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
| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -slp-vectorizer -slp-threshold=-18 < %s | FileCheck %s
; Make sure there's no SCEV assert when the indexes are for different
; sized address spaces
target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"
define void @slp_scev_assert(i32 %idx, i64 %tmp3) #0 {
; CHECK-LABEL: @slp_scev_assert(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[TMP:%.*]] = addrspacecast i8 addrspace(5)* undef to i8*
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, i8 addrspace(5)* undef, i32 [[IDX:%.*]]
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, i8* [[TMP]], i64 [[TMP3:%.*]]
; CHECK-NEXT: store i8 0, i8 addrspace(5)* [[TMP2]]
; CHECK-NEXT: store i8 0, i8* [[TMP4]]
; CHECK-NEXT: ret void
;
bb:
%tmp = addrspacecast i8 addrspace(5)* undef to i8*
%tmp2 = getelementptr inbounds i8, i8 addrspace(5)* undef, i32 %idx
%tmp4 = getelementptr inbounds i8, i8* %tmp, i64 %tmp3
store i8 0, i8 addrspace(5)* %tmp2
store i8 0, i8* %tmp4
ret void
}
define void @multi_as_reduction_different_sized(i32 addrspace(3)* %lds, i32 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_different_sized(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[FLAT:%.*]] = addrspacecast i32 addrspace(3)* [[LDS:%.*]] to i32*
; CHECK-NEXT: [[ADD0:%.*]] = add i32 [[IDX0:%.*]], 2
; CHECK-NEXT: [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT: [[LDS_1:%.*]] = getelementptr inbounds i32, i32 addrspace(3)* [[LDS]], i32 [[ADD0]]
; CHECK-NEXT: [[FLAT_1:%.*]] = getelementptr inbounds i32, i32* [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT: [[LOAD_LDS_0:%.*]] = load i32, i32 addrspace(3)* [[LDS]], align 4
; CHECK-NEXT: [[LOAD_LDS_1:%.*]] = load i32, i32 addrspace(3)* [[LDS_1]], align 4
; CHECK-NEXT: [[LOAD_FLAT_0:%.*]] = load i32, i32* [[FLAT]], align 4
; CHECK-NEXT: [[LOAD_FLAT_1:%.*]] = load i32, i32* [[FLAT_1]], align 4
; CHECK-NEXT: [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_LDS_0]]
; CHECK-NEXT: [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_LDS_1]]
; CHECK-NEXT: store i32 [[SUB0]], i32* undef
; CHECK-NEXT: store i32 [[SUB1]], i32* undef
; CHECK-NEXT: ret void
;
bb:
%flat = addrspacecast i32 addrspace(3)* %lds to i32*
%add0 = add i32 %idx0, 2
%add1 = add i64 %idx1, 1
%lds.1 = getelementptr inbounds i32, i32 addrspace(3)* %lds, i32 %add0
%flat.1 = getelementptr inbounds i32, i32* %flat, i64 %add1
%load.lds.0 = load i32, i32 addrspace(3)* %lds, align 4
%load.lds.1 = load i32, i32 addrspace(3)* %lds.1, align 4
%load.flat.0 = load i32, i32* %flat, align 4
%load.flat.1 = load i32, i32* %flat.1, align 4
%sub0 = sub i32 %load.flat.0, %load.lds.0
%sub1 = sub i32 %load.flat.1, %load.lds.1
store i32 %sub0, i32* undef
store i32 %sub1, i32* undef
ret void
}
; This should vectorize if using GetUnderlyingObject
define void @multi_as_reduction_same_size(i32 addrspace(1)* %global, i64 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_same_size(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[FLAT:%.*]] = addrspacecast i32 addrspace(1)* [[GLOBAL:%.*]] to i32*
; CHECK-NEXT: [[ADD0:%.*]] = add i64 [[IDX0:%.*]], 2
; CHECK-NEXT: [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT: [[GLOBAL_1:%.*]] = getelementptr inbounds i32, i32 addrspace(1)* [[GLOBAL]], i64 [[ADD0]]
; CHECK-NEXT: [[FLAT_1:%.*]] = getelementptr inbounds i32, i32* [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT: [[LOAD_GLOBAL_0:%.*]] = load i32, i32 addrspace(1)* [[GLOBAL]], align 4
