diff options
Diffstat (limited to 'dev-util/Tensile')
18 files changed, 929 insertions, 0 deletions
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest new file mode 100644 index 000000000000..4d93ffa204a1 --- /dev/null +++ b/dev-util/Tensile/Manifest @@ -0,0 +1,4 @@ +DIST Tensile-5.0.2-PR1419.patch.gz 43754 BLAKE2B 3b1d7333ed0115f77816806e7735e25d897dbae3f7a40ece61d655c8f016d9d1c804b586b4e0e75661e20c517f5937a61d5ead25a545f0ce2e4a178bcbb3e7be SHA512 0154d8e14d335f4ecd86e5db335289f8662cede7e5d4b93bf6c56d8e48c1d7d0457271c078c9abbbca1f4aa91f028f6ce6a89efb783ba0207d0f39fd5de8a42c +DIST rocm-Tensile-5.1.3.tar.gz 14357517 BLAKE2B 7c5881cc2b1d25db39f54a99f9f0a08395a5f7a58deebe74d9868e8f7044db5ae2f3907196adf16dc6abeba0f1f47f5cce5c8900891276daff8e4ad614cf3b94 SHA512 4405cb6de65c1da1fd93910fc901402a24e996ff04eccadf483c25d8eb518e03ea9bbf00e372dd5fe377696e71c39355603095f3c5fc1e26d311ed15d8e3b734 +DIST rocm-Tensile-5.4.2.tar.gz 14413805 BLAKE2B ca1c13785164439f6fe83f16b36b7a0b06a63b70b9f0741d77b0fff98f38d1e1c7e9a54ef61897e6eb0a4eaf52ec9e5f14f42a7e6fd657f26306afa3de0c74a2 SHA512 d99736a202c7353979e732941199b4b20e53ff2b6c19a7175609440a840aa5624bc53a8013c435f31ac37933527771521ff02137c1c8fd53c5317eca8533be80 +DIST rocm-Tensile-5.7.1.tar.gz 12396002 BLAKE2B ec2c2bc3f07fe22abd800b642c998ea80b9e4e2d0d28c4763d574dbd2fcb14c735223ddc7571d4ab3a48f0bd819cc835c16df9882049c5bf56f25ef729e6d654 SHA512 8f356f36cfec74ee9b000d1180cf52077dc5c595e5c5a40792fc3eb13f3139980c7440f9bc912218f946cf8137e2481d23fe39f110511c1bb55fe85a933ffd36 diff --git a/dev-util/Tensile/Tensile-5.1.3-r3.ebuild b/dev-util/Tensile/Tensile-5.1.3-r3.ebuild new file mode 100644 index 000000000000..da1adcab4f7b --- /dev/null +++ b/dev-util/Tensile/Tensile-5.1.3-r3.ebuild @@ -0,0 +1,88 @@ +# Copyright 1999-2023 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +PYTHON_COMPAT=( python3_{9..11} ) +DISTUTILS_USE_PEP517=setuptools +inherit distutils-r1 llvm prefix + +LLVM_MAX_SLOT=15 + +DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions" +HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile" +SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz + https://github.com/littlewu2508/littlewu2508.github.io/raw/main/gentoo-distfiles/${PN}-5.0.2-PR1419.patch.gz" +S="${WORKDIR}/${PN}-rocm-${PV}" + +LICENSE="MIT" +KEYWORDS="~amd64" +SLOT="0/$(ver_cut 1-2)" + +# Not compatible with recent versions of pytest +RESTRICT="test" + +RDEPEND="${PYTHON_DEPS} + >=dev-cpp/msgpack-cxx-6.0.0 + dev-python/pyyaml[${PYTHON_USEDEP}] + dev-python/msgpack[${PYTHON_USEDEP}] + dev-util/hip + sys-devel/clang:${LLVM_MAX_SLOT} + >=dev-util/rocm-smi-4.3.0 +" +DEPEND="${RDEPEND}" + +PATCHES=( + "${FILESDIR}"/${PN}-change-cmake-name-for-msgpack-cxx-6-release.patch + "${FILESDIR}"/${PN}-4.3.0-output-commands.patch + "${FILESDIR}"/${PN}-5.0.2-gfx1031.patch + "${FILESDIR}"/${PN}-5.0.2-fix-arch-parse.patch + "${FILESDIR}"/${PN}-5.0.2-use-ninja.patch +) + +CMAKE_USE_DIR="${WORKDIR}/Source" + +src_prepare() { + distutils-r1_src_prepare + sed -e "s,\@LLVM_PATH\@,$(get_llvm_prefix ${LLVM_MAX_SLOT}),g" "${FILESDIR}"/${PN}-5.1.3-gentoopath.patch > "${S}"/gentoopath.patch || die + eapply $(prefixify_ro "${S}"/gentoopath.patch) + + pushd ${PN} || die + + sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \ + -i Source/cmake/FindROCmSMI.cmake || die + sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \ + -i Source/CMakeLists.txt || die + sed -e "/chmod 755/d" -i Source/TensileCreateLibrary.cmake || die # remove chmod 755 on + sed -e "s,\${Tensile_ROOT}/bin/,,g" -i Source/TensileCreateLibrary.cmake cmake/TensileConfig.cmake || die # ${Tensile_ROOT}/bin does not exists; call command directly + + local Tensile_share_dir="\"${EPREFIX}/usr/share/${PN}\"" + sed -e "/HipClangVersion/s/0,0,0/$(hipconfig -v)/" -i Common.py || die + + sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" -i ReplacementKernels.py Common.py ${PN}.py || die + + sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile/Source\", end='')|" -i __init__.py || die + + popd || die + + sed -e "/package_data/d" -e "/data_files/d" -i setup.py || die +} + +python_install() { + distutils-r1_python_install + + python_moduleinto Tensile + pushd Tensile || die + python_domodule Components + python_newexe Utilities/merge.py ${PN}-merge +} + +src_install() { + distutils-r1_src_install + + pushd ${PN} || die + insinto /usr/share/${PN} + doins -r Configs Perf ReplacementKernels ReplacementKernels-cov3 Source CustomKernels + insinto /usr/$(get_libdir)/cmake/${PN} + doins cmake/*.cmake +} diff --git a/dev-util/Tensile/Tensile-5.4.2-r2.ebuild b/dev-util/Tensile/Tensile-5.4.2-r2.ebuild new file mode 100644 index 000000000000..c846dce60997 --- /dev/null +++ b/dev-util/Tensile/Tensile-5.4.2-r2.ebuild @@ -0,0 +1,119 @@ +# Copyright 1999-2023 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +PYTHON_COMPAT=( python3_{10..11} ) +DISTUTILS_USE_PEP517=setuptools +ROCM_VERSION=${PV} +inherit cmake distutils-r1 llvm prefix rocm + +LLVM_MAX_SLOT=15 + +DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions" +HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile" +SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz" +S="${WORKDIR}/${PN}-rocm-${PV}" + +LICENSE="MIT" +KEYWORDS="~amd64" +SLOT="0/$(ver_cut 1-2)" +IUSE="client" +REQUIRED_USE="client? ( ${ROCM_REQUIRED_USE} )" + +# Not compatible with recent versions of pytest +RESTRICT="test" + +RDEPEND="${PYTHON_DEPS} + >=dev-cpp/msgpack-cxx-6.0.0 + dev-python/pyyaml[${PYTHON_USEDEP}] + dev-python/msgpack[${PYTHON_USEDEP}] + dev-util/hip + >=dev-util/rocm-smi-4.3.0 + sys-devel/clang:${LLVM_MAX_SLOT} +" +DEPEND="${RDEPEND}" + +PATCHES=( + "${FILESDIR}"/${PN}-change-cmake-name-for-msgpack-cxx-6-release.patch + "${FILESDIR}"/${PN}-4.3.0-output-commands.patch + "${FILESDIR}"/${PN}-5.4.2-gfx1031.patch + "${FILESDIR}"/${PN}-5.4.2-fix-arch-parse.patch + "${FILESDIR}"/${PN}-5.4.2-use-ninja.patch +) + +CMAKE_USE_DIR="${S}/${PN}/Source" + +src_prepare() { + distutils-r1_src_prepare + sed -e "s,\@LLVM_PATH\@,$(get_llvm_prefix ${LLVM_MAX_SLOT}),g" \ + "${FILESDIR}"/${PN}-5.4.2-gentoopath.patch > "${S}"/gentoopath.patch || die + eapply $(prefixify_ro "${S}"/gentoopath.patch) + + pushd ${PN} || die + + sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \ + -i Source/cmake/FindROCmSMI.cmake || die + sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \ + -i Source/CMakeLists.txt || die + sed -e "/chmod 755/d" -i Source/TensileCreateLibrary.cmake || die # remove chmod 755 on + + # ${Tensile_ROOT}/bin does not exists; call command directly + sed -e "s,\${Tensile_ROOT}/bin/,,g" -i Source/TensileCreateLibrary.cmake cmake/TensileConfig.cmake || die + + local Tensile_share_dir="\"${EPREFIX}/usr/share/${PN}\"" + sed -e "/HipClangVersion/s/0.0.0/$(hipconfig -v)/" -i Common.py || die + + sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" \ + -i ReplacementKernels.py Common.py ${PN}.py || die + + sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile/Source\", end='')|" -i __init__.py || die + + popd || die + + sed -e "/package_data/d" -e "/data_files/d" -i setup.py || die + use client && PATCHES= cmake_src_prepare # do not apply patches again in cmake_src_prepare +} + +src_configure() { + distutils-r1_src_configure + if use client; then + local mycmakeargs=( + -DCMAKE_SKIP_RPATH=ON + -DTENSILE_USE_MSGPACK=ON + -DTENSILE_USE_LLVM=ON + -DTensile_LIBRARY_FORMAT=msgpack + -DAMDGPU_TARGETS="$(get_amdgpu_flags)" + ) + CXX=hipcc cmake_src_configure + fi +} + +src_compile() { + distutils-r1_src_compile + use client && cmake_src_compile +} + +python_install() { + distutils-r1_python_install + + python_moduleinto Tensile + pushd Tensile || die + python_domodule Components + python_newexe Utilities/merge.py ${PN}-merge +} + +src_install() { + distutils-r1_src_install + + pushd ${PN} || die + insinto /usr/share/${PN} + doins -r Configs Perf ReplacementKernels ReplacementKernels-cov3 Source CustomKernels + insinto /usr/$(get_libdir)/cmake/${PN} + doins cmake/*.cmake + + if use client; then + pushd "${BUILD_DIR}" || die + dobin client/tensile_client + fi +} diff --git a/dev-util/Tensile/Tensile-5.7.1-r2.ebuild b/dev-util/Tensile/Tensile-5.7.1-r2.ebuild new file mode 100644 index 000000000000..94b4aac75b1b --- /dev/null +++ b/dev-util/Tensile/Tensile-5.7.1-r2.ebuild @@ -0,0 +1,133 @@ +# Copyright 1999-2024 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +PYTHON_COMPAT=( python3_{10..12} ) +DISTUTILS_USE_PEP517=setuptools +ROCM_VERSION=${PV} +inherit cmake distutils-r1 llvm prefix rocm + +LLVM_MAX_SLOT=17 + +DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions" +HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile" +SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz" +S="${WORKDIR}/${PN}-rocm-${PV}" + +LICENSE="MIT" +KEYWORDS="~amd64" +SLOT="0/$(ver_cut 1-2)" +IUSE="client test" +REQUIRED_USE="client? ( ${ROCM_REQUIRED_USE} )" + +RESTRICT="!test? ( test )" + +RDEPEND="${PYTHON_DEPS} + client? ( dev-libs/boost ) + >=dev-cpp/msgpack-cxx-6.0.0 + dev-python/pyyaml[${PYTHON_USEDEP}] + dev-python/msgpack[${PYTHON_USEDEP}] + =dev-util/hip-5* + >=dev-util/rocm-smi-4.3.0 + sys-devel/clang:${LLVM_MAX_SLOT} +" +DEPEND="${RDEPEND}" +BDEPEND=" + test? ( + dev-python/pytest-forked[${PYTHON_USEDEP}] + dev-python/pytest-xdist[${PYTHON_USEDEP}] + dev-python/filelock[${PYTHON_USEDEP}] + ) +" + +distutils_enable_tests pytest + +PATCHES=( + "${FILESDIR}"/${PN}-4.3.0-output-commands.patch + "${FILESDIR}"/${PN}-5.4.2-fix-arch-parse.patch + "${FILESDIR}"/${PN}-5.4.2-use-ninja.patch + "${FILESDIR}"/${PN}-5.7.1-expand-isa-compatibility.patch + "${FILESDIR}"/${PN}-5.7.1-fix-msgpack-dependency.patch +) + +CMAKE_USE_DIR="${S}/${PN}/Source" + +src_prepare() { + distutils-r1_src_prepare + sed -e "s,\@LLVM_PATH\@,$(get_llvm_prefix ${LLVM_MAX_SLOT}),g" \ + "${FILESDIR}"/${PN}-5.7.1-gentoopath.patch > "${S}"/gentoopath.patch || die + eapply $(prefixify_ro "${S}"/gentoopath.patch) + + pushd ${PN} || die + + sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \ + -i Source/cmake/FindROCmSMI.cmake || die + sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \ + -i Source/CMakeLists.txt || die + + # ${Tensile_ROOT}/bin