lowerMUL-newload.ll
5.96 KB
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
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=arm-eabi -mcpu=krait | FileCheck %s
define void @func1(i16* %a, i16* %b, i16* %c) {
; CHECK-LABEL: func1:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: add r3, r1, #16
; CHECK-NEXT: vldr d18, [r2, #16]
; CHECK-NEXT: vld1.16 {d16}, [r3:64]
; CHECK-NEXT: vmovl.u16 q8, d16
; CHECK-NEXT: vaddw.s16 q10, q8, d18
; CHECK-NEXT: vmovn.i32 d19, q10
; CHECK-NEXT: vldr d20, [r0, #16]
; CHECK-NEXT: vstr d19, [r0, #16]
; CHECK-NEXT: vldr d19, [r2, #16]
; CHECK-NEXT: vmull.s16 q11, d18, d19
; CHECK-NEXT: vmovl.s16 q9, d19
; CHECK-NEXT: vmla.i32 q11, q8, q9
; CHECK-NEXT: vmovn.i32 d16, q11
; CHECK-NEXT: vstr d16, [r1, #16]
; CHECK-NEXT: vldr d16, [r2, #16]
; CHECK-NEXT: vmlal.s16 q11, d16, d20
; CHECK-NEXT: vmovn.i32 d16, q11
; CHECK-NEXT: vstr d16, [r0, #16]
; CHECK-NEXT: bx lr
entry:
; The test case trying to vectorize the pseudo code below.
; a[i] = b[i] + c[i];
; b[i] = a[i] * c[i];
; a[i] = b[i] + a[i] * c[i];
; Checking that vector load a[i] for "a[i] = b[i] + a[i] * c[i]" is
; scheduled before the first vector store to "a[i] = b[i] + c[i]".
; Checking that there is no vector load a[i] scheduled between the vector
; stores to a[i], otherwise the load of a[i] will be polluted by the first
; vector store to a[i].
; This test case check that the chain information is updated during
; lowerMUL for the new created Load SDNode.
%scevgep0 = getelementptr i16, i16* %a, i32 8
%vector_ptr0 = bitcast i16* %scevgep0 to <4 x i16>*
%vec0 = load <4 x i16>, <4 x i16>* %vector_ptr0, align 8
%scevgep1 = getelementptr i16, i16* %b, i32 8
%vector_ptr1 = bitcast i16* %scevgep1 to <4 x i16>*
%vec1 = load <4 x i16>, <4 x i16>* %vector_ptr1, align 8
%0 = zext <4 x i16> %vec1 to <4 x i32>
%scevgep2 = getelementptr i16, i16* %c, i32 8
%vector_ptr2 = bitcast i16* %scevgep2 to <4 x i16>*
%vec2 = load <4 x i16>, <4 x i16>* %vector_ptr2, align 8
%1 = sext <4 x i16> %vec2 to <4 x i32>
%vec3 = add <4 x i32> %1, %0
%2 = trunc <4 x i32> %vec3 to <4 x i16>
%scevgep3 = getelementptr i16, i16* %a, i32 8
%vector_ptr3 = bitcast i16* %scevgep3 to <4 x i16>*
store <4 x i16> %2, <4 x i16>* %vector_ptr3, align 8
%vector_ptr4 = bitcast i16* %scevgep2 to <4 x i16>*
%vec4 = load <4 x i16>, <4 x i16>* %vector_ptr4, align 8
%3 = sext <4 x i16> %vec4 to <4 x i32>
%vec5 = mul <4 x i32> %3, %vec3
%4 = trunc <4 x i32> %vec5 to <4 x i16>
%vector_ptr5 = bitcast i16* %scevgep1 to <4 x i16>*
store <4 x i16> %4, <4 x i16>* %vector_ptr5, align 8
%5 = sext <4 x i16> %vec0 to <4 x i32>
%vector_ptr6 = bitcast i16* %scevgep2 to <4 x i16>*
%vec6 = load <4 x i16>, <4 x i16>* %vector_ptr6, align 8
%6 = sext <4 x i16> %vec6 to <4 x i32>
%vec7 = mul <4 x i32> %6, %5
%vec8 = add <4 x i32> %vec7, %vec5
%7 = trunc <4 x i32> %vec8 to <4 x i16>
%vector_ptr7 = bitcast i16* %scevgep3 to <4 x i16>*
store <4 x i16> %7, <4 x i16>* %vector_ptr7, align 8
