* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2022-03-05 20:51 Andrew Ammerlaan
0 siblings, 0 replies; 6+ messages in thread
From: Andrew Ammerlaan @ 2022-03-05 20:51 UTC (permalink / raw
To: gentoo-commits
commit: db81ee2a9a65c0aaebb4df461f8be388c66cdd16
Author: Andrew Ammerlaan <andrewammerlaan <AT> gentoo <DOT> org>
AuthorDate: Sat Mar 5 20:49:57 2022 +0000
Commit: Andrew Ammerlaan <andrewammerlaan <AT> gentoo <DOT> org>
CommitDate: Sat Mar 5 20:51:52 2022 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=db81ee2a
dev-util/Tensile: add new dependency of rocBLAS
Package-Manager: Portage-3.0.30, Repoman-3.0.3
Signed-off-by: Andrew Ammerlaan <andrewammerlaan <AT> gentoo.org>
dev-util/Tensile/Manifest | 1 +
dev-util/Tensile/Tensile-4.3.0.ebuild | 74 +++++++++++++++++
...ensile-4.3.0-hsaco-compile-specified-arch.patch | 96 ++++++++++++++++++++++
.../files/Tensile-4.3.0-output-commands.patch | 23 ++++++
dev-util/Tensile/metadata.xml | 11 +++
5 files changed, 205 insertions(+)
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
new file mode 100644
index 000000000000..fd4cac044af8
--- /dev/null
+++ b/dev-util/Tensile/Manifest
@@ -0,0 +1 @@
+DIST rocm-Tensile-4.3.0.tar.gz 14250149 BLAKE2B 030138eaca2a0aadd96801e6dbd72e510716dd90553ef3795c5e04e00a34a05ecae82b24f755e4033a4acfcdb1cf26291da1e7902bb090f89a010d403e832beb SHA512 126db0b413c716fba8d5be9bff7a44fd1badacbf32f3db8d0db649819177db37ebd56fd22dd3c809655f5d29675be115e698cd10bc3d0b4b23878ae3726fce47
diff --git a/dev-util/Tensile/Tensile-4.3.0.ebuild b/dev-util/Tensile/Tensile-4.3.0.ebuild
new file mode 100644
index 000000000000..cf33fb19cf10
--- /dev/null
+++ b/dev-util/Tensile/Tensile-4.3.0.ebuild
@@ -0,0 +1,74 @@
+# Copyright 1999-2022 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{8..10} )
+DISTUTILS_USE_PEP517=setuptools
+inherit distutils-r1
+
+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)"
+
+# Not compatible with recent versions of pytest
+RESTRICT="test"
+
+RDEPEND="${PYTHON_DEPS}
+ dev-python/pyyaml[${PYTHON_USEDEP}]
+ dev-python/msgpack[${PYTHON_USEDEP}]
+"
+DEPEND="${RDEPEND}
+ dev-util/hip:${SLOT}
+"
+BDEPEND="test? (
+ dev-util/rocminfo:${SLOT}
+)"
+
+PATCHES=(
+ "${FILESDIR}/Tensile-${PV}-hsaco-compile-specified-arch.patch" # backported from upstream, should remove after 4.3.0
+ "${FILESDIR}/Tensile-4.3.0-output-commands.patch"
+)
+
+CMAKE_USE_DIR="${WORKDIR}/Source"
+
+distutils_enable_tests pytest
+
+src_prepare() {
+ distutils-r1_src_prepare
+
+ mv ${PN}/Source "${WORKDIR}"/ || die
+ sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \
+ -i "${WORKDIR}"/Source/cmake/FindROCmSMI.cmake || die
+ sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \
+ -i "${WORKDIR}"/Source/CMakeLists.txt || die
+
+ sed -e "/HipClangVersion/s/0,0,0/$(ver_rs 1-3 ,)/" \
+ -e "/SourcePath/s,os\.path\.join.*$,\"${EPREFIX}/usr/share/${PN}\"," \
+ -i ${PN}/Common.py || die
+
+ sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile\", end='')|" \
+ -i ${PN}/__init__.py || die
+}
+
+src_test() {
+ ROCM_PATH="${EPREFIX}/usr/" distutils-r1_src_test
+}
+
+src_install() {
+ distutils-r1_src_install
+
+ # Move the cmake files to the correct directory
+ mkdir -p "${ED}/usr/$(get_libdir)/cmake/${PN}" || die
+ mv "${ED}/usr/cmake/"* "${ED}/usr/$(get_libdir)/cmake/${PN}" || die
+ rm -r "${ED}/usr/cmake" || die
+
+ insinto /usr/share/${PN}
+ doins -r "${WORKDIR}"/Source/*
+ dosym . /usr/share/${PN}/Source
+}
diff --git a/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch b/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch
new file mode 100644
index 000000000000..8e6753781a2a
--- /dev/null
+++ b/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch
@@ -0,0 +1,96 @@
+https://github.com/ROCmSoftwarePlatform/Tensile/issues/1395
+https://github.com/ROCmSoftwarePlatform/Tensile/pull/1398
+
+--- a/Tensile/TensileCreateLibrary.py
++++ b/Tensile/TensileCreateLibrary.py
+@@ -136,6 +136,35 @@ def which(p):
+ return candidate
+ return None
+
++def splitArchs():
++ # Helper for architecture
++ def isSupported(arch):
++ return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
++ globalParameters["AsmCaps"][arch]["SupportedSource"]
++
++ if ";" in globalParameters["Architecture"]:
++ wantedArchs = globalParameters["Architecture"].split(";")
++ else:
++ wantedArchs = globalParameters["Architecture"].split("_")
