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
|
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -infer-address-spaces %s | FileCheck %s
; Ports of most of test/CodeGen/NVPTX/access-non-generic.ll
@scalar = internal addrspace(3) global float 0.0, align 4
@array = internal addrspace(3) global [10 x float] zeroinitializer, align 4
; CHECK-LABEL: @load_store_lds_f32(
; CHECK: %tmp = load float, float addrspace(3)* @scalar, align 4
; CHECK: call void @use(float %tmp)
; CHECK: store float %v, float addrspace(3)* @scalar, align 4
; CHECK: call void @llvm.amdgcn.s.barrier()
; CHECK: %tmp2 = load float, float addrspace(3)* @scalar, align 4
; CHECK: call void @use(float %tmp2)
; CHECK: store float %v, float addrspace(3)* @scalar, align 4
; CHECK: call void @llvm.amdgcn.s.barrier()
; CHECK: %tmp3 = load float, float addrspace(3)* getelementptr inbounds ([10 x float], [10 x float] addrspace(3)* @array, i32 0, i32 5), align 4
; CHECK: call void @use(float %tmp3)
; CHECK: store float %v, float addrspace(3)* getelementptr inbounds ([10 x float], [10 x float] addrspace(3)* @array, i32 0, i32 5), align 4
; CHECK: call void @llvm.amdgcn.s.barrier()
; CHECK: %tmp4 = getelementptr inbounds [10 x float], [10 x float] addrspace(3)* @array, i32 0, i32 5
; CHECK: %tmp5 = load float, float addrspace(3)* %tmp4, align 4
; CHECK: call void @use(float %tmp5)
; CHECK: store float %v, float addrspace(3)* %tmp4, align 4
; CHECK: call void @llvm.amdgcn.s.barrier()
; CHECK: %tmp7 = getelementptr inbounds [10 x float], [10 x float] addrspace(3)* @array, i32 0, i32 %i
; CHECK: %tmp8 = load float, float addrspace(3)* %tmp7, align 4
; CHECK: call void @use(float %tmp8)
; CHECK: store float %v, float addrspace(3)* %tmp7, align 4
; CHECK: call void @llvm.amdgcn.s.barrier()
; CHECK: ret void
define amdgpu_kernel void @load_store_lds_f32(i32 %i, float %v) #0 {
bb:
%tmp = load float, float addrspace(4)* addrspacecast (float addrspace(3)* @scalar to float addrspace(4)*), align 4
call void @use(float %tmp)
store float %v, float addrspace(4)* addrspacecast (float addrspace(3)* @scalar to float addrspace(4)*), align 4
call void @llvm.amdgcn.s.barrier()
%tmp1 = addrspacecast float addrspace(3)* @scalar to float addrspace(4)*
%tmp2 = load float, float addrspace(4)* %tmp1, align 4
call void @use(float %tmp2)
store float %v, float addrspace(4)* %tmp1, align 4
call void @llvm.amdgcn.s.barrier()
%tmp3 = load float, float addrspace(4)* getelementptr inbounds ([10 x float], [10 x float] addrspace(4)* addrspacecast ([10 x float] addrspace(3)* @array to [10 x float] addrspace(4)*), i32 0, i32 5), align 4
call void @use(float %tmp3)
store float %v, float addrspace(4)* getelementptr inbounds ([10 x float], [10 x float] addrspace(4)* addrspacecast ([10 x float] addrspace(3)* @array to [10 x float] addrspace(4)*), i32 0, i32 5), align 4
call void @llvm.amdgcn.s.barrier()
%tmp4 = getelementptr inbounds [10 x float], [10 x float] addrspace(4)* addrspacecast ([10 x float] addrspace(3)* @array to [10 x float] addrspace(4)*), i32 0, i32 5
%tmp5 = load float, float addrspace(4)* %tmp4, align 4
call void @use(float %tmp5)
store float %v, float addrspace(4)* %tmp4, align 4
call void @llvm.amdgcn.s.barrier()
%tmp6 = addrspacecast [10 x float] addrspace(3)* @array to [10 x float] addrspace(4)*
%tmp7 = getelementptr inbounds [10 x float], [10 x float] addrspace(4)* %tmp6, i32 0, i32 %i
%tmp8 = load float, float addrspace(4)* %tmp7, align 4
call void @use(float %tmp8)
store float %v, float addrspace(4)* %tmp7, align 4
call void @llvm.amdgcn.s.barrier()
ret void
}
; CHECK-LABEL: @constexpr_load_int_from_float_lds(
; CHECK: %tmp = load i32, i32 addrspace(3)* bitcast (float addrspace(3)* @scalar to i32 addrspace(3)*), align 4
define i32 @constexpr_load_int_from_float_lds() #0 {
bb:
%tmp = load i32, i32 addrspace(4)* addrspacecast (i32 addrspace(3)* bitcast (float addrspace(3)* @scalar to i32 addrspace(3)*) to i32 addrspace(4)*), align 4
ret i32 %tmp
}
; CHECK-LABEL: @load_int_from_global_float(
; CHECK: %tmp1 = getelementptr float, float addrspace(1)* %input, i32 %i
; CHECK: %tmp2 = getelementptr float, float addrspace(1)* %tmp1, i32 %j
