| 1 |
commit: f4aae2060ce38589b7fc56c5fcd165b291ffa986 |
| 2 |
Author: YiyangWu <xgreenlandforwyy <AT> gmail <DOT> com> |
| 3 |
AuthorDate: Tue Mar 22 09:27:48 2022 +0000 |
| 4 |
Commit: Benda XU <heroxbd <AT> gentoo <DOT> org> |
| 5 |
CommitDate: Sun Mar 27 03:12:51 2022 +0000 |
| 6 |
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=f4aae206 |
| 7 |
|
| 8 |
dev-util/Tensile: bump version to 5.0.2 |
| 9 |
|
| 10 |
Package-Manager: Portage-3.0.30, Repoman-3.0.3 |
| 11 |
Signed-off-by: Yiyang Wu <xgreenlandforwyy <AT> gmail.com> |
| 12 |
Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org> |
| 13 |
|
| 14 |
dev-util/Tensile/Manifest | 2 + |
| 15 |
dev-util/Tensile/Tensile-5.0.2.ebuild | 83 ++++++++++++++++ |
| 16 |
.../files/Tensile-5.0.2-fix-arch-parse.patch | 12 +++ |
| 17 |
.../Tensile/files/Tensile-5.0.2-gentoopath.patch | 75 +++++++++++++++ |
| 18 |
dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch | 104 +++++++++++++++++++++ |
| 19 |
.../Tensile/files/Tensile-5.0.2-use-ninja.patch | 21 +++++ |
| 20 |
6 files changed, 297 insertions(+) |
| 21 |
|
| 22 |
diff --git a/dev-util/Tensile/Manifest b/dev-util/Tensile/Manifest |
| 23 |
index fd4cac044af8..c3e8f48bab20 100644 |
| 24 |
--- a/dev-util/Tensile/Manifest |
| 25 |
+++ b/dev-util/Tensile/Manifest |
| 26 |
@@ -1 +1,3 @@ |
| 27 |
+DIST Tensile-5.0.2-PR1419.patch.gz 43754 BLAKE2B 3b1d7333ed0115f77816806e7735e25d897dbae3f7a40ece61d655c8f016d9d1c804b586b4e0e75661e20c517f5937a61d5ead25a545f0ce2e4a178bcbb3e7be SHA512 0154d8e14d335f4ecd86e5db335289f8662cede7e5d4b93bf6c56d8e48c1d7d0457271c078c9abbbca1f4aa91f028f6ce6a89efb783ba0207d0f39fd5de8a42c |
| 28 |
DIST rocm-Tensile-4.3.0.tar.gz 14250149 BLAKE2B 030138eaca2a0aadd96801e6dbd72e510716dd90553ef3795c5e04e00a34a05ecae82b24f755e4033a4acfcdb1cf26291da1e7902bb090f89a010d403e832beb SHA512 126db0b413c716fba8d5be9bff7a44fd1badacbf32f3db8d0db649819177db37ebd56fd22dd3c809655f5d29675be115e698cd10bc3d0b4b23878ae3726fce47 |
| 29 |
+DIST rocm-Tensile-5.0.2.tar.gz 14362090 BLAKE2B d44c3135496db41c4eb2545f6054c84493e89ef6357da881dae6859763338417e7a9e142cdce7e2de0536684bb641049ad0440964adbe68d03ca329f01efabc9 SHA512 f5ab7d7cbc5ba25498a02a6b7c07d5a9a548edcd44b439b92c8718072382adc18ae78274db1a7174e06436ac1396f3b4edafc7f491a2c87086c368e7270e64e3 |
| 30 |
|
| 31 |
diff --git a/dev-util/Tensile/Tensile-5.0.2.ebuild b/dev-util/Tensile/Tensile-5.0.2.ebuild |
| 32 |
new file mode 100644 |
| 33 |
index 000000000000..1210cd769aa5 |
| 34 |
--- /dev/null |
| 35 |
+++ b/dev-util/Tensile/Tensile-5.0.2.ebuild |
| 36 |
@@ -0,0 +1,83 @@ |
| 37 |
+# Copyright 1999-2022 Gentoo Authors |
| 38 |
+# Distributed under the terms of the GNU General Public License v2 |
| 39 |
+ |
| 40 |
+EAPI=8 |
| 41 |
+ |
| 42 |
+PYTHON_COMPAT=( python3_{8..10} ) |
| 43 |
+DISTUTILS_USE_PEP517=setuptools |
| 44 |
+inherit distutils-r1 prefix |
| 45 |
+ |
| 46 |
+DESCRIPTION="Stretching GPU performance for GEMMs and tensor contractions" |
| 47 |
+HOMEPAGE="https://github.com/ROCmSoftwarePlatform/Tensile" |
| 48 |
+SRC_URI="https://github.com/ROCmSoftwarePlatform/Tensile/archive/rocm-${PV}.tar.gz -> rocm-Tensile-${PV}.tar.gz |
