1; Verify that we can create unaligned loads and stores from VSX intrinsics.
2
3; RUN: opt < %s -instcombine -S | FileCheck %s
4
5target triple = "powerpc64-unknown-linux-gnu"
6
7@vf = common global <4 x float> zeroinitializer, align 1
8@res_vf = common global <4 x float> zeroinitializer, align 1
9@vd = common global <2 x double> zeroinitializer, align 1
10@res_vd = common global <2 x double> zeroinitializer, align 1
11
12define void @test1() {
13entry:
14  %t1 = alloca <4 x float>*, align 8
15  %t2 = alloca <2 x double>*, align 8
16  store <4 x float>* @vf, <4 x float>** %t1, align 8
17  %0 = load <4 x float>*, <4 x float>** %t1, align 8
18  %1 = bitcast <4 x float>* %0 to i8*
19  %2 = call <4 x i32> @llvm.ppc.vsx.lxvw4x(i8* %1)
20  store <4 x float>* @res_vf, <4 x float>** %t1, align 8
21  %3 = load <4 x float>*, <4 x float>** %t1, align 8
22  %4 = bitcast <4 x float>* %3 to i8*
23  call void @llvm.ppc.vsx.stxvw4x(<4 x i32> %2, i8* %4)
24  store <2 x double>* @vd, <2 x double>** %t2, align 8
25  %5 = load <2 x double>*, <2 x double>** %t2, align 8
26  %6 = bitcast <2 x double>* %5 to i8*
27  %7 = call <2 x double> @llvm.ppc.vsx.lxvd2x(i8* %6)
28  store <2 x double>* @res_vd, <2 x double>** %t2, align 8
29  %8 = load <2 x double>*, <2 x double>** %t2, align 8
30  %9 = bitcast <2 x double>* %8 to i8*
31  call void @llvm.ppc.vsx.stxvd2x(<2 x double> %7, i8* %9)
32  ret void
33}
34
35; CHECK-LABEL: @test1
36; CHECK: %0 = load <4 x i32>, <4 x i32>* bitcast (<4 x float>* @vf to <4 x i32>*), align 1
37; CHECK: store <4 x i32> %0, <4 x i32>* bitcast (<4 x float>* @res_vf to <4 x i32>*), align 1
38; CHECK: %1 = load <2 x double>, <2 x double>* @vd, align 1
39; CHECK: store <2 x double> %1, <2 x double>* @res_vd, align 1
40
41declare <4 x i32> @llvm.ppc.vsx.lxvw4x(i8*)
42declare void @llvm.ppc.vsx.stxvw4x(<4 x i32>, i8*)
43declare <2 x double> @llvm.ppc.vsx.lxvd2x(i8*)
44declare void @llvm.ppc.vsx.stxvd2x(<2 x double>, i8*)
45