summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorYiyang Wu <xgreenlandforwyy@gmail.com>2022-06-21 16:38:05 +0800
committerBenda Xu <heroxbd@gentoo.org>2022-09-22 09:06:33 +0800
commita31808f3051758ccde8604944e1559ce92ef8ec4 (patch)
tree4392a8dbf50b9ed95858af077ec3792c6228b36c /dev-util/Tensile
parentsys-apps/gptfdisk: fix dev-libs/popt-1.19 incompatibility (diff)
downloadgentoo-a31808f3051758ccde8604944e1559ce92ef8ec4.tar.gz
gentoo-a31808f3051758ccde8604944e1559ce92ef8ec4.tar.bz2
gentoo-a31808f3051758ccde8604944e1559ce92ef8ec4.zip
dev-util/Tensile: add 5.1.3, switch to vanilla clang-14
enable py3.11 as well. No need to rebuild when hip upgrade since hip is just used at runtime. Signed-off-by: Yiyang Wu <xgreenlandforwyy@gmail.com> Signed-off-by: Benda Xu <heroxbd@gentoo.org>
Diffstat (limited to 'dev-util/Tensile')
-rw-r--r--dev-util/Tensile/Manifest1
-rw-r--r--dev-util/Tensile/Tensile-5.0.2-r2.ebuild2
-rw-r--r--dev-util/Tensile/Tensile-5.1.3.ebuild86
-rw-r--r--dev-util/Tensile/files/Tensile-5.1.3-gentoopath.patch75
4 files changed, 163 insertions, 1 deletions
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index c3e8f48bab20..3b51748fa904 100644
--- a/dev-util/Tensile/Manifest
+++ b/dev-util/Tensile/Manifest
@@ -1,3 +1,4 @@
DIST Tensile-5.0.2-PR1419.patch.gz 43754 BLAKE2B 3b1d7333ed0115f77816806e7735e25d897dbae3f7a40ece61d655c8f016d9d1c804b586b4e0e75661e20c517f5937a61d5ead25a545f0ce2e4a178bcbb3e7be SHA512 0154d8e14d335f4ecd86e5db335289f8662cede7e5d4b93bf6c56d8e48c1d7d0457271c078c9abbbca1f4aa91f028f6ce6a89efb783ba0207d0f39fd5de8a42c
DIST rocm-Tensile-4.3.0.tar.gz 14250149 BLAKE2B 030138eaca2a0aadd96801e6dbd72e510716dd90553ef3795c5e04e00a34a05ecae82b24f755e4033a4acfcdb1cf26291da1e7902bb090f89a010d403e832beb SHA512 126db0b413c716fba8d5be9bff7a44fd1badacbf32f3db8d0db649819177db37ebd56fd22dd3c809655f5d29675be115e698cd10bc3d0b4b23878ae3726fce47
DIST rocm-Tensile-5.0.2.tar.gz 14362090 BLAKE2B d44c3135496db41c4eb2545f6054c84493e89ef6357da881dae6859763338417e7a9e142cdce7e2de0536684bb641049ad0440964adbe68d03ca329f01efabc9 SHA512 f5ab7d7cbc5ba25498a02a6b7c07d5a9a548edcd44b439b92c8718072382adc18ae78274db1a7174e06436ac1396f3b4edafc7f491a2c87086c368e7270e64e3
+DIST rocm-Tensile-5.1.3.tar.gz 14357517 BLAKE2B 7c5881cc2b1d25db39f54a99f9f0a08395a5f7a58deebe74d9868e8f7044db5ae2f3907196adf16dc6abeba0f1f47f5cce5c8900891276daff8e4ad614cf3b94 SHA512 4405cb6de65c1da1fd93910fc901402a24e996ff04eccadf483c25d8eb518e03ea9bbf00e372dd5fe377696e71c39355603095f3c5fc1e26d311ed15d8e3b734
diff --git a/dev-util/Tensile/Tensile-5.0.2-r2.ebuild b/dev-util/Tensile/Tensile-5.0.2-r2.ebuild
index 38854fd0c269..22d3a6cee451 100644
--- a/dev-util/Tensile/Tensile-5.0.2-r2.ebuild
+++ b/dev-util/Tensile/Tensile-5.0.2-r2.ebuild
@@ -3,7 +3,7 @@
EAPI=8
-PYTHON_COMPAT=( python3_{8..10} )
+PYTHON_COMPAT=( python3_{8..11} )
DISTUTILS_USE_PEP517=setuptools
inherit distutils-r1 prefix
diff --git a/dev-util/Tensile/Tensile-5.1.3.ebuild b/dev-util/Tensile/Tensile-5.1.3.ebuild
new file mode 100644
index 000000000000..f2a39a52dbbb
--- /dev/null
+++ b/dev-util/Tensile/Tensile-5.1.3.ebuild
@@ -0,0 +1,86 @@
+# Copyright 1999-2022 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{8..11} )
+DISTUTILS_USE_PEP517=setuptools
+inherit distutils-r1 llvm prefix
+
+LLVM_MAX_SLOT=14
+
+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-python/pyyaml[${PYTHON_USEDEP}]
+ dev-python/msgpack[${PYTHON_USEDEP}]
+ >=dev-util/rocm-smi-4.3.0
+ >=sys-devel/clang-14.0.6-r1:${LLVM_MAX_SLOT}=
+"
+DEPEND="${RDEPEND}
+ dev-util/hip
+"
+
+PATCHES=( "${FILESDIR}"/${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/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)