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
| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -basicaa -slp-vectorizer -S -slp-schedule-budget=16 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.9.0"
; Test if the budget for the scheduling region size works.
; We test with a reduced budget of 16 which should prevent vectorizing the loads.
declare void @unknown()
define void @test(float * %a, float * %b, float * %c, float * %d) {
; CHECK-LABEL: @test(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[L0:%.*]] = load float, float* [[A:%.*]]
; CHECK-NEXT: [[A1:%.*]] = getelementptr inbounds float, float* [[A]], i64 1
; CHECK-NEXT: [[L1:%.*]] = load float, float* [[A1]]
; CHECK-NEXT: [[A2:%.*]] = getelementptr inbounds float, float* [[A]], i64 2
; CHECK-NEXT: [[L2:%.*]] = load float, float* [[A2]]
; CHECK-NEXT: [[A3:%.*]] = getelementptr inbounds float, float* [[A]], i64 3
; CHECK-NEXT: [[L3:%.*]] = load float, float* [[A3]]
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown()
; CHECK-NEXT: store float [[L0]], float* [[B:%.*]]
; CHECK-NEXT: [[B1:%.*]] = getelementptr inbounds float, float* [[B]], i64 1
; CHECK-NEXT: store float [[L1]], float* [[B1]]
; CHECK-NEXT: [[B2:%.*]] = getelementptr inbounds float, float* [[B]], i64 2
; CHECK-NEXT: store float [[L2]], float* [[B2]]
; CHECK-NEXT: [[B3:%.*]] = getelementptr inbounds float, float* [[B]], i64 3
; CHECK-NEXT: store float [[L3]], float* [[B3]]
; CHECK-NEXT: [[C1:%.*]] = getelementptr inbounds float, float* [[C:%.*]], i64 1
; CHECK-NEXT: [[C2:%.*]] = getelementptr inbounds float, float* [[C]], i64 2
; CHECK-NEXT: [[C3:%.*]] = getelementptr inbounds float, float* [[C]], i64 3
; CHECK-NEXT: [[TMP0:%.*]] = bitcast float* [[C]] to <4 x float>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
; CHECK-NEXT: [[D1:%.*]] = getelementptr inbounds float, float* [[D:%.*]], i64 1
; CHECK-NEXT: [[D2:%.*]] = getelementptr inbounds float, float* [[D]], i64 2
; CHECK-NEXT: [[D3:%.*]] = getelementptr inbounds float, float* [[D]], i64 3
; CHECK-NEXT: [[TMP2:%.*]] = bitcast float* [[D]] to <4 x float>*
; CHECK-NEXT: store <4 x float> [[TMP1]], <4 x float>* [[TMP2]], align 4
; CHECK-NEXT: ret void
;
entry:
; Don't vectorize these loads.
%l0 = load float, float* %a
%a1 = getelementptr inbounds float, float* %a, i64 1
%l1 = load float, float* %a1
%a2 = getelementptr inbounds float, float* %a, i64 2
%l2 = load float, float* %a2
%a3 = getelementptr inbounds float, float* %a, i64 3
%l3 = load float, float* %a3
; some unrelated instructions inbetween to enlarge the scheduling region
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
call void @unknown()
; Don't vectorize these stores because their operands are too far away.
store float %l0, float* %b
%b1 = getelementptr inbounds float, float* %b, i64 1
store float %l1, float* %b1
%b2 = getelementptr inbounds float, float* %b, i64 2
store float %l2, float* %b2
%b3 = getelementptr inbounds float, float* %b, i64 3
store float %l3, float* %b3
; But still vectorize the following instructions, because even if the budget
; is exceeded there is a minimum region size.
%l4 = load float, float* %c
%c1 = getelementptr inbounds float, float* %c, i64 1
%l5 = load float, float* %c1
%c2 = getelementptr inbounds float, float* %c, i64 2
%l6 = load float, float* %c2
%c3 = getelementptr inbounds float, float* %c, i64 3
%l7 = load float, float* %c3
store float %l4, float* %d
%d1 = getelementptr inbounds float, float* %d, i64 1
store float %l5, float* %d1
%d2 = getelementptr inbounds float, float* %d, i64 2
store float %l6, float* %d2
%d3 = getelementptr inbounds float, float* %d, i64 3
store float %l7, float* %d3
ret void
}
|