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
|
;=========================== begin_copyright_notice ============================
;
; Copyright (C) 2017-2021 Intel Corporation
;
; SPDX-License-Identifier: MIT
;
;============================ end_copyright_notice =============================
; RUN: igc_opt -igc-fix-alignment -S %s -o %t.ll
; RUN: FileCheck %s --input-file=%t.ll
target triple = "igil_32_GEN9"
; Function Attrs: alwaysinline nounwind
; CHECK: @regular
define void @regular(i8 addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %3
; CHECK: load
; CHECK: align 1
%4 = load i8, i8 addrspace(1)* %arrayidx
%add = add nsw i32 %3, 1
%arrayidx1 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add
; CHECK: load
; CHECK: align 1
%5 = load i8, i8 addrspace(1)* %arrayidx1
%add3 = add nsw i32 %3, 2
%arrayidx4 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add3
; CHECK: load
; CHECK: align 1
%6 = load i8, i8 addrspace(1)* %arrayidx4
%add6 = add nsw i32 %3, 3
%arrayidx7 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add6
; CHECK: load
; CHECK: align 1
%7 = load i8, i8 addrspace(1)* %arrayidx7
%assembled.vect = insertelement <4 x i8> undef, i8 %4, i32 0
%assembled.vect8 = insertelement <4 x i8> %assembled.vect, i8 %5, i32 1
%assembled.vect9 = insertelement <4 x i8> %assembled.vect8, i8 %6, i32 2
%assembled.vect10 = insertelement <4 x i8> %assembled.vect9, i8 %7, i32 3
%arrayidx9 = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3
; CHECK: store
; CHECK: align 4
store <4 x i8> %assembled.vect10, <4 x i8> addrspace(1)* %arrayidx9
ret void
}
; Function Attrs: alwaysinline nounwind
; CHECK: @virtload
define void @virtload(i8 addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %3
%4 = bitcast i8 addrspace(1)* %add.ptr to <4 x i8> addrspace(1)*
; CHECK: load
; CHECK: align {{1|4}}
%5 = load <4 x i8>, <4 x i8> addrspace(1)* %4
%arrayidx = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3
; CHECK: store
; CHECK: align 4
store <4 x i8> %5, <4 x i8> addrspace(1)* %arrayidx
ret void
}
; Function Attrs: alwaysinline nounwind
; CHECK: @illegal
define void @illegal(i8 addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %3
%4 = bitcast i8 addrspace(1)* %add.ptr to <4 x i8> addrspace(1)*
; CHECK: load
; CHECK: align {{1|4}}
%5 = load <4 x i8>, <4 x i8> addrspace(1)* %4
%arrayidx = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3
; CHECK: store
; CHECK: align 4
store <4 x i8> %5, <4 x i8> addrspace(1)* %arrayidx
ret void
}
; Function Attrs: alwaysinline nounwind
; CHECK: @virtload_aligned
define void @virtload_aligned(<4 x i8> addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%4 = bitcast <4 x i8> addrspace(1)* %src to i8 addrspace(1)*
%5 = shl i32 %3, 2
%6 = getelementptr inbounds i8, i8 addrspace(1)* %4, i32 %5
%7 = bitcast i8 addrspace(1)* %6 to <4 x i8> addrspace(1)*
; CHECK: load
; CHECK: align 4
%8 = load <4 x i8>, <4 x i8> addrspace(1)* %7
%arrayidx = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3
; CHECK: store
; CHECK: align 4
store <4 x i8> %8, <4 x i8> addrspace(1)* %arrayidx
ret void
}
; Function Attrs: alwaysinline nounwind
; CHECK: @aligned
define void @aligned(<4 x i8> addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%add.ptr = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %src, i32 %3
; CHECK: load
; CHECK: align 4
%4 = load <4 x i8>, <4 x i8> addrspace(1)* %add.ptr
%arrayidx = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3
; CHECK: store
; CHECK: align 4
store <4 x i8> %4, <4 x i8> addrspace(1)* %arrayidx
ret void
}
; CHECK: @regular_scalarized
define void @regular_scalarized(i8 addrspace(1)* %src, <4 x i8> addrspace(1)* %dst, <8 x i32> %r0, <8 x i32> %payloadHeader, i16 %localIdX, i16 %localIdY, i16 %localIdZ) #0 {
entry:
%scalar = extractelement <8 x i32> %payloadHeader, i32 0
%groupId = extractelement <8 x i32> %r0, i32 1
%localSize = extractelement <8 x i32> %payloadHeader, i32 3
%0 = mul i32 %localSize, %groupId
%1 = zext i16 %localIdX to i32
%2 = add i32 %0, %1
%3 = add i32 %2, %scalar
%arrayidx = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %3
; CHECK: load
; CHECK: align 1
%4 = load i8, i8 addrspace(1)* %arrayidx
%add = add nsw i32 %3, 1
%arrayidx1 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add
; CHECK: load
; CHECK: align 1
%5 = load i8, i8 addrspace(1)* %arrayidx1
%add3 = add nsw i32 %3, 2
%arrayidx4 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add3
; CHECK: load
; CHECK: align 1
%6 = load i8, i8 addrspace(1)* %arrayidx4
%add6 = add nsw i32 %3, 3
%arrayidx7 = getelementptr inbounds i8, i8 addrspace(1)* %src, i32 %add6
; CHECK: load
; CHECK: align 1
%7 = load i8, i8 addrspace(1)* %arrayidx7
%ptrVec2ptrScl = getelementptr inbounds <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3, i32 0
; CHECK: store
; CHECK: align 4
store i8 %4, i8 addrspace(1)* %ptrVec2ptrScl
%GEP_lane8 = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3, i32 1
; CHECK: store
; CHECK: align 1
store i8 %5, i8 addrspace(1)* %GEP_lane8
%GEP_lane9 = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3, i32 2
; CHECK: store
; CHECK: align 2
store i8 %6, i8 addrspace(1)* %GEP_lane9
%GEP_lane10 = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %dst, i32 %3, i32 3
; CHECK: store
; CHECK: align 1
store i8 %7, i8 addrspace(1)* %GEP_lane10
ret void
}
attributes #0 = { alwaysinline nounwind }
|
