# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_USE_PEP517=standalone
PYTHON_COMPAT=( python3_{12..14} )
DISTUTILS_SINGLE_IMPL=1
inherit distutils-r1 pypi
DESCRIPTION="FlashInfer: kernel library for LLM serving (Python frontend)"
HOMEPAGE="
https://github.com/flashinfer-ai/flashinfer
https://pypi.org/project/flashinfer-python/
"
LICENSE="Apache-2.0"
SLOT="0"
KEYWORDS="~amd64"
# Pure-Python wheel install at packaging time; the C++ csrc/include
# and vendored cutlass/spdlog ride along as data files for runtime
# JIT compilation. No nvcc invocation here, so the gcc-15 host-pin
# isn't needed at install-time. (At first JIT invocation by a user,
# nvcc will be exercised — that path remains the user's responsibility
# to align with the gcc-15 / nvcc compatibility window.)
#
# 0.6.15 added cuda-python and nccl4py as base deps for the moe_ep EP
# transport backend. cuda-python is imported broadly (comm, gemm kernels,
# JIT utils, green_ctx) so it's declared here; nccl4py is imported only by
# moe_ep/nccl_ep and isn't packaged, so that CUDA-13 EP-transport path
# stays unsupported — moe_ep/_validators.py raises a clear runtime error
# if it's selected, and the core paths vllm exercises don't need it.
# Upstream floors cuda-python at >=12.0 (a >=13.0 floor would evict CUDA-12
# torch stacks); our overlay only ships 13.3.x, which satisfies it.
# verified 2026-07-19
RDEPEND="
sci-ml/pytorch[${PYTHON_SINGLE_USEDEP}]
app-alternatives/ninja
$(python_gen_cond_dep '
>=dev-python/apache-tvm-ffi-0.1.6[${PYTHON_USEDEP}]
<dev-python/apache-tvm-ffi-0.2[${PYTHON_USEDEP}]
dev-python/click[${PYTHON_USEDEP}]
>=dev-python/cuda-python-12.0[${PYTHON_USEDEP}]
dev-python/cuda-tile-bin[${PYTHON_USEDEP}]
dev-python/einops[${PYTHON_USEDEP}]
dev-python/numpy[${PYTHON_USEDEP}]
>=dev-python/nvidia-cudnn-frontend-1.13.0[${PYTHON_USEDEP}]
>=dev-python/nvidia-cutlass-dsl-4.5.0[${PYTHON_USEDEP}]
dev-python/nvidia-ml-py[${PYTHON_USEDEP}]
>=dev-python/packaging-24.2[${PYTHON_USEDEP}]
dev-python/requests[${PYTHON_USEDEP}]
dev-python/tabulate[${PYTHON_USEDEP}]
dev-python/tqdm[${PYTHON_USEDEP}]
~dev-python/flashinfer-cubin-'${PV}'[${PYTHON_USEDEP}]
')
"
DEPEND="${RDEPEND}"
BDEPEND="
$(python_gen_cond_dep '
>=dev-python/setuptools-77[${PYTHON_USEDEP}]
>=dev-python/packaging-24[${PYTHON_USEDEP}]
>=dev-python/apache-tvm-ffi-0.1.6[${PYTHON_USEDEP}]
')
"
python_install_all() {
distutils-r1_python_install_all
# Upstream's [tool.setuptools] py-modules = ["build_backend",
# "build_utils"] would leak both PEP-517 backend wrappers into
# top-level site-packages, polluting the global namespace and
# pulling setuptools into runtime. Same shape as the
# dev-python/torch-c-dlpack-ext fix; drop both in install_all.
# verified 2026-05-07 against 0.6.8.post1.
rm -f "${ED}"/usr/lib/python*/site-packages/build_backend.py || die
rm -f "${ED}"/usr/lib/python*/site-packages/build_utils.py || die
rm -rf "${ED}"/usr/lib/python*/site-packages/__pycache__/build_backend.* || die
rm -rf "${ED}"/usr/lib/python*/site-packages/__pycache__/build_utils.* || die
}