1 ; RUN: llc -march=hexagon < %s | FileCheck %s
3 ; Check that we no longer generate 4 inserts.
4 ; CHECK: combine(r{{[0-9]+}}.l,r{{[0-9]+}}.l)
5 ; CHECK: combine(r{{[0-9]+}}.l,r{{[0-9]+}}.l)
8 target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a0:0-n16:32"
9 target triple = "hexagon"
11 %struct.a = type { i16 }
13 define i32 @fun(%struct.a* nocapture %pData, i64 %c, i64* nocapture %d, i64* nocapture %e, i64* nocapture %f) #0 {
15 %g = getelementptr inbounds %struct.a, %struct.a* %pData, i32 0, i32 0
16 %0 = load i16, i16* %g, align 2, !tbaa !0
17 %conv185 = sext i16 %0 to i32
18 %shr86 = ashr i32 %conv185, 2
19 %cmp87 = icmp sgt i32 %shr86, 0
20 br i1 %cmp87, label %for.body.lr.ph, label %for.end
22 for.body.lr.ph: ; preds = %entry
23 %h.sroa.0.0.extract.trunc = trunc i64 %c to i32
24 %sext = shl i32 %h.sroa.0.0.extract.trunc, 16
25 %conv8 = ashr exact i32 %sext, 16
26 %l.sroa.2.4.extract.shift = lshr i64 %c, 32
27 %sext76 = ashr i32 %h.sroa.0.0.extract.trunc, 16
28 %m.sroa.2.6.extract.shift = lshr i64 %c, 48
29 %sext7980 = shl nuw nsw i64 %l.sroa.2.4.extract.shift, 16
30 %sext79 = trunc i64 %sext7980 to i32
31 %conv38 = ashr exact i32 %sext79, 16
32 %sext8283 = shl nuw nsw i64 %m.sroa.2.6.extract.shift, 16
33 %sext82 = trunc i64 %sext8283 to i32
34 %conv53 = ashr exact i32 %sext82, 16
37 for.body: ; preds = %for.body.lr.ph, %for.body
38 %arrayidx.phi = phi i64* [ %d, %for.body.lr.ph ], [ %arrayidx.inc, %for.body ]
39 %arrayidx30.phi = phi i64* [ %f, %for.body.lr.ph ], [ %arrayidx30.inc, %for.body ]
40 %arrayidx60.phi = phi i64* [ %e, %for.body.lr.ph ], [ %arrayidx60.inc, %for.body ]
41 %j.088.pmt = phi i32 [ 0, %for.body.lr.ph ], [ %inc.pmt, %for.body ]
42 %1 = load i64, i64* %arrayidx.phi, align 8, !tbaa !1
43 %n_union3.sroa.0.0.extract.trunc = trunc i64 %1 to i32
44 %n_union3.sroa.1.4.extract.shift = lshr i64 %1, 32
45 %2 = tail call i64 @llvm.hexagon.M2.dpmpyss.s0(i32 %n_union3.sroa.0.0.extract.trunc, i32 %conv8)
46 %3 = tail call i64 @llvm.hexagon.S2.asl.r.p(i64 %2, i32 -25)
47 %conv9 = trunc i64 %3 to i32
48 %4 = tail call i32 @llvm.hexagon.A2.sath(i32 %conv9)
49 %n_union13.sroa.1.4.extract.trunc = trunc i64 %n_union3.sroa.1.4.extract.shift to i32
50 %5 = tail call i64 @llvm.hexagon.M2.dpmpyss.s0(i32 %n_union13.sroa.1.4.extract.trunc, i32 %sext76)
51 %6 = tail call i64 @llvm.hexagon.S2.asl.r.p(i64 %5, i32 -25)
52 %conv24 = trunc i64 %6 to i32
53 %7 = tail call i32 @llvm.hexagon.A2.sath(i32 %conv24)
54 %8 = load i64, i64* %arrayidx30.phi, align 8, !tbaa !1
55 %n_union28.sroa.0.0.extract.trunc = trunc i64 %8 to i32
56 %n_union28.sroa.1.4.extract.shift = lshr i64 %8, 32
57 %9 = tail call i64 @llvm.hexagon.M2.dpmpyss.s0(i32 %n_union28.sroa.0.0.extract.trunc, i32 %conv38)
