Mercurial > hg > CbC > CbC_llvm

comparison llvm/test/Transforms/LoopDistribute/basic.ll @ 252:1f2b6ac9f198 llvm-original

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM16-1
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Fri, 18 Aug 2023 09:04:13 +0900
parents c4bab56944e8
children
comparison
equal deleted inserted replaced
237:c80f45b162ad 252:1f2b6ac9f198
17 17
18 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 18 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
19 target triple = "x86_64-apple-macosx10.10.0" 19 target triple = "x86_64-apple-macosx10.10.0"
20 20
21 ; CHECK-LABEL: @f( 21 ; CHECK-LABEL: @f(
22 define void @f(i32* noalias %a, 22 define void @f(ptr noalias %a,
23 i32* noalias %b, 23 ptr noalias %b,
24 i32* noalias %c, 24 ptr noalias %c,
25 i32* noalias %d, 25 ptr noalias %d,
26 i32* noalias %e) { 26 ptr noalias %e) {
27 entry: 27 entry:
28 br label %for.body 28 br label %for.body
29 29
30 ; Verify the two distributed loops. 30 ; Verify the two distributed loops.
31 31
51 ; VECTORIZE: mul <4 x i32> 51 ; VECTORIZE: mul <4 x i32>
52 52
53 for.body: ; preds = %for.body, %entry 53 for.body: ; preds = %for.body, %entry
54 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] 54 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
55 55
56 %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind 56 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %ind
57 %loadA = load i32, i32* %arrayidxA, align 4 57 %loadA = load i32, ptr %arrayidxA, align 4
58 58
59 %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind 59 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %ind
60 %loadB = load i32, i32* %arrayidxB, align 4 60 %loadB = load i32, ptr %arrayidxB, align 4
61 61
62 %mulA = mul i32 %loadB, %loadA 62 %mulA = mul i32 %loadB, %loadA
63 63
64 %add = add nuw nsw i64 %ind, 1 64 %add = add nuw nsw i64 %ind, 1
65 %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add 65 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
66 store i32 %mulA, i32* %arrayidxA_plus_4, align 4 66 store i32 %mulA, ptr %arrayidxA_plus_4, align 4
67 67
68 %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind 68 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %ind
69 %loadD = load i32, i32* %arrayidxD, align 4 69 %loadD = load i32, ptr %arrayidxD, align 4
70 70
71 %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind 71 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %ind
72 %loadE = load i32, i32* %arrayidxE, align 4 72 %loadE = load i32, ptr %arrayidxE, align 4
73 73
74 %mulC = mul i32 %loadD, %loadE 74 %mulC = mul i32 %loadD, %loadE
75 75
76 %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind 76 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %ind
77 store i32 %mulC, i32* %arrayidxC, align 4 77 store i32 %mulC, ptr %arrayidxC, align 4
78 78
79 %exitcond = icmp eq i64 %add, 20 79 %exitcond = icmp eq i64 %add, 20
80 br i1 %exitcond, label %for.end, label %for.body 80 br i1 %exitcond, label %for.end, label %for.body
81 81
82 for.end: ; preds = %for.body 82 for.end: ; preds = %for.body
86 declare i32 @llvm.convergent(i32) #0 86 declare i32 @llvm.convergent(i32) #0
87 87
88 ; It is OK to distribute with a convergent operation, since in each 88 ; It is OK to distribute with a convergent operation, since in each
89 ; new loop the convergent operation has the ssame control dependency. 89 ; new loop the convergent operation has the ssame control dependency.
90 ; CHECK-LABEL: @f_with_convergent( 90 ; CHECK-LABEL: @f_with_convergent(
91 define void @f_with_convergent(i32* noalias %a, 91 define void @f_with_convergent(ptr noalias %a,
92 i32* noalias %b, 92 ptr noalias %b,
93 i32* noalias %c, 93 ptr noalias %c,
94 i32* noalias %d, 94 ptr noalias %d,
95 i32* noalias %e) { 95 ptr noalias %e) {
96 entry: 96 entry:
97 br label %for.body 97 br label %for.body
98 98
99 ; Verify the two distributed loops. 99 ; Verify the two distributed loops.
100 100
123 ; VECTORIZE: mul i32 123 ; VECTORIZE: mul i32
124 124
125 for.body: ; preds = %for.body, %entry 125 for.body: ; preds = %for.body, %entry
126 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] 126 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
127 127
128 %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind 128 %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %ind
129 %loadA = load i32, i32* %arrayidxA, align 4 129 %loadA = load i32, ptr %arrayidxA, align 4
130 130
131 %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind 131 %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %ind
132 %loadB = load i32, i32* %arrayidxB, align 4 132 %loadB = load i32, ptr %arrayidxB, align 4
133 133
134 %mulA = mul i32 %loadB, %loadA 134 %mulA = mul i32 %loadB, %loadA
135 135
136 %add = add nuw nsw i64 %ind, 1 136 %add = add nuw nsw i64 %ind, 1
137 %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add 137 %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
138 store i32 %mulA, i32* %arrayidxA_plus_4, align 4 138 store i32 %mulA, ptr %arrayidxA_plus_4, align 4
139 139
140 %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind 140 %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %ind
141 %loadD = load i32, i32* %arrayidxD, align 4 141 %loadD = load i32, ptr %arrayidxD, align 4
142 142
143 %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind 143 %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %ind
144 %loadE = load i32, i32* %arrayidxE, align 4 144 %loadE = load i32, ptr %arrayidxE, align 4
145 145
146 %convergentD = call i32 @llvm.convergent(i32 %loadD) 146 %convergentD = call i32 @llvm.convergent(i32 %loadD)
147 %mulC = mul i32 %convergentD, %loadE 147 %mulC = mul i32 %convergentD, %loadE
148 148
149 %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind 149 %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %ind
150 store i32 %mulC, i32* %arrayidxC, align 4 150 store i32 %mulC, ptr %arrayidxC, align 4
151 151
152 %exitcond = icmp eq i64 %add, 20 152 %exitcond = icmp eq i64 %add, 20
153 br i1 %exitcond, label %for.end, label %for.body 153 br i1 %exitcond, label %for.end, label %for.body
154 154
155 for.end: ; preds = %for.body 155 for.end: ; preds = %for.body