1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -slp-vectorizer -mtriple=aarch64--linux-gnu < %s | FileCheck %s
3
4target datalayout = "e-m:e-i32:64-i128:128-n32:64-S128"
5
6declare void @foo(i64, i64, i64, i64)
7
8define void @test1(<4 x i16> %a, <4 x i16> %b, i64* %p) {
9; Make sure types of sub and its sources are not extended.
10; CHECK-LABEL: @test1(
11; CHECK-NEXT:  entry:
12; CHECK-NEXT:    [[Z0:%.*]] = zext <4 x i16> [[A:%.*]] to <4 x i32>
13; CHECK-NEXT:    [[Z1:%.*]] = zext <4 x i16> [[B:%.*]] to <4 x i32>
14; CHECK-NEXT:    [[SUB0:%.*]] = sub <4 x i32> [[Z0]], [[Z1]]
15; CHECK-NEXT:    [[E0:%.*]] = extractelement <4 x i32> [[SUB0]], i32 0
16; CHECK-NEXT:    [[S0:%.*]] = sext i32 [[E0]] to i64
17; CHECK-NEXT:    [[GEP0:%.*]] = getelementptr inbounds i64, i64* [[P:%.*]], i64 [[S0]]
18; CHECK-NEXT:    [[LOAD0:%.*]] = load i64, i64* [[GEP0]], align 4
19; CHECK-NEXT:    [[E1:%.*]] = extractelement <4 x i32> [[SUB0]], i32 1
20; CHECK-NEXT:    [[S1:%.*]] = sext i32 [[E1]] to i64
21; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[S1]]
22; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, i64* [[GEP1]], align 4
23; CHECK-NEXT:    [[E2:%.*]] = extractelement <4 x i32> [[SUB0]], i32 2
24; CHECK-NEXT:    [[S2:%.*]] = sext i32 [[E2]] to i64
25; CHECK-NEXT:    [[GEP2:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[S2]]
26; CHECK-NEXT:    [[LOAD2:%.*]] = load i64, i64* [[GEP2]], align 4
27; CHECK-NEXT:    [[E3:%.*]] = extractelement <4 x i32> [[SUB0]], i32 3
28; CHECK-NEXT:    [[S3:%.*]] = sext i32 [[E3]] to i64
29; CHECK-NEXT:    [[GEP3:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[S3]]
30; CHECK-NEXT:    [[LOAD3:%.*]] = load i64, i64* [[GEP3]], align 4
31; CHECK-NEXT:    call void @foo(i64 [[LOAD0]], i64 [[LOAD1]], i64 [[LOAD2]], i64 [[LOAD3]])
32; CHECK-NEXT:    ret void
33;
34entry:
35  %z0 = zext <4 x i16> %a to <4 x i32>
36  %z1 = zext <4 x i16> %b to <4 x i32>
37  %sub0 = sub <4 x i32> %z0, %z1
38  %e0 = extractelement <4 x i32> %sub0, i32 0
39  %s0 = sext i32 %e0 to i64
40  %gep0 = getelementptr inbounds i64, i64* %p, i64 %s0
41  %load0 = load i64, i64* %gep0
42  %e1 = extractelement <4 x i32> %sub0, i32 1
43  %s1 = sext i32 %e1 to i64
44  %gep1 = getelementptr inbounds i64, i64* %p, i64 %s1
45  %load1 = load i64, i64* %gep1
46  %e2 = extractelement <4 x i32> %sub0, i32 2
47  %s2 = sext i32 %e2 to i64
48  %gep2 = getelementptr inbounds i64, i64* %p, i64 %s2
49  %load2 = load i64, i64* %gep2
50  %e3 = extractelement <4 x i32> %sub0, i32 3
51  %s3 = sext i32 %e3 to i64
52  %gep3 = getelementptr inbounds i64, i64* %p, i64 %s3
53  %load3 = load i64, i64* %gep3
54  call void @foo(i64 %load0, i64 %load1, i64 %load2, i64 %load3)
55  ret void
56}
57
58define void @test2(<4 x i16> %a, <4 x i16> %b, i64 %c0, i64 %c1, i64 %c2, i64 %c3, i64* %p) {
