xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/AArch64/ext-trunc.ll (revision 31eaf86a1e8f1870e6ee4c42088a5213bde294b8) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -passes=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, ptr %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, ptr [[P:%.*]], i64 [[S0]]
18; CHECK-NEXT:    [[LOAD0:%.*]] = load i64, ptr [[GEP0]], align 4
19; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i32> [[SUB0]], <4 x i32> poison, <2 x i32> <i32 1, i32 2>
20; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[TMP0]], i32 0
21; CHECK-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
22; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[TMP2]]
23; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, ptr [[GEP1]], align 4
24; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
25; CHECK-NEXT:    [[TMP4:%.*]] = sext i32 [[TMP3]] to i64
26; CHECK-NEXT:    [[GEP2:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[TMP4]]
27; CHECK-NEXT:    [[LOAD2:%.*]] = load i64, ptr [[GEP2]], align 4
28; CHECK-NEXT:    [[E3:%.*]] = extractelement <4 x i32> [[SUB0]], i32 3
29; CHECK-NEXT:    [[S3:%.*]] = sext i32 [[E3]] to i64
30; CHECK-NEXT:    [[GEP3:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[S3]]
31; CHECK-NEXT:    [[LOAD3:%.*]] = load i64, ptr [[GEP3]], align 4
32; CHECK-NEXT:    call void @foo(i64 [[LOAD0]], i64 [[LOAD1]], i64 [[LOAD2]], i64 [[LOAD3]])
33; CHECK-NEXT:    ret void
34;
35entry:
36  %z0 = zext <4 x i16> %a to <4 x i32>
37  %z1 = zext <4 x i16> %b to <4 x i32>
38  %sub0 = sub <4 x i32> %z0, %z1
39  %e0 = extractelement <4 x i32> %sub0, i32 0
40  %s0 = sext i32 %e0 to i64
41  %gep0 = getelementptr inbounds i64, ptr %p, i64 %s0
42  %load0 = load i64, ptr %gep0
43  %e1 = extractelement <4 x i32> %sub0, i32 1
44  %s1 = sext i32 %e1 to i64
45  %gep1 = getelementptr inbounds i64, ptr %p, i64 %s1
46  %load1 = load i64, ptr %gep1
47  %e2 = extractelement <4 x i32> %sub0, i32 2
48  %s2 = sext i32 %e2 to i64
49  %gep2 = getelementptr inbounds i64, ptr %p, i64 %s2
50  %load2 = load i64, ptr %gep2
51  %e3 = extractelement <4 x i32> %sub0, i32 3
52  %s3 = sext i32 %e3 to i64
53  %gep3 = getelementptr inbounds i64, ptr %p, i64 %s3
54  %load3 = load i64, ptr %gep3
55  call void @foo(i64 %load0, i64 %load1, i64 %load2, i64 %load3)
56  ret void
57}
58
59define void @test2(<4 x i16> %a, <4 x i16> %b, i64 %c0, i64 %c1, i64 %c2, i64 %c3, ptr %p) {
60; CHECK-LABEL: @test2(
61; CHECK-NEXT:  entry:
62; CHECK-NEXT:    [[Z0:%.*]] = zext <4 x i16> [[A:%.*]] to <4 x i32>
63; CHECK-NEXT:    [[Z1:%.*]] = zext <4 x i16> [[B:%.*]] to <4 x i32>
64; CHECK-NEXT:    [[SUB0:%.*]] = sub <4 x i32> [[Z0]], [[Z1]]
65; CHECK-NEXT:    [[E0:%.*]] = extractelement <4 x i32> [[SUB0]], i32 0
66; CHECK-NEXT:    [[S0:%.*]] = sext i32 [[E0]] to i64
67; CHECK-NEXT:    [[A0:%.*]] = add i64 [[S0]], [[C0:%.*]]
68; CHECK-NEXT:    [[GEP0:%.*]] = getelementptr inbounds i64, ptr [[P:%.*]], i64 [[A0]]
69; CHECK-NEXT:    [[LOAD0:%.*]] = load i64, ptr [[GEP0]], align 4
70; CHECK-NEXT:    [[E1:%.*]] = extractelement <4 x i32> [[SUB0]], i32 1
71; CHECK-NEXT:    [[S1:%.*]] = sext i32 [[E1]] to i64
72; CHECK-NEXT:    [[A1:%.*]] = add i64 [[S1]], [[C1:%.*]]
73; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[A1]]
74; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, ptr [[GEP1]], align 4
75; CHECK-NEXT:    [[E2:%.*]] = extractelement <4 x i32> [[SUB0]], i32 2
76; CHECK-NEXT:    [[S2:%.*]] = sext i32 [[E2]] to i64
77; CHECK-NEXT:    [[A2:%.*]] = add i64 [[S2]], [[C2:%.*]]
78; CHECK-NEXT:    [[GEP2:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[A2]]
79; CHECK-NEXT:    [[LOAD2:%.*]] = load i64, ptr [[GEP2]], align 4
80; CHECK-NEXT:    [[E3:%.*]] = extractelement <4 x i32> [[SUB0]], i32 3
81; CHECK-NEXT:    [[S3:%.*]] = sext i32 [[E3]] to i64
82; CHECK-NEXT:    [[A3:%.*]] = add i64 [[S3]], [[C3:%.*]]
83; CHECK-NEXT:    [[GEP3:%.*]] = getelementptr inbounds i64, ptr [[P]], i64 [[A3]]
84; CHECK-NEXT:    [[LOAD3:%.*]] = load i64, ptr [[GEP3]], align 4
85; CHECK-NEXT:    call void @foo(i64 [[LOAD0]], i64 [[LOAD1]], i64 [[LOAD2]], i64 [[LOAD3]])
86; CHECK-NEXT:    ret void
87;
88entry:
89  %z0 = zext <4 x i16> %a to <4 x i32>
90  %z1 = zext <4 x i16> %b to <4 x i32>
91  %sub0 = sub <4 x i32> %z0, %z1
92  %e0 = extractelement <4 x i32> %sub0, i32 0
93  %s0 = sext i32 %e0 to i64
94  %a0 = add i64 %s0, %c0
95  %gep0 = getelementptr inbounds i64, ptr %p, i64 %a0
96  %load0 = load i64, ptr %gep0
97  %e1 = extractelement <4 x i32> %sub0, i32 1
98  %s1 = sext i32 %e1 to i64
99  %a1 = add i64 %s1, %c1
100  %gep1 = getelementptr inbounds i64, ptr %p, i64 %a1
101  %load1 = load i64, ptr %gep1
102  %e2 = extractelement <4 x i32> %sub0, i32 2
103  %s2 = sext i32 %e2 to i64
104  %a2 = add i64 %s2, %c2
105  %gep2 = getelementptr inbounds i64, ptr %p, i64 %a2
106  %load2 = load i64, ptr %gep2
107  %e3 = extractelement <4 x i32> %sub0, i32 3
108  %s3 = sext i32 %e3 to i64
109  %a3 = add i64 %s3, %c3
110  %gep3 = getelementptr inbounds i64, ptr %p, i64 %a3
111  %load3 = load i64, ptr %gep3
112  call void @foo(i64 %load0, i64 %load1, i64 %load2, i64 %load3)
113  ret void
114}
115