| 49 |
+ https://github.com/littlewu2508/littlewu2508.github.io/raw/main/gentoo-distfiles/${P}-PR1419.patch.gz" |
| 50 |
+S="${WORKDIR}/${PN}-rocm-${PV}" |
| 51 |
+ |
| 52 |
+LICENSE="MIT" |
| 53 |
+KEYWORDS="~amd64" |
| 54 |
+SLOT="0/$(ver_cut 1-2)" |
| 55 |
+ |
| 56 |
+# Not compatible with recent versions of pytest |
| 57 |
+RESTRICT="test" |
| 58 |
+ |
| 59 |
+RDEPEND="${PYTHON_DEPS} |
| 60 |
+ dev-python/pyyaml[${PYTHON_USEDEP}] |
| 61 |
+ dev-python/msgpack[${PYTHON_USEDEP}] |
| 62 |
+ >=dev-util/rocm-smi-4.3.0 |
| 63 |
+" |
| 64 |
+DEPEND="${RDEPEND} |
| 65 |
+ dev-util/hip:${SLOT} |
| 66 |
+" |
| 67 |
+ |
| 68 |
+PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.patch |
| 69 |
+ "${FILESDIR}"/${PN}-5.0.2-gfx1031.patch |
| 70 |
+ "${FILESDIR}"/${PN}-5.0.2-fix-arch-parse.patch |
| 71 |
+ "${FILESDIR}"/${PN}-5.0.2-use-ninja.patch |
| 72 |
+ "${WORKDIR}"/${PN}-5.0.2-PR1419.patch |
| 73 |
+ ) |
| 74 |
+ |
| 75 |
+CMAKE_USE_DIR="${WORKDIR}/Source" |
| 76 |
+ |
| 77 |
+src_prepare() { |
| 78 |
+ distutils-r1_src_prepare |
| 79 |
+ eapply $(prefixify_ro "${FILESDIR}"/${PN}-5.0.2-gentoopath.patch) |
| 80 |
+ |
| 81 |
+ pushd ${PN} || die |
| 82 |
+ |
| 83 |
+ sed -e "/ROCM_SMI_ROOT/s,lib,$(get_libdir)," \ |
| 84 |
+ -i Source/cmake/FindROCmSMI.cmake || die |
| 85 |
+ sed -r -e "/TENSILE_USE_LLVM/s/ON/OFF/" \ |
| 86 |
+ -i Source/CMakeLists.txt || die |
| 87 |
+ sed -e "/chmod 755/d" -i Source/TensileCreateLibrary.cmake || die # remove chmod 755 on |
| 88 |
+ sed -e "s,\${Tensile_ROOT}/bin/,,g" -i Source/TensileCreateLibrary.cmake cmake/TensileConfig.cmake || die # ${Tensile_ROOT}/bin does not exists; call command directly |
| 89 |
+ |
| 90 |
+ local Tensile_share_dir="\"${EPREFIX}/usr/share/${PN}\"" |
| 91 |
+ sed -e "/HipClangVersion/s/0,0,0/$(hipconfig -v)/" -i Common.py || die |
| 92 |
+ |
| 93 |
+ sed -e "s,os.path.dirname(os.path.realpath(__file__)),${Tensile_share_dir},g" -i ReplacementKernels.py Common.py ${PN}.py || die |
| 94 |
+ |
| 95 |
+ sed -e "s|os\.path\.dirname.*$|\"${EPREFIX}/usr/share/Tensile/Source\", end='')|" -i __init__.py || die |
| 96 |
+ |
| 97 |
+ popd || die |
| 98 |
+ |
| 99 |
+ sed -e "/package_data/d" -e "/data_files/d" -i setup.py || die |
| 100 |
+} |
| 101 |
+ |
| 102 |
+python_install() { |
| 103 |
+ distutils-r1_python_install |
| 104 |
+ |
| 105 |
+ python_moduleinto Tensile |
| 106 |
+ pushd Tensile |
| 107 |
+ python_domodule Components |
| 108 |
+ python_newexe Utilities/merge.py ${PN}-merge |
| 109 |
+} |
| 110 |
+ |
| 111 |
+src_install() { |
| 112 |
+ distutils-r1_src_install |
| 113 |
+ |
| 114 |
+ pushd ${PN} || die |
| 115 |
+ insinto /usr/share/${PN} |
| 116 |
+ doins -r Configs Perf ReplacementKernels ReplacementKernels-cov3 Source CustomKernels |
| 117 |
+ insinto /usr/$(get_libdir)/cmake/${PN} |
| 118 |
+ doins cmake/*.cmake |
| 119 |
+} |
| 120 |
|
| 121 |
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 |
| 122 |
new file mode 100644 |
| 123 |
index 000000000000..47bee8819c44 |
| 124 |
--- /dev/null |
| 125 |
+++ b/dev-util/Tensile/files/Tensile-5.0.2-fix-arch-parse.patch |