++ archs = []
++ cmdlineArchs = []
++ if "all" in wantedArchs:
++ for arch in globalParameters['SupportedISA']:
++ if isSupported(arch):
++ if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
++ if (arch == (9,0,10)):
++ archs += [gfxName(arch) + '-xnack+']
++ cmdlineArchs += [gfxName(arch) + ':xnack+']
++ archs += [gfxName(arch) + '-xnack-']
++ cmdlineArchs += [gfxName(arch) + ':xnack-']
++ else:
++ archs += [gfxName(arch)]
++ cmdlineArchs += [gfxName(arch)]
++ else:
++ for arch in wantedArchs:
++ archs += [re.sub(":", "-", arch)]
++ cmdlineArchs += [arch]
++ return archs, cmdlineArchs
+
+ def buildSourceCodeObjectFile(CxxCompiler, outputPath, kernelFile):
+ buildPath = ensurePath(os.path.join(globalParameters['WorkingPath'], 'code_object_tmp'))
+@@ -149,24 +178,8 @@ def buildSourceCodeObjectFile(CxxCompiler, outputPath, kernelFile):
+ objectFilename = base + '.o'
+ soFilename = base + '.so'
+
+- def isSupported(arch):
+- return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
+- globalParameters["AsmCaps"][arch]["SupportedSource"]
+-
+ if (CxxCompiler == "hipcc"):
+- archs = []
+- cmdlineArchs = []
+- for arch in globalParameters['SupportedISA']:
+- if isSupported(arch):
+- if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
+- if (arch == (9,0,10)):
+- archs += [gfxName(arch) + '-xnack+']
+- cmdlineArchs += [gfxName(arch) + ':xnack+']
+- archs += [gfxName(arch) + '-xnack-']
+- cmdlineArchs += [gfxName(arch) + ':xnack-']
+- else:
+- archs += [gfxName(arch)]
+- cmdlineArchs += [gfxName(arch)]
++ archs, cmdlineArchs = splitArchs()
+
+ archFlags = ['--offload-arch=' + arch for arch in cmdlineArchs]
+
+@@ -1063,11 +1076,6 @@ def buildObjectFileNames(solutionWriter, kernelWriterSource, kernelWriterAssembl
+ sourceKernels = list([k for k in kernels if k['KernelLanguage'] == 'Source'])
+ asmKernels = list([k for k in kernels if k['KernelLanguage'] == 'Assembly'])
+
+- # Helper for architecture
+- def isSupported(arch):
+- return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
+- globalParameters["AsmCaps"][arch]["SupportedSource"]
+-
+ # Build a list of kernel object names.
+ for kernel in sourceKernels:
+ sourceKernelNames += [kernelWriterSource.getKernelFileBase(kernel)]
+@@ -1081,15 +1089,7 @@ def buildObjectFileNames(solutionWriter, kernelWriterSource, kernelWriterAssembl
+
+ # Source based kernels are built for all supported architectures
+ if (cxxCompiler == 'hipcc'):
+- sourceArchs = []
+- for arch in globalParameters['SupportedISA']:
+- if isSupported(arch):
+- if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
+- if (arch == (9,0,10)):
+- sourceArchs += [gfxName(arch) + '-xnack+']
+- sourceArchs += [gfxName(arch) + '-xnack-']
+- else:
+- sourceArchs += [gfxName(arch)]
++ sourceArchs, _ = splitArchs()
+ else:
+ raise RuntimeError("Unknown compiler %s" % cxxCompiler)
+
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/metadata.xml b/dev-util/Tensile/metadata.xml
new file mode 100644
index 000000000000..9bbebec502ca
--- /dev/null
+++ b/dev-util/Tensile/metadata.xml
@@ -0,0 +1,11 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE pkgmetadata SYSTEM 'http://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>
+</pkgmetadata>
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2022-12-19 14:00 Andreas Sturmlechner
0 siblings, 0 replies; 6+ messages in thread
From: Andreas Sturmlechner @ 2022-12-19 14:00 UTC (permalink / raw
To: gentoo-commits
commit: 3a572264e28322d9b09f6f9475cca0eede4d5f3f
Author: Andreas Sturmlechner <asturm <AT> gentoo <DOT> org>
AuthorDate: Sun Dec 18 13:09:21 2022 +0000
Commit: Andreas Sturmlechner <asturm <AT> gentoo <DOT> org>
CommitDate: Mon Dec 19 14:00:02 2022 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=3a572264
dev-util/Tensile: drop 4.3.0-r1
Signed-off-by: Andreas Sturmlechner <asturm <AT> gentoo.org>
dev-util/Tensile/Manifest | 1 -
dev-util/Tensile/Tensile-4.3.0-r1.ebuild | 86 ---------------
.../files/Tensile-4.3.0-fix-arch-parse.patch | 12 ---
.../Tensile/files/Tensile-4.3.0-gentoopath.patch | 62 -----------
dev-util/Tensile/files/Tensile-4.3.0-gfx1031.patch | 119 ---------------------
...ensile-4.3.0-hsaco-compile-specified-arch.patch | 96 -----------------
.../Tensile/files/Tensile-4.3.0-use-ninja.patch | 21 ----
7 files changed, 397 deletions(-)
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index 3b51748fa904..c3f43afdabe4 100644