; CHECK-NEXT: [[LOAD_GLOBAL_1:%.*]] = load i32, i32 addrspace(1)* [[GLOBAL_1]], align 4
; CHECK-NEXT: [[LOAD_FLAT_0:%.*]] = load i32, i32* [[FLAT]], align 4
; CHECK-NEXT: [[LOAD_FLAT_1:%.*]] = load i32, i32* [[FLAT_1]], align 4
; CHECK-NEXT: [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_GLOBAL_0]]
; CHECK-NEXT: [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_GLOBAL_1]]
; CHECK-NEXT: store i32 [[SUB0]], i32* undef
; CHECK-NEXT: store i32 [[SUB1]], i32* undef
; CHECK-NEXT: ret void
;
bb:
%flat = addrspacecast i32 addrspace(1)* %global to i32*
%add0 = add i64 %idx0, 2
%add1 = add i64 %idx1, 1
%global.1 = getelementptr inbounds i32, i32 addrspace(1)* %global, i64 %add0
%flat.1 = getelementptr inbounds i32, i32* %flat, i64 %add1
%load.global.0 = load i32, i32 addrspace(1)* %global, align 4
%load.global.1 = load i32, i32 addrspace(1)* %global.1, align 4
%load.flat.0 = load i32, i32* %flat, align 4
%load.flat.1 = load i32, i32* %flat.1, align 4
%sub0 = sub i32 %load.flat.0, %load.global.0
%sub1 = sub i32 %load.flat.1, %load.global.1
store i32 %sub0, i32* undef
store i32 %sub1, i32* undef
ret void
}
; This should vectorize if using GetUnderlyingObject
; The add is done in the same width, even though the address space size is smaller
define void @multi_as_reduction_different_sized_noncanon(i32 addrspace(3)* %lds, i64 %idx0, i64 %idx1) #0 {
; CHECK-LABEL: @multi_as_reduction_different_sized_noncanon(
; CHECK-NEXT: bb:
; CHECK-NEXT: [[FLAT:%.*]] = addrspacecast i32 addrspace(3)* [[LDS:%.*]] to i32*
; CHECK-NEXT: [[ADD0:%.*]] = add i64 [[IDX0:%.*]], 2
; CHECK-NEXT: [[ADD1:%.*]] = add i64 [[IDX1:%.*]], 1
; CHECK-NEXT: [[LDS_1:%.*]] = getelementptr inbounds i32, i32 addrspace(3)* [[LDS]], i64 [[ADD0]]
; CHECK-NEXT: [[FLAT_1:%.*]] = getelementptr inbounds i32, i32* [[FLAT]], i64 [[ADD1]]
; CHECK-NEXT: [[LOAD_LDS_0:%.*]] = load i32, i32 addrspace(3)* [[LDS]], align 4
; CHECK-NEXT: [[LOAD_LDS_1:%.*]] = load i32, i32 addrspace(3)* [[LDS_1]], align 4
; CHECK-NEXT: [[LOAD_FLAT_0:%.*]] = load i32, i32* [[FLAT]], align 4
; CHECK-NEXT: [[LOAD_FLAT_1:%.*]] = load i32, i32* [[FLAT_1]], align 4
; CHECK-NEXT: [[SUB0:%.*]] = sub i32 [[LOAD_FLAT_0]], [[LOAD_LDS_0]]
; CHECK-NEXT: [[SUB1:%.*]] = sub i32 [[LOAD_FLAT_1]], [[LOAD_LDS_1]]
; CHECK-NEXT: store i32 [[SUB0]], i32* undef
; CHECK-NEXT: store i32 [[SUB1]], i32* undef
; CHECK-NEXT: ret void
;
bb:
%flat = addrspacecast i32 addrspace(3)* %lds to i32*
%add0 = add i64 %idx0, 2
%add1 = add i64 %idx1, 1
%lds.1 = getelementptr inbounds i32, i32 addrspace(3)* %lds, i64 %add0
%flat.1 = getelementptr inbounds i32, i32* %flat, i64 %add1
%load.lds.0 = load i32, i32 addrspace(3)* %lds, align 4
%load.lds.1 = load i32, i32 addrspace(3)* %lds.1, align 4
%load.flat.0 = load i32, i32* %flat, align 4
%load.flat.1 = load i32, i32* %flat.1, align 4
%sub0 = sub i32 %load.flat.0, %load.lds.0
%sub1 = sub i32 %load.flat.1, %load.lds.1
store i32 %sub0, i32* undef
store i32 %sub1, i32* undef
ret void
}
; CHECK-LABEL: slp_crash_on_addrspacecast
; CHECK: ret void
define void @slp_crash_on_addrspacecast() {
entry:
%0 = getelementptr inbounds i64, i64 addrspace(3)* undef, i32 undef
%p0 = addrspacecast i64 addrspace(3)* %0 to i64*
store i64 undef, i64* %p0, align 8
%1 = getelementptr inbounds i64, i64 addrspace(3)* undef, i32 undef
%p1 = addrspacecast i64 addrspace(3)* %1 to i64*
store i64 undef, i64* %p1, align 8
ret void
}
|