ret void
}
define void @func2(i16* %a, i16* %b, i16* %c) {
; CHECK-LABEL: func2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: add r3, r1, #16
; CHECK-NEXT: vldr d18, [r2, #16]
; CHECK-NEXT: vld1.16 {d16}, [r3:64]
; CHECK-NEXT: vmovl.u16 q8, d16
; CHECK-NEXT: vaddw.s16 q10, q8, d18
; CHECK-NEXT: vmovn.i32 d19, q10
; CHECK-NEXT: vldr d20, [r0, #16]
; CHECK-NEXT: vstr d19, [r0, #16]
; CHECK-NEXT: vldr d19, [r2, #16]
; CHECK-NEXT: vmull.s16 q11, d18, d19
; CHECK-NEXT: vmovl.s16 q9, d19
; CHECK-NEXT: vmla.i32 q11, q8, q9
; CHECK-NEXT: vmovn.i32 d16, q11
; CHECK-NEXT: vstr d16, [r1, #16]
; CHECK-NEXT: vldr d16, [r2, #16]
; CHECK-NEXT: vmlal.s16 q11, d16, d20
; CHECK-NEXT: vaddw.s16 q8, q11, d20
; CHECK-NEXT: vmovn.i32 d16, q8
; CHECK-NEXT: vstr d16, [r0, #16]
; CHECK-NEXT: bx lr
entry:
; The test case trying to vectorize the pseudo code below.
; a[i] = b[i] + c[i];
; b[i] = a[i] * c[i];
; a[i] = b[i] + a[i] * c[i] + a[i];
; Checking that vector load a[i] for "a[i] = b[i] + a[i] * c[i] + a[i]"
; is scheduled before the first vector store to "a[i] = b[i] + c[i]".
; Checking that there is no vector load a[i] scheduled between the first
; vector store to a[i] and the vector add of a[i], otherwise the load of
; a[i] will be polluted by the first vector store to a[i].
; This test case check that both the chain and value of the new created
; Load SDNode are updated during lowerMUL.
%scevgep0 = getelementptr i16, i16* %a, i32 8
%vector_ptr0 = bitcast i16* %scevgep0 to <4 x i16>*
%vec0 = load <4 x i16>, <4 x i16>* %vector_ptr0, align 8
%scevgep1 = getelementptr i16, i16* %b, i32 8
%vector_ptr1 = bitcast i16* %scevgep1 to <4 x i16>*
%vec1 = load <4 x i16>, <4 x i16>* %vector_ptr1, align 8
%0 = zext <4 x i16> %vec1 to <4 x i32>
%scevgep2 = getelementptr i16, i16* %c, i32 8
%vector_ptr2 = bitcast i16* %scevgep2 to <4 x i16>*
%vec2 = load <4 x i16>, <4 x i16>* %vector_ptr2, align 8
%1 = sext <4 x i16> %vec2 to <4 x i32>
%vec3 = add <4 x i32> %1, %0
%2 = trunc <4 x i32> %vec3 to <4 x i16>
%scevgep3 = getelementptr i16, i16* %a, i32 8
%vector_ptr3 = bitcast i16* %scevgep3 to <4 x i16>*
store <4 x i16> %2, <4 x i16>* %vector_ptr3, align 8
%vector_ptr4 = bitcast i16* %scevgep2 to <4 x i16>*
%vec4 = load <4 x i16>, <4 x i16>* %vector_ptr4, align 8
%3 = sext <4 x i16> %vec4 to <4 x i32>
%vec5 = mul <4 x i32> %3, %vec3
%4 = trunc <4 x i32> %vec5 to <4 x i16>
%vector_ptr5 = bitcast i16* %scevgep1 to <4 x i16>*
store <4 x i16> %4, <4 x i16>* %vector_ptr5, align 8
%5 = sext <4 x i16> %vec0 to <4 x i32>
%vector_ptr6 = bitcast i16* %scevgep2 to <4 x i16>*
%vec6 = load <4 x i16>, <4 x i16>* %vector_ptr6, align 8
%6 = sext <4 x i16> %vec6 to <4 x i32>
%vec7 = mul <4 x i32> %6, %5
%vec8 = add <4 x i32> %vec7, %vec5
%vec9 = add <4 x i32> %vec8, %5
%7 = trunc <4 x i32> %vec9 to <4 x i16>
%vector_ptr7 = bitcast i16* %scevgep3 to <4 x i16>*
store <4 x i16> %7, <4 x i16>* %vector_ptr7, align 8
ret void
}