--- a/dev-util/Tensile/Manifest
+++ b/dev-util/Tensile/Manifest
@@ -1,4 +1,3 @@
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-4.3.0-r1.ebuild b/dev-util/Tensile/Tensile-4.3.0-r1.ebuild
deleted file mode 100644
index 7cb7e94cd256..000000000000
--- a/dev-util/Tensile/Tensile-4.3.0-r1.ebuild
+++ /dev/null
@@ -1,86 +0,0 @@
-# Copyright 1999-2022 Gentoo Authors
-# Distributed under the terms of the GNU General Public License v2
-
-EAPI=8
-
-PYTHON_COMPAT=( python3_{8..10} )
-DISTUTILS_USE_PEP517=setuptools
-inherit distutils-r1 prefix
-
-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/$(ver_cut 1-2)"
-
-# Not compatible with recent versions of pytest
-RESTRICT="test"
-
-RDEPEND="${PYTHON_DEPS}
- dev-python/pyyaml[${PYTHON_USEDEP}]
- dev-python/msgpack[${PYTHON_USEDEP}]
- >=dev-util/rocm-smi-4.3.0
-"
-DEPEND="${RDEPEND}
- dev-util/hip:${SLOT}
-"
-
-PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.patch
- "${FILESDIR}"/${PN}-4.3.0-hsaco-compile-specified-arch.patch
- "${FILESDIR}"/${PN}-4.3.0-gfx1031.patch
- "${FILESDIR}"/${PN}-4.3.0-fix-arch-parse.patch
- "${FILESDIR}"/${PN}-4.3.0-use-ninja.patch
- )
-
-src_prepare() {
- distutils-r1_src_prepare
- eapply $(prefixify_ro "${FILESDIR}"/${PN}-4.3.0-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)/" \
- -e "/SourcePath/s,globalParameters\[\"ScriptPath\"\],${Tensile_share_dir}," \
- -i Common.py || die
-
- sed -e "/CMAKE_CXX_COMPILER/s,globalParameters\[\"ROCmBinPath\"\],\"${EPREFIX}/usr/lib/hip/bin\"," -i ClientExecutable.py || die
-
- sed -e "/scriptDir/s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir}," -i ReplacementKernels.py || die
-
- sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" -i ${PN}.py || die
-
- sed -e "s|os\.path\.dirname.*$|\"${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
- insinto /usr/$(get_libdir)/cmake/${PN}
- doins cmake/*.cmake
-}
diff --git a/dev-util/Tensile/files/Tensile-4.3.0-fix-arch-parse.patch b/dev-util/Tensile/files/Tensile-4.3.0-fix-arch-parse.patch
deleted file mode 100644
index 01b94b32cb6e..000000000000
--- a/dev-util/Tensile/files/Tensile-4.3.0-fix-arch-parse.patch
+++ /dev/null
@@ -1,12 +0,0 @@
-Index: Tensile-rocm-4.3.0/Tensile/Tensile.py
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/Tensile.py
-+++ Tensile-rocm-4.3.0/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-4.3.0-gentoopath.patch b/dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch
deleted file mode 100644
index da135e3b38ee..000000000000
--- a/dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch
+++ /dev/null
@@ -1,62 +0,0 @@
-Index: Tensile-rocm-4.3.0/Tensile/Source/client/CMakeLists.txt
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/Source/client/CMakeLists.txt
-+++ Tensile-rocm-4.3.0/Tensile/Source/client/CMakeLists.txt
-@@ -50,3 +50,5 @@ 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-4.3.0/Tensile/Source/lib/CMakeLists.txt
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/Source/lib/CMakeLists.txt
-+++ Tensile-rocm-4.3.0/Tensile/Source/lib/CMakeLists.txt
-@@ -48,7 +48,7 @@ set(tensile_sources ${tensile_sources}
- )
-
- if(TENSILE_USE_LLVM)
-- find_package(LLVM 13.0 QUIET CONFIG)
-+ find_package(LLVM PATHS @GENTOO_PORTAGE_EPREFIX@/usr/lib/llvm/roc QUIET CONFIG)
- if(NOT LLVM_FOUND)
- find_package(LLVM 12.0 QUIET CONFIG)
- if(NOT LLVM_FOUND)
-Index: Tensile-rocm-4.3.0/Tensile/Common.py
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/Common.py
-+++ Tensile-rocm-4.3.0/Tensile/Common.py
-@@ -1667,7 +1667,7 @@ 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-4.3.0/Tensile/ClientWriter.py
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/ClientWriter.py
-+++ Tensile-rocm-4.3.0/Tensile/ClientWriter.py
-@@ -284,7 +284,7 @@ def getBuildNewClientLibraryScript(build
- import io
- runScriptFile = io.StringIO()
-
-- callCreateLibraryCmd = globalParameters["ScriptPath"] + "/bin/TensileCreateLibrary"
-+ callCreateLibraryCmd = "TensileCreateLibrary"
-
-
- if globalParameters["MergeFiles"]:
-Index: Tensile-rocm-4.3.0/Tensile/GenerateSummations.py
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/GenerateSummations.py
-+++ Tensile-rocm-4.3.0/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", "--no-legacy-components", \
- "--new-client-only", "--no-short-file-names", "--no-library-print-debug", "--architecture=all", \
