| blob | 59e17e31e89d9ca271816628527b434f1c2ce28e |
1 ; REQUIRES: aarch64-registered-target
3 ; This test needs to be target specific due to the cost estimate in the output.
5 ; RUN: opt -lower-matrix-intrinsics -pass-remarks-output=%t -pass-remarks=lower-matrix-intrinsics -mtriple=arm64-apple-iphoneos %s 2>&1 -disable-output | FileCheck --check-prefix=STDERR %s
6 ; RUN: FileCheck --input-file=%t --check-prefix=YAML %s
8 ; YAML-LABEL: --- !Passed
9 ; YAML-NEXT: Pass: lower-matrix-intrinsics
10 ; YAML-NEXT: Name: matrix-lowered
11 ; YAML-NEXT: DebugLoc: { File: test.cpp, Line: 35, Column: 71 }
12 ; YAML-NEXT: Function: test_2leafs
13 ; YAML-NEXT: Args:
14 ; YAML-NEXT: - String: 'Lowered with '
15 ; YAML-NEXT: - NumStores: '4'
16 ; YAML-NEXT: - String: ' stores, '
17 ; YAML-NEXT: - NumLoads: '0'
18 ; YAML-NEXT: - String: ' loads, '
19 ; YAML-NEXT: - NumComputeOps: '0'
20 ; YAML-NEXT: - String: ' compute ops, '
21 ; YAML-NEXT: - NumExposedTransposes: '0'
22 ; YAML-NEXT: - String: ' exposed transposes'
23 ; YAML-NEXT: - String: ",\nadditionally "
24 ; YAML-NEXT: - NumStores: '0'
25 ; YAML-NEXT: - String: ' stores, '
26 ; YAML-NEXT: - NumLoads: '4'
27 ; YAML-NEXT: - String: ' loads, '
28 ; YAML-NEXT: - NumFPOps: '16'
29 ; YAML-NEXT: - String: ' compute ops'
30 ; YAML-NEXT: - String: ' are shared with other expressions'
31 ; YAML-NEXT: - String: |
32 ; YAML: column.major.store.4x2.double(
33 ; YAML-NEXT: shared with remark at line 35 column 45 (transpose.2x4.double(column.major.load.2x4.double(addr %arg1,
34 ; YAML-NEXT: scalar)),
35 ; YAML-NEXT: addr %arg3,
36 ; YAML-NEXT: 10)
38 ; YAML-LABEL: --- !Passed
39 ; YAML-NEXT: Pass: lower-matrix-intrinsics
40 ; YAML-NEXT: Name: matrix-lowered
41 ; YAML-NEXT: DebugLoc: { File: test.cpp, Line: 35, Column: 45 }
42 ; YAML-NEXT: Function: test_2leafs
43 ; YAML-NEXT: Args:
44 ; YAML-NEXT: - String: 'Lowered with '
45 ; YAML-NEXT: - NumStores: '30'
46 ; YAML-NEXT: - String: ' stores, '
47 ; YAML-NEXT: - NumLoads: '45'
48 ; YAML-NEXT: - String: ' loads, '
49 ; YAML-NEXT: - NumComputeOps: '120'
50 ; YAML-NEXT: - String: ' compute ops, '
51 ; YAML-NEXT: - NumExposedTransposes: '0'
52 ; YAML-NEXT: - String: ' exposed transposes'
53 ; YAML-NEXT: - String: ",\nadditionally "
54 ; YAML-NEXT: - NumStores: '0'
55 ; YAML-NEXT: - String: ' stores, '
56 ; YAML-NEXT: - NumLoads: '4'
57 ; YAML-NEXT: - String: ' loads, '
58 ; YAML-NEXT: - NumFPOps: '16'
59 ; YAML-NEXT: - String: ' compute ops'
60 ; YAML-NEXT: - String: ' are shared with other expressions'
61 ; YAML-NEXT: - String: |
62 ; YAML: column.major.store.4x15.double(
63 ; YAML-NEXT: fsub(
64 ; YAML-NEXT: column.major.load.4x15.double(addr %arg2, 20),
65 ; YAML-NEXT: multiply.4x2.2x15.double(
66 ; YAML-NEXT: shared with remark at line 35 column 71 (transpose.2x4.double(column.major.load.2x4.double(addr %arg1,
67 ; YAML-NEXT: scalar)),
68 ; YAML-NEXT: column.major.load.2x15.double(addr %arg3, scalar))),
69 ; YAML-NEXT: addr %arg2,
70 ; YAML-NEXT: 10)
73 ; STDERR-LABEL: remark: test.cpp:35:71: Lowered with 4 stores, 0 loads, 0 compute ops,
74 ; STDERR-NEXT: additionally 0 stores, 4 loads, 16 compute ops are shared with other expressions
75 ; STDERR-NEXT: column.major.store.4x2.double(
76 ; STDERR-NEXT: shared with remark at line 35 column 45 (transpose.2x4.double(column.major.load.2x4.double(addr %arg1,
77 ; STDERR-NEXT: scalar)),
78 ; STDERR-NEXT: addr %arg3,
79 ; STDERR-NEXT: 10)
81 ; STDERR-LABEL: remark: test.cpp:35:45: Lowered with 30 stores, 45 loads, 120 compute ops,
82 ; STDERR-NEXT: additionally 0 stores, 4 loads, 16 compute ops are shared with other expressions
