Install this version:
emerge -a =dev-python/nvidia-cutlass-dsl-4.6.0
If this version is masked, you can unmask it using the autounmask tool or standard emerge options:
autounmask =dev-python/nvidia-cutlass-dsl-4.6.0
Or alternatively:
emerge --autounmask-write -a =dev-python/nvidia-cutlass-dsl-4.6.0
| Version | EAPI | Keywords | Slot |
|---|---|---|---|
| 4.6.0 | 8 | ~amd64 | 0 |
# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_USE_PEP517=no
PYTHON_COMPAT=( python3_{12..14} )
inherit distutils-r1
MY_WHEEL="${PN//-/_}-${PV}-py3-none-any.whl"
DESCRIPTION="NVIDIA CUTLASS Python DSL — metapackage for libs-base + libs-cu13"
HOMEPAGE="
https://github.com/NVIDIA/cutlass
https://docs.nvidia.com/cutlass/
https://pypi.org/project/nvidia-cutlass-dsl/
"
SRC_URI="
https://files.pythonhosted.org/packages/8b/1c/fbddb760a0228df87a9e9d1e60b76ecbe6e18035f5853efe0b4563651b2b/${MY_WHEEL}
"
S="${WORKDIR}"
LICENSE="NVIDIA-CUDA"
SLOT="0"
KEYWORDS="~amd64"
RESTRICT="bindist mirror"
# Trivial metapackage: empty py3-none-any wheel that just pulls libs-base
# and (via the cu13 extra) libs-cu13. No PyPI source release — cutlass-dsl
# ships a PyPI-only metawheel from NVIDIA/cutlass — so packaging the wheel
# directly is byte-equivalent to an empty source.
#
# Always the cu13 path (amd64 + CUDA 13.2 at /opt/cuda); cu12 would need a
# libs-cu12 sibling we haven't packaged. Add a USE flag if a cu12 user emerges.
RDEPEND="
~dev-python/nvidia-cutlass-dsl-libs-base-${PV}[${PYTHON_USEDEP}]
~dev-python/nvidia-cutlass-dsl-libs-cu13-${PV}[${PYTHON_USEDEP}]
"
src_unpack() {
mkdir -p "${S}/wheel" || die
cp "${DISTDIR}/${MY_WHEEL}" "${S}/wheel/" || die
}
src_install() {
python_foreach_impl install_wheel
}
install_wheel() {
${EPYTHON} -m installer --destdir="${D}" "${S}/wheel/${MY_WHEEL}" || die
}
~dev-python/nvidia-cutlass-dsl-libs-base-4.6.0[] ~dev-python/nvidia-cutlass-dsl-libs-cu13-4.6.0[]