does not exists; call command directly + sed -e "s,\${Tensile_ROOT}/bin/,,g" -i cmake/TensileConfig.cmake || die + + local Tensile_share_dir="\"${EPREFIX}/usr/share/${PN}\"" + sed -e "/HipClangVersion/s/0.0.0/$(hipconfig -v)/" -i Common.py || die + + sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" \ + -i ReplacementKernels.py Common.py ${PN}.py || die + + sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile/Source\", end='')|" -i __init__.py || die + + popd || die + + sed -e "/package_data/d" -e "/data_files/d" -i setup.py || die + use client && PATCHES= cmake_src_prepare # do not apply patches again in cmake_src_prepare +} + +src_configure() { + distutils-r1_src_configure + if use client; then + local mycmakeargs=( + -DCMAKE_SKIP_RPATH=ON + -DTENSILE_USE_MSGPACK=ON + -DTENSILE_USE_LLVM=ON + -DTensile_LIBRARY_FORMAT=msgpack + -DAMDGPU_TARGETS="$(get_amdgpu_flags)" + ) + CXX=hipcc cmake_src_configure + fi +} + +src_compile() { + distutils-r1_src_compile + use client && cmake_src_compile +} + +python_install() { + distutils-r1_python_install + + python_moduleinto Tensile + pushd Tensile || die + python_domodule Components + python_newexe Utilities/merge.py ${PN}-merge +} + +src_install() { + distutils-r1_src_install + + pushd ${PN} || die + insinto /usr/share/${PN} + doins -r Configs Perf ReplacementKernels-cov3 Source CustomKernels + insinto /usr/$(get_libdir)/cmake/${PN} + doins cmake/*.cmake + + if use client; then + pushd "${BUILD_DIR}" || die + dobin client/tensile_client + fi +} + +# Test suite fails to start without this +python_test() { + export ROCM_PATH="${EPREFIX}/usr" + epytest +} diff --git a/dev-util/Tensile/files/Tensile-4.3.0-output-commands.patch b/dev-util/Tensile/files/Tensile-4.3.0-output-commands.patch new file mode 100644 index 000000000000..be5a4db21429 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-4.3.0-output-commands.patch @@ -0,0 +1,23 @@ +diff --color -uprN orig/Tensile/cmake/TensileConfig.cmake Tensile-rocm-4.3.0/Tensile/cmake/TensileConfig.cmake +--- orig/Tensile/cmake/TensileConfig.cmake 2021-08-18 17:48:00.115478470 +0800 ++++ Tensile-rocm-4.3.0/Tensile/cmake/TensileConfig.cmake 2021-08-18 17:48:49.963478801 +0800 +@@ -234,6 +234,7 @@ function(TensileCreateLibraryFiles + COMMENT "Generating Tensile Libraries" + OUTPUT ${Tensile_EMBED_LIBRARY_SOURCE};${Tensile_MANIFEST_CONTENTS} + COMMAND ${CommandLine} ++ USES_TERMINAL + ) + + set("${Tensile_VAR_PREFIX}_ALL_FILES" ${Tensile_MANIFEST_CONTENTS} PARENT_SCOPE) +diff --color -uprN orig/Tensile/Common.py Tensile-rocm-4.3.0/Tensile/Common.py +--- orig/Tensile/Common.py 2021-08-18 17:48:00.075478470 +0800 ++++ Tensile-rocm-4.3.0/Tensile/Common.py 2021-08-18 17:48:23.287478624 +0800 +@@ -179,7 +179,7 @@ globalParameters["PrintTensorD"] = 0 + globalParameters["PrintTensorRef"] = 0 # Print reference tensor. 0x1=after init; 0x2=after copy-back; 0x3=both + globalParameters["PrintIndexAssignments"] = 0 # Print the tensor index assignment info + globalParameters["PrintWinnersOnly"] = False # Only print the solutions which become the fastest +-globalParameters["PrintCodeCommands"] = False # print the commands used to generate the code objects (asm,link,hip-clang, etc) ++globalParameters["PrintCodeCommands"] = True # print the commands used to generate the code objects (asm,link,hip-clang, etc) + globalParameters["DumpTensors"] = False # If True, dump tensors to binary files instead of printing them. + + # TODO - remove this when NewClient is mainstream diff --git a/dev-util/Tensile/files/Tensile-5.0.2-fix-arch-parse.patch b/dev-util/Tensile/files/Tensile-5.0.2-fix-arch-parse.patch new file mode 100644 index 000000000000..47bee8819c44 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.0.2-fix-arch-parse.patch @@ -0,0 +1,12 @@ +Index: Tensile-rocm-5.0.1/Tensile/Tensile.py +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Tensile.py ++++ Tensile-rocm-5.0.1/Tensile/Tensile.py +@@ -97,7 +97,6 @@ def addCommonArguments(argParser): + """ + + (key, value) = par.split("=") +- value = eval(value) + return (key, value) + + argParser.add_argument("-d", "--device", dest="device", type=int, \ diff --git a/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch b/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch new file mode 100644 index 000000000000..17925cfd49e0 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch @@ -0,0 +1,104 @@ +Index: Tensile-rocm-5.0.1/pytest.ini +=================================================================== +--- Tensile-rocm-5.0.1.orig/pytest.ini ++++ Tensile-rocm-5.0.1/pytest.ini +@@ -87,6 +87,7 @@ markers = + xfail-gfx1011: architecture + xfail-gfx1012: architecture + xfail-gfx1030: architecture ++ xfail-gfx1031: architecture + skip-gfx000: architecture + skip-gfx900: architecture + skip-gfx906: architecture +@@ -96,3 +97,4 @@ markers = + skip-gfx1011: architecture + skip-gfx1012: architecture + skip-gfx1030: architecture ++ skip-gfx1031: architecture +Index: Tensile-rocm-5.0.1/Tensile/Common.py +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Common.py ++++ Tensile-rocm-5.0.1/Tensile/Common.py +@@ -198,7 +198,7 @@ globalParameters["MergeFiles"] = True + globalParameters["NumMergedFiles"] = 1 # The number of files that kernels should be split between when merging + + globalParameters["MaxFileName"] = 64 # If a file name would be longer than this, shorten it with a hash. +-globalParameters["SupportedISA"] = [(8,0,3), (9,0,0), (9,0,6), (9,0,8), (9,0,10), (10,1,0), (10,1,1), (10,1,2), (10,3,0)] # assembly kernels writer supports these architectures ++globalParameters["SupportedISA"] = [(8,0,3), (9,0,0), (9,0,6), (9,0,8), (9,0,10), (10,1,0), (10,1,1), (10,1,2), (10,3,0),(10,3,1)] # assembly kernels writer supports these architectures + + globalParameters["GenerateManifestAndExit"] = False # Output manifest file with list of expected library objects and exit + globalParameters["NewClient"] = 2 # Old client deprecated: NewClient must be set to 2. +@@ -264,7 +264,7 @@ architectureMap = { + 'gfx906':'vega20', 'gfx906:xnack+':'vega20', 'gfx906:xnack-':'vega20', + 'gfx908':'arcturus','gfx908:xnack+':'arcturus', 'gfx908:xnack-':'arcturus', + 'gfx90a':'aldebaran', 'gfx90a:xnack+':'aldebaran', 'gfx90a:xnack-':'aldebaran', +- 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21' ++ 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', 'gfx1031':'navi22' + } + + def getArchitectureName(gfxName): +Index: Tensile-rocm-5.0.1/Tensile/KernelWriterAssembly.py +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/KernelWriterAssembly.py ++++ Tensile-rocm-5.0.1/Tensile/KernelWriterAssembly.py +@@ -2364,7 +2364,7 @@ class KernelWriterAssembly(KernelWriter) + + # integer i8x4 + elif kernel["ProblemType"]["DataType"].isInt8x4(): +- if self.version == (9,0,6) or self.version == (9,0,8) or self.version == (9,0,10) or self.version == (10,3,0): ++ if self.version == (9,0,6) or self.version == (9,0,8) or self.version == (9,0,10) or self.version == (10,3,0) or self.version == (10,3,1): + for b in range(0, kernel["ThreadTile1"]): + for a in range(0, kernel["ThreadTile0"]): + for iui in range(0, innerUnroll): +Index: Tensile-rocm-5.0.1/Tensile/Source/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Source/CMakeLists.txt ++++ Tensile-rocm-5.0.1/Tensile/Source/CMakeLists.txt +@@ -47,7 +47,7 @@ if(NOT DEFINED CXX_VERSION_STRING) + endif() + + if(CMAKE_CXX_COMPILER STREQUAL "hipcc") +- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 CACHE STRING "GPU architectures") ++ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 CACHE STRING "GPU architectures") + else() + set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906 gfx908 gfx90a CACHE STRING "GPU architectures") + endif() +Index: Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/AMDGPU.hpp +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/include/Tensile/AMDGPU.hpp ++++ Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/AMDGPU.hpp +@@ -62,7 +62,8 @@ namespace Tensile + gfx1010 = 1010, + gfx1011 = 1011, + gfx1012 = 1012, +- gfx1030 = 1030 ++ gfx1030 = 1030, ++ gfx1031 = 1031 + }; + + AMDGPU(); +Index: Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp ++++ Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp +@@ -215,6 +215,7 @@ namespace Tensile + iot::enumCase(io, value, "gfx1011", AMDGPU::Processor::gfx1011); + iot::enumCase(io, value, "gfx1012", AMDGPU::Processor::gfx1012); + iot::enumCase(io, value, "gfx1030", AMDGPU::Processor::gfx1030); ++ iot::enumCase(io, value, "gfx1031", AMDGPU::Processor::gfx1031); + } + }; + } // namespace Serialization +Index: Tensile-rocm-5.0.1/Tensile/Source/lib/source/AMDGPU.cpp +=================================================================== +--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/source/AMDGPU.cpp ++++ Tensile-rocm-5.0.1/Tensile/Source/lib/source/AMDGPU.cpp +@@ -82,6 +82,8 @@ namespace Tensile + return stream << "gfx1012"; + case AMDGPU::Processor::gfx1030: + return stream << "gfx1030"; ++ case AMDGPU::Processor::gfx1031: ++ return stream << "gfx1031"; + } + return stream; + } diff --git a/dev-util/Tensile/files/Tensile-5.0.2-use-ninja.patch b/dev-util/Tensile/files/Tensile-5.0.2-use-ninja.patch new file mode 100644 index 000000000000..485ec992d808 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.0.2-use-ninja.patch @@ -0,0 +1,21 @@ +diff --git a/Tensile/ClientExecutable.py b/Tensile/ClientExecutable.py +index ca3809f..2d1d83b 100644 +--- a/Tensile/ClientExecutable.py ++++ b/Tensile/ClientExecutable.py +@@ -36,6 +36,7 @@ class CMakeEnvironment: + + args = ['cmake'] + args += itertools.chain.from_iterable([ ['-D', '{}={}'.format(key, value)] for key,value in self.options.items()]) ++ args += itertools.chain.from_iterable([ ['-G', 'Ninja'], [ '-D', 'CMAKE_EXPORT_COMPILE_COMMANDS=ON' ] ]) + args += [self.sourceDir] + + Common.print2(' '.join(args)) +@@ -43,7 +44,7 @@ class CMakeEnvironment: + subprocess.check_call(args, cwd=Common.ensurePath(self.buildDir)) + + def build(self): +- args = ['make', '-j'] ++ args = ['ninja'] + Common.print2(' '.join(args)) + with Common.ClientExecutionLock(): + subprocess.check_call(args, cwd=self.buildDir) diff --git