| 126 |
@@ -0,0 +1,12 @@ |
| 127 |
+Index: Tensile-rocm-5.0.1/Tensile/Tensile.py |
| 128 |
+=================================================================== |
| 129 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Tensile.py |
| 130 |
++++ Tensile-rocm-5.0.1/Tensile/Tensile.py |
| 131 |
+@@ -97,7 +97,6 @@ def addCommonArguments(argParser): |
| 132 |
+ """ |
| 133 |
+ |
| 134 |
+ (key, value) = par.split("=") |
| 135 |
+- value = eval(value) |
| 136 |
+ return (key, value) |
| 137 |
+ |
| 138 |
+ argParser.add_argument("-d", "--device", dest="device", type=int, \ |
| 139 |
|
| 140 |
diff --git a/dev-util/Tensile/files/Tensile-5.0.2-gentoopath.patch b/dev-util/Tensile/files/Tensile-5.0.2-gentoopath.patch |
| 141 |
new file mode 100644 |
| 142 |
index 000000000000..b8b59af5184e |
| 143 |
--- /dev/null |
| 144 |
+++ b/dev-util/Tensile/files/Tensile-5.0.2-gentoopath.patch |
| 145 |
@@ -0,0 +1,75 @@ |
| 146 |
+Fix incorrect path in scripts and CMakeLists, in order to build tensile_client |
| 147 |
+=================================================================== |
| 148 |
+--- Tensile-rocm-5.0.2.orig/Tensile/Source/client/CMakeLists.txt |
| 149 |
++++ Tensile-rocm-5.0.2/Tensile/Source/client/CMakeLists.txt |
| 150 |
+@@ -56,3 +56,5 @@ endif() |
| 151 |
+ foreach(arch IN LISTS TENSILE_GPU_ARCHS) |
| 152 |
+ target_link_libraries(tensile_client PRIVATE "--amdgpu-target=${arch}") |
| 153 |
+ endforeach(arch) |
| 154 |
++ |
| 155 |
++target_link_libraries(tensile_client PRIVATE LLVMSupport) |
| 156 |
+Index: Tensile-rocm-5.0.2/Tensile/Source/lib/CMakeLists.txt |
| 157 |
+=================================================================== |
| 158 |
+--- Tensile-rocm-5.0.2.orig/Tensile/Source/lib/CMakeLists.txt |
| 159 |
++++ Tensile-rocm-5.0.2/Tensile/Source/lib/CMakeLists.txt |
| 160 |
+@@ -47,7 +47,7 @@ set(tensile_sources ${tensile_sources} |
| 161 |
+ ) |
| 162 |
+ |
| 163 |
+ if(TENSILE_USE_LLVM) |
| 164 |
+- find_package(LLVM 13.0 QUIET CONFIG) |
| 165 |
++ find_package(LLVM PATHS @GENTOO_PORTAGE_EPREFIX@/usr/lib/llvm/roc QUIET CONFIG) |
| 166 |
+ if(NOT LLVM_FOUND) |
| 167 |
+ find_package(LLVM 12.0 QUIET CONFIG) |
| 168 |
+ if(NOT LLVM_FOUND) |
| 169 |
+Index: Tensile-rocm-5.0.2/Tensile/Common.py |
| 170 |
+=================================================================== |
| 171 |
+--- Tensile-rocm-5.0.2.orig/Tensile/Common.py |
| 172 |
++++ Tensile-rocm-5.0.2/Tensile/Common.py |
| 173 |
+@@ -1873,7 +1873,7 @@ def assignGlobalParameters( config ): |
| 174 |
+ else: |
| 175 |
+ print2(" %24s: %8s (unspecified)" % (key, defaultValue)) |
| 176 |
+ |
| 177 |
+- globalParameters["ROCmPath"] = "/opt/rocm" |
| 178 |
++ globalParameters["ROCmPath"] = "@GENTOO_PORTAGE_EPREFIX@/usr" |
| 179 |
+ if "ROCM_PATH" in os.environ: |
| 180 |
+ globalParameters["ROCmPath"] = os.environ.get("ROCM_PATH") |
| 181 |
+ if "TENSILE_ROCM_PATH" in os.environ: |
| 182 |
+Index: Tensile-rocm-5.0.2/Tensile/ClientWriter.py |
| 183 |
+=================================================================== |
| 184 |
+--- Tensile-rocm-5.0.2.orig/Tensile/ClientWriter.py |