diff --git a/dev-util/Tensile/files/Tensile-4.3.0-gfx1031.patch b/dev-util/Tensile/files/Tensile-4.3.0-gfx1031.patch
deleted file mode 100644
index 99abceb7c44d..000000000000
--- a/dev-util/Tensile/files/Tensile-4.3.0-gfx1031.patch
+++ /dev/null
@@ -1,119 +0,0 @@
---- Tensile-rocm-4.3.0/pytest.ini 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/pytest.ini 2021-09-30 17:32:21.382806851 +0800
-@@ -88,6 +88,7 @@
- xfail-gfx1011: architecture
- xfail-gfx1012: architecture
- xfail-gfx1030: architecture
-+ xfail-gfx1031: architecture
- skip-gfx000: architecture
- skip-gfx900: architecture
- skip-gfx906: architecture
-@@ -97,3 +98,4 @@
- skip-gfx1011: architecture
- skip-gfx1012: architecture
- skip-gfx1030: architecture
-+ skip-gfx1031: architecture
---- Tensile-rocm-4.3.0/Tensile/Common.py 2021-10-05 14:41:22.490754124 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Common.py 2021-09-30 17:33:19.041724015 +0800
-@@ -200,7 +200,7 @@
- globalParameters["MergeFiles"] = True # F=store every solution and kernel in separate file; T=store all solutions in single file
-
- 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),(10,3,1)] # 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)] # assembly kernels writer supports these architectures
-
- globalParameters["GenerateManifestAndExit"] = False # Output manifest file with list of expected library objects and exit
- globalParameters["ClientBuildPath"] = "0_Build" # subdirectory for host code build directory
-@@ -265,7 +265,7 @@
- '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':'navi11', 'gfx1012':'navi12', 'gfx1030':'navi21', 'gfx1031':'navi22',
-- 'gfx1010':'navi10', 'gfx1011':'navi11', 'gfx1012':'navi12', 'gfx1030':'navi21'
- }
-
- def getArchitectureName(gfxName):
---- Tensile-rocm-4.3.0/Tensile/KernelWriterAssembly.py 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/KernelWriterAssembly.py 2021-09-30 17:27:06.696711144 +0800
-@@ -2257,7 +2257,7 @@
-
- # 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) or self.version == (10,3,1):
-- if self.version == (9,0,6) or self.version == (9,0,8) or self.version == (9,0,10) or self.version == (10,3,0):
- for b in range(0, kernel["ThreadTile1"]):
- for a in range(0, kernel["ThreadTile0"]):
- for iui in range(0, innerUnroll):
---- Tensile-rocm-4.3.0/Tensile/Source/CMakeLists.txt 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Source/CMakeLists.txt 2021-09-30 17:29:12.846345541 +0800
-@@ -51,7 +51,7 @@
- endif()
-
- if(CMAKE_CXX_COMPILER STREQUAL "hipcc")
-+ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 CACHE STRING "GPU architectures")
-- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 CACHE STRING "GPU architectures")
- else()
- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906 gfx908 gfx90a CACHE STRING "GPU architectures")
- endif()
-@@ -104,7 +104,7 @@
- set_property( CACHE Tensile_CODE_OBJECT_VERSION PROPERTY STRINGS V2 V3)
- set_property( CACHE Tensile_COMPILER PROPERTY STRINGS hipcc)
- set(Tensile_ARCHITECTURE all CACHE STRING "Which GPU architecture to use")
-+ set_property( CACHE Tensile_ARCHITECTURE PROPERTY STRINGS all gfx000 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031)
-- set_property( CACHE Tensile_ARCHITECTURE PROPERTY STRINGS all gfx000 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030)
- set_property( CACHE Tensile_LIBRARY_FORMAT PROPERTY STRINGS yaml msgpack)
-
- message( STATUS "Tensile_CODE_OBJECT_VERSION from Tensile/Source/CMakeLists.txt: ${Tensile_CODE_OBJECT_VERSION}")
-@@ -206,7 +206,7 @@
- ${Tensile_RUNTIME_LANGUAGE} # HIP or OCL
- ${Tensile_COMPILER} # hipcc
- ${Tensile_CODE_OBJECT_VERSION} # V2 or V3
-+ ${Tensile_ARCHITECTURE} # all, gfx000, gfx803, gfx900, gfx906:xnack-, gfx908:xnack-, gfx90a:xnack-, gfx1010, gfx1011, gfx1012, gfx1030, gfx1031
-- ${Tensile_ARCHITECTURE} # all, gfx000, gfx803, gfx900, gfx906:xnack-, gfx908:xnack-, gfx90a:xnack-, gfx1010, gfx1011, gfx1012, gfx1030
- ${Tensile_LIBRARY_FORMAT} # yaml or msgpack
- ${Tensile_MERGE_FILES} # ON or OFF
- ${Tensile_SHORT_FILE_NAMES} # ON or OFF
---- Tensile-rocm-4.3.0/Tensile/Source/lib/include/Tensile/AMDGPU.hpp 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Source/lib/include/Tensile/AMDGPU.hpp 2021-09-30 17:48:11.277066515 +0800
-@@ -62,7 +62,8 @@
- gfx1010 = 1010,
- gfx1011 = 1011,
- gfx1012 = 1012,