83 ; STDERR-NEXT: column.major.store.4x15.double(
84 ; STDERR-NEXT: fsub(
85 ; STDERR-NEXT: column.major.load.4x15.double(addr %arg2, 20),
86 ; STDERR-NEXT: multiply.4x2.2x15.double(
87 ; STDERR-NEXT: shared with remark at line 35 column 71 (transpose.2x4.double(column.major.load.2x4.double(addr %arg1,
88 ; STDERR-NEXT: scalar)),
89 ; STDERR-NEXT: column.major.load.2x15.double(addr %arg3, scalar))),
90 ; STDERR-NEXT: addr %arg2,
91 ; STDERR-NEXT: 10)
92 define void @test_2leafs(double* %arg1, double* %arg2, double* %arg3, i64 %stride) !dbg !8 {
93 bb:
94 %shared.load = tail call <8 x double> @llvm.matrix.column.major.load.v8f64.p0f64(double* %arg1, i64 %stride, i1 false, i32 2, i32 4), !dbg !10, !noalias !10
95 %shared.load.2 = tail call <30 x double> @llvm.matrix.column.major.load.v30f64.p0f64(double* %arg3, i64 %stride, i1 false, i32 2, i32 15), !dbg !10, !noalias !10
96 %tmp17 = tail call <8 x double> @llvm.matrix.transpose.v8f64(<8 x double> %shared.load, i32 2, i32 4), !dbg !10
97 tail call void @llvm.matrix.column.major.store.v8f64.p0f64(<8 x double> %tmp17, double* %arg3, i64 10, i1 false, i32 4, i32 2), !dbg !10
98 %tmp18 = tail call <60 x double> @llvm.matrix.column.major.load.v60f64.p0f64(double* %arg2, i64 20, i1 false, i32 4, i32 15), !dbg !11
99 %tmp48 = tail call <60 x double> @llvm.matrix.multiply.v60f64.v8f64.v30f64(<8 x double> %tmp17, <30 x double> %shared.load.2, i32 4, i32 2, i32 15), !dbg !11
100 %tmp49 = fsub <60 x double> %tmp18, %tmp48, !dbg !11
101 tail call void @llvm.matrix.column.major.store.v60f64.p0f64(<60 x double> %tmp49, double* %arg2, i64 10, i1 false, i32 4, i32 15), !dbg !11
102 ret void
103 }
105 declare <8 x double> @llvm.matrix.transpose.v8f64(<8 x double>, i32 immarg, i32 immarg)
106 declare <8 x double> @llvm.matrix.column.major.load.v8f64.p0f64(double*, i64, i1 immarg, i32 immarg, i32 immarg)
107 declare <30 x double> @llvm.matrix.column.major.load.v30f64.p0f64(double*, i64, i1 immarg, i32 immarg, i32 immarg)
108 declare <60 x double> @llvm.matrix.column.major.load.v60f64.p0f64(double*, i64, i1 immarg, i32 immarg, i32 immarg)
109 declare void @llvm.matrix.column.major.store.v60f64.p0f64(<60 x double>, double* writeonly, i64, i1 immarg, i32 immarg, i32 immarg)
110 declare void @llvm.matrix.column.major.store.v8f64.p0f64(<8 x double>, double* writeonly, i64, i1 immarg, i32 immarg, i32 immarg)
111 declare <60 x double> @llvm.matrix.multiply.v60f64.v8f64.v30f64(<8 x double>, <30 x double>, i32 immarg, i32 immarg, i32 immarg)
113 !llvm.module.flags = !{!0, !1, !2, !3}
114 !llvm.dbg.cu = !{!4}
115 !llvm.ident = !{!7}
117 !0 = !{i32 2, !"SDK Version", [2 x i32] [i32 13, i32 0]}
118 !1 = !{i32 2, !"Debug Info Version", i32 3}
119 !2 = !{i32 1, !"wchar_size", i32 4}
120 !3 = !{i32 7, !"PIC Level", i32 2}
121 !4 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !5, producer: "clang", isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug, enums: !6, nameTableKind: GNU)
122 !5 = !DIFile(filename: "test.cpp", directory: "")
123 !6 = !{}
124 !7 = !{!"clang"}
125 !8 = distinct !DISubprogram(name: "test", scope: !5, file: !5, line: 26, type: !9, scopeLine: 27, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
126 !9 = !DISubroutineType(types: !6)
127 !10 = distinct !DILocation(line: 35, column: 71, scope: !8)
128 !11 = distinct !DILocation(line: 35, column: 45, scope: !8)
129 !12 = !DILocation(line: 800, column: 17, scope: !13, inlinedAt: !15)
130 !13 = distinct !DISubprogram(name: "foo", scope: !14, file: !14, line: 789, type: !9, scopeLine: 790, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
