1; RUN: llc -march=hexagon -enable-pipeliner=false < %s | FileCheck %s
2
3; Test that we generate a .cur
4
5; CHECK: v{{[0-9]*}}.cur{{ *}}
6; CHECK: v{{[0-9]*}}.cur{{ *}}
7
8define void @conv3x3_i(i8* noalias nocapture readonly %iptr0, i32 %shift, i32 %width) #0 {
9entry:
10  br i1 undef, label %for.body.lr.ph, label %for.end
11
12for.body.lr.ph:
13  br label %for.body
14
15for.body:
16  %iptr0.pn = phi i8* [ %iptr0, %for.body.lr.ph ], [ %iptr0.addr.0121, %for.body ]
17  %j.0115 = phi i32 [ 0, %for.body.lr.ph ], [ %add, %for.body ]
18  %sline000.0114 = phi <16 x i32> [ zeroinitializer, %for.body.lr.ph ], [ %1, %for.body ]
19  %sline100.0113 = phi <16 x i32> [ zeroinitializer, %for.body.lr.ph ], [ zeroinitializer, %for.body ]
20  %iptr0.addr.0121 = getelementptr inbounds i8, i8* %iptr0.pn, i32 64
21  %0 = bitcast i8* %iptr0.addr.0121 to <16 x i32>*
22  %1 = load <16 x i32>, <16 x i32>* %0, align 64, !tbaa !1
23  %2 = load <16 x i32>, <16 x i32>* null, align 64, !tbaa !1
24  %3 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %1, <16 x i32> %sline000.0114, i32 4)
25  %4 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> zeroinitializer, <16 x i32> %sline100.0113, i32 4)
26  %5 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %2, <16 x i32> zeroinitializer, i32 4)
27  %6 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %3, <16 x i32> %sline000.0114)
28  %7 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %5, <16 x i32> zeroinitializer)
29  %8 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi(<32 x i32> %6, i32 0, i32 0)
30  %9 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> %8, <32 x i32> zeroinitializer, i32 undef, i32 0)
31  %10 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> %9, <32 x i32> undef, i32 undef, i32 0)
32  %11 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %10)
33  %12 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %11, <16 x i32> undef, i32 %shift)
34  %13 = tail call <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32> undef, <16 x i32> %12)
35  store <16 x i32> %13, <16 x i32>* undef, align 64, !tbaa !1
36  %14 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> zeroinitializer, <32 x i32> %7, i32 undef, i32 1)
37  %15 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %14)
38  %16 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %15, <16 x i32> undef, i32 %shift)
39  %17 = tail call <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32> %16, <16 x i32> undef)
40  store <16 x i32> %17, <16 x i32>* undef, align 64, !tbaa !1
41  %add = add nsw i32 %j.0115, 64
42  %cmp = icmp slt i32 %add, %width
43  br i1 %cmp, label %for.body, label %for.end
44
45for.end:
46  ret void
47}
48
49declare <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32>, <16 x i32>, i32) #1
50declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #1
51declare <32 x i32> @llvm.hexagon.V6.vrmpybusi(<32 x i32>, i32, i32) #1
52declare <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32>, <32 x i32>, i32, i32) #1
53declare <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32>, <16 x i32>, i32) #1
54declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #1
55declare <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32>, <16 x i32>) #1
56
57attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvx" }
58attributes #1 = { nounwind readnone }
59
60!1 = !{!2, !2, i64 0}
61!2 = !{!"omnipotent char", !3, i64 0}
62!3 = !{!"Simple C/C++ TBAA"}
63