1 ; RUN: llc -march=hexagon < %s | FileCheck %s
3 ; CHECK: if (p{{[0-3]}}) v{{[0-9]+}} = v{{[0-9]+}}
5 target triple = "hexagon"
7 ; Function Attrs: nounwind
8 define void @fast9_detect_coarse(ptr nocapture readnone %img, i32 %xsize, i32 %stride, i32 %barrier, ptr nocapture %bitmask, i32 %boundary) #0 {
10 %0 = mul i32 %boundary, -2
11 %sub = add i32 %0, %xsize
12 %rem = and i32 %boundary, 63
13 %add = add i32 %sub, %rem
14 %1 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 -1)
15 %2 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 1)
16 %3 = tail call <64 x i1> @llvm.hexagon.V6.pred.scalar2(i32 %add)
17 %4 = tail call <16 x i32> @llvm.hexagon.V6.vandqrt.acc(<16 x i32> %2, <64 x i1> %3, i32 12)
18 %and4 = and i32 %add, 511
19 %cmp = icmp eq i32 %and4, 0
20 %sMaskR.0 = select i1 %cmp, <16 x i32> %1, <16 x i32> %4
21 %cmp547 = icmp sgt i32 %add, 0
22 br i1 %cmp547, label %for.body.lr.ph, label %for.end
24 for.body.lr.ph: ; preds = %entry
25 %5 = tail call <64 x i1> @llvm.hexagon.V6.pred.scalar2(i32 %boundary)
26 %6 = tail call <16 x i32> @llvm.hexagon.V6.vandqrt(<64 x i1> %5, i32 16843009)
27 %7 = tail call <16 x i32> @llvm.hexagon.V6.vnot(<16 x i32> %6)
28 %8 = add i32 %rem, %xsize
33 %13 = add nuw nsw i32 %12, 16
34 %scevgep = getelementptr i32, ptr %bitmask, i32 %13
37 for.body: ; preds = %for.body.lr.ph, %for.body
38 %i.050 = phi i32 [ %add, %for.body.lr.ph ], [ %sub6, %for.body ]
39 %sMask.049 = phi <16 x i32> [ %7, %for.body.lr.ph ], [ %1, %for.body ]
40 %optr.048 = phi ptr [ %bitmask, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
41 %14 = tail call <16 x i32> @llvm.hexagon.V6.vand(<16 x i32> undef, <16 x i32> %sMask.049)
42 %incdec.ptr = getelementptr inbounds <16 x i32>, ptr %optr.048, i32 1
43 store <16 x i32> %14, ptr %optr.048, align 64
44 %sub6 = add nsw i32 %i.050, -512
45 %cmp5 = icmp sgt i32 %sub6, 0
46 br i1 %cmp5, label %for.body, label %for.cond.for.end_crit_edge
48 for.cond.for.end_crit_edge: ; preds = %for.body
51 for.end: ; preds = %for.cond.for.end_crit_edge, %entry
52 %optr.0.lcssa = phi ptr [ %scevgep, %for.cond.for.end_crit_edge ], [ %bitmask, %entry ]
53 %15 = load <16 x i32>, ptr %optr.0.lcssa, align 64
54 %16 = tail call <16 x i32> @llvm.hexagon.V6.vand(<16 x i32> %15, <16 x i32> %sMaskR.0)
55 store <16 x i32> %16, ptr %optr.0.lcssa, align 64
59 declare <16 x i32> @llvm.hexagon.V6.lvsplatw(i32) #1
60 declare <64 x i1> @llvm.hexagon.V6.pred.scalar2(i32) #1
61 declare <16 x i32> @llvm.hexagon.V6.vandqrt.acc(<16 x i32>, <64 x i1>, i32) #1
62 declare <16 x i32> @llvm.hexagon.V6.vandqrt(<64 x i1>, i32) #1
63 declare <16 x i32> @llvm.hexagon.V6.vnot(<16 x i32>) #1
64 declare <16 x i32> @llvm.hexagon.V6.vand(<16 x i32>, <16 x i32>) #1
66 attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
67 attributes #1 = { nounwind readnone }