2 ; RUN: llc -march=hexagon --stats -o - 2>&1 < %s | FileCheck %s
4 ; Check that the compilation succeeded and that some code was generated.
7 ; Check that the loop is pipelined and that a valid node order is used.
8 ; CHECK-NOT: Number of node order issues found
9 ; CHECK: Number of loops software pipelined
10 ; CHECK-NOT: Number of node order issues found
12 target triple = "hexagon"
14 define void @fred(i16* noalias nocapture readonly %p0, i32 %p1, i32 %p2, i16* noalias nocapture %p3, i32 %p4) local_unnamed_addr #1 {
16 %mul = mul i32 %p4, %p1
17 %add.ptr = getelementptr inbounds i16, i16* %p0, i32 %mul
18 %add = add nsw i32 %p4, 1
19 %rem = srem i32 %add, 5
20 %mul1 = mul i32 %rem, %p1
21 %add.ptr2 = getelementptr inbounds i16, i16* %p0, i32 %mul1
22 %add.ptr6 = getelementptr inbounds i16, i16* %p0, i32 0
23 %add7 = add nsw i32 %p4, 3
24 %rem8 = srem i32 %add7, 5
25 %mul9 = mul i32 %rem8, %p1
26 %add.ptr10 = getelementptr inbounds i16, i16* %p0, i32 %mul9
27 %add.ptr14 = getelementptr inbounds i16, i16* %p0, i32 0
28 %incdec.ptr18 = getelementptr inbounds i16, i16* %add.ptr14, i32 32
29 %0 = bitcast i16* %incdec.ptr18 to <16 x i32>*
30 %incdec.ptr17 = getelementptr inbounds i16, i16* %add.ptr10, i32 32
31 %1 = bitcast i16* %incdec.ptr17 to <16 x i32>*
32 %incdec.ptr16 = getelementptr inbounds i16, i16* %add.ptr6, i32 32
33 %2 = bitcast i16* %incdec.ptr16 to <16 x i32>*
34 %incdec.ptr15 = getelementptr inbounds i16, i16* %add.ptr2, i32 32
35 %3 = bitcast i16* %incdec.ptr15 to <16 x i32>*
36 %incdec.ptr = getelementptr inbounds i16, i16* %add.ptr, i32 32
37 %4 = bitcast i16* %incdec.ptr to <16 x i32>*
38 %5 = bitcast i16* %p3 to <16 x i32>*
39 br i1 undef, label %for.end.loopexit.unr-lcssa, label %for.body
41 for.body: ; preds = %for.body, %entry
42 %optr.0102 = phi <16 x i32>* [ %incdec.ptr24.3, %for.body ], [ %5, %entry ]
43 %iptr4.0101 = phi <16 x i32>* [ %incdec.ptr23.3, %for.body ], [ %0, %entry ]
44 %iptr3.0100 = phi <16 x i32>* [ %incdec.ptr22.3, %for.body ], [ %1, %entry ]
45 %iptr2.099 = phi <16 x i32>* [ undef, %for.body ], [ %2, %entry ]
46 %iptr1.098 = phi <16 x i32>* [ %incdec.ptr20.3, %for.body ], [ %3, %entry ]
47 %iptr0.097 = phi <16 x i32>* [ %incdec.ptr19.3, %for.body ], [ %4, %entry ]
48 %dVsumv1.096 = phi <32 x i32> [ %66, %for.body ], [ undef, %entry ]
49 %niter = phi i32 [ %niter.nsub.3, %for.body ], [ undef, %entry ]
50 %6 = load <16 x i32>, <16 x i32>* %iptr0.097, align 64, !tbaa !1
51 %7 = load <16 x i32>, <16 x i32>* %iptr1.098, align 64, !tbaa !1
52 %8 = load <16 x i32>, <16 x i32>* %iptr2.099, align 64, !tbaa !1
53 %9 = load <16 x i32>, <16 x i32>* %iptr3.0100, align 64, !tbaa !1
54 %10 = load <16 x i32>, <16 x i32>* %iptr4.0101, align 64, !tbaa !1
55 %11 = tail call <32 x i32> @llvm.hexagon.V6.vaddhw(<16 x i32> %6, <16 x i32> %10)
56 %12 = tail call <32 x i32> @llvm.hexagon.V6.vmpyhsat.acc(<32 x i32> %11, <16 x i32> %8, i32 393222)
57 %13 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %9, <16 x i32> %7)
58 %14 = tail call <32 x i32> @llvm.hexagon.V6.vmpahb.acc(<32 x i32> %12, <32 x i32> %13, i32 67372036)
59 %15 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %dVsumv1.096)
60 %16 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %14)
61 %17 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %16, <16 x i32> %15, i32 4)
62 %18 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %14)
63 %19 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %16, <16 x i32> %15, i32 8)
64 %20 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %18, <16 x i32> undef, i32 8)
65 %21 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %17, <16 x i32> %19)
66 %22 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %15, <16 x i32> %19)
67 %23 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %22, <16 x i32> %17, i32 101058054)
68 %24 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %23, <16 x i32> zeroinitializer, i32 67372036)
