Install this version:
emerge -a =dev-python/spconv-cu126-2.3.8
If this version is masked, you can unmask it using the autounmask tool or standard emerge options:
autounmask =dev-python/spconv-cu126-2.3.8
Or alternatively:
emerge --autounmask-write -a =dev-python/spconv-cu126-2.3.8
| Version | EAPI | Keywords | Slot |
|---|---|---|---|
| 2.3.8 | 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
DISTUTILS_SINGLE_IMPL=1
PYTHON_COMPAT=( python3_13 )
inherit distutils-r1
MY_PN="${PN/-/_}"
DESCRIPTION="Spatially sparse convolution library (prebuilt cu126 wheel)"
HOMEPAGE="
https://github.com/traveller59/spconv
https://pypi.org/project/spconv-cu126/
"
# Prebuilt manylinux CUDA-12.6 wheel. spconv is the sparse-conv backend that
# matches the TRELLIS pretrained weights (they are spconv-format; torchsparse
# uses incompatible weight naming/layout). Verified to load TRELLIS and run
# image->3D on a CUDA-13.3 / torch-2.11 host. spconv 2.x bundles its own CUDA
# kernels and does not link libtorch, so it is torch-version-independent.
SRC_URI="
https://files.pythonhosted.org/packages/af/fd/c52d71468849d09b333123f9d0cac27b4a3815a8faecff21ebd66d9c7b45/${MY_PN}-${PV}-cp313-cp313-manylinux_2_28_x86_64.whl
"
S="${WORKDIR}"
LICENSE="Apache-2.0"
SLOT="0"
KEYWORDS="-* ~amd64"
RESTRICT="bindist mirror strip"
QA_PREBUILT="usr/lib/python3.*/site-packages/spconv/*"
# Upstream also lists 'fire'; it is only used by spconv's CLI tools, not when
# importing the prebuilt ops at runtime. verified 2026-06-17
RDEPEND="
~dev-python/cumm-cu126-0.7.11[${PYTHON_SINGLE_USEDEP}]
$(python_gen_cond_dep '
dev-python/pccm[${PYTHON_USEDEP}]
dev-python/ccimport[${PYTHON_USEDEP}]
dev-python/pybind11[${PYTHON_USEDEP}]
dev-python/numpy[${PYTHON_USEDEP}]
')
"
src_unpack() {
cp "${DISTDIR}/${A}" "${WORKDIR}/" || die
}
src_compile() { :; }
src_install() {
python_setup
"${EPYTHON}" -m installer --destdir="${D}" "${WORKDIR}/${A}" || die
python_optimize
}
~dev-python/cumm-cu126-0.7.11[${PYTHON_SINGLE_USEDEP}] $(python_gen_cond_dep ' dev-python/pccm[${PYTHON_USEDEP}] dev-python/ccimport[${PYTHON_USEDEP}] dev-python/pybind11[${PYTHON_USEDEP}] dev-python/numpy[${PYTHON_USEDEP}] ')