131 !14 = !DIFile(filename: "bar.h", directory: "bar")
132 !15 = distinct !DILocation(line: 1280, column: 5, scope: !16, inlinedAt: !18)
133 !16 = distinct !DISubprogram(name: "zar", scope: !17, file: !17, line: 1275, type: !9, scopeLine: 1278, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
134 !17 = !DIFile(filename: "file1.h", directory: "dir1")
135 !18 = distinct !DILocation(line: 1278, column: 1, scope: !19, inlinedAt: !20)
136 !19 = distinct !DISubprogram(name: "yo", scope: !17, file: !17, line: 1275, type: !9, scopeLine: 1278, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
137 !20 = distinct !DILocation(line: 2514, column: 26, scope: !21, inlinedAt: !22)
138 !21 = distinct !DISubprogram(name: "zzzz", scope: !14, file: !14, line: 2505, type: !9, scopeLine: 2506, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
139 !22 = distinct !DILocation(line: 1263, column: 5, scope: !23, inlinedAt: !24)
140 !23 = distinct !DISubprogram(name: "ppppp", scope: !17, file: !17, line: 1258, type: !9, scopeLine: 1261, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
141 !24 = distinct !DILocation(line: 1261, column: 1, scope: !25, inlinedAt: !26)
142 !25 = distinct !DISubprogram(name: "qqqq", scope: !17, file: !17, line: 1258, type: !9, scopeLine: 1261, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
143 !26 = distinct !DILocation(line: 168, column: 7, scope: !27, inlinedAt: !29)
144 !27 = distinct !DISubprogram(name: "lll", scope: !28, file: !28, line: 166, type: !9, scopeLine: 169, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
145 !28 = !DIFile(filename: "file2.h", directory: "dir2")
146 !29 = distinct !DILocation(line: 169, column: 1, scope: !30, inlinedAt: !31)
147 !30 = distinct !DISubprogram(name: "Expr1", scope: !28, file: !28, line: 166, type: !9, scopeLine: 169, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
148 !31 = distinct !DILocation(line: 368, column: 12, scope: !32, inlinedAt: !33)
149 !32 = distinct !DISubprogram(name: "yyyyy", scope: !14, file: !14, line: 364, type: !9, scopeLine: 365, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
150 !33 = distinct !DILocation(line: 1297, column: 34, scope: !34, inlinedAt: !35)
151 !34 = distinct !DISubprogram(name: "eeeee", scope: !14, file: !14, line: 1290, type: !9, scopeLine: 1291, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
152 !35 = distinct !DILocation(line: 2306, column: 5, scope: !36, inlinedAt: !11)
153 !36 = distinct !DISubprogram(name: "aaaaa", scope: !37, file: !37, line: 2304, type: !9, scopeLine: 2305, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
154 !37 = !DIFile(filename: "foo.c", directory: "/")
155 !38 = distinct !DISubprogram(name: "test2", scope: !5, file: !5, line: 90, type: !9, scopeLine: 27, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
156 !39 = distinct !DILocation(line: 44, column: 44, scope: !38)
157 !40 = distinct !DILocation(line: 55, column: 55, scope: !38)
158 !41 = distinct !DILocation(line: 66, column: 66, scope: !38)
159 !42 = distinct !DISubprogram(name: "test2", scope: !5, file: !5, line: 90, type: !9, scopeLine: 27, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
160 !43 = distinct !DILocation(line: 77, column: 77, scope: !42)
161 !44 = distinct !DILocation(line: 88, column: 88, scope: !42)
162 !45 = distinct !DISubprogram(name: "test2", scope: !5, file: !5, line: 90, type: !9, scopeLine: 27, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !4, retainedNodes: !6)
163 !46 = distinct !DILocation(line: 99, column: 99, scope: !45)
164 !47 = distinct !DILocation(line: 111, column: 111, scope: !45)