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
|
; RUN: llc -O0 -march=r600 -mcpu=bonaire -mattr=-promote-alloca < %s | FileCheck -check-prefix=CHECK -check-prefix=CHECK-NO-PROMOTE %s
; RUN: llc -O0 -march=r600 -mcpu=bonaire -mattr=+promote-alloca < %s | FileCheck -check-prefix=CHECK -check-prefix=CHECK-PROMOTE %s
; Disable optimizations in case there are optimizations added that
; specialize away generic pointer accesses.
; CHECK-LABEL: {{^}}branch_use_flat_i32:
; CHECK: FLAT_STORE_DWORD {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, [M0, FLAT_SCRATCH]
; CHECK: S_ENDPGM
define void @branch_use_flat_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* %gptr, i32 addrspace(3)* %lptr, i32 %x, i32 %c) #0 {
entry:
%cmp = icmp ne i32 %c, 0
br i1 %cmp, label %local, label %global
local:
%flat_local = addrspacecast i32 addrspace(3)* %lptr to i32 addrspace(4)*
br label %end
global:
%flat_global = addrspacecast i32 addrspace(1)* %gptr to i32 addrspace(4)*
br label %end
end:
%fptr = phi i32 addrspace(4)* [ %flat_local, %local ], [ %flat_global, %global ]
store i32 %x, i32 addrspace(4)* %fptr, align 4
; %val = load i32 addrspace(4)* %fptr, align 4
; store i32 %val, i32 addrspace(1)* %out, align 4
ret void
}
; These testcases might become useless when there are optimizations to
; remove generic pointers.
; CHECK-LABEL: {{^}}store_flat_i32:
; CHECK: V_MOV_B32_e32 v[[DATA:[0-9]+]], {{s[0-9]+}}
; CHECK: V_MOV_B32_e32 v[[LO_VREG:[0-9]+]], {{s[0-9]+}}
; CHECK: V_MOV_B32_e32 v[[HI_VREG:[0-9]+]], {{s[0-9]+}}
; CHECK: FLAT_STORE_DWORD v[[DATA]], v{{\[}}[[LO_VREG]]:[[HI_VREG]]{{\]}}
define void @store_flat_i32(i32 addrspace(1)* %gptr, i32 %x) #0 {
%fptr = addrspacecast i32 addrspace(1)* %gptr to i32 addrspace(4)*
store i32 %x, i32 addrspace(4)* %fptr, align 4
ret void
}
; CHECK-LABEL: {{^}}store_flat_i64:
; CHECK: FLAT_STORE_DWORDX2
define void @store_flat_i64(i64 addrspace(1)* %gptr, i64 %x) #0 {
%fptr = addrspacecast i64 addrspace(1)* %gptr to i64 addrspace(4)*
store i64 %x, i64 addrspace(4)* %fptr, align 8
ret void
}
; CHECK-LABEL: {{^}}store_flat_v4i32:
; CHECK: FLAT_STORE_DWORDX4
define void @store_flat_v4i32(<4 x i32> addrspace(1)* %gptr, <4 x i32> %x) #0 {
%fptr = addrspacecast <4 x i32> addrspace(1)* %gptr to <4 x i32> addrspace(4)*
store <4 x i32> %x, <4 x i32> addrspace(4)* %fptr, align 16
ret void
}
; CHECK-LABEL: {{^}}store_flat_trunc_i16:
; CHECK: FLAT_STORE_SHORT
define void @store_flat_trunc_i16(i16 addrspace(1)* %gptr, i32 %x) #0 {
%fptr = addrspacecast i16 addrspace(1)* %gptr to i16 addrspace(4)*
%y = trunc i32 %x to i16
store i16 %y, i16 addrspace(4)* %fptr, align 2
ret void
}
; CHECK-LABEL: {{^}}store_flat_trunc_i8:
; CHECK: FLAT_STORE_BYTE
define void @store_flat_trunc_i8(i8 addrspace(1)* %gptr, i32 %x) #0 {
%fptr = addrspacecast i8 addrspace(1)* %gptr to i8 addrspace(4)*
%y = trunc i32 %x to i8
store i8 %y, i8 addrspace(4)* %fptr, align 2
ret void
}
; CHECK-LABEL @load_flat_i32:
; CHECK: FLAT_LOAD_DWORD
define void @load_flat_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i32 addrspace(1)* %gptr to i32 addrspace(4)*