69 %25 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> undef, <16 x i32> %20)
70 %26 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %25, <16 x i32> undef, i32 101058054)
71 %27 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %26, <16 x i32> %21, i32 67372036)
72 %28 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %27, <16 x i32> %24, i32 8)
73 %incdec.ptr24 = getelementptr inbounds <16 x i32>, <16 x i32>* %optr.0102, i32 1
74 store <16 x i32> %28, <16 x i32>* %optr.0102, align 64, !tbaa !1
75 %incdec.ptr19.1 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr0.097, i32 2
76 %incdec.ptr23.1 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr4.0101, i32 2
77 %29 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %14)
78 %30 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %14)
79 %31 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> undef, <16 x i32> %29, i32 4)
80 %32 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> undef, <16 x i32> %30, i32 4)
81 %33 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> undef, <16 x i32> %29, i32 8)
82 %34 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> undef, <16 x i32> %30, i32 8)
83 %35 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %31, <16 x i32> %33)
84 %36 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %29, <16 x i32> %33)
85 %37 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %36, <16 x i32> %31, i32 101058054)
86 %38 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %37, <16 x i32> undef, i32 67372036)
87 %39 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %30, <16 x i32> %34)
88 %40 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %39, <16 x i32> %32, i32 101058054)
89 %41 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %40, <16 x i32> %35, i32 67372036)
90 %42 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %41, <16 x i32> %38, i32 8)
91 %incdec.ptr24.1 = getelementptr inbounds <16 x i32>, <16 x i32>* %optr.0102, i32 2
92 store <16 x i32> %42, <16 x i32>* %incdec.ptr24, align 64, !tbaa !1
93 %incdec.ptr19.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr0.097, i32 3
94 %43 = load <16 x i32>, <16 x i32>* %incdec.ptr19.1, align 64, !tbaa !1
95 %incdec.ptr20.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr1.098, i32 3
96 %incdec.ptr21.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr2.099, i32 3
97 %incdec.ptr22.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr3.0100, i32 3
98 %incdec.ptr23.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr4.0101, i32 3
99 %44 = load <16 x i32>, <16 x i32>* %incdec.ptr23.1, align 64, !tbaa !1
100 %45 = tail call <32 x i32> @llvm.hexagon.V6.vaddhw(<16 x i32> %43, <16 x i32> %44)
101 %46 = tail call <32 x i32> @llvm.hexagon.V6.vmpyhsat.acc(<32 x i32> %45, <16 x i32> undef, i32 393222)
102 %47 = tail call <32 x i32> @llvm.hexagon.V6.vmpahb.acc(<32 x i32> %46, <32 x i32> undef, i32 67372036)
103 %48 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %47)
104 %49 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %48, <16 x i32> undef, i32 4)
105 %50 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %48, <16 x i32> undef, i32 8)
106 %51 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> zeroinitializer, <16 x i32> undef)
107 %52 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %49, <16 x i32> %50)
108 %53 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> undef, <16 x i32> %50)
109 %54 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %53, <16 x i32> %49, i32 101058054)
110 %55 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %54, <16 x i32> %51, i32 67372036)
111 %56 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> undef, <16 x i32> %52, i32 67372036)
112 %57 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %56, <16 x i32> %55, i32 8)
113 %incdec.ptr24.2 = getelementptr inbounds <16 x i32>, <16 x i32>* %optr.0102, i32 3
114 store <16 x i32> %57, <16 x i32>* %incdec.ptr24.1, align 64, !tbaa !1
115 %incdec.ptr19.3 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr0.097, i32 4
116 %58 = load <16 x i32>, <16 x i32>* %incdec.ptr19.2, align 64, !tbaa !1
117 %incdec.ptr20.3 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr1.098, i32 4
118 %59 = load <16 x i32>, <16 x i32>* %incdec.ptr20.2, align 64, !tbaa !1
119 %60 = load <16 x i32>, <16 x i32>* %incdec.ptr21.2, align 64, !tbaa !1
120 %incdec.ptr22.3 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr3.0100, i32 4
121 %61 = load <16 x i32>, <16 x i32>* %incdec.ptr22.2, align 64, !tbaa !1
122 %incdec.ptr23.3 = getelementptr inbounds <16 x i32>, <16 x i32>* %iptr4.0101, i32 4
123 %62 = load <16 x i32>, <16 x i32>* %incdec.ptr23.2, align 64, !tbaa !1
124 %63 = tail call <32 x i32> @llvm.hexagon.V6.vaddhw(<16 x i32> %58, <16 x i32> %62)
125 %64 = tail call <32 x i32> @llvm.hexagon.V6.vmpyhsat.acc(<32 x i32> %63, <16 x i32> %60, i32 393222)
126 %65 = tail call <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32> %61, <16 x i32> %59)
127 %66 = tail call <32 x i32> @llvm.hexagon.V6.vmpahb.acc(<32 x i32> %64, <32 x i32> %65, i32 67372036)
128 %67 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %47)
129 %68 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %66)
130 %69 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %68, <16 x i32> undef, i32 4)
131 %70 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %66)
132 %71 = tail call <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32> %70, <16 x i32> %67, i32 4)
133 %72 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %70, <16 x i32> %67, i32 8)
134 %73 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %67, <16 x i32> %71)
135 %74 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> undef, <16 x i32> %69, i32 101058054)
136 %75 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %74, <16 x i32> %73, i32 67372036)
137 %76 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %67, <16 x i32> %72)
138 %77 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %76, <16 x i32> %71, i32 101058054)
139 %78 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32> %77, <16 x i32> undef, i32 67372036)
140 %79 = tail call <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32> %78, <16 x i32> %75, i32 8)
141 %incdec.ptr24.3 = getelementptr inbounds <16 x i32>, <16 x i32>* %optr.0102, i32 4
142 store <16 x i32> %79, <16 x i32>* %incdec.ptr24.2, align 64, !tbaa !1
143 %niter.nsub.3 = add i32 %niter, -4
144 %niter.ncmp.3 = icmp eq i32 %niter.nsub.3, 0
145 br i1 %niter.ncmp.3, label %for.end.loopexit.unr-lcssa, label %for.body
147 for.end.loopexit.unr-lcssa: ; preds = %for.body, %entry
151 declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #0
152 declare <16 x i32> @llvm.hexagon.V6.lo(<32 x i32>) #0
153 declare <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32>, <16 x i32>) #0
154 declare <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32>, <16 x i32>, i32) #0
155 declare <16 x i32> @llvm.hexagon.V6.valignbi(<16 x i32>, <16 x i32>, i32) #0
156 declare <16 x i32> @llvm.hexagon.V6.vasrwh(<16 x i32>, <16 x i32>, i32) #0
157 declare <16 x i32> @llvm.hexagon.V6.vmpyiwb.acc(<16 x i32>, <16 x i32>, i32) #0
158 declare <32 x i32> @llvm.hexagon.V6.vaddhw(<16 x i32>, <16 x i32>) #0
159 declare <32 x i32> @llvm.hexagon.V6.vcombine(<16 x i32>, <16 x i32>) #0
160 declare <32 x i32> @llvm.hexagon.V6.vmpahb.acc(<32 x i32>, <32 x i32>, i32) #0
161 declare <32 x i32> @llvm.hexagon.V6.vmpyhsat.acc(<32 x i32>, <16 x i32>, i32) #0
163 attributes #0 = { nounwind readnone }
164 attributes #1 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
166 !1 = !{!2, !2, i64 0}
167 !2 = !{!"omnipotent char", !3, i64 0}
168 !3 = !{!"Simple C/C++ TBAA"}