1 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
2 ; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -instcombine -gvn -S | FileCheck %s
3 ; RUN: opt < %s -bb-vectorize -bb-vectorize-req-chain-depth=3 -bb-vectorize-aligned-only -instcombine -gvn -S | FileCheck %s -check-prefix=CHECK-AO
5 ; FIXME: re-enable this once pointer vectors work properly
8 ; Simple 3-pair chain also with loads and stores (using ptrs and gep)
9 define double @test1(i64* %a, i64* %b, i64* %c) nounwind uwtable readonly {
11 %i0 = load i64, i64* %a, align 8
12 %i1 = load i64, i64* %b, align 8
13 %mul = mul i64 %i0, %i1
14 %arrayidx3 = getelementptr inbounds i64, i64* %a, i64 1
15 %i3 = load i64, i64* %arrayidx3, align 8
16 %arrayidx4 = getelementptr inbounds i64, i64* %b, i64 1
17 %i4 = load i64, i64* %arrayidx4, align 8
18 %mul5 = mul i64 %i3, %i4
19 %ptr = inttoptr i64 %mul to double*
20 %ptr5 = inttoptr i64 %mul5 to double*
21 %aptr = getelementptr inbounds double, double* %ptr, i64 2
22 %aptr5 = getelementptr inbounds double, double* %ptr5, i64 3
23 %av = load double, double* %aptr, align 16
24 %av5 = load double, double* %aptr5, align 16
25 %r = fmul double %av, %av5
26 store i64 %mul, i64* %c, align 8
27 %arrayidx5 = getelementptr inbounds i64, i64* %c, i64 1
28 store i64 %mul5, i64* %arrayidx5, align 8
30 ; CHECK-LABEL: @test1(
31 ; CHECK: %i0.v.i0 = bitcast i64* %a to <2 x i64>*
32 ; CHECK: %i1.v.i0 = bitcast i64* %b to <2 x i64>*
33 ; CHECK: %i0 = load <2 x i64>, <2 x i64>* %i0.v.i0, align 8
34 ; CHECK: %i1 = load <2 x i64>, <2 x i64>* %i1.v.i0, align 8
35 ; CHECK: %mul = mul <2 x i64> %i0, %i1
36 ; CHECK: %ptr = inttoptr <2 x i64> %mul to <2 x double*>
37 ; CHECK: %aptr = getelementptr inbounds double, <2 x double*> %ptr, <2 x i64> <i64 2, i64 3>
38 ; CHECK: %aptr.v.r1 = extractelement <2 x double*> %aptr, i32 0
39 ; CHECK: %aptr.v.r2 = extractelement <2 x double*> %aptr, i32 1
40 ; CHECK: %av = load double, double* %aptr.v.r1, align 16
41 ; CHECK: %av5 = load double, double* %aptr.v.r2, align 16
42 ; CHECK: %r = fmul double %av, %av5
43 ; CHECK: %0 = bitcast i64* %c to <2 x i64>*
44 ; CHECK: store <2 x i64> %mul, <2 x i64>* %0, align 8
45 ; CHECK: ret double %r
46 ; CHECK-AO-LABEL: @test1(
47 ; CHECK-AO-NOT: load <2 x
50 ; Simple 3-pair chain with loads and stores (using ptrs and gep)
51 define void @test2(i64** %a, i64** %b, i64** %c) nounwind uwtable readonly {
53 %i0 = load i64*, i64** %a, align 8
54 %i1 = load i64*, i64** %b, align 8
55 %arrayidx3 = getelementptr inbounds i64*, i64** %a, i64 1
56 %i3 = load i64*, i64** %arrayidx3, align 8
57 %arrayidx4 = getelementptr inbounds i64*, i64** %b, i64 1
58 %i4 = load i64*, i64** %arrayidx4, align 8
59 %o1 = load i64, i64* %i1, align 8
60 %o4 = load i64, i64* %i4, align 8
61 %ptr0 = getelementptr inbounds i64, i64* %i0, i64 %o1
62 %ptr3 = getelementptr inbounds i64, i64* %i3, i64 %o4
63 store i64* %ptr0, i64** %c, align 8
64 %arrayidx5 = getelementptr inbounds i64*, i64** %c, i64 1
65 store i64* %ptr3, i64** %arrayidx5, align 8
67 ; CHECK-LABEL: @test2(
68 ; CHECK: %i0.v.i0 = bitcast i64** %a to <2 x i64*>*
69 ; CHECK: %i1 = load i64*, i64** %b, align 8
70 ; CHECK: %i0 = load <2 x i64*>, <2 x i64*>* %i0.v.i0, align 8
71 ; CHECK: %arrayidx4 = getelementptr inbounds i64*, i64** %b, i64 1
72 ; CHECK: %i4 = load i64*, i64** %arrayidx4, align 8
73 ; CHECK: %o1 = load i64, i64* %i1, align 8
74 ; CHECK: %o4 = load i64, i64* %i4, align 8
75 ; CHECK: %ptr0.v.i1.1 = insertelement <2 x i64> undef, i64 %o1, i32 0
76 ; CHECK: %ptr0.v.i1.2 = insertelement <2 x i64> %ptr0.v.i1.1, i64 %o4, i32 1
77 ; CHECK: %ptr0 = getelementptr inbounds i64, <2 x i64*> %i0, <2 x i64> %ptr0.v.i1.2
78 ; CHECK: %0 = bitcast i64** %c to <2 x i64*>*
79 ; CHECK: store <2 x i64*> %ptr0, <2 x i64*>* %0, align 8
81 ; CHECK-AO-LABEL: @test2(
85 ; Simple 3-pair chain with loads and stores (using ptrs and gep)
86 ; using pointer vectors.
87 define void @test3(<2 x i64*>* %a, <2 x i64*>* %b, <2 x i64*>* %c) nounwind uwtable readonly {
89 %i0 = load <2 x i64*>, <2 x i64*>* %a, align 8
90 %i1 = load <2 x i64*>, <2 x i64*>* %b, align 8
91 %arrayidx3 = getelementptr inbounds <2 x i64*>, <2 x i64*>* %a, i64 1
92 %i3 = load <2 x i64*>, <2 x i64*>* %arrayidx3, align 8
93 %arrayidx4 = getelementptr inbounds <2 x i64*>, <2 x i64*>* %b, i64 1
94 %i4 = load <2 x i64*>, <2 x i64*>* %arrayidx4, align 8
95 %j1 = extractelement <2 x i64*> %i1, i32 0
96 %j4 = extractelement <2 x i64*> %i4, i32 0
97 %o1 = load i64, i64* %j1, align 8
98 %o4 = load i64, i64* %j4, align 8
99 %j0 = extractelement <2 x i64*> %i0, i32 0
100 %j3 = extractelement <2 x i64*> %i3, i32 0
101 %ptr0 = getelementptr inbounds i64, i64* %j0, i64 %o1
102 %ptr3 = getelementptr inbounds i64, i64* %j3, i64 %o4
103 %qtr0 = insertelement <2 x i64*> undef, i64* %ptr0, i32 0
104 %rtr0 = insertelement <2 x i64*> %qtr0, i64* %ptr0, i32 1
105 %qtr3 = insertelement <2 x i64*> undef, i64* %ptr3, i32 0
106 %rtr3 = insertelement <2 x i64*> %qtr3, i64* %ptr3, i32 1
107 store <2 x i64*> %rtr0, <2 x i64*>* %c, align 8
108 %arrayidx5 = getelementptr inbounds <2 x i64*>, <2 x i64*>* %c, i64 1
109 store <2 x i64*> %rtr3, <2 x i64*>* %arrayidx5, align 8
111 ; CHECK-LABEL: @test3(
112 ; CHECK: %i0.v.i0 = bitcast <2 x i64*>* %a to <4 x i64*>*
113 ; CHECK: %i1 = load <2 x i64*>, <2 x i64*>* %b, align 8
114 ; CHECK: %i0 = load <4 x i64*>, <4 x i64*>* %i0.v.i0, align 8
115 ; CHECK: %arrayidx4 = getelementptr inbounds <2 x i64*>, <2 x i64*>* %b, i64 1
116 ; CHECK: %i4 = load <2 x i64*>, <2 x i64*>* %arrayidx4, align 8
117 ; CHECK: %j1 = extractelement <2 x i64*> %i1, i32 0
118 ; CHECK: %j4 = extractelement <2 x i64*> %i4, i32 0
119 ; CHECK: %o1 = load i64, i64* %j1, align 8
120 ; CHECK: %o4 = load i64, i64* %j4, align 8
121 ; CHECK: %ptr0.v.i1.1 = insertelement <2 x i64> undef, i64 %o1, i32 0
122 ; CHECK: %ptr0.v.i1.2 = insertelement <2 x i64> %ptr0.v.i1.1, i64 %o4, i32 1
123 ; CHECK: %ptr0.v.i0 = shufflevector <4 x i64*> %i0, <4 x i64*> undef, <2 x i32> <i32 0, i32 2>
124 ; CHECK: %ptr0 = getelementptr inbounds i64, <2 x i64*> %ptr0.v.i0, <2 x i64> %ptr0.v.i1.2
125 ; CHECK: %rtr0 = shufflevector <2 x i64*> %ptr0, <2 x i64*> undef, <2 x i32> zeroinitializer
126 ; CHECK: %rtr3 = shufflevector <2 x i64*> %ptr0, <2 x i64*> undef, <2 x i32> <i32 1, i32 1>
127 ; CHECK: %0 = bitcast <2 x i64*>* %c to <4 x i64*>*
128 ; CHECK: %1 = shufflevector <2 x i64*> %rtr0, <2 x i64*> %rtr3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
129 ; CHECK: store <4 x i64*> %1, <4 x i64*>* %0, align 8
131 ; CHECK-AO-LABEL: @test3(