%fload = load i32 addrspace(4)* %fptr, align 4
store i32 %fload, i32 addrspace(1)* %out, align 4
ret void
}
; CHECK-LABEL @load_flat_i64:
; CHECK: FLAT_LOAD_DWORDX2
define void @load_flat_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i64 addrspace(1)* %gptr to i64 addrspace(4)*
%fload = load i64 addrspace(4)* %fptr, align 4
store i64 %fload, i64 addrspace(1)* %out, align 8
ret void
}
; CHECK-LABEL @load_flat_v4i32:
; CHECK: FLAT_LOAD_DWORDX4
define void @load_flat_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast <4 x i32> addrspace(1)* %gptr to <4 x i32> addrspace(4)*
%fload = load <4 x i32> addrspace(4)* %fptr, align 4
store <4 x i32> %fload, <4 x i32> addrspace(1)* %out, align 8
ret void
}
; CHECK-LABEL @sextload_flat_i8:
; CHECK: FLAT_LOAD_SBYTE
define void @sextload_flat_i8(i32 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i8 addrspace(1)* %gptr to i8 addrspace(4)*
%fload = load i8 addrspace(4)* %fptr, align 4
%ext = sext i8 %fload to i32
store i32 %ext, i32 addrspace(1)* %out, align 4
ret void
}
; CHECK-LABEL @zextload_flat_i8:
; CHECK: FLAT_LOAD_UBYTE
define void @zextload_flat_i8(i32 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i8 addrspace(1)* %gptr to i8 addrspace(4)*
%fload = load i8 addrspace(4)* %fptr, align 4
%ext = zext i8 %fload to i32
store i32 %ext, i32 addrspace(1)* %out, align 4
ret void
}
; CHECK-LABEL @sextload_flat_i16:
; CHECK: FLAT_LOAD_SSHORT
define void @sextload_flat_i16(i32 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i16 addrspace(1)* %gptr to i16 addrspace(4)*
%fload = load i16 addrspace(4)* %fptr, align 4
%ext = sext i16 %fload to i32
store i32 %ext, i32 addrspace(1)* %out, align 4
ret void
}
; CHECK-LABEL @zextload_flat_i16:
; CHECK: FLAT_LOAD_USHORT
define void @zextload_flat_i16(i32 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %gptr) #0 {
%fptr = addrspacecast i16 addrspace(1)* %gptr to i16 addrspace(4)*
%fload = load i16 addrspace(4)* %fptr, align 4
%ext = zext i16 %fload to i32
store i32 %ext, i32 addrspace(1)* %out, align 4
ret void
}
; TODO: This should not be zero when registers are used for small
; scratch allocations again.
; Check for prologue initializing special SGPRs pointing to scratch.
; CHECK-LABEL: {{^}}store_flat_scratch:
; CHECK: S_MOVK_I32 flat_scratch_lo, 0
; CHECK-NO-PROMOTE: S_MOVK_I32 flat_scratch_hi, 40
; CHECK-PROMOTE: S_MOVK_I32 flat_scratch_hi, 0
; CHECK: FLAT_STORE_DWORD
; CHECK: S_BARRIER
; CHECK: FLAT_LOAD_DWORD
define void @store_flat_scratch(i32 addrspace(1)* noalias %out, i32) #0 {
%alloca = alloca i32, i32 9, align 4
%x = call i32 @llvm.r600.read.tidig.x() #3
%pptr = getelementptr i32* %alloca, i32 %x
%fptr = addrspacecast i32* %pptr to i32 addrspace(4)*
store i32 %x, i32 addrspace(4)* %fptr
; Dummy call
call void @llvm.AMDGPU.barrier.local() #1
%reload = load i32 addrspace(4)* %fptr, align 4
store i32 %reload, i32 addrspace(1)* %out, align 4
ret void
}
declare void @llvm.AMDGPU.barrier.local() #1
declare i32 @llvm.r600.read.tidig.x() #3
attributes #0 = { nounwind }
attributes #1 = { nounwind noduplicate }
attributes #3 = { nounwind readnone }
|
