1 |
commit: 2b7c870b48d225a9ade51eac426310de68f41069 |
2 |
Author: Yiyang Wu <xgreenlandforwyy <AT> gmail <DOT> com> |
3 |
AuthorDate: Tue Jan 31 14:57:06 2023 +0000 |
4 |
Commit: Benda XU <heroxbd <AT> gentoo <DOT> org> |
5 |
CommitDate: Wed Feb 1 01:51:35 2023 +0000 |
6 |
URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=2b7c870b |
7 |
|
8 |
dev-util/Tensile: add missing patches |
9 |
|
10 |
Closes: https://bugs.gentoo.org/892736 |
11 |
Closes: https://github.com/gentoo/gentoo/pull/29356 |
12 |
Signed-off-by: Yiyang Wu <xgreenlandforwyy <AT> gmail.com> |
13 |
Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org> |
14 |
|
15 |
dev-util/Tensile/Tensile-5.4.2.ebuild | 4 +- |
16 |
.../files/Tensile-5.4.2-fix-arch-parse.patch | 12 +++ |
17 |
dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch | 88 ++++++++++++++++++++++ |
18 |
3 files changed, 102 insertions(+), 2 deletions(-) |
19 |
|
20 |
diff --git a/dev-util/Tensile/Tensile-5.4.2.ebuild b/dev-util/Tensile/Tensile-5.4.2.ebuild |
21 |
index 09bccd0ab2dd..1e347fdcee56 100644 |
22 |
--- a/dev-util/Tensile/Tensile-5.4.2.ebuild |
23 |
+++ b/dev-util/Tensile/Tensile-5.4.2.ebuild |
24 |
@@ -32,8 +32,8 @@ DEPEND="${RDEPEND} |
25 |
" |
26 |
|
27 |
PATCHES=( "${FILESDIR}"/${PN}-4.3.0-output-commands.patch |
28 |
- "${FILESDIR}"/${PN}-5.3.3-gfx1031.patch |
29 |
- "${FILESDIR}"/${PN}-5.3.3-fix-arch-parse.patch |
30 |
+ "${FILESDIR}"/${PN}-5.4.2-gfx1031.patch |
31 |
+ "${FILESDIR}"/${PN}-5.4.2-fix-arch-parse.patch |
32 |
"${FILESDIR}"/${PN}-5.4.2-use-ninja.patch |
33 |
) |
34 |
|
35 |
|
36 |
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 |
37 |
new file mode 100644 |
38 |
index 000000000000..14263e89ef84 |
39 |
--- /dev/null |
40 |
+++ b/dev-util/Tensile/files/Tensile-5.4.2-fix-arch-parse.patch |
41 |
@@ -0,0 +1,12 @@ |
42 |
+Index: Tensile-rocm-5.3.3/Tensile/Tensile.py |
43 |
+=================================================================== |
44 |
+--- Tensile-rocm-5.3.3.orig/Tensile/Tensile.py |
45 |
++++ Tensile-rocm-5.3.3/Tensile/Tensile.py |
46 |
+@@ -101,7 +101,6 @@ def addCommonArguments(argParser): |
47 |
+ Allows the --global-parameters option to specify any parameters from the command line. |
48 |
+ """ |
49 |
+ (key, value) = par.split("=") |
50 |
+- value = eval(value) |
51 |
+ return (key, value) |
52 |
+ |
53 |
+ argParser.add_argument("-d", "--device", dest="device", type=int, \ |
54 |
|
55 |
diff --git a/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch b/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch |
56 |
new file mode 100644 |
57 |
index 000000000000..1f703c07c3ec |
58 |
--- /dev/null |
59 |
+++ b/dev-util/Tensile/files/Tensile-5.4.2-gfx1031.patch |
60 |
@@ -0,0 +1,88 @@ |
61 |
+Index: Tensile-rocm-5.3.3/pytest.ini |
62 |
+=================================================================== |
63 |
+--- Tensile-rocm-5.3.3.orig/pytest.ini |
64 |
++++ Tensile-rocm-5.3.3/pytest.ini |
65 |
+@@ -88,6 +88,7 @@ markers = |
66 |
+ xfail-gfx1011: architecture |
67 |
+ xfail-gfx1012: architecture |
68 |
+ xfail-gfx1030: architecture |
69 |
++ xfail-gfx1031: architecture |
70 |
+ xfail-gfx1100: architecture |
71 |
+ xfail-gfx1101: architecture |
72 |
+ xfail-gfx1102: architecture |
73 |
+@@ -100,6 +101,7 @@ markers = |
74 |
+ skip-gfx1011: architecture |
75 |
+ skip-gfx1012: architecture |
76 |
+ skip-gfx1030: architecture |
77 |
++ skip-gfx1031: architecture |
78 |
+ skip-gfx1100: architecture |
79 |
+ skip-gfx1101: architecture |
80 |
+ skip-gfx1102: architecture |
81 |
+Index: Tensile-rocm-5.3.3/Tensile/Common.py |
82 |
+=================================================================== |
83 |
+--- Tensile-rocm-5.3.3.orig/Tensile/Common.py |
84 |
++++ Tensile-rocm-5.3.3/Tensile/Common.py |
85 |
+@@ -201,7 +201,7 @@ globalParameters["MergeFiles"] = True |