| 185 |
++++ Tensile-rocm-5.0.2/Tensile/ClientWriter.py |
| 186 |
+@@ -205,7 +205,7 @@ def getBuildClientLibraryScript(buildPat |
| 187 |
+ import io |
| 188 |
+ runScriptFile = io.StringIO() |
| 189 |
+ |
| 190 |
+- callCreateLibraryCmd = globalParameters["ScriptPath"] + "/bin/TensileCreateLibrary" |
| 191 |
++ callCreateLibraryCmd = "TensileCreateLibrary" |
| 192 |
+ |
| 193 |
+ |
| 194 |
+ if globalParameters["MergeFiles"]: |
| 195 |
+Index: Tensile-rocm-5.0.2/Tensile/GenerateSummations.py |
| 196 |
+=================================================================== |
| 197 |
+--- Tensile-rocm-5.0.2.orig/Tensile/GenerateSummations.py |
| 198 |
++++ Tensile-rocm-5.0.2/Tensile/GenerateSummations.py |
| 199 |
+@@ -57,7 +57,7 @@ def createLibraryForBenchmark(logicPath, |
| 200 |
+ Selection. |
| 201 |
+ """ |
| 202 |
+ |
| 203 |
+- pythonExePath = os.path.join(os.path.dirname(os.path.realpath(__file__)), "bin", "TensileCreateLibrary") |
| 204 |
++ pythonExePath = "TensileCreateLibrary" |
| 205 |
+ args = [pythonExePath, \ |
| 206 |
+ "--merge-files", "--new-client-only", "--no-short-file-names", "--no-library-print-debug", \ |
| 207 |
+ "--architecture=all", "--code-object-version=V3", "--cxx-compiler=hipcc", "--library-format=yaml", \ |
| 208 |
+Index: Tensile-rocm-5.0.2/HostLibraryTests/CMakeLists.txt |
| 209 |
+=================================================================== |
| 210 |
+--- Tensile-rocm-5.0.2.orig/HostLibraryTests/CMakeLists.txt |
| 211 |
++++ Tensile-rocm-5.0.2/HostLibraryTests/CMakeLists.txt |
| 212 |
+@@ -135,7 +135,7 @@ set(test_sources ${test_sources} |
| 213 |
+ ) |
| 214 |
+ |
| 215 |
+ if(TENSILE_USE_LLVM) |
| 216 |
+- find_package(LLVM 13.0 QUIET CONFIG) |
| 217 |
++ find_package(LLVM PATHS @GENTOO_PORTAGE_EPREFIX@/usr/lib/llvm/roc QUIET CONFIG) |
| 218 |
+ if(NOT LLVM_FOUND) |
| 219 |
+ find_package(LLVM 12.0 QUIET CONFIG) |
| 220 |
+ if(NOT LLVM_FOUND) |
| 221 |
|
| 222 |
diff --git a/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch b/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch |
| 223 |
new file mode 100644 |
| 224 |
index 000000000000..17925cfd49e0 |
| 225 |
--- /dev/null |
| 226 |
+++ b/dev-util/Tensile/files/Tensile-5.0.2-gfx1031.patch |
| 227 |
@@ -0,0 +1,104 @@ |
| 228 |
+Index: Tensile-rocm-5.0.1/pytest.ini |
| 229 |
+=================================================================== |
| 230 |
+--- Tensile-rocm-5.0.1.orig/pytest.ini |
| 231 |
++++ Tensile-rocm-5.0.1/pytest.ini |
| 232 |
+@@ -87,6 +87,7 @@ markers = |
| 233 |
+ xfail-gfx1011: architecture |
| 234 |
+ xfail-gfx1012: architecture |
| 235 |
+ xfail-gfx1030: architecture |
| 236 |
++ xfail-gfx1031: architecture |
| 237 |
+ skip-gfx000: architecture |
| 238 |
+ skip-gfx900: architecture |
| 239 |
+ skip-gfx906: architecture |
| 240 |
+@@ -96,3 +97,4 @@ markers = |
| 241 |
+ skip-gfx1011: architecture |
| 242 |
+ skip-gfx1012: architecture |
| 243 |
+ skip-gfx1030: architecture |
| 244 |
++ skip-gfx1031: architecture |
| 245 |
+Index: Tensile-rocm-5.0.1/Tensile/Common.py |
| 246 |
