summaryrefslogtreecommitdiff
path: root/dev-util/Tensile
diff options
context:
space:
mode:
authorV3n3RiX <venerix@koprulu.sector>2022-04-06 22:33:41 +0100
committerV3n3RiX <venerix@koprulu.sector>2022-04-06 22:33:41 +0100
commite68d405c5d712af4387159df07e226217bdda049 (patch)
tree009ab0f3d427f0813e62930d71802cb054c07e30 /dev-util/Tensile
parent401101f9c8077911929d3f2b60a37098460a5d89 (diff)
gentoo resync : 06.04.2022
Diffstat (limited to 'dev-util/Tensile')
-rw-r--r--dev-util/Tensile/Manifest11
-rw-r--r--dev-util/Tensile/Tensile-4.3.0-r1.ebuild6
-rw-r--r--dev-util/Tensile/Tensile-5.0.2.ebuild83
-rw-r--r--dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch4
-rw-r--r--dev-util/Tensile/files/Tensile-5.0.2-fix-arch-parse.patch12
-rw-r--r--dev-util/Tensile/files/Tensile-5.0.2-gentoopath.patch75
-rw-r--r--dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch104
-rw-r--r--dev-util/Tensile/files/Tensile-5.0.2-use-ninja.patch21
8 files changed, 309 insertions, 7 deletions
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest
index 6972e0b87371..a7bde65036ef 100644
--- a/dev-util/Tensile/Manifest
+++ b/dev-util/Tensile/Manifest
@@ -1,9 +1,16 @@
AUX Tensile-4.3.0-fix-arch-parse.patch 424 BLAKE2B ac89228abe3173a4cabcb90a679be3706a39cf6bebc448fb17dc9a94be5c5c75289e53fb867c17a3bc36330079905302a83736663d2f743c9805c97253b9fec0 SHA512 131e9fa7c39f12fbbbc37ace0d29162ff2805e8435985d3edab7e678d89587a2ee9cc1549eee21842c42e99e4a4fba3125fa34fec3dbc9f61cff96d8eeaf5b25
-AUX Tensile-4.3.0-gentoopath.patch 2770 BLAKE2B 72f5d6ae4f447603f2e02bba8ac25bbc85382327ac4e15c2978c2bdd5ab7ff937e71e1dab006d548fd104eb4401b878ba5a67acfc535b456eee532836c90eb1c SHA512 030c616fc12fc25d4749f6bd9914afd6be6844ba44796e928183031e6f5dd98df238379a7b26225306233798b5d32ab1b4cf28f39775b2b7a17040c4f704106f
+AUX Tensile-4.3.0-gentoopath.patch 2796 BLAKE2B 10c90d69ea4c83cd36f1f61be8c716e66eba889bd10a221c4af895d30f04d851c1eba4c3d6c845d96d0e14a78f617e55a4acdd2f93db9c5fcbaeffd50ab5c6f1 SHA512 e14ca9c94f302863c1b0f9eca86ab3a7dbed9c61d1488255a5c9c0d458c800e7329a41f7b91694558a733632bee025dee5bb59a8478982a3ab1d591f4cb3a1e4
AUX Tensile-4.3.0-gfx1031.patch 7037 BLAKE2B ea2e5dec9d5f6749a78cae4d1659a08b712e548cdca29dabde29f52a60f443eac0c10bac27667545816e5b6b89835ee5e79d766028bad79ba17e5b14a4bf5ea9 SHA512 2b1936a6e5b52bf2f6aaa70255be9f78e82329f7855a3db56d769fed13934ca2c2671f089bc5aea43c4a0e67d3ec8ad185e32f7ade03c515b3c74a683568b192
AUX Tensile-4.3.0-hsaco-compile-specified-arch.patch 3859 BLAKE2B 4590978cff0fdb0501ba2405615775cb3f8900ce164c3d57cc59dfe82e1adc25376b6463855a045f5e0652cc985b0015566b95ff28327c59f2db875a84cb49b3 SHA512 57520f1d46f46c7cfd5b8c3ae307da07695c19479b687ce994ad9fba2e755ed7c776c3d8bee8e0642c7613a5d53d8469f329b1eb07563c13d2f7bda8b3d36687
AUX Tensile-4.3.0-output-commands.patch 1629 BLAKE2B c43395d8e8aeb956d5eb84c8a34ad4b995ed32499ceababeb22d5ca3ac5ee6e8ef353a1467a77943a441f0840b9a6ad1f53202ec8673ef3753aa2bec1395fec4 SHA512 45bdae1d87f6e31d337b22f146b9dee25f4f87357c6ac52bd843556dd742431db10efb55fb8ef815cc9941b5b9f05970d43a6fec0f2d9195c30c633e33d138bf
AUX Tensile-4.3.0-use-ninja.patch 957 BLAKE2B cb919b224aaeb3b9d65a755ff0688aa823b70361d2ee940d62f33828c118a60c88cd0f108b6e34b22097d9c4c5122a4db2808332a97cd6e7685a4447570b7552 SHA512 3d206a093e870fc95f0ed42e08691144589112d4581f73ba1c4c74ed9bdad0812ec4f67a6076a8a3ac9ab89c61c70f1baaedb774e0908177473be2d206858bf4
