTHON_USEDEP}]
"
python_prepare_all() {
# Use the wrapped scikit-build-core backend directly, as in librmm/dask-cuda.
sed -i \
-e 's/build-backend = "rapids_build_backend.build"/build-backend = "scikit_build_core.build"/' \
-e '/"rapids-build-backend>=0.4.0,<0.5.0",/d' \
pyproject.toml || die
# Silence CCCL's deprecated stream_ref header under RMM's -Werror build.
# verified 2026-06-10
sed -i \
-e '/^ LANGUAGES CXX)/a add_compile_definitions(CCCL_IGNORE_DEPRECATED_STREAM_REF_HEADER)' \
CMakeLists.txt || die
distutils-r1_python_prepare_all
}
python_compile() {
# rmm checks rapids-cmake out from its moving release/X.Y branch; pin the
# matching release tag instead. # verified 2026-10-08
local -x CMAKE_ARGS="${CMAKE_ARGS}
-DFETCHCONTENT_SOURCE_DIR_RAPIDS-CMAKE=${WORKDIR}/rapids-cmake-${MY_PV}
"
distutils-r1_python_compile
}
>=dev-python/cuda-bindings-13.0.1[] <dev-python/cuda-bindings-14[] ~dev-python/librmm-26.10.0[] >=dev-python/numpy-2.0[] <dev-python/numpy-3[] dev-util/nvidia-cuda-toolkit:=
>=dev-python/cuda-bindings-13.0.1[] <dev-python/cuda-bindings-14[] ~dev-python/librmm-26.10.0[] >=dev-python/numpy-2.0[] <dev-python/numpy-3[]
>=dev-build/cmake-4 dev-build/ninja >=dev-python/cuda-bindings-13.0.1[] <dev-python/cuda-bindings-14[] >=dev-python/cython-3.2.2[] ~dev-python/librmm-26.10.0[] >=dev-python/scikit-build-core-0.11.0[]
| Type | File | Size | Source URLs |
|---|---|---|---|
| DIST | rapids-cmake-26.10.00.gh.tar.gz | 240055 bytes | https://github.com/rapidsai/rapids-cmake/archive/refs/tags/v26.10.00.tar.gz |
| DIST | rmm-26.10.00.gh.tar.gz | 380209 bytes | https://github.com/rapidsai/rmm/archive/refs/tags/v26.10.00.tar.gz |