-+ gfx1030 = 1030,
-+ gfx1031 = 1031
-- gfx1030 = 1030
- };
-
- AMDGPU();
---- Tensile-rocm-4.3.0/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp 2021-09-30 17:31:26.075845249 +0800
-@@ -215,6 +215,7 @@
- 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
---- Tensile-rocm-4.3.0/Tensile/Source/lib/source/AMDGPU.cpp 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Source/lib/source/AMDGPU.cpp 2021-09-30 17:31:45.195486318 +0800
-@@ -82,6 +82,8 @@
- return stream << "gfx1012";
- case AMDGPU::Processor::gfx1030:
- return stream << "gfx1030";
-+ case AMDGPU::Processor::gfx1031:
-+ return stream << "gfx1031";
- }
- return stream;
- }
---- Tensile-rocm-4.3.0/Tensile/Source/TensileCreateLibrary.cmake 2021-04-22 22:35:33.000000000 +0800
-+++ Tensile-rocm-4.3.0/Tensile/Source/TensileCreateLibrary.cmake 2021-09-30 17:29:24.766121912 +0800
-@@ -138,7 +138,7 @@
- add_library(Tensile ${options} ${Tensile_SOURCE_FILES})
- # specify gpu targets
- if( Tensile_ARCHITECTURE MATCHES "all" )
-+ set( Tensile_HIP_ISA "gfx803" "gfx900" "gfx906" "gfx908" "gfx1010" "gfx1011" "gfx1012" "gfx1030", "gfx1031")
-- set( Tensile_HIP_ISA "gfx803" "gfx900" "gfx906" "gfx908" "gfx1010" "gfx1011" "gfx1012" "gfx1030")
- else()
- set( Tensile_HIP_ISA ${Tensile_ARCHITECTURE})
- endif()
diff --git a/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch b/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch
deleted file mode 100644
index 8e6753781a2a..000000000000
--- a/dev-util/Tensile/files/Tensile-4.3.0-hsaco-compile-specified-arch.patch
+++ /dev/null
@@ -1,96 +0,0 @@
-https://github.com/ROCmSoftwarePlatform/Tensile/issues/1395
-https://github.com/ROCmSoftwarePlatform/Tensile/pull/1398
-
---- a/Tensile/TensileCreateLibrary.py
-+++ b/Tensile/TensileCreateLibrary.py
-@@ -136,6 +136,35 @@ def which(p):
- return candidate
- return None
-
-+def splitArchs():
-+ # Helper for architecture
-+ def isSupported(arch):
-+ return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
-+ globalParameters["AsmCaps"][arch]["SupportedSource"]
-+
-+ if ";" in globalParameters["Architecture"]:
-+ wantedArchs = globalParameters["Architecture"].split(";")
-+ else:
-+ wantedArchs = globalParameters["Architecture"].split("_")
-+ archs = []
-+ cmdlineArchs = []
-+ if "all" in wantedArchs:
-+ for arch in globalParameters['SupportedISA']:
-+ if isSupported(arch):
-+ if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
-+ if (arch == (9,0,10)):
-+ archs += [gfxName(arch) + '-xnack+']
-+ cmdlineArchs += [gfxName(arch) + ':xnack+']
-+ archs += [gfxName(arch) + '-xnack-']
-+ cmdlineArchs += [gfxName(arch) + ':xnack-']
-+ else:
-+ archs += [gfxName(arch)]
-+ cmdlineArchs += [gfxName(arch)]
-+ else:
-+ for arch in wantedArchs:
-+ archs += [re.sub(":", "-", arch)]
-+ cmdlineArchs += [arch]
-+ return archs, cmdlineArchs
-
- def buildSourceCodeObjectFile(CxxCompiler, outputPath, kernelFile):
- buildPath = ensurePath(os.path.join(globalParameters['WorkingPath'], 'code_object_tmp'))
-@@ -149,24 +178,8 @@ def buildSourceCodeObjectFile(CxxCompiler, outputPath, kernelFile):
- objectFilename = base + '.o'
- soFilename = base + '.so'
-
-- def isSupported(arch):
-- return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
-- globalParameters["AsmCaps"][arch]["SupportedSource"]
--
- if (CxxCompiler == "hipcc"):
-- archs = []
-- cmdlineArchs = []
-- for arch in globalParameters['SupportedISA']:
-- if isSupported(arch):
-- if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
-- if (arch == (9,0,10)):
-- archs += [gfxName(arch) + '-xnack+']
-- cmdlineArchs += [gfxName(arch) + ':xnack+']
-- archs += [gfxName(arch) + '-xnack-']
-- cmdlineArchs += [gfxName(arch) + ':xnack-']
-- else:
-- archs += [gfxName(arch)]
-- cmdlineArchs += [gfxName(arch)]
-+ archs, cmdlineArchs = splitArchs()
-
- archFlags = ['--offload-arch=' + arch for arch in cmdlineArchs]
-
-@@ -1063,11 +1076,6 @@ def buildObjectFileNames(solutionWriter, kernelWriterSource, kernelWriterAssembl
- sourceKernels = list([k for k in kernels if k['KernelLanguage'] == 'Source'])
- asmKernels = list([k for k in kernels if k['KernelLanguage'] == 'Assembly'])
-
-- # Helper for architecture
-- def isSupported(arch):
-- return globalParameters["AsmCaps"][arch]["SupportedISA"] and \
-- globalParameters["AsmCaps"][arch]["SupportedSource"]
--
- # Build a list of kernel object names.
