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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
|
; RUN: opt -S -march=r600 -mcpu=cayman -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine < %s | FileCheck %s
; Check vectorization that would ordinarily require a runtime bounds
; check on the pointers when mixing address spaces. For now we cannot
; assume address spaces do not alias, and we can't assume that
; different pointers are directly comparable.
;
; These all test this basic loop for different combinations of address
; spaces, and swapping in globals or adding noalias.
;
;void foo(int addrspace(N)* [noalias] a, int addrspace(M)* [noalias] b, int n)
;{
; for (int i = 0; i < n; ++i)
; {
; a[i] = 3 * b[i];
; }
;}
; Artificial datalayout
target datalayout = "e-p:32:32:32-p1:16:16:16-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-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-v2048:2048:2048-n32:64"
@g_as1 = common addrspace(1) global [1024 x i32] zeroinitializer, align 16
@q_as2 = common addrspace(2) global [1024 x i32] zeroinitializer, align 16
; Both parameters are unidentified objects with the same address
; space, so this should vectorize normally.
define void @foo(i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 %n) #0 {
; CHECK-LABEL: @foo(
; CHECK: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32 addrspace(1)* %b, i64 %idxprom
%0 = load i32 addrspace(1)* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds i32 addrspace(1)* %a, i64 %idxprom1
store i32 %mul, i32 addrspace(1)* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; Parameters are unidentified and different address spaces, so cannot vectorize.
define void @bar0(i32* %a, i32 addrspace(1)* %b, i32 %n) #0 {
; CHECK-LABEL: @bar0(
; CHECK-NOT: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32 addrspace(1)* %b, i64 %idxprom
%0 = load i32 addrspace(1)* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds i32* %a, i64 %idxprom1
store i32 %mul, i32* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; Swapped arguments should be the same
define void @bar1(i32 addrspace(1)* %a, i32* %b, i32 %n) #0 {
; CHECK-LABEL: @bar1(
; CHECK-NOT: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32* %b, i64 %idxprom
%0 = load i32* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds i32 addrspace(1)* %a, i64 %idxprom1
store i32 %mul, i32 addrspace(1)* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; We should still be able to vectorize with noalias even if the
; address spaces are different.
define void @bar2(i32* noalias %a, i32 addrspace(1)* noalias %b, i32 %n) #0 {
; CHECK-LABEL: @bar2(
; CHECK: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32 addrspace(1)* %b, i64 %idxprom
%0 = load i32 addrspace(1)* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds i32* %a, i64 %idxprom1
store i32 %mul, i32* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; Store to identified global with different address space. This isn't
; generally safe and shouldn't be vectorized.
define void @arst0(i32* %b, i32 %n) #0 {
; CHECK-LABEL: @arst0(
; CHECK-NOT: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds i32* %b, i64 %idxprom
%0 = load i32* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds [1024 x i32] addrspace(1)* @g_as1, i64 0, i64 %idxprom1
store i32 %mul, i32 addrspace(1)* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; Load from identified global with different address space.
; This isn't generally safe and shouldn't be vectorized.
define void @arst1(i32* %b, i32 %n) #0 {
; CHECK-LABEL: @arst1(
; CHECK-NOT: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds [1024 x i32] addrspace(1)* @g_as1, i64 0, i64 %idxprom
%0 = load i32 addrspace(1)* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds i32* %b, i64 %idxprom1
store i32 %mul, i32* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
; Read and write to 2 identified globals in different address
; spaces. This should be vectorized.
define void @aoeu(i32 %n) #0 {
; CHECK-LABEL: @aoeu(
; CHECK: <4 x i32>
; CHECK: ret
entry:
br label %for.cond
for.cond: ; preds = %for.body, %entry
%i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
%cmp = icmp slt i32 %i.0, %n
br i1 %cmp, label %for.body, label %for.end
for.body: ; preds = %for.cond
%idxprom = sext i32 %i.0 to i64
%arrayidx = getelementptr inbounds [1024 x i32] addrspace(2)* @q_as2, i64 0, i64 %idxprom
%0 = load i32 addrspace(2)* %arrayidx, align 4
%mul = mul nsw i32 %0, 3
%idxprom1 = sext i32 %i.0 to i64
%arrayidx2 = getelementptr inbounds [1024 x i32] addrspace(1)* @g_as1, i64 0, i64 %idxprom1
store i32 %mul, i32 addrspace(1)* %arrayidx2, align 4
%inc = add nsw i32 %i.0, 1
br label %for.cond
for.end: ; preds = %for.cond
ret void
}
attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
|
