commit: ef3519c7e5f5c64612a85b5b341e51fc6b8380e8 Author: Sv. Lockal <lockalsash <AT> gmail <DOT> com> AuthorDate: Mon Feb 16 12:32:05 2026 +0000 Commit: Sam James <sam <AT> gentoo <DOT> org> CommitDate: Fri Mar 6 00:44:11 2026 +0000 URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=ef3519c7
sci-libs/hipsparselt: new package, add 7.2.0 Closes: https://bugs.gentoo.org/968487 Signed-off-by: Sv. Lockal <lockalsash <AT> gmail.com> Part-of: https://github.com/gentoo/gentoo/pull/45826 Signed-off-by: Sam James <sam <AT> gentoo.org> sci-libs/hipsparselt/Manifest | 3 + .../files/hipBLASLt-7.1.0-rocisa-nanobind.patch | 19 +++ sci-libs/hipsparselt/hipsparselt-7.2.0.ebuild | 185 +++++++++++++++++++++ sci-libs/hipsparselt/metadata.xml | 29 ++++ 4 files changed, 236 insertions(+) diff --git a/sci-libs/hipsparselt/Manifest b/sci-libs/hipsparselt/Manifest new file mode 100644 index 000000000000..9b55321634a3 --- /dev/null +++ b/sci-libs/hipsparselt/Manifest @@ -0,0 +1,3 @@ +DIST hipblaslt-7.2.0.tar.gz 429012541 BLAKE2B 40476fddfc021fbed97e755d884229e1f7a79f9f53d615e8f582bfe702a2f71f02fa490aad636b17ac71ed9bf0deb8ab21742ca4599f8c0859fea0ef50c756b4 SHA512 1866113e64b5ad3c5a31531b295413c4d03ffec6adc70b588590d77ee48713792a7d592fcee026193eb3fa6d43f09c573d54c475025b2f16e2b46787be448aaa +DIST hipsparselt-7.2.0.tar.gz 6972883 BLAKE2B 13d2ae6a3b6a100ebfcf6662ce7ef9d2411ed4462399edceb37771abab20da3580bde0acd25572d0d257e82489062842a0439b66d9310dd82dd24a0b1c227a16 SHA512 39009eead7b21f968726339be7df3c7f5e67de683364ed023b080191314e1a030f8f621263d292b6d49f4a44a3299ac3afb4e6cb441cffeaeaebd399fad3f0f6 +DIST origami-7.2.0.tar.gz 52203 BLAKE2B ea6c5ce23102280bd70003e541748d0c7b73c2179484c1eb182e5a3946f4dfe998a87e312d200209299dd1d2dfba42eed390fc31c37d6e9a8b06fecca84f1ff8 SHA512 ed6326f1ef8eec00e11109a392ec72ca2d28b8106b0ebf08f150a76937974b7b33f4e972799400c34a68a528935ea6234366d2d063079fad7bdb8852d13afe26 diff --git a/sci-libs/hipsparselt/files/hipBLASLt-7.1.0-rocisa-nanobind.patch b/sci-libs/hipsparselt/files/hipBLASLt-7.1.0-rocisa-nanobind.patch new file mode 100644 index 000000000000..e08158786a8a --- /dev/null +++ b/sci-libs/hipsparselt/files/hipBLASLt-7.1.0-rocisa-nanobind.patch @@ -0,0 +1,19 @@ +Use system nanobind +--- a/tensilelite/rocisa/CMakeLists.txt 2025-11-09 09:49:11.051195769 -0000 ++++ b/tensilelite/rocisa/CMakeLists.txt 2025-11-09 09:50:56.393579853 -0000 +@@ -17,14 +17,7 @@ + ) + + if(HIPBLASLT_BUNDLE_PYTHON_DEPS) +- include(FetchContent) +- FetchContent_Declare( +- nanobind +- GIT_REPOSITORY https://github.com/wjakob/nanobind.git +- GIT_TAG 9b3afa9dbdc23641daf26fadef7743e7127ff92f # v2.6.1 +- ) +- FetchContent_MakeAvailable(nanobind) +- ++ find_package(nanobind CONFIG REQUIRED) + set(ROCISAINST_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/rocisa/src/instruction/instruction.cpp" + "${CMAKE_CURRENT_SOURCE_DIR}/rocisa/src/instruction/common.cpp" + "${CMAKE_CURRENT_SOURCE_DIR}/rocisa/src/instruction/branch.cpp" diff --git a/sci-libs/hipsparselt/hipsparselt-7.2.0.ebuild b/sci-libs/hipsparselt/hipsparselt-7.2.0.ebuild new file mode 100644 index 000000000000..2557e58a7b6a --- /dev/null +++ b/sci-libs/hipsparselt/hipsparselt-7.2.0.ebuild @@ -0,0 +1,185 @@ +# Copyright 1999-2026 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +ROCM_SKIP_GLOBALS=1 +PYTHON_COMPAT=( python3_{10..14} ) + +LLVM_COMPAT=( 22 ) + +inherit cmake flag-o-matic multiprocessing llvm-r2 python-any-r1 rocm +DESCRIPTION="Sparse GEMM operations library for AMD Instinct accelerators" +HOMEPAGE="https://github.com/ROCm/rocm-libraries/tree/develop/projects/hipsparselt"; +SRC_URI=" + https://github.com/ROCm/rocm-libraries/releases/download/rocm-${PV}/hipsparselt.tar.gz -> hipsparselt-${PV}.tar.gz + https://github.com/ROCm/rocm-libraries/releases/download/rocm-${PV}/origami.tar.gz -> origami-${PV}.tar.gz + https://github.com/ROCm/rocm-libraries/releases/download/rocm-${PV}/hipblaslt.tar.gz -> hipblaslt-${PV}.tar.gz +" + +S="${WORKDIR}/hipsparselt" +ORIGAMI_S="${WORKDIR}/origami" +HIPBLASLT_S="${WORKDIR}/hipblaslt" + +LICENSE="MIT" +SLOT="0/$(ver_cut 1-2)" +KEYWORDS="~amd64" + +# From hipsparselt_supported_architectures.cmake +SUPPORTED_GPUS=( gfx942 gfx950 ) +IUSE_TARGETS=( "${SUPPORTED_GPUS[@]/#/amdgpu_targets_}" ) +IUSE="${IUSE_TARGETS[*]/#/+} benchmark roctracer test" +RESTRICT="!test? ( test )" + +RDEPEND=" + dev-util/hip:${SLOT} + roctracer? ( dev-util/roctracer:${SLOT} ) + benchmark? ( sci-libs/flexiblas ) +" + +DEPEND=" + ${RDEPEND} + dev-cpp/msgpack-cxx + dev-libs/rocm-comgr:${SLOT} + dev-libs/rocr-runtime:${SLOT} + sci-libs/hipSPARSE:${SLOT} + dev-util/rocm-smi:${SLOT} + llvm-runtimes/openmp +" + +BDEPEND=" + ${PYTHON_DEPS} + dev-build/rocm-cmake:${SLOT} + dev-util/hipcc:${SLOT} + $(python_gen_any_dep " + dev-python/msgpack[\${PYTHON_USEDEP}] + dev-python/pyyaml[\${PYTHON_USEDEP}] + dev-python/joblib[\${PYTHON_USEDEP}] + dev-python/nanobind[\${PYTHON_USEDEP}] + dev-python/setuptools[\${PYTHON_USEDEP}] + ") + $(llvm_gen_dep "llvm-core/clang:\${LLVM_SLOT}") + test? ( + dev-cpp/gtest + sci-libs/flexiblas + ) +" + +python_check_deps() { + python_has_version "dev-python/msgpack[${PYTHON_USEDEP}]" && + python_has_version "dev-python/pyyaml[${PYTHON_USEDEP}]" && + python_has_version "dev-python/joblib[${PYTHON_USEDEP}]" && + python_has_version "dev-python/nanobind[${PYTHON_USEDEP}]" && + python_has_version "dev-python/setuptools[${PYTHON_USEDEP}]" +} + +pkg_setup() { + QA_FLAGS_IGNORED="usr/$(get_libdir)/hipsparselt/library/.*" + python-any-r1_pkg_setup +} + +pkg_pretend() { + if [[ "${AMDGPU_TARGETS[*]}" = "" ]]; then + ewarn "hipSPARSELt supports only some GPUs: ${SUPPORTED_GPUS[*]}," + ewarn "but none of them were defined in AMDGPU_TARGETS USE_EXPAND variable." + ewarn + ewarn "Library will continue to be built in \"dummy\" mode," + ewarn "serving as a non-functional placeholder for end-user applications." + fi +} + +src_prepare() { + rocm_use_clang + + # too many warnings + append-cxxflags -Wno-explicit-specialization-storage-class + + # Do not install tests + sed -e 's/COMPONENT "tests"/COMPONENT "tests" EXCLUDE_FROM_ALL/' \ + -i CMakeLists.txt || die + + pushd "${HIPBLASLT_S}" || die + eapply "${FILESDIR}"/hipBLASLt-7.1.0-rocisa-nanobind.patch + + sed -e 's:../../shared/origami:../origami:' -i CMakeLists.txt || die + + local shebangs=($(grep -rl "#!/usr/bin/env python3" tensilelite/Tensile || die)) + python_fix_shebang -q "${shebangs[@]}" + + # Fix compiler validation (just a validation) + sed -e "s/amdclang/$(basename "$CC")/g" \ + -i tensilelite/Tensile/Toolchain/Validators.py || die + sed -e "s:\$(ROCM_PATH)/bin/amdclang++:$(get_llvm_prefix)/bin/clang++:g" \ + -i tensilelite/Makefile || die + popd || die + + cmake_src_prepare +} + +src_configure() { + rocm_use_clang + + # Tensile guesses weirdly how to compile things, ld.bfd won't work, so force lld + append-cxxflags -DCMAKE_CXX_FLAGS="-fuse-ld=lld" + + local targets="$(get_amdgpu_flags)" + local HIPSPARSELT_ENABLE_DEVICE=$([ "${AMDGPU_TARGETS[*]}" != "" ] && echo ON || echo OFF ) + + # targets has a trailing semicolon, this trips up Tensile's input parser, so carefully prune + local mycmakeargs=( + -DGPU_TARGETS="${targets::-1}" + -DHIPSPARSELT_ENABLE_SAMPLES=OFF + -DHIPSPARSELT_ENABLE_DEVICE=${HIPSPARSELT_ENABLE_DEVICE} + -DHIPSPARSELT_BUILD_TESTING="$(usex test ON OFF)" + -DHIPSPARSELT_ENABLE_BENCHMARKS="$(usex benchmark ON OFF)" + + -DHIPSPARSELT_ENABLE_MARKER="$(usex roctracer ON OFF)" + -Dnanobind_DIR="$(python_get_sitedir)/nanobind/cmake" + -DPython_EXECUTABLE="${PYTHON}" + -DROCM_SYMLINK_LIBS=OFF + -Wno-dev + ) + + if [[ "${AMDGPU_TARGETS[*]}" != "" ]]; then + mycmakeargs+=( + -DTENSILELITE_BUILD_PARALLEL_LEVEL=$(makeopts_jobs) + ) + fi + + if use test || use benchmark; then + mycmakeargs+=( + -DHIPSPARSELT_ENABLE_CLIENT=ON + -DBLA_VENDOR=FlexiBLAS + ) + else + mycmakeargs+=( + -DHIPSPARSELT_ENABLE_CLIENT=OFF + ) + fi + + cmake_src_configure +} + +src_compile() { + local -x ROCM_PATH="${EPREFIX}/usr" + # set PYTHONPATH to load Tensile from virtualenv, not the system-wide one + local -x PYTHONPATH="${S}_build/virtualenv/lib/${EPYTHON}/site-packages" + local -x TENSILE_ROCM_ASSEMBLER_PATH="$(get_llvm_prefix)/bin/clang++" + # TensileCreateLibrary reads CMAKE_CXX_COMPILER again + local -x CMAKE_CXX_COMPILER="$(get_llvm_prefix)/bin/clang++" + cmake_src_compile +} + +src_install() { + cmake_src_install + + # Stop llvm-strip from removing .strtab section from *.hsaco files, + # otherwise rocclr/elf/elf.cpp complains with "failed: null sections(STRTAB)" and crashes + dostrip -x /usr/$(get_libdir)/hipsparselt/library/ +} + +src_test() { + check_amdgpu + # Non-instinct GPUs: tests just succeed without testing anything + HIP_VISIBLE_DEVICES=0 cmake_src_test +} diff --git a/sci-libs/hipsparselt/metadata.xml b/sci-libs/hipsparselt/metadata.xml new file mode 100644 index 000000000000..99d611403637 --- /dev/null +++ b/sci-libs/hipsparselt/metadata.xml @@ -0,0 +1,29 @@ +<?xml version="1.0" encoding="UTF-8"?> +<!DOCTYPE pkgmetadata SYSTEM "https://www.gentoo.org/dtd/metadata.dtd";> +<pkgmetadata> + <maintainer type="person" proxied="yes"> + <email>[email protected]</email> + <name>Sv. Lockal</name> + <description>Primary maintainer</description> + </maintainer> + <maintainer type="person" proxied="yes"> + <email>[email protected]</email> + <name>Yiyang Wu</name> + </maintainer> + <maintainer type="person" proxied="yes"> + <email>[email protected]</email> + <name>Paul Zander</name> + <description>General GPGPU stuff</description> + </maintainer> + <maintainer type="project" proxied="proxy"> + <email>[email protected]</email> + <name>Gentoo Science Project</name> + </maintainer> + <upstream> + <remote-id type="github">ROCm/hipBLASLt</remote-id> + </upstream> + <use> + <flag name="benchmark">Build and install rocblas-bench</flag> + <flag name="roctracer">Enable roctracer (roctx) marker</flag> + </use> +</pkgmetadata>