- for kernel in sourceKernels:
- sourceKernelNames += [kernelWriterSource.getKernelFileBase(kernel)]
-@@ -1081,15 +1089,7 @@ def buildObjectFileNames(solutionWriter, kernelWriterSource, kernelWriterAssembl
-
- # Source based kernels are built for all supported architectures
- if (cxxCompiler == 'hipcc'):
-- sourceArchs = []
-- for arch in globalParameters['SupportedISA']:
-- if isSupported(arch):
-- if (arch == (9,0,6) or arch == (9,0,8) or arch == (9,0,10)):
-- if (arch == (9,0,10)):
-- sourceArchs += [gfxName(arch) + '-xnack+']
-- sourceArchs += [gfxName(arch) + '-xnack-']
-- else:
-- sourceArchs += [gfxName(arch)]
-+ sourceArchs, _ = splitArchs()
- else:
- raise RuntimeError("Unknown compiler %s" % cxxCompiler)
-
diff --git a/dev-util/Tensile/files/Tensile-4.3.0-use-ninja.patch b/dev-util/Tensile/files/Tensile-4.3.0-use-ninja.patch
deleted file mode 100644
index 6645efd72fb1..000000000000
--- a/dev-util/Tensile/files/Tensile-4.3.0-use-ninja.patch
+++ /dev/null
@@ -1,21 +0,0 @@
-Index: Tensile-rocm-4.3.0/Tensile/ClientExecutable.py
-===================================================================
---- Tensile-rocm-4.3.0.orig/Tensile/ClientExecutable.py
-+++ Tensile-rocm-4.3.0/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)
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2023-01-31 14:30 Benda XU
0 siblings, 0 replies; 6+ messages in thread
From: Benda XU @ 2023-01-31 14:30 UTC (permalink / raw
To: gentoo-commits
commit: 3b32497070812f973e5064f9ada07ab5d404b402
Author: Yiyang Wu <xgreenlandforwyy <AT> gmail <DOT> com>
AuthorDate: Fri Jan 20 09:58:17 2023 +0000
Commit: Benda XU <heroxbd <AT> gentoo <DOT> org>
CommitDate: Tue Jan 31 14:26:37 2023 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=3b324970
dev-util/Tensile: add 5.4.2
Signed-off-by: Yiyang Wu <xgreenlandforwyy <AT> gmail.com>
Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org>
dev-util/Tensile/Manifest | 1 +
dev-util/Tensile/Tensile-5.4.2.ebuild | 89 ++++++++++++++++++++++
.../Tensile/files/Tensile-5.4.2-gentoopath.patch | 67 ++++++++++++++++
.../Tensile/files/Tensile-5.4.2-use-ninja.patch | 21 +++++
4 files changed, 178 insertions(+)
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index c3f43afdabe4..dd1df20ec814 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-5.0.2.tar.gz 14362090 BLAKE2B d44c3135496db41c4eb2545f6054c84493e89ef6357da881dae6859763338417e7a9e142cdce7e2de0536684bb641049ad0440964adbe68d03ca329f01efabc9 SHA512 f5ab7d7cbc5ba25498a02a6b7c07d5a9a548edcd44b439b92c8718072382adc18ae78274db1a7174e06436ac1396f3b4edafc7f491a2c87086c368e7270e64e3
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
diff --git a/dev-util/Tensile/Tensile-5.4.2.ebuild b/dev-util/Tensile/Tensile-5.4.2.ebuild
new file mode 100644
index 000000000000..09bccd0ab2dd
--- /dev/null
+++ b/dev-util/Tensile/Tensile-5.4.2.ebuild
@@ -0,0 +1,89 @@
+# 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
+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-${PV}.tar.gz"
+S="${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}
+ sys-devel/clang:${LLVM_MAX_SLOT}
+ dev-python/pyyaml[${PYTHON_USEDEP}]
+ dev-python/msgpack[${PYTHON_USEDEP}]
+ >=dev-util/rocm-smi-4.3.0
+"
+DEPEND="${RDEPEND}
+ dev-util/hip
+"
+
+PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.patch
+ "${FILESDIR}"/${PN}-5.3.3-gfx1031.patch
+ "${FILESDIR}"/${PN}-5.3.3-fix-arch-parse.patch
+ "${FILESDIR}"/${PN}-5.4.2-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}"/${PN}-5.4.2-gentoopath.patch > "${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
+}
+
+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-5.4.2-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.4.2-gentoopath.patch
new file mode 100644
index 000000000000..2aed8082de05
--- /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-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