59; CHECK-LABEL: @test2(
60; CHECK-NEXT:  entry:
61; CHECK-NEXT:    [[Z0:%.*]] = zext <4 x i16> [[A:%.*]] to <4 x i32>
62; CHECK-NEXT:    [[Z1:%.*]] = zext <4 x i16> [[B:%.*]] to <4 x i32>
63; CHECK-NEXT:    [[SUB0:%.*]] = sub <4 x i32> [[Z0]], [[Z1]]
64; CHECK-NEXT:    [[TMP0:%.*]] = sext <4 x i32> [[SUB0]] to <4 x i64>
65; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i64> poison, i64 [[C0:%.*]], i32 0
66; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x i64> [[TMP1]], i64 [[C1:%.*]], i32 1
67; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x i64> [[TMP2]], i64 [[C2:%.*]], i32 2
68; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x i64> [[TMP3]], i64 [[C3:%.*]], i32 3
69; CHECK-NEXT:    [[TMP5:%.*]] = add <4 x i64> [[TMP0]], [[TMP4]]
70; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x i64> [[TMP5]], i32 0
71; CHECK-NEXT:    [[GEP0:%.*]] = getelementptr inbounds i64, i64* [[P:%.*]], i64 [[TMP6]]
72; CHECK-NEXT:    [[LOAD0:%.*]] = load i64, i64* [[GEP0]], align 4
73; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[TMP5]], i32 1
74; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[TMP7]]
75; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, i64* [[GEP1]], align 4
76; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x i64> [[TMP5]], i32 2
77; CHECK-NEXT:    [[GEP2:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[TMP8]]
78; CHECK-NEXT:    [[LOAD2:%.*]] = load i64, i64* [[GEP2]], align 4
79; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i64> [[TMP5]], i32 3
80; CHECK-NEXT:    [[GEP3:%.*]] = getelementptr inbounds i64, i64* [[P]], i64 [[TMP9]]
81; CHECK-NEXT:    [[LOAD3:%.*]] = load i64, i64* [[GEP3]], align 4
82; CHECK-NEXT:    call void @foo(i64 [[LOAD0]], i64 [[LOAD1]], i64 [[LOAD2]], i64 [[LOAD3]])
83; CHECK-NEXT:    ret void
84;
85entry:
86  %z0 = zext <4 x i16> %a to <4 x i32>
87  %z1 = zext <4 x i16> %b to <4 x i32>
88  %sub0 = sub <4 x i32> %z0, %z1
89  %e0 = extractelement <4 x i32> %sub0, i32 0
90  %s0 = sext i32 %e0 to i64
91  %a0 = add i64 %s0, %c0
92  %gep0 = getelementptr inbounds i64, i64* %p, i64 %a0
93  %load0 = load i64, i64* %gep0
94  %e1 = extractelement <4 x i32> %sub0, i32 1
95  %s1 = sext i32 %e1 to i64
96  %a1 = add i64 %s1, %c1
97  %gep1 = getelementptr inbounds i64, i64* %p, i64 %a1
98  %load1 = load i64, i64* %gep1
99  %e2 = extractelement <4 x i32> %sub0, i32 2
100  %s2 = sext i32 %e2 to i64
101  %a2 = add i64 %s2, %c2
102  %gep2 = getelementptr inbounds i64, i64* %p, i64 %a2
103  %load2 = load i64, i64* %gep2
104  %e3 = extractelement <4 x i32> %sub0, i32 3
105  %s3 = sext i32 %e3 to i64
106  %a3 = add i64 %s3, %c3
107  %gep3 = getelementptr inbounds i64, i64* %p, i64 %a3
108  %load3 = load i64, i64* %gep3
109  call void @foo(i64 %load0, i64 %load1, i64 %load2, i64 %load3)
110  ret void
111}
112