a/dev-util/Tensile/files/Tensile-5.1.3-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.1.3-gentoopath.patch new file mode 100644 index 000000000000..2f41bf3aa715 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.1.3-gentoopath.patch @@ -0,0 +1,75 @@ +Fix incorrect path in scripts and CMakeLists, in order to build tensile_client +=================================================================== +--- Tensile-rocm-5.0.2.orig/Tensile/Source/client/CMakeLists.txt ++++ Tensile-rocm-5.0.2/Tensile/Source/client/CMakeLists.txt +@@ -56,3 +56,5 @@ endif() + foreach(arch IN LISTS TENSILE_GPU_ARCHS) + target_link_libraries(tensile_client PRIVATE "--amdgpu-target=${arch}") + endforeach(arch) ++ ++target_link_libraries(tensile_client PRIVATE LLVMSupport) +Index: Tensile-rocm-5.0.2/Tensile/Source/lib/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.0.2.orig/Tensile/Source/lib/CMakeLists.txt ++++ Tensile-rocm-5.0.2/Tensile/Source/lib/CMakeLists.txt +@@ -47,7 +47,7 @@ set(tensile_sources ${tensile_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM 13.0 QUIET CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ QUIET CONFIG) + if(NOT LLVM_FOUND) + find_package(LLVM 12.0 QUIET CONFIG) + if(NOT LLVM_FOUND) +Index: Tensile-rocm-5.0.2/Tensile/Common.py +=================================================================== +--- Tensile-rocm-5.0.2.orig/Tensile/Common.py ++++ Tensile-rocm-5.0.2/Tensile/Common.py +@@ -1873,7 +1873,7 @@ def assignGlobalParameters( config ): + else: + print2(" %24s: %8s (unspecified)" % (key, defaultValue)) + +- globalParameters["ROCmPath"] = "/opt/rocm" ++ globalParameters["ROCmPath"] = "@GENTOO_PORTAGE_EPREFIX@/usr" + if "ROCM_PATH" in os.environ: + globalParameters["ROCmPath"] = os.environ.get("ROCM_PATH") + if "TENSILE_ROCM_PATH" in os.environ: +Index: Tensile-rocm-5.0.2/Tensile/ClientWriter.py +=================================================================== +--- Tensile-rocm-5.0.2.orig/Tensile/ClientWriter.py ++++ Tensile-rocm-5.0.2/Tensile/ClientWriter.py +@@ -205,7 +205,7 @@ def getBuildClientLibraryScript(buildPat + import io + runScriptFile = io.StringIO() + +- callCreateLibraryCmd = globalParameters["ScriptPath"] + "/bin/TensileCreateLibrary" ++ callCreateLibraryCmd = "TensileCreateLibrary" + + + if globalParameters["MergeFiles"]: +Index: Tensile-rocm-5.0.2/Tensile/GenerateSummations.py +=================================================================== +--- Tensile-rocm-5.0.2.orig/Tensile/GenerateSummations.py ++++ Tensile-rocm-5.0.2/Tensile/GenerateSummations.py +@@ -57,7 +57,7 @@ def createLibraryForBenchmark(logicPath, + Selection. + """ + +- pythonExePath = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bin", "TensileCreateLibrary") ++ pythonExePath = "TensileCreateLibrary" + args = [pythonExePath, \ + "--merge-files", "--new-client-only", "--no-short-file-names", "--no-library-print-debug", \ + "--architecture=all", "--code-object-version=V3", "--cxx-compiler=hipcc", "--library-format=yaml", \ +Index: Tensile-rocm-5.0.2/HostLibraryTests/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.0.2.orig/HostLibraryTests/CMakeLists.txt ++++ Tensile-rocm-5.0.2/HostLibraryTests/CMakeLists.txt +@@ -135,7 +135,7 @@ set(test_sources ${test_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM 13.0 QUIET CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ QUIET CONFIG) + if(NOT LLVM_FOUND) + find_package(LLVM 12.0 QUIET CONFIG) + if(NOT LLVM_FOUND) diff --git a/dev-util/Tensile/files/Tensile-5.4.2-fix-arch-parse.patch b/dev-util/Tensile/files/Tensile-5.4.2-fix-arch-parse.patch new file mode 100644 index 000000000000..14263e89ef84 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.4.2-fix-arch-parse.patch @@ -0,0 +1,12 @@ +Index: Tensile-rocm-5.3.3/Tensile/Tensile.py +=================================================================== +--- Tensile-rocm-5.3.3.orig/Tensile/Tensile.py ++++ Tensile-rocm-5.3.3/Tensile/Tensile.py +@@ -101,7 +101,6 @@ def addCommonArguments(argParser): + Allows the --global-parameters option to specify any parameters from the command line. + """ + (key, value) = par.split("=") +- value = eval(value) + return (key, value) + + argParser.add_argument("-d", "--device", dest="device", type=int, \ diff --git a/dev-util/Tensile/files/Tensile-5.4.2-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.4.2-gentoopath.patch new file mode 100644 index 000000000000..eda9b0307c52 --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.4.2-gentoopath.patch @@ -0,0 +1,67 @@ +Fix incorrect path in scripts and CMakeLists, in order to build tensile_client +=================================================================== +Index: Tensile-rocm-5.4.2/Tensile/Source/lib/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.4.2.orig/Tensile/Source/lib/CMakeLists.txt ++++ Tensile-rocm-5.4.2/Tensile/Source/lib/CMakeLists.txt +@@ -51,7 +51,7 @@ set(tensile_sources ${tensile_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM 13.0 QUIET CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ QUIET CONFIG) + if(NOT LLVM_FOUND) + find_package(LLVM 12.0 QUIET