1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -interleaved-access -S %s | FileCheck %s
3
4target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-apple-macosx10.15.0"
6
7; No interleaved load instruction is generated, but the shuffle is moved just
8; after the load.
9define <2 x double> @shuffle_binop_fol(<4 x double>* %ptr) {
10; CHECK-LABEL: @shuffle_binop_fol(
11; CHECK-NEXT:  vector.body.preheader:
12; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x double>, <4 x double>* [[PTR:%.*]], align 8
13; CHECK-NEXT:    [[EXTRACTED1:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> poison, <2 x i32> <i32 0, i32 2>
14; CHECK-NEXT:    [[EXTRACTED2:%.*]] = shufflevector <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, <4 x double> poison, <2 x i32> <i32 0, i32 2>
15; CHECK-NEXT:    [[FADD3:%.*]] = fadd <2 x double> [[EXTRACTED1]], [[EXTRACTED2]]
16; CHECK-NEXT:    ret <2 x double> [[FADD3]]
17;
18vector.body.preheader:
19  %wide.load = load <4 x double>, <4 x double>* %ptr, align 8
20  %fadd = fadd <4 x double> %wide.load, <double 1.0, double 1.0, double 1.0, double 1.0>
21  %extracted = shufflevector <4 x double> %fadd, <4 x double> undef, <2 x i32> <i32 0, i32 2>
22  ret <2 x double> %extracted
23}
24
25define <2 x double> @shuffle_binop_fol_oob(<4 x double>* %ptr) {
26; CHECK-LABEL: @shuffle_binop_fol_oob(
27; CHECK-NEXT:  vector.body.preheader:
28; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x double>, <4 x double>* [[PTR:%.*]], align 8
29; CHECK-NEXT:    [[FADD:%.*]] = fadd <4 x double> [[WIDE_LOAD]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>
30; CHECK-NEXT:    [[EXTRACTED:%.*]] = shufflevector <4 x double> [[FADD]], <4 x double> undef, <2 x i32> <i32 0, i32 4>
31; CHECK-NEXT:    ret <2 x double> [[EXTRACTED]]
32;
33vector.body.preheader:
34  %wide.load = load <4 x double>, <4 x double>* %ptr, align 8
35  %fadd = fadd <4 x double> %wide.load, <double 1.0, double 1.0, double 1.0, double 1.0>
36  %extracted = shufflevector <4 x double> %fadd, <4 x double> undef, <2 x i32> <i32 0, i32 4>
37  ret <2 x double> %extracted
38}
39
40; No interleaved load instruction is generated, but the extractelement
41; instructions are updated to use the shuffle instead of the load.
42define void @shuffle_extract(<4 x double>* %ptr, i1 %c) {
43; CHECK-LABEL: @shuffle_extract(
44; CHECK-NEXT:  vector.body.preheader:
45; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x double>, <4 x double>* [[PTR:%.*]], align 8
46; CHECK-NEXT:    [[EXTRACTED:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> undef, <2 x i32> <i32 0, i32 2>
47; CHECK-NEXT:    br i1 [[C:%.*]], label [[IF_THEN:%.*]], label [[IF_MERGE:%.*]]
48; CHECK:       if.then:
49; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <2 x double> [[EXTRACTED]], i64 0
50; CHECK-NEXT:    call void @use(double [[TMP0]])
51; CHECK-NEXT:    br label [[IF_MERGE]]
52; CHECK:       if.merge:
53; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[EXTRACTED]], i64 1
54; CHECK-NEXT:    call void @use(double [[TMP1]])
55; CHECK-NEXT:    ret void
56;
57vector.body.preheader:
58  %wide.load = load <4 x double>, <4 x double>* %ptr, align 8
59  %extracted = shufflevector <4 x double> %wide.load, <4 x double> undef, <2 x i32> <i32 0, i32 2>
60  br i1 %c, label %if.then, label %if.merge
61
62if.then:
63  %e0 = extractelement <4 x double> %wide.load, i32 0
64  call void @use(double %e0)
65  br label %if.merge
66
67if.merge:
68  %e1 = extractelement <4 x double> %wide.load, i32 2
69  call void @use(double %e1)
70  ret void
71}
72
73declare void @use(double)
74