+AUX Tensile-5.0.2-fix-arch-parse.patch 424 BLAKE2B 43e716ccc9256c2f98cc1d2807e776599b384875161233d4a010a53f11a5764fd7a5b4c50ded1762737e9fbe7923b64e9353ee4590fd1b6b9b03ac621af35c9b SHA512 5913e7fc42bc5b43e1dcca531e4ef88b524325b497601171a9844013906c2ffc87469a4c55037c721ace5757c6aaf4d8bf8ca683f659f9d8988555c90425802a
+AUX Tensile-5.0.2-gentoopath.patch 3410 BLAKE2B bbf2c406bbcf2c34efd04667788719f6f10aebda3f9d9096e6798c1a8924768df4fed30223e2f4675af88c41841d148e43e3cd63edc7523dab4618ae245b0339 SHA512 c7b99eab1f0c979f54fcac3b2dcb4a2094b9a536bb392fcf4fb62050b1dc6e61f5280d3d1c7e3ed21617ebe693a17eb2bb7535a93f700123fe6aa6f28f256514
+AUX Tensile-5.0.2-gfx1031.patch 5561 BLAKE2B 244bb644a90ba34f500f4d0cc634acf041cce42f9a5a9f10ee7bd1feff3b78b626b56f3b8cfa4005cc422ef643cd60a7da954618ff9f8f460568ff4b011136bc SHA512 1cd1dd95e481a76fde21dd09aac1ac02f1372e552eeb45eb100dc970639e28f52a709735aca6480d2286b8c747a75435aa7f2e2a288ffe2a596d6f75e4bd17ec
+AUX Tensile-5.0.2-use-ninja.patch 897 BLAKE2B f06a3831a43a4b6c00e7eb231cc13901f7b4a3a924a50d85ba81f9def535951c450a01504506476c1ee88317cbdf0c0c00a57cb5cd198ab44e8e0b94aefda25d SHA512 15404d3cd5dbc7b74e81092d86f9c49cd45fe4df645602849ff515baafa3890fd68ca808cbebd6ae37e951b382ff4446a31089306c950f8549431dfbe75c3b53
+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
-EBUILD Tensile-4.3.0-r1.ebuild 2730 BLAKE2B ecbed69974d2d179cf0309c8265d394dfbfcd4ea9ef77448f4d4553aa51569bb5dced2c3404b387621e42037e62ed28b4de9ca0db472e4265f0f6b28e1dd630a SHA512 136eb9243bde86b66509d6e91b78a74a1e50727387eef515726efa9258f9e94a503a272bad4e5eeab30867b58522441d50a3dbcfec496197c3d2feee8469b2dd
+DIST rocm-Tensile-5.0.2.tar.gz 14362090 BLAKE2B d44c3135496db41c4eb2545f6054c84493e89ef6357da881dae6859763338417e7a9e142cdce7e2de0536684bb641049ad0440964adbe68d03ca329f01efabc9 SHA512 f5ab7d7cbc5ba25498a02a6b7c07d5a9a548edcd44b439b92c8718072382adc18ae78274db1a7174e06436ac1396f3b4edafc7f491a2c87086c368e7270e64e3
+EBUILD Tensile-4.3.0-r1.ebuild 2764 BLAKE2B e24ddb516d8dfd9e5e691f567f5b97b279e518b38be49a2cc69e8eae100d4d2ce7dc1432247c0bd5c94656ff1a7faf88a8e8f2e8df3dbbbc491eb4323cc20706 SHA512 79859e7dd5647bc4c1eb766faa9b3116994ed74aeb900caaf1b524a1216dcadd8d0c608d097a0c4c1d51b3117b75ccf7a061b7624cad6390fd5ce37494d67987
+EBUILD Tensile-5.0.2.ebuild 2588 BLAKE2B 6c96777b8e08bf2cc70ec1d4bc77a97ea311dd3612d21ae3c8ea364d14287e982f0d2efe0c41b14e1108f64842c56f7ea202cddbb89cceaa57a8990d866ca6cb SHA512 606b503214c13747877380b6b91f80e780d83e9ade7ae983a4e350bf485b7bb5f90deaf9bc2844b4f94afb8c5d155103c2147ac402a0503edbafa0cbbf4ac11d
MISC metadata.xml 349 BLAKE2B 363f31b1d597e4c4959c588c9765154049c7c9692bbd64a0736226261b6232c75af7cbf255de59813cd10b9d45b30cf08d4ad0eb1422bf956264377a58002340 SHA512 d83250993581f9b23b43b8553cf46fe75411232ca7ec18c32c070d6acb0f9c7a01769755a3f11b4c5aafe40a63ee0c3817fc45dd01325ce02bbf557d5ab0cdea
diff --git a/dev-util/Tensile/Tensile-4.3.0-r1.ebuild b/dev-util/Tensile/Tensile-4.3.0-r1.ebuild
index 143867819d04..7cb7e94cd256 100644
--- a/dev-util/Tensile/Tensile-4.3.0-r1.ebuild
+++ b/dev-util/Tensile/Tensile-4.3.0-r1.ebuild
@@ -5,7 +5,7 @@ EAPI=8
PYTHON_COMPAT=( python3_{8..10} )
DISTUTILS_USE_PEP517=setuptools
-inherit distutils-r1
+inherit distutils-r1 prefix
DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions"
HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile"
@@ -33,11 +33,11 @@ PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.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
- "${FILESDIR}"/${PN}-4.3.0-gentoopath.patch
)
src_prepare() {
distutils-r1_src_prepare
+ eapply $(prefixify_ro "${FILESDIR}"/${PN}-4.3.0-gentoopath.patch)
pushd ${PN} || die
@@ -70,7 +70,7 @@ python_install() {
distutils-r1_python_install
python_moduleinto Tensile
- pushd Tensile
+ pushd Tensile || die
python_domodule Components
python_newexe Utilities/merge.py ${PN}-merge
}
diff --git a/dev-util/Tensile/Tensile-5.0.2.ebuild b/dev-util/Tensile/Tensile-5.0.2.ebuild
new file mode 100644
index 000000000000..c9240e233281
--- /dev/null
+++ b/dev-util/Tensile/Tensile-5.0.2.ebuild
@@ -0,0 +1,83 @@
+# Copyright 1999-2022 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+PYTHON_COMPAT=( python3_{8..10} )
+DISTUTILS_USE_PEP517=setuptools
+inherit distutils-r1 prefix
+
+DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions"
+HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile"
+SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz
+ https://github.com/littlewu2508/littlewu2508.github.io/raw/main/gentoo-distfiles/${P}-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[${PYTHON_USEDEP}]
+ >=dev-util/rocm-smi-4.3.0
+"
+DEPEND="${RDEPEND}
+ dev-util/hip:${SLOT}
+"
+
+PATCHES=( "${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}"/${PN}-5.0.2-use-ninja.patch
+ "${WORKDIR}"/${PN}-5.0.2-PR1419.patch
+ )
+
+CMAKE_USE_DIR="${WORKDIR}/Source"
+
+src_prepare() {
+ distutils-r1_src_prepare
+ eapply $(prefixify_ro "${FILESDIR}"/${PN}-5.0.2-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-4.3.0-gentoopath.patch b/dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch
index 6fda0ddc6a8d..da135e3b38ee 100644
--- a/dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch
+++ b/dev-util/Tensile/files/Tensile-4.3.0-gentoopath.patch
@@ -17,7 +17,7 @@ Index: Tensile-rocm-4.3.0/Tensile/Source/lib/CMakeLists.txt
if(TENSILE_USE_LLVM)
- find_package(LLVM 13.0 QUIET CONFIG)
-+ find_package(LLVM PATHS /opt/gentoo/usr/lib/llvm/roc 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)
@@ -30,7 +30,7 @@ Index: Tensile-rocm-4.3.0/Tensile/Common.py
print2(" %24s: %8s (unspecified)" % (key, defaultValue))
- globalParameters["ROCmPath"] = "/opt/rocm"
-+ globalParameters["ROCmPath"] = "/opt/gentoo/usr"
++ 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:
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-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.0.2-gentoopath.patch
new file mode 100644
index 000000000000..b8b59af5184e
--- /dev/null
+++ b/dev-util/Tensile/files/Tensile-5.0.2-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 @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-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 @GENTOO_PORTAGE_EPREFIX@/usr/lib/llvm/roc 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-5.0.2-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)