; CHECK: %tmp3 = bitcast float addrspace(1)* %tmp2 to i32 addrspace(1)*
; CHECK: %tmp4 = load i32, i32 addrspace(1)* %tmp3
; CHECK: ret i32 %tmp4
define i32 @load_int_from_global_float(float addrspace(1)* %input, i32 %i, i32 %j) #0 {
bb:
%tmp = addrspacecast float addrspace(1)* %input to float addrspace(4)*
%tmp1 = getelementptr float, float addrspace(4)* %tmp, i32 %i
%tmp2 = getelementptr float, float addrspace(4)* %tmp1, i32 %j
%tmp3 = bitcast float addrspace(4)* %tmp2 to i32 addrspace(4)*
%tmp4 = load i32, i32 addrspace(4)* %tmp3
ret i32 %tmp4
}
; CHECK-LABEL: @nested_const_expr(
; CHECK: store i32 1, i32 addrspace(3)* bitcast (float addrspace(3)* getelementptr inbounds ([10 x float], [10 x float] addrspace(3)* @array, i64 0, i64 1) to i32 addrspace(3)*), align 4
define amdgpu_kernel void @nested_const_expr() #0 {
store i32 1, i32 addrspace(4)* bitcast (float addrspace(4)* getelementptr ([10 x float], [10 x float] addrspace(4)* addrspacecast ([10 x float] addrspace(3)* @array to [10 x float] addrspace(4)*), i64 0, i64 1) to i32 addrspace(4)*), align 4
ret void
}
; CHECK-LABEL: @rauw(
; CHECK: %addr = getelementptr float, float addrspace(1)* %input, i64 10
; CHECK-NEXT: %v = load float, float addrspace(1)* %addr
; CHECK-NEXT: store float %v, float addrspace(1)* %addr
; CHECK-NEXT: ret void
define amdgpu_kernel void @rauw(float addrspace(1)* %input) #0 {
bb:
%generic_input = addrspacecast float addrspace(1)* %input to float addrspace(4)*
%addr = getelementptr float, float addrspace(4)* %generic_input, i64 10
%v = load float, float addrspace(4)* %addr
store float %v, float addrspace(4)* %addr
ret void
}
; FIXME: Should be able to eliminate the cast inside the loop
; CHECK-LABEL: @loop(
; CHECK: %p = bitcast [10 x float] addrspace(3)* @array to float addrspace(3)*
; CHECK: %end = getelementptr float, float addrspace(3)* %p, i64 10
; CHECK: br label %loop
; CHECK: loop: ; preds = %loop, %entry
; CHECK: %i = phi float addrspace(3)* [ %p, %entry ], [ %i2, %loop ]
; CHECK: %v = load float, float addrspace(3)* %i
; CHECK: call void @use(float %v)
; CHECK: %i2 = getelementptr float, float addrspace(3)* %i, i64 1
; CHECK: %exit_cond = icmp eq float addrspace(3)* %i2, %end
; CHECK: br i1 %exit_cond, label %exit, label %loop
define amdgpu_kernel void @loop() #0 {
entry:
%p = addrspacecast [10 x float] addrspace(3)* @array to float addrspace(4)*
%end = getelementptr float, float addrspace(4)* %p, i64 10
br label %loop
loop: ; preds = %loop, %entry
%i = phi float addrspace(4)* [ %p, %entry ], [ %i2, %loop ]
%v = load float, float addrspace(4)* %i
call void @use(float %v)
%i2 = getelementptr float, float addrspace(4)* %i, i64 1
%exit_cond = icmp eq float addrspace(4)* %i2, %end
br i1 %exit_cond, label %exit, label %loop
exit: ; preds = %loop
ret void
}
@generic_end = external addrspace(1) global float addrspace(4)*
; CHECK-LABEL: @loop_with_generic_bound(
; CHECK: %p = bitcast [10 x float] addrspace(3)* @array to float addrspace(3)*
; CHECK: %end = load float addrspace(4)*, float addrspace(4)* addrspace(1)* @generic_end
; CHECK: br label %loop
; CHECK: loop:
; CHECK: %i = phi float addrspace(3)* [ %p, %entry ], [ %i2, %loop ]
; CHECK: %v = load float, float addrspace(3)* %i
; CHECK: call void @use(float %v)
; CHECK: %i2 = getelementptr float, float addrspace(3)* %i, i64 1
; CHECK: %0 = addrspacecast float addrspace(3)* %i2 to float addrspace(4)*
; CHECK: %exit_cond = icmp eq float addrspace(4)* %0, %end
; CHECK: br i1 %exit_cond, label %exit, label %loop
define amdgpu_kernel void @loop_with_generic_bound() #0 {
entry:
%p = addrspacecast [10 x float] addrspace(3)* @array to float addrspace(4)*
%end = load float addrspace(4)*, float addrspace(4)* addrspace(1)* @generic_end
br label %loop
loop: ; preds = %loop, %entry
%i = phi float addrspace(4)* [ %p, %entry ], [ %i2, %loop ]
%v = load float, float addrspace(4)* %i
call void @use(float %v)
%i2 = getelementptr float, float addrspace(4)* %i, i64 1
%exit_cond = icmp eq float addrspace(4)* %i2, %end
br i1 %exit_cond, label %exit, label %loop
exit: ; preds = %loop
ret void
}
declare void @llvm.amdgcn.s.barrier() #1
declare void @use(float) #0
attributes #0 = { nounwind }
attributes #1 = { convergent nounwind }
|
