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 ; Simple 3-pair chain also with loads and stores (using ptrs and gep)
6 define double @test1(i64* %a, i64* %b, i64* %c) nounwind uwtable readonly {
8 %i0 = load i64* %a, align 8
9 %i1 = load i64* %b, align 8
10 %mul = mul i64 %i0, %i1
11 %arrayidx3 = getelementptr inbounds i64* %a, i64 1
12 %i3 = load i64* %arrayidx3, align 8
13 %arrayidx4 = getelementptr inbounds i64* %b, i64 1
14 %i4 = load i64* %arrayidx4, align 8
15 %mul5 = mul i64 %i3, %i4
16 %ptr = inttoptr i64 %mul to double*
17 %ptr5 = inttoptr i64 %mul5 to double*
18 %aptr = getelementptr inbounds double* %ptr, i64 2
19 %aptr5 = getelementptr inbounds double* %ptr5, i64 3
20 %av = load double* %aptr, align 16
21 %av5 = load double* %aptr5, align 16
22 %r = fmul double %av, %av5
23 store i64 %mul, i64* %c, align 8
24 %arrayidx5 = getelementptr inbounds i64* %c, i64 1
25 store i64 %mul5, i64* %arrayidx5, align 8
28 ; CHECK: %i0.v.i0 = bitcast i64* %a to <2 x i64>*
29 ; CHECK: %i1.v.i0 = bitcast i64* %b to <2 x i64>*
30 ; CHECK: %i0 = load <2 x i64>* %i0.v.i0, align 8
31 ; CHECK: %i1 = load <2 x i64>* %i1.v.i0, align 8
32 ; CHECK: %mul = mul <2 x i64> %i0, %i1
33 ; CHECK: %ptr = inttoptr <2 x i64> %mul to <2 x double*>
34 ; CHECK: %aptr = getelementptr inbounds <2 x double*> %ptr, <2 x i64> <i64 2, i64 3>
35 ; CHECK: %aptr.v.r1 = extractelement <2 x double*> %aptr, i32 0
36 ; CHECK: %aptr.v.r2 = extractelement <2 x double*> %aptr, i32 1
37 ; CHECK: %av = load double* %aptr.v.r1, align 16
38 ; CHECK: %av5 = load double* %aptr.v.r2, align 16
39 ; CHECK: %r = fmul double %av, %av5
40 ; CHECK: %0 = bitcast i64* %c to <2 x i64>*
41 ; CHECK: store <2 x i64> %mul, <2 x i64>* %0, align 8
42 ; CHECK: ret double %r
44 ; CHECK-AO-NOT: load <2 x
47 ; Simple 3-pair chain with loads and stores (using ptrs and gep)
48 define void @test2(i64** %a, i64** %b, i64** %c) nounwind uwtable readonly {
50 %i0 = load i64** %a, align 8
51 %i1 = load i64** %b, align 8
52 %arrayidx3 = getelementptr inbounds i64** %a, i64 1
53 %i3 = load i64** %arrayidx3, align 8
54 %arrayidx4 = getelementptr inbounds i64** %b, i64 1
55 %i4 = load i64** %arrayidx4, align 8
56 %o1 = load i64* %i1, align 8
57 %o4 = load i64* %i4, align 8
58 %ptr0 = getelementptr inbounds i64* %i0, i64 %o1
59 %ptr3 = getelementptr inbounds i64* %i3, i64 %o4
60 store i64* %ptr0, i64** %c, align 8
61 %arrayidx5 = getelementptr inbounds i64** %c, i64 1
62 store i64* %ptr3, i64** %arrayidx5, align 8
65 ; CHECK: %i0.v.i0 = bitcast i64** %a to <2 x i64*>*
66 ; CHECK: %i1 = load i64** %b, align 8
67 ; CHECK: %i0 = load <2 x i64*>* %i0.v.i0, align 8
68 ; CHECK: %arrayidx4 = getelementptr inbounds i64** %b, i64 1
69 ; CHECK: %i4 = load i64** %arrayidx4, align 8
70 ; CHECK: %o1 = load i64* %i1, align 8
71 ; CHECK: %o4 = load i64* %i4, align 8
72 ; CHECK: %ptr0.v.i1.1 = insertelement <2 x i64> undef, i64 %o1, i32 0
73 ; CHECK: %ptr0.v.i1.2 = insertelement <2 x i64> %ptr0.v.i1.1, i64 %o4, i32 1
74 ; CHECK: %ptr0 = getelementptr inbounds <2 x i64*> %i0, <2 x i64> %ptr0.v.i1.2
75 ; CHECK: %0 = bitcast i64** %c to <2 x i64*>*
76 ; CHECK: store <2 x i64*> %ptr0, <2 x i64*>* %0, align 8
82 ; Simple 3-pair chain with loads and stores (using ptrs and gep)
83 ; using pointer vectors.
84 define void @test3(<2 x i64*>* %a, <2 x i64*>* %b, <2 x i64*>* %c) nounwind uwtable readonly {
86 %i0 = load <2 x i64*>* %a, align 8
87 %i1 = load <2 x i64*>* %b, align 8
88 %arrayidx3 = getelementptr inbounds <2 x i64*>* %a, i64 1
89 %i3 = load <2 x i64*>* %arrayidx3, align 8
90 %arrayidx4 = getelementptr inbounds <2 x i64*>* %b, i64 1
91 %i4 = load <2 x i64*>* %arrayidx4, align 8
92 %j1 = extractelement <2 x i64*> %i1, i32 0
93 %j4 = extractelement <2 x i64*> %i4, i32 0
94 %o1 = load i64* %j1, align 8
95 %o4 = load i64* %j4, align 8
96 %j0 = extractelement <2 x i64*> %i0, i32 0
97 %j3 = extractelement <2 x i64*> %i3, i32 0
98 %ptr0 = getelementptr inbounds i64* %j0, i64 %o1
99 %ptr3 = getelementptr inbounds i64* %j3, i64 %o4
100 %qtr0 = insertelement <2 x i64*> undef, i64* %ptr0, i32 0
101 %rtr0 = insertelement <2 x i64*> %qtr0, i64* %ptr0, i32 1
102 %qtr3 = insertelement <2 x i64*> undef, i64* %ptr3, i32 0
103 %rtr3 = insertelement <2 x i64*> %qtr3, i64* %ptr3, i32 1
104 store <2 x i64*> %rtr0, <2 x i64*>* %c, align 8
105 %arrayidx5 = getelementptr inbounds <2 x i64*>* %c, i64 1
106 store <2 x i64*> %rtr3, <2 x i64*>* %arrayidx5, align 8
109 ; CHECK: %i0.v.i0 = bitcast <2 x i64*>* %a to <4 x i64*>*
110 ; CHECK: %i1 = load <2 x i64*>* %b, align 8
111 ; CHECK: %i0 = load <4 x i64*>* %i0.v.i0, align 8
112 ; CHECK: %arrayidx4 = getelementptr inbounds <2 x i64*>* %b, i64 1
113 ; CHECK: %i4 = load <2 x i64*>* %arrayidx4, align 8
114 ; CHECK: %j1 = extractelement <2 x i64*> %i1, i32 0
115 ; CHECK: %j4 = extractelement <2 x i64*> %i4, i32 0
116 ; CHECK: %o1 = load i64* %j1, align 8
117 ; CHECK: %o4 = load i64* %j4, align 8
118 ; CHECK: %ptr0.v.i1.1 = insertelement <2 x i64> undef, i64 %o1, i32 0
119 ; CHECK: %ptr0.v.i1.2 = insertelement <2 x i64> %ptr0.v.i1.1, i64 %o4, i32 1
120 ; CHECK: %ptr0.v.i0 = shufflevector <4 x i64*> %i0, <4 x i64*> undef, <2 x i32> <i32 0, i32 2>
121 ; CHECK: %ptr0 = getelementptr inbounds <2 x i64*> %ptr0.v.i0, <2 x i64> %ptr0.v.i1.2
122 ; CHECK: %rtr0 = shufflevector <2 x i64*> %ptr0, <2 x i64*> undef, <2 x i32> zeroinitializer
123 ; CHECK: %rtr3 = shufflevector <2 x i64*> %ptr0, <2 x i64*> undef, <2 x i32> <i32 1, i32 1>
124 ; CHECK: %0 = bitcast <2 x i64*>* %c to <4 x i64*>*
125 ; CHECK: %1 = shufflevector <2 x i64*> %rtr0, <2 x i64*> %rtr3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
126 ; CHECK: store <4 x i64*> %1, <4 x i64*>* %0, align 8