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2023-02-01 1:52 Benda XU
0 siblings, 0 replies; 6+ messages in thread
From: Benda XU @ 2023-02-01 1:52 UTC (permalink / raw
To: gentoo-commits
commit: 2b7c870b48d225a9ade51eac426310de68f41069
Author: Yiyang Wu <xgreenlandforwyy <AT> gmail <DOT> com>
AuthorDate: Tue Jan 31 14:57:06 2023 +0000
Commit: Benda XU <heroxbd <AT> gentoo <DOT> org>
CommitDate: Wed Feb 1 01:51:35 2023 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=2b7c870b
dev-util/Tensile: add missing patches
Closes: https://bugs.gentoo.org/892736
Closes: https://github.com/gentoo/gentoo/pull/29356
Signed-off-by: Yiyang Wu <xgreenlandforwyy <AT> gmail.com>
Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org>
dev-util/Tensile/Tensile-5.4.2.ebuild | 4 +-
.../files/Tensile-5.4.2-fix-arch-parse.patch | 12 +++
dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch | 88 ++++++++++++++++++++++
3 files changed, 102 insertions(+), 2 deletions(-)
diff --git a/dev-util/Tensile/Tensile-5.4.2.ebuild b/dev-util/Tensile/Tensile-5.4.2.ebuild
index 09bccd0ab2dd..1e347fdcee56 100644
--- a/dev-util/Tensile/Tensile-5.4.2.ebuild
+++ b/dev-util/Tensile/Tensile-5.4.2.ebuild
@@ -32,8 +32,8 @@ DEPEND="${RDEPEND}
"
PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.patch
- "${FILESDIR}"/${PN}-5.3.3-gfx1031.patch
- "${FILESDIR}"/${PN}-5.3.3-fix-arch-parse.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
)
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-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);
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2023-03-30 22:41 Marek Szuba
0 siblings, 0 replies; 6+ messages in thread
From: Marek Szuba @ 2023-03-30 22:41 UTC (permalink / raw
To: gentoo-commits
commit: a0035c9a825799ec5df594b8b7e3184fbf8cdd4f
Author: Marek Szuba <marecki <AT> gentoo <DOT> org>
AuthorDate: Thu Mar 30 22:37:21 2023 +0000
Commit: Marek Szuba <marecki <AT> gentoo <DOT> org>
CommitDate: Thu Mar 30 22:41:42 2023 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=a0035c9a
dev-util/Tensile: adapt for msgpack-cxx-6.0.0
The name of the cmake module has changed again.
Signed-off-by: Marek Szuba <marecki <AT> gentoo.org>
...ile-5.1.3-r2.ebuild => Tensile-5.1.3-r3.ebuild} | 15 +++++++------
...ile-5.4.2-r1.ebuild => Tensile-5.4.2-r2.ebuild} | 15 +++++++------
...1-Change-cmake-name-for-msgpack-5-release.patch | 25 ----------------------
...ange-cmake-name-for-msgpack-cxx-6-release.patch | 13 +++++++++++
4 files changed, 29 insertions(+), 39 deletions(-)
diff --git a/dev-util/Tensile/Tensile-5.1.3-r2.ebuild b/dev-util/Tensile/Tensile-5.1.3-r3.ebuild
similarity index 89%
rename from dev-util/Tensile/Tensile-5.1.3-r2.ebuild
rename to dev-util/Tensile/Tensile-5.1.3-r3.ebuild
index 2f487059341f..da1adcab4f7b 100644
--- a/dev-util/Tensile/Tensile-5.1.3-r2.ebuild
+++ b/dev-util/Tensile/Tensile-5.1.3-r3.ebuild
@@ -23,7 +23,7 @@ SLOT="0/$(ver_cut 1-2)"
RESTRICT="test"
RDEPEND="${PYTHON_DEPS}
- dev-cpp/msgpack-cxx
+ >=dev-cpp/msgpack-cxx-6.0.0
dev-python/pyyaml[${PYTHON_USEDEP}]
dev-python/msgpack[${PYTHON_USEDEP}]
dev-util/hip
@@ -32,12 +32,13 @@ RDEPEND="${PYTHON_DEPS}
"
DEPEND="${RDEPEND}"
-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
- "${FILESDIR}"/0001-Change-cmake-name-for-msgpack-5-release.patch
- )
+PATCHES=(
+ "${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"
diff --git a/dev-util/Tensile/Tensile-5.4.2-r1.ebuild b/dev-util/Tensile/Tensile-5.4.2-r2.ebuild
similarity index 91%
rename from dev-util/Tensile/Tensile-5.4.2-r1.ebuild
rename to dev-util/Tensile/Tensile-5.4.2-r2.ebuild
index 8c23e68fdb21..c846dce60997 100644
--- a/dev-util/Tensile/Tensile-5.4.2-r1.ebuild
+++ b/dev-util/Tensile/Tensile-5.4.2-r2.ebuild
@@ -25,7 +25,7 @@ REQUIRED_USE="client? ( ${ROCM_REQUIRED_USE} )"
RESTRICT="test"
RDEPEND="${PYTHON_DEPS}
