diff options
author | Sv. Lockal <lockalsash@gmail.com> | 2024-03-19 18:49:11 +0000 |
---|---|---|
committer | Sam James <sam@gentoo.org> | 2024-06-26 10:19:20 +0100 |
commit | e41b4bb57f3cd8010b6cd0e52fe0fae8e646901b (patch) | |
tree | 121d64d2def274f2edd779b1bedd055e4ec46782 /sci-libs | |
parent | sci-libs/miopen: add 6.1.1 (diff) | |
download | gentoo-e41b4bb57f3cd8010b6cd0e52fe0fae8e646901b.tar.gz gentoo-e41b4bb57f3cd8010b6cd0e52fe0fae8e646901b.tar.bz2 gentoo-e41b4bb57f3cd8010b6cd0e52fe0fae8e646901b.zip |
sci-libs/rocBLAS: add 6.1.1
Changes:
* in updated expand-isa-compatibility patch do not coerce gfx1011 and gfx1012 to gfx1010, as Gentoo users can build rocBLAS for gfx1011 and gfx1012 with USE flags
* add myself to maintainers
Signed-off-by: Sv. Lockal <lockalsash@gmail.com>
Signed-off-by: Sam James <sam@gentoo.org>
Diffstat (limited to 'sci-libs')
-rw-r--r-- | sci-libs/rocBLAS/Manifest | 1 | ||||
-rw-r--r-- | sci-libs/rocBLAS/files/rocBLAS-6.0.2-expand-isa-compatibility.patch | 96 | ||||
-rw-r--r-- | sci-libs/rocBLAS/metadata.xml | 4 | ||||
-rw-r--r-- | sci-libs/rocBLAS/rocBLAS-6.1.1.ebuild | 111 |
4 files changed, 212 insertions, 0 deletions
diff --git a/sci-libs/rocBLAS/Manifest b/sci-libs/rocBLAS/Manifest index f17d3a47c686..ca306e9c9efb 100644 --- a/sci-libs/rocBLAS/Manifest +++ b/sci-libs/rocBLAS/Manifest @@ -3,3 +3,4 @@ DIST rocBLAS-5.4.2-Tensile-asm_full-navi22.tar.gz 1057686 BLAKE2B aa755595bbcd48 DIST rocm-rocBLAS-5.1.3.tar.gz 13741546 BLAKE2B af10d8cb69bee44ef5aab2dc350e1d9f3b6fddb3a840c17230c4b55b2649a4724c9f2e58c1eafdb3e2716eea016a72a35c3ddb498881b1ce682d780baa8d8b07 SHA512 5ab71838fdd0e9c5848cbf28a19d113353b619a878d8c7d05f64feb32faae2054169c95ed6e9dd6b05a2e807b57229dd2c361c4d289b6e6f17c196558640890f DIST rocm-rocBLAS-5.4.2.tar.gz 15938434 BLAKE2B 14ebafa944fdac443800bb7f9b16f8ecf0f420b168d3c6534f68ad7d14bf058a4cc1673fce8f4b9be53e4a6c1cf05011e01853cd901bce0b59827d2aca4d029f SHA512 e62bb80457c1e89454885499bdce9d60beecd706806724418983c78c65c2ae303550f9670b5a6e71dae6a61c0e42b223ab01b36b8406430731ebcbff54c4a8f4 DIST rocm-rocBLAS-5.7.1.tar.gz 55389700 BLAKE2B 9e58b1d29c8a04aa58ce17fdd9b0c2504934261851619dc5b696a04e8c6a79e7c44d86a6db7a9c76a434d76593de8c191cba6f163cf398d01cfc001e5fa32b5a SHA512 cfc73f6c490595a3c153c2ac2a3cb5926c48c2c19baa07ff3dcdf3a7ccaed82cfdff64ed5a3be64c349cb43d654d4b104d9090e2ec9f2061049d6a4921c07722 +DIST rocm-rocBLAS-6.1.1.tar.gz 100103895 BLAKE2B df89a21e47730714b80c62717a86d65ac0d819c2d33b2a3c6252b60b8d7cc9eb5f8f1db27159b0a8dfc87aba7ccad97e2f832bc0044e92d0b69c526d382c745a SHA512 74e847c4ddee16cae7f41e268a815ea03875519c365abe7dfce313a9545df245baa73301b432f64a36cc988281fec7530629a0d72583e9ea287c98ec8a0b3907 diff --git a/sci-libs/rocBLAS/files/rocBLAS-6.0.2-expand-isa-compatibility.patch b/sci-libs/rocBLAS/files/rocBLAS-6.0.2-expand-isa-compatibility.patch new file mode 100644 index 000000000000..ca895c4b4ff5 --- /dev/null +++ b/sci-libs/rocBLAS/files/rocBLAS-6.0.2-expand-isa-compatibility.patch @@ -0,0 +1,96 @@ +--- a/library/src/handle.cpp ++++ b/library/src/handle.cpp +@@ -21,6 +21,7 @@ + * ************************************************************************ */ + #include "handle.hpp" + #include <cstdarg> ++#include <cstring> + #include <limits> + #ifdef WIN32 + #include <windows.h> +@@ -80,6 +81,20 @@ static Processor getActiveArch(int deviceId) + // strip out xnack/ecc from name + std::string deviceFullString(deviceProperties.gcnArchName); + std::string deviceString = deviceFullString.substr(0, deviceFullString.find(":")); ++ // coerce to compatible arch ++ int gcnArch = deviceString.substr(0, 3) == "gfx" ? std::stoi(deviceString.substr(3)) : 0; ++ switch(gcnArch) ++ { ++ case 902: case 909: case 912: ++ std::strcpy(deviceProperties.gcnArchName, "gfx900"); ++ break; ++ case 1013: ++ std::strcpy(deviceProperties.gcnArchName, "gfx1010"); ++ break; ++ case 1031: case 1032: case 1033: case 1034: case 1035: case 1036: ++ std::strcpy(deviceProperties.gcnArchName, "gfx1030"); ++ break; ++ } + + if(deviceString.find("gfx803") != std::string::npos) + { +--- a/library/src/rocblas_auxiliary.cpp ++++ b/library/src/rocblas_auxiliary.cpp +@@ -24,6 +24,7 @@ + #include "logging.hpp" + #include "rocblas-auxiliary.h" + #include <cctype> ++#include <cstring> + #include <cstdlib> + #include <memory> + #include <string> +@@ -1242,6 +1243,20 @@ std::string rocblas_internal_get_arch_name() + hipGetDevice(&deviceId); + hipDeviceProp_t deviceProperties; + hipGetDeviceProperties(&deviceProperties, deviceId); ++ // coerce to compatible arch ++ int gcnArch = strncmp(deviceProperties.gcnArchName, "gfx", 3) == 0 ? std::stoi(deviceProperties.gcnArchName + 3) : 0; ++ switch(gcnArch) ++ { ++ case 902: case 909: case 912: ++ std::strcpy(deviceProperties.gcnArchName, "gfx900"); ++ break; ++ case 1013: ++ std::strcpy(deviceProperties.gcnArchName, "gfx1010"); ++ break; ++ case 1031: case 1032: case 1033: case 1034: case 1035: case 1036: ++ std::strcpy(deviceProperties.gcnArchName, "gfx1030"); ++ break; ++ } + return ArchName<hipDeviceProp_t>{}(deviceProperties); + } + +--- a/library/src/tensile_host.cpp ++++ b/library/src/tensile_host.cpp +@@ -45,6 +45,7 @@ extern "C" void rocblas_shutdown(); + #include <Tensile/hip/HipUtils.hpp> + #include <atomic> + #include <complex> ++#include <cstring> + #include <exception> + #include <future> + #include <iomanip> +@@ -802,6 +803,23 @@ namespace + std::string deviceFullString(prop.gcnArchName); + std::string deviceString + = deviceFullString.substr(0, deviceFullString.find(":")); ++ // coerce to compatible arch ++ int gcnArch = deviceString.substr(0, 3) == "gfx" ? std::stoi(deviceString.substr(3)) : 0; ++ switch(gcnArch) ++ { ++ case 902: case 909: case 912: ++ std::strcpy(prop.gcnArchName, "gfx900"); ++ deviceString = prop.gcnArchName; ++ break; ++ case 1013: ++ std::strcpy(prop.gcnArchName, "gfx1010"); ++ deviceString = prop.gcnArchName; ++ break; ++ case 1031: case 1032: case 1033: case 1034: case 1035: case 1036: ++ std::strcpy(prop.gcnArchName, "gfx1030"); ++ deviceString = prop.gcnArchName; ++ break; ++ } + m_devicePropMap[deviceString] = std::make_shared<hipDeviceProp_t>(prop); + } + } diff --git a/sci-libs/rocBLAS/metadata.xml b/sci-libs/rocBLAS/metadata.xml index 066f5c749e3f..0e8f76b2a76e 100644 --- a/sci-libs/rocBLAS/metadata.xml +++ b/sci-libs/rocBLAS/metadata.xml @@ -12,6 +12,10 @@ <email>xgreenlandforwyy@gmail.com</email> <name>Yiyang Wu</name> </maintainer> + <maintainer type="person"> + <email>lockalsash@gmail.com</email> + <name>Sv. Lockal</name> + </maintainer> <use> <flag name="test">Perform rocblas-test to compare the result between rocBLAS and system BLAS.</flag> <flag name="benchmark">Build and install rocblas-bench.</flag> diff --git a/sci-libs/rocBLAS/rocBLAS-6.1.1.ebuild b/sci-libs/rocBLAS/rocBLAS-6.1.1.ebuild new file mode 100644 index 000000000000..fed17e82b14e --- /dev/null +++ b/sci-libs/rocBLAS/rocBLAS-6.1.1.ebuild @@ -0,0 +1,111 @@ +# Copyright 1999-2024 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +DOCS_BUILDER="doxygen" +DOCS_DIR="docs/.doxygen" +DOCS_DEPEND="media-gfx/graphviz" +ROCM_VERSION=${PV} +inherit cmake docs edo multiprocessing rocm + +DESCRIPTION="AMD's library for BLAS on ROCm" +HOMEPAGE="https://github.com/ROCmSoftwarePlatform/rocBLAS" +SRC_URI="https://github.com/ROCmSoftwarePlatform/rocBLAS/archive/rocm-${PV}.tar.gz -> rocm-${P}.tar.gz" +S="${WORKDIR}/${PN}-rocm-${PV}" + +LICENSE="BSD" +SLOT="0/$(ver_cut 1-2)" +KEYWORDS="~amd64" +IUSE="benchmark test" +RESTRICT="!test? ( test )" +REQUIRED_USE="${ROCM_REQUIRED_USE}" + +BDEPEND=" + >=dev-build/rocm-cmake-5.3 + dev-util/Tensile:${SLOT} + dev-python/joblib + test? ( dev-cpp/gtest ) +" + +DEPEND=" + >=dev-cpp/msgpack-cxx-6.0.0 + =dev-util/hip-6* + test? ( + virtual/blas + dev-cpp/gtest + sys-libs/libomp + ) + benchmark? ( + virtual/blas + sys-libs/libomp + ) +" + +QA_FLAGS_IGNORED="/usr/lib64/rocblas/library/.*" + +PATCHES=( + "${FILESDIR}"/${PN}-5.4.2-cpp_lib_filesystem.patch + "${FILESDIR}"/${PN}-5.4.2-add-missing-header.patch + "${FILESDIR}"/${PN}-5.4.2-link-cblas.patch + "${FILESDIR}"/${PN}-6.0.2-expand-isa-compatibility.patch + ) + +src_prepare() { + cmake_src_prepare + sed -e "s:,-rpath=.*\":\":" -i clients/CMakeLists.txt || die +} + +src_configure() { + addpredict /dev/random + addpredict /dev/kfd + addpredict /dev/dri/ + + local mycmakeargs=( + -DCMAKE_SKIP_RPATH=ON + -DBUILD_FILE_REORG_BACKWARD_COMPATIBILITY=OFF + -DROCM_SYMLINK_LIBS=OFF + -DAMDGPU_TARGETS="$(get_amdgpu_flags)" + -DTensile_LOGIC="asm_full" + -DTensile_COMPILER="hipcc" + -DTensile_LIBRARY_FORMAT="msgpack" + -DTensile_CODE_OBJECT_VERSION="default" + -DTensile_ROOT="${EPREFIX}/usr/share/Tensile" + -DBUILD_WITH_TENSILE=ON + -DCMAKE_INSTALL_INCLUDEDIR="include/rocblas" + -DBUILD_CLIENTS_SAMPLES=OFF + -DBUILD_CLIENTS_TESTS=$(usex test ON OFF) + -DBUILD_CLIENTS_BENCHMARKS=$(usex benchmark ON OFF) + -DTensile_CPU_THREADS=$(makeopts_jobs) + -DBUILD_WITH_PIP=OFF + ) + + CXX=hipcc cmake_src_configure +} + +src_compile() { + docs_compile + cmake_src_compile +} + +src_test() { + check_amdgpu + cd "${BUILD_DIR}"/clients/staging || die + export ROCBLAS_TEST_TIMEOUT=3600 ROCBLAS_TENSILE_LIBPATH="${BUILD_DIR}/Tensile/library" + export LD_LIBRARY_PATH="${BUILD_DIR}/clients:${BUILD_DIR}/library/src" + edob ./${PN,,}-test +} + +src_install() { + cmake_src_install + + if use benchmark; then + cd "${BUILD_DIR}" || die + dolib.a clients/librocblas_fortran_client.a + dobin clients/staging/rocblas-bench + fi + + # 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)/rocblas/library/ +} |