+=================================================================== |
| 247 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Common.py |
| 248 |
++++ Tensile-rocm-5.0.1/Tensile/Common.py |
| 249 |
+@@ -198,7 +198,7 @@ globalParameters["MergeFiles"] = True |
| 250 |
+ globalParameters["NumMergedFiles"] = 1 # The number of files that kernels should be split between when merging |
| 251 |
+ |
| 252 |
+ globalParameters["MaxFileName"] = 64 # If a file name would be longer than this, shorten it with a hash. |
| 253 |
+-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 |
| 254 |
++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 |
| 255 |
+ |
| 256 |
+ globalParameters["GenerateManifestAndExit"] = False # Output manifest file with list of expected library objects and exit |
| 257 |
+ globalParameters["NewClient"] = 2 # Old client deprecated: NewClient must be set to 2. |
| 258 |
+@@ -264,7 +264,7 @@ architectureMap = { |
| 259 |
+ 'gfx906':'vega20', 'gfx906:xnack+':'vega20', 'gfx906:xnack-':'vega20', |
| 260 |
+ 'gfx908':'arcturus','gfx908:xnack+':'arcturus', 'gfx908:xnack-':'arcturus', |
| 261 |
+ 'gfx90a':'aldebaran', 'gfx90a:xnack+':'aldebaran', 'gfx90a:xnack-':'aldebaran', |
| 262 |
+- 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21' |
| 263 |
++ 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', 'gfx1031':'navi22' |
| 264 |
+ } |
| 265 |
+ |
| 266 |
+ def getArchitectureName(gfxName): |
| 267 |
+Index: Tensile-rocm-5.0.1/Tensile/KernelWriterAssembly.py |
| 268 |
+=================================================================== |
| 269 |
+--- Tensile-rocm-5.0.1.orig/Tensile/KernelWriterAssembly.py |
| 270 |
++++ Tensile-rocm-5.0.1/Tensile/KernelWriterAssembly.py |
| 271 |
+@@ -2364,7 +2364,7 @@ class KernelWriterAssembly(KernelWriter) |
| 272 |
+ |
| 273 |
+ # integer i8x4 |
| 274 |
+ elif kernel["ProblemType"]["DataType"].isInt8x4(): |
| 275 |
+- if self.version == (9,0,6) or self.version == (9,0,8) or self.version == (9,0,10) or self.version == (10,3,0): |
| 276 |
++ 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): |
| 277 |
+ for b in range(0, kernel["ThreadTile1"]): |
| 278 |
+ for a in range(0, kernel["ThreadTile0"]): |
| 279 |
+ for iui in range(0, innerUnroll): |
| 280 |
+Index: Tensile-rocm-5.0.1/Tensile/Source/CMakeLists.txt |
| 281 |
+=================================================================== |
| 282 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Source/CMakeLists.txt |
| 283 |
++++ Tensile-rocm-5.0.1/Tensile/Source/CMakeLists.txt |
| 284 |
+@@ -47,7 +47,7 @@ if(NOT DEFINED CXX_VERSION_STRING) |
| 285 |
+ endif() |
| 286 |
+ |
| 287 |
+ if(CMAKE_CXX_COMPILER STREQUAL "hipcc") |
| 288 |
+- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 CACHE STRING "GPU architectures") |
| 289 |
++ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 CACHE STRING "GPU architectures") |
| 290 |
+ else() |
| 291 |
+ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906 gfx908 gfx90a CACHE STRING "GPU architectures") |
| 292 |
+ endif() |
| 293 |
+Index: Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
| 294 |