CONFIG) + if(NOT LLVM_FOUND) +Index: Tensile-rocm-5.4.2/Tensile/Common.py +=================================================================== +--- Tensile-rocm-5.4.2.orig/Tensile/Common.py ++++ Tensile-rocm-5.4.2/Tensile/Common.py +@@ -1950,7 +1950,7 @@ def assignGlobalParameters( config ): + else: + print2(" %24s: %8s (unspecified)" % (key, defaultValue)) + +- globalParameters["ROCmPath"] = "/opt/rocm" ++ globalParameters["ROCmPath"] = "@GENTOO_PORTAGE_EPREFIX@/usr" + if "ROCM_PATH" in os.environ: + globalParameters["ROCmPath"] = os.environ.get("ROCM_PATH") + if "TENSILE_ROCM_PATH" in os.environ: +Index: Tensile-rocm-5.4.2/Tensile/ClientWriter.py +=================================================================== +--- Tensile-rocm-5.4.2.orig/Tensile/ClientWriter.py ++++ Tensile-rocm-5.4.2/Tensile/ClientWriter.py +@@ -180,7 +180,7 @@ def runClient(libraryLogicPath, forBench + def getBuildClientLibraryScript(buildPath, libraryLogicPath): + callCreateLibraryCmd = ["python"] if os.name == "nt" else [] + +- callCreateLibraryCmd += [os.path.join(globalParameters["ScriptPath"] , "bin", "TensileCreateLibrary")] ++ callCreateLibraryCmd += ["TensileCreateLibrary"] + + if globalParameters["MergeFiles"]: + callCreateLibraryCmd += ["--merge-files"] +Index: Tensile-rocm-5.4.2/Tensile/GenerateSummations.py +=================================================================== +--- Tensile-rocm-5.4.2.orig/Tensile/GenerateSummations.py ++++ Tensile-rocm-5.4.2/Tensile/GenerateSummations.py +@@ -59,7 +59,7 @@ def createLibraryForBenchmark(logicPath, + Selection. + """ + +- pythonExePath = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bin", "TensileCreateLibrary") ++ pythonExePath = "TensileCreateLibrary" + args = [pythonExePath, \ + "--merge-files", "--new-client-only", "--no-short-file-names", "--no-library-print-debug", \ + "--architecture=all", "--code-object-version=V3", "--cxx-compiler=hipcc", "--library-format=yaml", \ +Index: Tensile-rocm-5.4.2/HostLibraryTests/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.4.2.orig/HostLibraryTests/CMakeLists.txt ++++ Tensile-rocm-5.4.2/HostLibraryTests/CMakeLists.txt +@@ -139,7 +139,7 @@ set(test_sources ${test_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM 13.0 QUIET CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ QUIET CONFIG) + if(NOT LLVM_FOUND) + find_package(LLVM 12.0 QUIET CONFIG) + if(NOT LLVM_FOUND) diff --git a/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch b/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch new file mode 100644 index 000000000000..1f703c07c3ec --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch @@ -0,0 +1,88 @@ +Index: Tensile-rocm-5.3.3/pytest.ini +=================================================================== +--- Tensile-rocm-5.3.3.orig/pytest.ini ++++ Tensile-rocm-5.3.3/pytest.ini +@@ -88,6 +88,7 @@ markers = + xfail-gfx1011: architecture + xfail-gfx1012: architecture + xfail-gfx1030: architecture ++ xfail-gfx1031: architecture + xfail-gfx1100: architecture + xfail-gfx1101: architecture + xfail-gfx1102: architecture +@@ -100,6 +101,7 @@ markers = + skip-gfx1011: architecture + skip-gfx1012: architecture + skip-gfx1030: architecture ++ skip-gfx1031: architecture + skip-gfx1100: architecture + skip-gfx1101: architecture + skip-gfx1102: architecture +Index: Tensile-rocm-5.3.3/Tensile/Common.py +=================================================================== +--- Tensile-rocm-5.3.3.orig/Tensile/Common.py ++++ Tensile-rocm-5.3.3/Tensile/Common.py +@@ -201,7 +201,7 @@ globalParameters["MergeFiles"] = True + globalParameters["NumMergedFiles"] = 1 # The number of files that kernels should be split between when merging + + globalParameters["MaxFileName"] = 64 # If a file name would be longer than this, shorten it with a hash. +-globalParameters["SupportedISA"] = [(8,0,3), (9,0,0), (9,0,6), (9,0,8), (9,0,10), (10,1,0), (10,1,1), (10,1,2), (10,3,0), (11,0,0), (11,0,1), (11,0,2)] # assembly kernels writer supports these architectures ++globalParameters["SupportedISA"] = [(8,0,3), (9,0,0), (9,0,6), (9,0,8), (9,0,10), (10,1,0), (10,1,1), (10,1,2), (10,3,0), (10,3,1), (11,0,0), (11,0,1), (11,0,2)] # assembly kernels writer supports these architectures + + globalParameters["GenerateManifestAndExit"] = False # Output manifest file with list of expected library objects and exit + globalParameters["NewClient"] = 2 # Old client deprecated: NewClient must be set to 2. +@@ -271,7 +271,7 @@ architectureMap = { + 'gfx906':'vega20', 'gfx906:xnack+':'vega20', 'gfx906:xnack-':'vega20', + 'gfx908':'arcturus','gfx908:xnack+':'arcturus', 'gfx908:xnack-':'arcturus', + 'gfx90a':'aldebaran', 'gfx90a:xnack+':'aldebaran', 'gfx90a:xnack-':'aldebaran', +- 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', ++ 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', 'gfx1031':'navi22', + 'gfx1100':'navi31', 'gfx1101':'navi32', 'gfx1102':'navi33' + } + +Index: Tensile-rocm-5.3.3/Tensile/Source/CMakeLists.txt +=================================================================== +--- Tensile-rocm-5.3.3.orig/Tensile/Source/CMakeLists.txt ++++ Tensile-rocm-5.3.3/Tensile/Source/CMakeLists.txt +@@ -51,7 +51,7 @@ if(NOT DEFINED CXX_VERSION_STRING) + endif() + + if(CMAKE_CXX_COMPILER STREQUAL "hipcc") +- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") ++ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") + else() + set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906 gfx908 gfx90a gfx1010 gfx1011 gfx1012 gfx1030 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") + endif() +Index: Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/AMDGPU.hpp +=================================================================== +--- Tensile-rocm-5.3.3.orig/Tensile/Source/lib/include/Tensile/AMDGPU.hpp ++++ Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/AMDGPU.hpp +@@ -63,6 +63,7 @@ namespace Tensile + gfx1011 = 1011, + gfx1012 = 1012, + gfx1030 = 1030, ++ gfx1031 = 1031, + gfx1100 = 1100, + gfx1101 = 1101, + gfx1102 = 1102 +@@ -90,6 +91,8 @@ namespace Tensile + return "gfx1012"; + case AMDGPU::Processor::gfx1030: + return "gfx1030"; ++ case AMDGPU::Processor::gfx1031: ++ return "gfx1032"; + case AMDGPU::Processor::gfx1100: + return "gfx1100"; + case AMDGPU::Processor::gfx1101: +Index: Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp +=================================================================== +--- Tensile-rocm-5.3.3.orig/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp ++++ Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp +@@ -215,6 +215,7 @@ namespace Tensile + iot::enumCase(io, value, "gfx1011", AMDGPU::Processor::gfx1011); + iot::enumCase(io, value, "gfx1012", AMDGPU::Processor::gfx1012); + iot::enumCase(io, value, "gfx1030", AMDGPU::Processor::gfx1030); ++ iot::enumCase(io, value, "gfx1031", AMDGPU::Processor::gfx1031); + iot::enumCase(io, value, "gfx1100", AMDGPU::Processor::gfx1100); + iot::enumCase(io, value, "gfx1101", AMDGPU::Processor::gfx1101); + iot::enumCase(io, value, "gfx1102", AMDGPU::Processor::gfx1102); diff --git a/dev-util/Tensile/files/Tensile-5.4.2-use-ninja.patch b/dev-util/Tensile/files/Tensile-5.4.2-use-ninja.patch new file mode 100644 index 000000000000..f88bf00665cb --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.4.2-use-ninja.patch @@ -0,0 +1,21 @@ +Index: Tensile-rocm-5.4.2/Tensile/ClientExecutable.py +=================================================================== +--- Tensile-rocm-5.4.2.orig/Tensile/ClientExecutable.py ++++ Tensile-rocm-5.4.2/Tensile/ClientExecutable.py +@@ -44,6 +44,7 @@ class CMakeEnvironment: + args = ['cmake'] + args += ['-G', 'Ninja'] if (os.name == 'nt') else [] + args += itertools.chain.from_iterable([ ['-D{}={}'.format(key, value)] for key,value in self.options.items()]) ++ args += itertools.chain.from_iterable([ ['-G', 'Ninja'], [ '-D', 'CMAKE_EXPORT_COMPILE_COMMANDS=ON' ] ]) + args += [self.sourceDir] + args = [cmake_path(arg) for arg in args] + +@@ -53,7 +54,7 @@ class CMakeEnvironment: + subprocess.check_output(args, stderr=subprocess.STDOUT, cwd=Common.ensurePath(self.buildDir)) + + def build(self): +- args = [('ninja' if (os.name == "nt") else 'make'), f'-j{CPUThreadCount()}'] ++ args = [('ninja'), f'-j{CPUThreadCount()}'] + Common.print2(' '.join(args)) + with Common.ClientExecutionLock(): + # change to use check_output to force windows cmd block util command finish diff --git a/dev-util/Tensile/files/Tensile-5.7.1-expand-isa-compatibility.patch b/dev-util/Tensile/files/Tensile-5.7.1-expand-isa-compatibility.patch new file mode 100644 index 000000000000..ebfc537c5c4d --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.7.1-expand-isa-compatibility.patch @@ -0,0 +1,49 @@ +Allow Tensile to load the compatible kernels when running on +architectures compatible with those ISAs. + +Based on patch from Cordell Bloor <cgmb@slerp.xyz> +https://salsa.debian.org/rocm-team/rocblas/-/blob/master/debian/patches/0012-expand-isa-compatibility.patch + +--- a/Tensile/Source/lib/source/hip/HipHardware.cpp ++++ b/Tensile/Source/lib/source/hip/HipHardware.cpp +@@ -24,6 +24,8 @@ + * + *******************************************************************************/ + ++#include <cstring> ++ + #include <Tensile/AMDGPU.hpp> + #include <Tensile/hip/HipHardware.hpp> + #include <Tensile/hip/HipUtils.hpp> +@@ -56,6 +58,31 @@ namespace Tensile + { + hipDeviceProp_t prop; + HIP_CHECK_EXC(hipGetDeviceProperties(&prop, deviceId)); ++ // coerce to compatible arch ++ switch(prop.gcnArch) ++ { ++ case 902: ++ case 909: ++ case 912: ++ prop.gcnArch = 900; ++ std::strcpy(prop.gcnArchName, "gfx900"); ++ break; ++ case 1011: ++ case 1012: ++ case 1013: ++ prop.gcnArch = 1010; ++ std::strcpy(prop.gcnArchName, "gfx1010"); ++ break; ++ case 1031: ++ case 1032: ++ case 1033: ++ case 1034: ++ case 1035: ++ case 1036: ++ prop.gcnArch = 1030; ++ std::strcpy(prop.gcnArchName, "gfx1030"); ++ break; ++ } + #if HIP_VERSION >= 50220730 + int hip_version; + HIP_CHECK_EXC(hipRuntimeGetVersion(&hip_version)); diff --git a/dev-util/Tensile/files/Tensile-5.7.1-fix-msgpack-dependency.patch b/dev-util/Tensile/files/Tensile-5.7.1-fix-msgpack-dependency.patch new file mode 