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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2017-2021 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; RUN: igc_opt %s -S -o - %enable-basic-aa% -igc-memopt -instcombine | FileCheck %s
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f80:128:128-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024-a:64:64-f80:128:128-n8:16:32:64"
define void @f0(i32* %dst, i32* %src) {
entry:
%0 = load i32, i32* %src, align 4
%arrayidx1 = getelementptr inbounds i32, i32* %src, i64 1
%1 = load i32, i32* %arrayidx1, align 4
%arrayidx2 = getelementptr inbounds i32, i32* %src, i64 2
%2 = load i32, i32* %arrayidx2, align 4
store i32 %0, i32* %dst, align 4
%arrayidx4 = getelementptr inbounds i32, i32* %dst, i64 1
store i32 %1, i32* %arrayidx4, align 4
%arrayidx5 = getelementptr inbounds i32, i32* %dst, i64 2
store i32 %2, i32* %arrayidx5, align 4
ret void
}
; CHECK-LABEL: define void @f0
; CHECK: %0 = bitcast i32* %src to <3 x i32>*
; CHECK: %1 = load <3 x i32>, <3 x i32>* %0, align 4
; CHECK: %2 = bitcast i32* %dst to <3 x i32>*
; CHECK: store <3 x i32> %1, <3 x i32>* %2, align 4
; CHECK: ret void
define void @f1(i32* %dst, i32* %src) {
entry:
%0 = load i32, i32* %src, align 4
store i32 %0, i32* %dst, align 4
%arrayidx2 = getelementptr inbounds i32, i32* %src, i64 1
%1 = load i32, i32* %arrayidx2, align 4
%arrayidx3 = getelementptr inbounds i32, i32* %dst, i64 1
store i32 %1, i32* %arrayidx3, align 4
%arrayidx4 = getelementptr inbounds i32, i32* %src, i64 2
%2 = load i32, i32* %arrayidx4, align 4
%arrayidx5 = getelementptr inbounds i32, i32* %dst, i64 2
store i32 %2, i32* %arrayidx5, align 4
ret void
}
; Without 'noalias' attribute, '%dst' may be alias to '%src'. Hence, we cannot
; merge loads/stores interleaved.
; CHECK-LABEL: define void @f1
; CHECK: %0 = load i32, i32* %src, align 4
; CHECK: store i32 %0, i32* %dst, align 4
; CHECK: %arrayidx2 = getelementptr inbounds i32, i32* %src, i32 1
; CHECK: %1 = load i32, i32* %arrayidx2, align 4
; CHECK: %arrayidx3 = getelementptr inbounds i32, i32* %dst, i32 1
; CHECK: store i32 %1, i32* %arrayidx3, align 4
; CHECK: %arrayidx4 = getelementptr inbounds i32, i32* %src, i32 2
; CHECK: %2 = load i32, i32* %arrayidx4, align 4
; CHECK: %arrayidx5 = getelementptr inbounds i32, i32* %dst, i32 2
; CHECK: store i32 %2, i32* %arrayidx5, align 4
; CHECK: ret void
define void @f2(i32* noalias %dst, i32* noalias %src) {
entry:
%0 = load i32, i32* %src, align 4
store i32 %0, i32* %dst, align 4
%arrayidx2 = getelementptr inbounds i32, i32* %src, i64 1
%1 = load i32, i32* %arrayidx2, align 4
%arrayidx3 = getelementptr inbounds i32, i32* %dst, i64 1
store i32 %1, i32* %arrayidx3, align 4
%arrayidx4 = getelementptr inbounds i32, i32* %src, i64 2
%2 = load i32, i32* %arrayidx4, align 4
%arrayidx5 = getelementptr inbounds i32, i32* %dst, i64 2
store i32 %2, i32* %arrayidx5, align 4
ret void
}
; CHECK-LABEL: define void @f2
; CHECK: %0 = bitcast i32* %src to <3 x i32>*
; CHECK: %1 = load <3 x i32>, <3 x i32>* %0, align 4
; CHECK: %2 = bitcast i32* %dst to <3 x i32>*
; CHECK: store <3 x i32> %1, <3 x i32>* %2, align 4
; CHECK: ret void
define void @f3(i32 %i, i32 %j, i32* noalias %dst, i32* noalias %src) {
entry:
%mul = mul nsw i32 %j, %i
%idxprom = sext i32 %mul to i64
%arrayidx = getelementptr inbounds i32, i32* %src, i64 %idxprom
%0 = load i32, i32* %arrayidx, align 4
%add1 = add nsw i32 %j, %i
%idxprom3 = sext i32 %add1 to i64
%arrayidx4 = getelementptr inbounds i32, i32* %dst, i64 %idxprom3
store i32 %0, i32* %arrayidx4, align 4
%add6 = add nsw i32 %mul, 1
%idxprom7 = sext i32 %add6 to i64
%arrayidx8 = getelementptr inbounds i32, i32* %src, i64 %idxprom7
%1 = load i32, i32* %arrayidx8, align 4
%add10 = add nsw i32 %add1, 1
%idxprom11 = sext i32 %add10 to i64
%arrayidx12 = getelementptr inbounds i32, i32* %dst, i64 %idxprom11
store i32 %1, i32* %arrayidx12, align 4
%add14 = add nsw i32 %mul, 2
%idxprom15 = sext i32 %add14 to i64
%arrayidx16 = getelementptr inbounds i32, i32* %src, i64 %idxprom15
%2 = load i32, i32* %arrayidx16, align 4
