1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE
3 ; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX
4 ; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX
6 define void @powof2div_uniform(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, ptr noalias nocapture readonly %c){
7 ; CHECK-LABEL: @powof2div_uniform(
9 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[B:%.*]], align 4
10 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[C:%.*]], align 4
11 ; CHECK-NEXT: [[TMP4:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP1]]
12 ; CHECK-NEXT: [[TMP5:%.*]] = sdiv <4 x i32> [[TMP4]], splat (i32 2)
13 ; CHECK-NEXT: store <4 x i32> [[TMP5]], ptr [[A:%.*]], align 4
14 ; CHECK-NEXT: ret void
17 %0 = load i32, ptr %b, align 4
18 %1 = load i32, ptr %c, align 4
19 %add = add nsw i32 %1, %0
20 %div = sdiv i32 %add, 2
21 store i32 %div, ptr %a, align 4
22 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 1
23 %2 = load i32, ptr %arrayidx3, align 4
24 %arrayidx4 = getelementptr inbounds i32, ptr %c, i64 1
25 %3 = load i32, ptr %arrayidx4, align 4
26 %add5 = add nsw i32 %3, %2
27 %div6 = sdiv i32 %add5, 2
28 %arrayidx7 = getelementptr inbounds i32, ptr %a, i64 1
29 store i32 %div6, ptr %arrayidx7, align 4
30 %arrayidx8 = getelementptr inbounds i32, ptr %b, i64 2
31 %4 = load i32, ptr %arrayidx8, align 4
32 %arrayidx9 = getelementptr inbounds i32, ptr %c, i64 2
33 %5 = load i32, ptr %arrayidx9, align 4
34 %add10 = add nsw i32 %5, %4
35 %div11 = sdiv i32 %add10, 2
36 %arrayidx12 = getelementptr inbounds i32, ptr %a, i64 2
37 store i32 %div11, ptr %arrayidx12, align 4
38 %arrayidx13 = getelementptr inbounds i32, ptr %b, i64 3
39 %6 = load i32, ptr %arrayidx13, align 4
40 %arrayidx14 = getelementptr inbounds i32, ptr %c, i64 3
41 %7 = load i32, ptr %arrayidx14, align 4
42 %add15 = add nsw i32 %7, %6
43 %div16 = sdiv i32 %add15, 2
44 %arrayidx17 = getelementptr inbounds i32, ptr %a, i64 3
45 store i32 %div16, ptr %arrayidx17, align 4
49 define void @powof2div_nonuniform(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, ptr noalias nocapture readonly %c){
50 ; SSE-LABEL: @powof2div_nonuniform(
52 ; SSE-NEXT: [[TMP0:%.*]] = load i32, ptr [[B:%.*]], align 4
53 ; SSE-NEXT: [[TMP1:%.*]] = load i32, ptr [[C:%.*]], align 4
54 ; SSE-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP0]]
55 ; SSE-NEXT: [[DIV:%.*]] = sdiv i32 [[ADD]], 2
56 ; SSE-NEXT: store i32 [[DIV]], ptr [[A:%.*]], align 4
57 ; SSE-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 1
58 ; SSE-NEXT: [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4
59 ; SSE-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 1
60 ; SSE-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4
61 ; SSE-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP3]], [[TMP2]]
62 ; SSE-NEXT: [[DIV6:%.*]] = sdiv i32 [[ADD5]], 4
63 ; SSE-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 1
64 ; SSE-NEXT: store i32 [[DIV6]], ptr [[ARRAYIDX7]], align 4
65 ; SSE-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 2
66 ; SSE-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
67 ; SSE-NEXT: [[ARRAYIDX9:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 2
68 ; SSE-NEXT: [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX9]], align 4
69 ; SSE-NEXT: [[ADD10:%.*]] = add nsw i32 [[TMP5]], [[TMP4]]
70 ; SSE-NEXT: [[DIV11:%.*]] = sdiv i32 [[ADD10]], 8
71 ; SSE-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 2
72 ; SSE-NEXT: store i32 [[DIV11]], ptr [[ARRAYIDX12]], align 4
73 ; SSE-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 3
74 ; SSE-NEXT: [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX13]], align 4
75 ; SSE-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 3
76 ; SSE-NEXT: [[TMP7:%.*]] = load i32, ptr [[ARRAYIDX14]], align 4
77 ; SSE-NEXT: [[ADD15:%.*]] = add nsw i32 [[TMP7]], [[TMP6]]
78 ; SSE-NEXT: [[DIV16:%.*]] = sdiv i32 [[ADD15]], 16
79 ; SSE-NEXT: [[ARRAYIDX17:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 3
80 ; SSE-NEXT: store i32 [[DIV16]], ptr [[ARRAYIDX17]], align 4
83 ; AVX-LABEL: @powof2div_nonuniform(
85 ; AVX-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[B:%.*]], align 4
86 ; AVX-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[C:%.*]], align 4
87 ; AVX-NEXT: [[TMP4:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP1]]
88 ; AVX-NEXT: [[TMP5:%.*]] = sdiv <4 x i32> [[TMP4]], <i32 2, i32 4, i32 8, i32 16>
89 ; AVX-NEXT: store <4 x i32> [[TMP5]], ptr [[A:%.*]], align 4
93 %0 = load i32, ptr %b, align 4
94 %1 = load i32, ptr %c, align 4
95 %add = add nsw i32 %1, %0
96 %div = sdiv i32 %add, 2
97 store i32 %div, ptr %a, align 4
98 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 1
99 %2 = load i32, ptr %arrayidx3, align 4
100 %arrayidx4 = getelementptr inbounds i32, ptr %c, i64 1
101 %3 = load i32, ptr %arrayidx4, align 4
102 %add5 = add nsw i32 %3, %2
103 %div6 = sdiv i32 %add5, 4
104 %arrayidx7 = getelementptr inbounds i32, ptr %a, i64 1
105 store i32 %div6, ptr %arrayidx7, align 4
106 %arrayidx8 = getelementptr inbounds i32, ptr %b, i64 2
107 %4 = load i32, ptr %arrayidx8, align 4
108 %arrayidx9 = getelementptr inbounds i32, ptr %c, i64 2
109 %5 = load i32, ptr %arrayidx9, align 4
110 %add10 = add nsw i32 %5, %4
111 %div11 = sdiv i32 %add10, 8
112 %arrayidx12 = getelementptr inbounds i32, ptr %a, i64 2
113 store i32 %div11, ptr %arrayidx12, align 4
114 %arrayidx13 = getelementptr inbounds i32, ptr %b, i64 3
115 %6 = load i32, ptr %arrayidx13, align 4
116 %arrayidx14 = getelementptr inbounds i32, ptr %c, i64 3
117 %7 = load i32, ptr %arrayidx14, align 4
118 %add15 = add nsw i32 %7, %6
119 %div16 = sdiv i32 %add15, 16
120 %arrayidx17 = getelementptr inbounds i32, ptr %a, i64 3
121 store i32 %div16, ptr %arrayidx17, align 4