100644 index 000000000000..f6f3f0e24c1a --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.7.1-fix-msgpack-dependency.patch @@ -0,0 +1,29 @@ +Gentoo uses "msgpack-cxx" as library name and adds important +MSGPACK_NO_BOOST compile definition when msgpack-cxx us installed with USE=-boost +--- a/Tensile/Source/lib/CMakeLists.txt ++++ b/Tensile/Source/lib/CMakeLists.txt +@@ -98,22 +98,9 @@ if(TENSILE_USE_LLVM OR TENSILE_USE_MSGPACK) + endif() + + if(TENSILE_USE_MSGPACK) +- find_package(msgpack REQUIRED) ++ find_package(msgpack-cxx REQUIRED) + target_compile_definitions(TensileHost PUBLIC -DTENSILE_MSGPACK=1) +- +- if(TARGET msgpackc-cxx) +- get_target_property(msgpack_inc msgpackc-cxx INTERFACE_INCLUDE_DIRECTORIES) +- elseif(TARGET msgpackc) +- get_target_property(msgpack_inc msgpackc INTERFACE_INCLUDE_DIRECTORIES) +- endif() +- +- if(DEFINED msgpack_inc) +- # include C++ headers manually +- # External header includes included as system files +- target_include_directories(TensileHost +- SYSTEM PRIVATE $<BUILD_INTERFACE:${msgpack_inc}> +- ) +- endif() ++ target_link_libraries(TensileHost PRIVATE msgpack-cxx) + endif() + + if(TENSILE_USE_LLVM) diff --git a/dev-util/Tensile/files/Tensile-5.7.1-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.7.1-gentoopath.patch new file mode 100644 index 000000000000..c8bc360cc39b --- /dev/null +++ b/dev-util/Tensile/files/Tensile-5.7.1-gentoopath.patch @@ -0,0 +1,57 @@ +Fix incorrect path in scripts and CMakeLists, in order to build tensile_client + +--- a/HostLibraryTests/CMakeLists.txt ++++ b/HostLibraryTests/CMakeLists.txt +@@ -139,7 +139,7 @@ set(test_sources ${test_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM REQUIRED CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ REQUIRED CONFIG) + + set(test_sources ${test_sources} + ContractionLibraryLoading_test.cpp +--- a/Tensile/ClientWriter.py ++++ b/Tensile/ClientWriter.py +@@ -176,7 +176,7 @@ def runClient(libraryLogicPath, forBenchmark, enableTileSelection, configPaths=N + def getBuildClientLibraryScript(buildPath, libraryLogicPath): + callCreateLibraryCmd = ["python"] if os.name == "nt" else [] + +- callCreateLibraryCmd += [os.path.join(globalParameters["ScriptPath"] , "bin", "TensileCreateLibrary")] ++ callCreateLibraryCmd += ["TensileCreateLibrary"] + + if globalParameters["MergeFiles"]: + callCreateLibraryCmd += ["--merge-files"] +--- a/Tensile/Common.py ++++ b/Tensile/Common.py +@@ -2108,7 +2108,7 @@ def assignGlobalParameters( config ): + else: + print2(" %24s: %8s (unspecified)" % (key, defaultValue)) + +- globalParameters["ROCmPath"] = "/opt/rocm" ++ globalParameters["ROCmPath"] = "@GENTOO_PORTAGE_EPREFIX@/usr" + if "ROCM_PATH" in os.environ: + globalParameters["ROCmPath"] = os.environ.get("ROCM_PATH") + if "TENSILE_ROCM_PATH" in os.environ: +--- a/Tensile/GenerateSummations.py ++++ b/Tensile/GenerateSummations.py +@@ -59,7 +59,7 @@ def createLibraryForBenchmark(logicPath, libraryPath, currentPath): + Selection. + """ + +- pythonExePath = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bin", "TensileCreateLibrary") ++ pythonExePath = "TensileCreateLibrary" + args = [pythonExePath, \ + "--merge-files", "--new-client-only", "--no-short-file-names", "--no-library-print-debug", \ + "--architecture=all", "--cxx-compiler=hipcc", "--library-format=yaml", \ +--- a/Tensile/Source/lib/CMakeLists.txt ++++ b/Tensile/Source/lib/CMakeLists.txt +@@ -52,7 +52,7 @@ set(tensile_sources ${tensile_sources} + ) + + if(TENSILE_USE_LLVM) +- find_package(LLVM REQUIRED CONFIG) ++ find_package(LLVM PATHS @LLVM_PATH@ REQUIRED CONFIG) + + set(tensile_sources ${tensile_sources} + source/llvm/YAML.cpp diff --git a/dev-util/Tensile/files/Tensile-change-cmake-name-for-msgpack-cxx-6-release.patch b/dev-util/Tensile/files/Tensile-change-cmake-name-for-msgpack-cxx-6-release.patch new file mode 100644 index 000000000000..d8ac203781da --- /dev/null +++ b/dev-util/Tensile/files/Tensile-change-cmake-name-for-msgpack-cxx-6-release.patch @@ -0,0 +1,13 @@ +Yiyang's original patch for msgpack-cxx-5 adapted to version 6. + +--- a/Tensile/Source/lib/CMakeLists.txt ++++ b/Tensile/Source/lib/CMakeLists.txt +@@ -97,7 +97,7 @@ if(TENSILE_USE_LLVM OR TENSILE_USE_MSGPACK) + endif() + + if(TENSILE_USE_MSGPACK) +- find_package(msgpack REQUIRED) ++ find_package(msgpack-cxx REQUIRED) + target_compile_definitions(TensileHost PUBLIC -DTENSILE_MSGPACK=1) + + if(TARGET msgpackc-cxx) diff --git a/dev-util/Tensile/metadata.xml b/dev-util/Tensile/metadata.xml new file mode 100644 index 000000000000..32364cb13b06 --- /dev/null +++ b/dev-util/Tensile/metadata.xml @@ -0,0 +1,14 @@ +<?xml version="1.0" encoding="UTF-8"?> +<!DOCTYPE pkgmetadata SYSTEM "https://www.gentoo.org/dtd/metadata.dtd"> +<pkgmetadata> + <maintainer type="project"> + <email>sci@gentoo.org</email> + <name>Science Project</name> + </maintainer> + <upstream> + <remote-id type="github">ROCmSoftwarePlatform/Tensile</remote-id> + </upstream> + <use> + <flag name="client">Build and install tensile_client executable to run benchmarks and tune GPU GEMM</flag> + </use> +</pkgmetadata> |