- dev-cpp/msgpack-cxx
+ >=dev-cpp/msgpack-cxx-6.0.0
dev-python/pyyaml[${PYTHON_USEDEP}]
dev-python/msgpack[${PYTHON_USEDEP}]
dev-util/hip
@@ -34,12 +34,13 @@ RDEPEND="${PYTHON_DEPS}
"
DEPEND="${RDEPEND}"
-PATCHES=( "${FILESDIR}"/${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
- "${FILESDIR}"/0001-Change-cmake-name-for-msgpack-5-release.patch
- )
+PATCHES=(
+ "${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"
diff --git a/dev-util/Tensile/files/0001-Change-cmake-name-for-msgpack-5-release.patch b/dev-util/Tensile/files/0001-Change-cmake-name-for-msgpack-5-release.patch
deleted file mode 100644
index 7f8fb95123ee..000000000000
--- a/dev-util/Tensile/files/0001-Change-cmake-name-for-msgpack-5-release.patch
+++ /dev/null
@@ -1,25 +0,0 @@
-From 66f5f943bbd0fd3b41b02ce7a563bed9f59f9c38 Mon Sep 17 00:00:00 2001
-From: Yiyang Wu <xgreenlandforwyy@gmail.com>
-Date: Sat, 11 Feb 2023 15:37:11 +0800
-Subject: [PATCH] Change cmake name for msgpack-5 release
-
----
- Tensile/Source/lib/CMakeLists.txt | 2 +-
- 1 file changed, 1 insertion(+), 1 deletion(-)
-
-diff --git a/Tensile/Source/lib/CMakeLists.txt b/Tensile/Source/lib/CMakeLists.txt
-index 47a5d15d..f1a6118b 100644
---- 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(msgpackc-cxx REQUIRED)
- target_compile_definitions(TensileHost PUBLIC -DTENSILE_MSGPACK=1)
-
- if(TARGET msgpackc-cxx)
---
-2.39.1
-
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)
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/
@ 2023-11-27 14:35 Benda XU
0 siblings, 0 replies; 6+ messages in thread
From: Benda XU @ 2023-11-27 14:35 UTC (permalink / raw
To: gentoo-commits
commit: 02cb2430fcb5dbc70df96be5380228a8e47c8d80
Author: Sv. Lockal <lockalsash <AT> gmail <DOT> com>
AuthorDate: Wed Oct 18 18:42:05 2023 +0000
Commit: Benda XU <heroxbd <AT> gentoo <DOT> org>
CommitDate: Mon Nov 27 14:34:53 2023 +0000
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=02cb2430
dev-util/Tensile: add 5.7.1
increase LLVM_MAX_SLOT to 17
Signed-off-by: Sv. Lockal <lockalsash <AT> gmail.com>
Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org>
dev-util/Tensile/Manifest | 1 +
dev-util/Tensile/Tensile-5.7.1-r1.ebuild | 132 +++++++++++++++++++++
.../Tensile-5.7.1-expand-isa-compatibility.patch | 49 ++++++++
.../Tensile/files/Tensile-5.7.1-gentoopath.patch | 57 +++++++++
4 files changed, 239 insertions(+)
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index 643f4fe5816c..4d93ffa204a1 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-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-5.7.1-r1.ebuild b/dev-util/Tensile/Tensile-5.7.1-r1.ebuild
new file mode 100644
index 000000000000..709189734d4a
--- /dev/null
+++ b/dev-util/Tensile/Tensile-5.7.1-r1.ebuild
@@ -0,0 +1,132 @@
+# Copyright 1999-2023 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}
+ >=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}"
+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}"/${PN}-change-cmake-name-for-msgpack-cxx-6-release.patch
+ "${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}"/${PN}-5.7.1-expand-isa-compatibility.patch
+)
+
+CMAKE_USE_DIR="${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-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-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
^ permalink raw reply related [flat|nested] 6+ messages in thread
end of thread, other threads:[~2023-11-27 14:35 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-01-31 14:30 [gentoo-commits] repo/gentoo:master commit in: dev-util/Tensile/, dev-util/Tensile/files/ Benda XU
-- strict thread matches above, loose matches on Subject: below --
2023-11-27 14:35 Benda XU
2023-03-30 22:41 Marek Szuba
2023-02-01 1:52 Benda XU
2022-12-19 14:00 Andreas Sturmlechner
2022-03-05 20:51 Andrew Ammerlaan
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox