1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
|
; RUN: llc < %s -mtriple=armv7-apple-darwin | FileCheck %s
; PR7158
define i32 @test_pr7158() nounwind {
bb.nph55.bb.nph55.split_crit_edge:
br label %bb3
bb3: ; preds = %bb3, %bb.nph55.bb.nph55.split_crit_edge
br i1 undef, label %bb.i19, label %bb3
bb.i19: ; preds = %bb.i19, %bb3
%0 = insertelement <4 x float> undef, float undef, i32 3 ; <<4 x float>> [#uses=3]
%1 = fmul <4 x float> %0, %0 ; <<4 x float>> [#uses=1]
%2 = bitcast <4 x float> %1 to <2 x double> ; <<2 x double>> [#uses=0]
%3 = fmul <4 x float> %0, undef ; <<4 x float>> [#uses=0]
br label %bb.i19
}
; Check that the DAG combiner does not arbitrarily modify BUILD_VECTORs
; after legalization.
define void @test_illegal_build_vector() nounwind {
entry:
store <2 x i64> undef, <2 x i64>* undef, align 16
%0 = load <16 x i8>* undef, align 16 ; <<16 x i8>> [#uses=1]
%1 = or <16 x i8> zeroinitializer, %0 ; <<16 x i8>> [#uses=1]
store <16 x i8> %1, <16 x i8>* undef, align 16
ret void
}
; Radar 8407927: Make sure that VMOVRRD gets optimized away when the result is
; converted back to be used as a vector type.
; CHECK: test_vmovrrd_combine
define <4 x i32> @test_vmovrrd_combine() nounwind {
entry:
br i1 undef, label %bb1, label %bb2
bb1:
%0 = bitcast <2 x i64> zeroinitializer to <2 x double>
%1 = extractelement <2 x double> %0, i32 0
%2 = bitcast double %1 to i64
%3 = insertelement <1 x i64> undef, i64 %2, i32 0
; CHECK-NOT: vmov s
; CHECK: vext.8
%4 = shufflevector <1 x i64> %3, <1 x i64> undef, <2 x i32> <i32 0, i32 1>
%tmp2006.3 = bitcast <2 x i64> %4 to <16 x i8>
%5 = shufflevector <16 x i8> %tmp2006.3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19>
%tmp2004.3 = bitcast <16 x i8> %5 to <4 x i32>
br i1 undef, label %bb2, label %bb1
bb2:
%result = phi <4 x i32> [ undef, %entry ], [ %tmp2004.3, %bb1 ]
ret <4 x i32> %result
}
; Test trying to do a ShiftCombine on illegal types.
; The vector should be split first.
define void @lshrIllegalType(<8 x i32>* %A) nounwind {
%tmp1 = load <8 x i32>* %A
%tmp2 = lshr <8 x i32> %tmp1, < i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
store <8 x i32> %tmp2, <8 x i32>* %A
ret void
}
; Test folding a binary vector operation with constant BUILD_VECTOR
; operands with i16 elements.
define void @test_i16_constant_fold() nounwind optsize {
entry:
%0 = sext <4 x i1> zeroinitializer to <4 x i16>
%1 = add <4 x i16> %0, zeroinitializer
%2 = shufflevector <4 x i16> %1, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%3 = add <8 x i16> %2, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
%4 = trunc <8 x i16> %3 to <8 x i8>
tail call void @llvm.arm.neon.vst1.v8i8(i8* undef, <8 x i8> %4, i32 1)
unreachable
}
declare void @llvm.arm.neon.vst1.v8i8(i8*, <8 x i8>, i32) nounwind
; Test that loads and stores of i64 vector elements are handled as f64 values
; so they are not split up into i32 values. Radar 8755338.
define void @i64_buildvector(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_buildvector
; CHECK: vldr
%t0 = load i64* %ptr, align 4
%t1 = insertelement <2 x i64> undef, i64 %t0, i32 0
store <2 x i64> %t1, <2 x i64>* %vp
ret void
}
define void @i64_insertelement(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_insertelement
; CHECK: vldr
%t0 = load i64* %ptr, align 4
%vec = load <2 x i64>* %vp
%t1 = insertelement <2 x i64> %vec, i64 %t0, i32 0
store <2 x i64> %t1, <2 x i64>* %vp
ret void
}
define void @i64_extractelement(i64* %ptr, <2 x i64>* %vp) nounwind {
; CHECK: i64_extractelement
; CHECK: vstr
%vec = load <2 x i64>* %vp
%t1 = extractelement <2 x i64> %vec, i32 0
store i64 %t1, i64* %ptr
ret void
}
; Test trying to do a AND Combine on illegal types.
define void @andVec(<3 x i8>* %A) nounwind {
%tmp = load <3 x i8>* %A, align 4
%and = and <3 x i8> %tmp, <i8 7, i8 7, i8 7>
store <3 x i8> %and, <3 x i8>* %A
ret void
}
; Test trying to do an OR Combine on illegal types.
define void @orVec(<3 x i8>* %A) nounwind {
%tmp = load <3 x i8>* %A, align 4
%or = or <3 x i8> %tmp, <i8 7, i8 7, i8 7>
store <3 x i8> %or, <3 x i8>* %A
ret void
}
; The following test was hitting an assertion in the DAG combiner when
; constant folding the multiply because the "sext undef" was translated to
; a BUILD_VECTOR with i32 0 operands, which did not match the i16 operands
; of the other BUILD_VECTOR.
define i16 @foldBuildVectors() {
%1 = sext <8 x i8> undef to <8 x i16>
%2 = mul <8 x i16> %1, <i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255, i16 255>
%3 = extractelement <8 x i16> %2, i32 0
ret i16 %3
}
; Test that we are generating vrev and vext for reverse shuffles of v8i16
; shuffles.
; CHECK: reverse_v8i16
define void @reverse_v8i16(<8 x i16>* %loadaddr, <8 x i16>* %storeaddr) {
%v0 = load <8 x i16>* %loadaddr
; CHECK: vrev64.16
; CHECK: vext.16
%v1 = shufflevector <8 x i16> %v0, <8 x i16> undef,
<8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
store <8 x i16> %v1, <8 x i16>* %storeaddr
ret void
}
; Test that we are generating vrev and vext for reverse shuffles of v16i8
; shuffles.
; CHECK: reverse_v16i8
define void @reverse_v16i8(<16 x i8>* %loadaddr, <16 x i8>* %storeaddr) {
%v0 = load <16 x i8>* %loadaddr
; CHECK: vrev64.8
; CHECK: vext.8
%v1 = shufflevector <16 x i8> %v0, <16 x i8> undef,
<16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8,
i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
store <16 x i8> %v1, <16 x i8>* %storeaddr
ret void
}
|
