|
134
|
1 ; RUN: llc -march=hexagon < %s | FileCheck %s
|
|
120
|
2
|
|
|
3 ; Test that we generate a .cur
|
|
|
4
|
|
134
|
5 ; CHECK: v{{[0-9]*}}.cur
|
|
120
|
6
|
|
|
7 define void @conv3x3_i(i8* noalias nocapture readonly %iptr0, i32 %shift, i32 %width) #0 {
|
|
|
8 entry:
|
|
|
9 br i1 undef, label %for.body.lr.ph, label %for.end
|
|
|
10
|
|
|
11 for.body.lr.ph:
|
|
|
12 br label %for.body
|
|
|
13
|
|
|
14 for.body:
|
|
|
15 %iptr0.pn = phi i8* [ %iptr0, %for.body.lr.ph ], [ %iptr0.addr.0121, %for.body ]
|
|
|
16 %j.0115 = phi i32 [ 0, %for.body.lr.ph ], [ %add, %for.body ]
|
|
|
17 %sline000.0114 = phi <16 x i32> [ zeroinitializer, %for.body.lr.ph ], [ %1, %for.body ]
|
|
|
18 %sline100.0113 = phi <16 x i32> [ zeroinitializer, %for.body.lr.ph ], [ zeroinitializer, %for.body ]
|
|
|
19 %iptr0.addr.0121 = getelementptr inbounds i8, i8* %iptr0.pn, i32 64
|
|
|
20 %0 = bitcast i8* %iptr0.addr.0121 to <16 x i32>*
|
|
|
21 %1 = load <16 x i32>, <16 x i32>* %0, align 64, !tbaa !1
|
|
|
22 %2 = load <16 x i32>, <16 x i32>* null, align 64, !tbaa !1
|
|
|
23 %3 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %1, <16 x i32> %sline000.0114, i32 4)
|
|
|
24 %4 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> zeroinitializer, <16 x i32> %sline100.0113, i32 4)
|
|
|
25 %5 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %2, <16 x i32> zeroinitializer, i32 4)
|
|
|
26 %6 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %3, <16 x i32> %sline000.0114)
|
|
|
27 %7 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %5, <16 x i32> zeroinitializer)
|
|
|
28 %8 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi(<32 x i32> %6, i32 0, i32 0)
|
|
|
29 %9 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> %8, <32 x i32> zeroinitializer, i32 undef, i32 0)
|
|
|
30 %10 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> %9, <32 x i32> undef, i32 undef, i32 0)
|
|
|
31 %11 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %10)
|
|
|
32 %12 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %11, <16 x i32> undef, i32 %shift)
|
|
|
33 %13 = tail call <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32> undef, <16 x i32> %12)
|
|
|
34 store <16 x i32> %13, <16 x i32>* undef, align 64, !tbaa !1
|
|
|
35 %14 = tail call <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32> zeroinitializer, <32 x i32> %7, i32 undef, i32 1)
|
|
|
36 %15 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %14)
|
|
|
37 %16 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %15, <16 x i32> undef, i32 %shift)
|
|
|
38 %17 = tail call <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32> %16, <16 x i32> undef)
|
|
|
39 store <16 x i32> %17, <16 x i32>* undef, align 64, !tbaa !1
|
|
|
40 %add = add nsw i32 %j.0115, 64
|
|
|
41 %cmp = icmp slt i32 %add, %width
|
|
|
42 br i1 %cmp, label %for.body, label %for.end
|
|
|
43
|
|
|
44 for.end:
|
|
|
45 ret void
|
|
|
46 }
|
|
|
47
|
|
|
48 declare <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32>, <16 x i32>, i32) #1
|
|
|
49 declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #1
|
|
|
50 declare <32 x i32> @llvm.hexagon.V6.vrmpybusi(<32 x i32>, i32, i32) #1
|
|
|
51 declare <32 x i32> @llvm.hexagon.V6.vrmpybusi.acc(<32 x i32>, <32 x i32>, i32, i32) #1
|
|
|
52 declare <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32>, <16 x i32>, i32) #1
|
|
|
53 declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #1
|
|
|
54 declare <16 x i32> @llvm.hexagon.V6.vsathub(<16 x i32>, <16 x i32>) #1
|
|
|
55
|
|
121
|
56 attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
|
|
120
|
57 attributes #1 = { nounwind readnone }
|
|
|
58
|
|
|
59 !1 = !{!2, !2, i64 0}
|
|
|
60 !2 = !{!"omnipotent char", !3, i64 0}
|
|
|
61 !3 = !{!"Simple C/C++ TBAA"}
|