86 |
+ globalParameters["NumMergedFiles"] = 1 # The number of files that kernels should be split between when merging |
87 |
+ |
88 |
+ globalParameters["MaxFileName"] = 64 # If a file name would be longer than this, shorten it with a hash. |
89 |
+-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 |
90 |
++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 |
91 |
+ |
92 |
+ globalParameters["GenerateManifestAndExit"] = False # Output manifest file with list of expected library objects and exit |
93 |
+ globalParameters["NewClient"] = 2 # Old client deprecated: NewClient must be set to 2. |
94 |
+@@ -271,7 +271,7 @@ architectureMap = { |
95 |
+ 'gfx906':'vega20', 'gfx906:xnack+':'vega20', 'gfx906:xnack-':'vega20', |
96 |
+ 'gfx908':'arcturus','gfx908:xnack+':'arcturus', 'gfx908:xnack-':'arcturus', |
97 |
+ 'gfx90a':'aldebaran', 'gfx90a:xnack+':'aldebaran', 'gfx90a:xnack-':'aldebaran', |
98 |
+- 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', |
99 |
++ 'gfx1010':'navi10', 'gfx1011':'navi12', 'gfx1012':'navi14', 'gfx1030':'navi21', 'gfx1031':'navi22', |
100 |
+ 'gfx1100':'navi31', 'gfx1101':'navi32', 'gfx1102':'navi33' |
101 |
+ } |
102 |
+ |
103 |
+Index: Tensile-rocm-5.3.3/Tensile/Source/CMakeLists.txt |
104 |
+=================================================================== |
105 |
+--- Tensile-rocm-5.3.3.orig/Tensile/Source/CMakeLists.txt |
106 |
++++ Tensile-rocm-5.3.3/Tensile/Source/CMakeLists.txt |
107 |
+@@ -51,7 +51,7 @@ if(NOT DEFINED CXX_VERSION_STRING) |
108 |
+ endif() |
109 |
+ |
110 |
+ if(CMAKE_CXX_COMPILER STREQUAL "hipcc") |
111 |
+- set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") |
112 |
++ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906:xnack- gfx908:xnack- gfx90a:xnack- gfx1010 gfx1011 gfx1012 gfx1030 gfx1031 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") |
113 |
+ else() |
114 |
+ set(TENSILE_GPU_ARCHS gfx803 gfx900 gfx906 gfx908 gfx90a gfx1010 gfx1011 gfx1012 gfx1030 gfx1100 gfx1101 gfx1102 CACHE STRING "GPU architectures") |
115 |
+ endif() |
116 |
+Index: Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
117 |
+=================================================================== |
118 |
+--- Tensile-rocm-5.3.3.orig/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
119 |
++++ Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/AMDGPU.hpp |
120 |
+@@ -63,6 +63,7 @@ namespace Tensile |
121 |
+ gfx1011 = 1011, |
122 |
+ gfx1012 = 1012, |
123 |
+ gfx1030 = 1030, |
124 |
++ gfx1031 = 1031, |
125 |
+ gfx1100 = 1100, |
126 |
+ gfx1101 = 1101, |
127 |
+ gfx1102 = 1102 |
128 |
+@@ -90,6 +91,8 @@ namespace Tensile |
129 |
+ return "gfx1012"; |
130 |
+ case AMDGPU::Processor::gfx1030: |
131 |
+ return "gfx1030"; |
132 |
++ case AMDGPU::Processor::gfx1031: |
133 |
++ return "gfx1032"; |
134 |
+ case AMDGPU::Processor::gfx1100: |
135 |
+ return "gfx1100"; |
136 |
+ case AMDGPU::Processor::gfx1101: |
137 |
+Index: Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
138 |
+=================================================================== |
139 |
+--- Tensile-rocm-5.3.3.orig/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
140 |
++++ Tensile-rocm-5.3.3/Tensile/Source/lib/include/Tensile/Serialization/Predicates.hpp |
141 |
+@@ -215,6 +215,7 @@ namespace Tensile |
142 |
+ iot::enumCase(io, value, "gfx1011", AMDGPU::Processor::gfx1011); |
143 |
+ iot::enumCase(io, value, "gfx1012", AMDGPU::Processor::gfx1012); |
144 |
+ iot::enumCase(io, value, "gfx1030", AMDGPU::Processor::gfx1030); |
145 |
++ iot::enumCase(io, value, "gfx1031", AMDGPU::Processor::gfx1031); |
146 |
+ iot::enumCase(io, value, "gfx1100", AMDGPU::Processor::gfx1100); |
147 |
+ iot::enumCase(io, value, "gfx1101", AMDGPU::Processor::gfx1101); |
148 |
+ iot::enumCase(io, value, "gfx1102", AMDGPU::Processor::gfx1102); |