+=================================================================== |
| 295 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
| 296 |
++++ Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
| 297 |
+@@ -62,7 +62,8 @@ namespace Tensile |
| 298 |
+ gfx1010 = 1010, |
| 299 |
+ gfx1011 = 1011, |
| 300 |
+ gfx1012 = 1012, |
| 301 |
+- gfx1030 = 1030 |
| 302 |
++ gfx1030 = 1030, |
| 303 |
++ gfx1031 = 1031 |
| 304 |
+ }; |
| 305 |
+ |
| 306 |
+ AMDGPU(); |
| 307 |
+Index: Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
| 308 |
+=================================================================== |
| 309 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
| 310 |
++++ Tensile-rocm-5.0.1/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
| 311 |
+@@ -215,6 +215,7 @@ namespace Tensile |
| 312 |
+ iot::enumCase(io, value, "gfx1011", AMDGPU::Processor::gfx1011); |
| 313 |
+ iot::enumCase(io, value, "gfx1012", AMDGPU::Processor::gfx1012); |
| 314 |
+ iot::enumCase(io, value, "gfx1030", AMDGPU::Processor::gfx1030); |
| 315 |
++ iot::enumCase(io, value, "gfx1031", AMDGPU::Processor::gfx1031); |
| 316 |
+ } |
| 317 |
+ }; |
| 318 |
+ } // namespace Serialization |
| 319 |
+Index: Tensile-rocm-5.0.1/Tensile/Source/lib/source/AMDGPU.cpp |
| 320 |
+=================================================================== |
| 321 |
+--- Tensile-rocm-5.0.1.orig/Tensile/Source/lib/source/AMDGPU.cpp |
| 322 |
++++ Tensile-rocm-5.0.1/Tensile/Source/lib/source/AMDGPU.cpp |
| 323 |
+@@ -82,6 +82,8 @@ namespace Tensile |
| 324 |
+ return stream << "gfx1012"; |
| 325 |
+ case AMDGPU::Processor::gfx1030: |
| 326 |
+ return stream << "gfx1030"; |
| 327 |
++ case AMDGPU::Processor::gfx1031: |
| 328 |
++ return stream << "gfx1031"; |
| 329 |
+ } |
| 330 |
+ return stream; |
| 331 |
+ } |
| 332 |
|
| 333 |
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 |
| 334 |
new file mode 100644 |
| 335 |
index 000000000000..485ec992d808 |
| 336 |
--- /dev/null |
| 337 |
+++ b/dev-util/Tensile/files/Tensile-5.0.2-use-ninja.patch |
| 338 |
@@ -0,0 +1,21 @@ |
| 339 |
+diff --git a/Tensile/ClientExecutable.py b/Tensile/ClientExecutable.py |
| 340 |
+index ca3809f..2d1d83b 100644 |
| 341 |
+--- a/Tensile/ClientExecutable.py |
| 342 |
++++ b/Tensile/ClientExecutable.py |
| 343 |
+@@ -36,6 +36,7 @@ class CMakeEnvironment: |
| 344 |
+ |
| 345 |
+ args = ['cmake'] |
| 346 |
+ args += itertools.chain.from_iterable([ ['-D', '{}={}'.format(key, value)] for key,value in self.options.items()]) |
| 347 |
++ args += itertools.chain.from_iterable([ ['-G', 'Ninja'], [ '-D', 'CMAKE_EXPORT_COMPILE_COMMANDS=ON' ] ]) |
| 348 |
+ args += [self.sourceDir] |
| 349 |
+ |
| 350 |
+ Common.print2(' '.join(args)) |
| 351 |
+@@ -43,7 +44,7 @@ class CMakeEnvironment: |
| 352 |
+ subprocess.check_call(args, cwd=Common.ensurePath(self.buildDir)) |
| 353 |
+ |
| 354 |
+ def build(self): |
| 355 |
+- args = ['make', '-j'] |
| 356 |
++ args = ['ninja'] |
| 357 |
+ Common.print2(' '.join(args)) |
| 358 |
+ with Common.ClientExecutionLock(): |
| 359 |
+ subprocess.check_call(args, cwd=self.buildDir) |
Archives