%add18 = add nsw i32 %add1, 2
%idxprom19 = sext i32 %add18 to i64
%arrayidx20 = getelementptr inbounds i32, i32* %dst, i64 %idxprom19
store i32 %2, i32* %arrayidx20, align 4
ret void
}
; CHECK-LABEL: define void @f3
; CHECK: %mul = mul nsw i32 %j, %i
; CHECK: %arrayidx = getelementptr inbounds i32, i32* %src, i32 %mul
; CHECK: %0 = bitcast i32* %arrayidx to <3 x i32>*
; CHECK: %1 = load <3 x i32>, <3 x i32>* %0, align 4
; CHECK: %add1 = add nsw i32 %j, %i
; CHECK: %arrayidx4 = getelementptr inbounds i32, i32* %dst, i32 %add1
; CHECK: %2 = bitcast i32* %arrayidx4 to <3 x i32>*
; CHECK: store <3 x i32> %1, <3 x i32>* %2, align 4
; CHECK: ret void
define void @f4(i32 %i, i32 %j, i32* noalias %dst, i32* noalias %src) {
entry:
%mul = mul nsw i32 %j, %i
%idxprom = sext i32 %mul to i64
%arrayidx = getelementptr inbounds i32, i32* %src, i64 %idxprom
%0 = load i32, i32* %arrayidx, align 4
%add1 = add nsw i32 %j, %i
%idxprom3 = sext i32 %add1 to i64
%arrayidx4 = getelementptr inbounds i32, i32* %dst, i64 %idxprom3
store i32 %0, i32* %arrayidx4, align 4
fence seq_cst
%add6 = add nsw i32 %mul, 1
%idxprom7 = sext i32 %add6 to i64
%arrayidx8 = getelementptr inbounds i32, i32* %src, i64 %idxprom7
%1 = load i32, i32* %arrayidx8, align 4
%add10 = add nsw i32 %add1, 1
%idxprom11 = sext i32 %add10 to i64
%arrayidx12 = getelementptr inbounds i32, i32* %dst, i64 %idxprom11
store i32 %1, i32* %arrayidx12, align 4
%add14 = add nsw i32 %mul, 2
%idxprom15 = sext i32 %add14 to i64
%arrayidx16 = getelementptr inbounds i32, i32* %src, i64 %idxprom15
%2 = load i32, i32* %arrayidx16, align 4
%add18 = add nsw i32 %add1, 2
%idxprom19 = sext i32 %add18 to i64
%arrayidx20 = getelementptr inbounds i32, i32* %dst, i64 %idxprom19
store i32 %2, i32* %arrayidx20, align 4
ret void
}
; As the fence is placed after the first pair of load/store, only the last 2
; consecutive loads/stores could be merged.
; CHECK-LABEL: define void @f4
; CHECK: %mul = mul nsw i32 %j, %i
; CHECK: %arrayidx = getelementptr inbounds i32, i32* %src, i32 %mul
; CHECK: %0 = load i32, i32* %arrayidx, align 4
; CHECK: %add1 = add nsw i32 %j, %i
; CHECK: %arrayidx4 = getelementptr inbounds i32, i32* %dst, i32 %add1
; CHECK: store i32 %0, i32* %arrayidx4, align 4
; CHECK: fence seq_cst
; CHECK: %add6 = add nsw i32 %mul, 1
; CHECK: %arrayidx8 = getelementptr inbounds i32, i32* %src, i32 %add6
; CHECK: %1 = bitcast i32* %arrayidx8 to [[i64_TYPE:i64|<2 x i32>]]*
; CHECK: %2 = load [[i64_TYPE]], [[i64_TYPE]]* %1, align 4
; CHECK: %add10 = add nsw i32 %add1, 1
; CHECK: %arrayidx12 = getelementptr inbounds i32, i32* %dst, i32 %add10
; CHECK: %3 = bitcast i32* %arrayidx12 to [[i64_TYPE]]*
; CHECK: store [[i64_TYPE]] %2, [[i64_TYPE]]* %3, align 4
; CHECK: ret void
define void @f5(i32* noalias %dst, i32* noalias %src) {
entry:
%0 = load i32, i32* %src, align 4
%arrayidx1 = getelementptr inbounds i32, i32* %dst, i64 3
store i32 %0, i32* %arrayidx1, align 4
%arrayidx2 = getelementptr inbounds i32, i32* %src, i64 1
%1 = load i32, i32* %arrayidx2, align 4
%arrayidx3 = getelementptr inbounds i32, i32* %dst, i64 2
store i32 %1, i32* %arrayidx3, align 4
%arrayidx4 = getelementptr inbounds i32, i32* %src, i64 2
%2 = load i32, i32* %arrayidx4, align 4
%arrayidx5 = getelementptr inbounds i32, i32* %dst, i64 1
store i32 %2, i32* %arrayidx5, align 4
%arrayidx6 = getelementptr inbounds i32, i32* %src, i64 3
%3 = load i32, i32* %arrayidx6, align 4
store i32 %3, i32* %dst, align 4
ret void
}
; CHECK-LABEL: define void @f5
; CHECK: %0 = bitcast i32* %src to <4 x i32>*
; CHECK: %1 = load <4 x i32>, <4 x i32>* %0, align 4
; CHECK: %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
; CHECK: %3 = bitcast i32* %dst to <4 x i32>*
; CHECK: store <4 x i32> %2, <4 x i32>* %3, align 4
; CHECK: ret void
!igc.functions = !{!0, !3, !4, !5, !6, !7}
!0 = !{void (i32*, i32*)* @f0, !1}
!3 = !{void (i32*, i32*)* @f1, !1}
!4 = !{void (i32*, i32*)* @f2, !1}
!5 = !{void (i32, i32, i32*, i32*)* @f3, !1}
!6 = !{void (i32, i32, i32*, i32*)* @f4, !1}
!7 = !{void (i32*, i32*)* @f5, !1}
!1 = !{!2}
!2 = !{!"function_type", i32 0}
|