58 %10 = tail call i64 @llvm.hexagon.S2.asl.r.p(i64 %9, i32 -25)
59 %conv39 = trunc i64 %10 to i32
60 %11 = tail call i32 @llvm.hexagon.A2.sath(i32 %conv39)
61 %n_union43.sroa.1.4.extract.trunc = trunc i64 %n_union28.sroa.1.4.extract.shift to i32
62 %12 = tail call i64 @llvm.hexagon.M2.dpmpyss.s0(i32 %n_union43.sroa.1.4.extract.trunc, i32 %conv53)
63 %13 = tail call i64 @llvm.hexagon.S2.asl.r.p(i64 %12, i32 -25)
64 %conv54 = trunc i64 %13 to i32
65 %14 = tail call i32 @llvm.hexagon.A2.sath(i32 %conv54)
66 %n_union.sroa.3.6.insert.ext = zext i32 %14 to i64
67 %n_union.sroa.3.6.insert.shift = shl i64 %n_union.sroa.3.6.insert.ext, 48
68 %conv40.mask = and i32 %11, 65535
69 %n_union.sroa.2.4.insert.ext = zext i32 %conv40.mask to i64
70 %n_union.sroa.2.4.insert.shift = shl nuw nsw i64 %n_union.sroa.2.4.insert.ext, 32
71 %conv25.mask = and i32 %7, 65535
72 %n_union.sroa.1.2.insert.ext = zext i32 %conv25.mask to i64
73 %n_union.sroa.1.2.insert.shift = shl nuw nsw i64 %n_union.sroa.1.2.insert.ext, 16
74 %conv10.mask = and i32 %4, 65535
75 %n_union.sroa.0.0.insert.ext = zext i32 %conv10.mask to i64
76 %n_union.sroa.2.4.insert.insert = or i64 %n_union.sroa.1.2.insert.shift, %n_union.sroa.0.0.insert.ext
77 %n_union.sroa.1.2.insert.insert = or i64 %n_union.sroa.2.4.insert.insert, %n_union.sroa.2.4.insert.shift
78 %n_union.sroa.0.0.insert.insert = or i64 %n_union.sroa.1.2.insert.insert, %n_union.sroa.3.6.insert.shift
79 %15 = load i64, i64* %arrayidx60.phi, align 8, !tbaa !1
80 %16 = tail call i64 @llvm.hexagon.A2.vaddhs(i64 %15, i64 %n_union.sroa.0.0.insert.insert)
81 store i64 %16, i64* %arrayidx60.phi, align 8, !tbaa !1
82 %inc.pmt = add i32 %j.088.pmt, 1
83 %17 = load i16, i16* %g, align 2, !tbaa !0
84 %conv1 = sext i16 %17 to i32
85 %shr = ashr i32 %conv1, 2
86 %cmp = icmp slt i32 %inc.pmt, %shr
87 %arrayidx.inc = getelementptr i64, i64* %arrayidx.phi, i32 1
88 %arrayidx30.inc = getelementptr i64, i64* %arrayidx30.phi, i32 1
89 %arrayidx60.inc = getelementptr i64, i64* %arrayidx60.phi, i32 1
90 br i1 %cmp, label %for.body, label %for.end.loopexit
92 for.end.loopexit: ; preds = %for.body
95 for.end: ; preds = %for.end.loopexit, %entry
99 declare i32 @llvm.hexagon.A2.sath(i32) #1
101 declare i64 @llvm.hexagon.S2.asl.r.p(i64, i32) #1
103 declare i64 @llvm.hexagon.M2.dpmpyss.s0(i32, i32) #1
105 declare i64 @llvm.hexagon.A2.vaddhs(i64, i64) #1
107 attributes #0 = { nounwind "fp-contract-model"="standard" "no-frame-pointer-elim-non-leaf" "realign-stack" "relocation-model"="static" "ssp-buffers-size"="8" }
108 attributes #1 = { nounwind readnone }
111 !1 = !{!"omnipotent char", !2}
112 !2 = !{!"Simple C/C++ TBAA"}