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
|
; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -show-mc-encoding -verify-machineinstrs < %s | FileCheck %s
;;;==========================================================================;;;
;;; MUBUF LOAD TESTS
;;;==========================================================================;;;
; MUBUF load with an immediate byte offset that fits into 12-bits
; CHECK-LABEL: {{^}}mubuf_load0:
; CHECK: buffer_load_dword v{{[0-9]}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:4 ; encoding: [0x04,0x00,0x30,0xe0
define amdgpu_kernel void @mubuf_load0(ptr addrspace(1) %out, ptr addrspace(1) %in) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %in, i64 1
%1 = load i32, ptr addrspace(1) %0
store i32 %1, ptr addrspace(1) %out
ret void
}
; MUBUF load with the largest possible immediate offset
; CHECK-LABEL: {{^}}mubuf_load1:
; CHECK: buffer_load_ubyte v{{[0-9]}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:4095 ; encoding: [0xff,0x0f,0x20,0xe0
define amdgpu_kernel void @mubuf_load1(ptr addrspace(1) %out, ptr addrspace(1) %in) {
entry:
%0 = getelementptr i8, ptr addrspace(1) %in, i64 4095
%1 = load i8, ptr addrspace(1) %0
store i8 %1, ptr addrspace(1) %out
ret void
}
; MUBUF load with an immediate byte offset that doesn't fit into 12-bits
; CHECK-LABEL: {{^}}mubuf_load2:
; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
; CHECK: buffer_load_dword v{{[0-9]}}, off, s[{{[0-9]+:[0-9]+}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x30,0xe0
define amdgpu_kernel void @mubuf_load2(ptr addrspace(1) %out, ptr addrspace(1) %in) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %in, i64 1024
%1 = load i32, ptr addrspace(1) %0
store i32 %1, ptr addrspace(1) %out
ret void
}
; MUBUF load with a 12-bit immediate offset and a register offset
; CHECK-LABEL: {{^}}mubuf_load3:
; CHECK-NOT: ADD
; CHECK: buffer_load_dword v{{[0-9]}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], 0 addr64 offset:4 ; encoding: [0x04,0x80,0x30,0xe0
define amdgpu_kernel void @mubuf_load3(ptr addrspace(1) %out, ptr addrspace(1) %in, i64 %offset) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %in, i64 %offset
%1 = getelementptr i32, ptr addrspace(1) %0, i64 1
%2 = load i32, ptr addrspace(1) %1
store i32 %2, ptr addrspace(1) %out
ret void
}
; CHECK-LABEL: {{^}}soffset_max_imm:
; CHECK: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], 64 offen glc
define amdgpu_gs void @soffset_max_imm(ptr addrspace(4) inreg, ptr addrspace(4) inreg, ptr addrspace(4) inreg, ptr addrspace(4) inreg, i32 inreg, i32 inreg, i32, i32, i32, i32, i32, i32, i32, i32) {
main_body:
%tmp1 = load ptr addrspace(8), ptr addrspace(4) %0
%tmp2 = shl i32 %6, 2
%tmp3 = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) %tmp1, i32 %tmp2, i32 64, i32 1)
%tmp4 = add i32 %6, 16
%tmp1.4xi32 = bitcast ptr addrspace(8) %tmp1 to ptr addrspace(8)
call void @llvm.amdgcn.raw.ptr.tbuffer.store.i32(i32 %tmp3, ptr addrspace(8) %tmp1.4xi32, i32 %tmp4, i32 %4, i32 68, i32 3)
ret void
}
; Make sure immediates that aren't inline constants don't get folded into
; the soffset operand.
; FIXME: for this test we should be smart enough to shift the immediate into
; the offset field.
; CHECK-LABEL: {{^}}soffset_no_fold:
; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x41
; CHECK: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+}}:{{[0-9]+}}], [[SOFFSET]] offen glc
define amdgpu_gs void @soffset_no_fold(ptr addrspace(4) inreg, ptr addrspace(4) inreg, ptr addrspace(4) inreg, ptr addrspace(4) inreg, i32 inreg, i32 inreg, i32, i32, i32, i32, i32, i32, i32, i32) {
main_body:
%tmp1 = load ptr addrspace(8), ptr addrspace(4) %0
%tmp2 = shl i32 %6, 2
%tmp3 = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) %tmp1, i32 %tmp2, i32 65, i32 1)
%tmp4 = add i32 %6, 16
%tmp1.4xi32 = bitcast ptr addrspace(8) %tmp1 to ptr addrspace(8)
call void @llvm.amdgcn.raw.ptr.tbuffer.store.i32(i32 %tmp3, ptr addrspace(8) %tmp1.4xi32, i32 %tmp4, i32 %4, i32 68, i32 3)
ret void
}
;;;==========================================================================;;;
;;; MUBUF STORE TESTS
;;;==========================================================================;;;
; MUBUF store with an immediate byte offset that fits into 12-bits
; CHECK-LABEL: {{^}}mubuf_store0:
; CHECK: buffer_store_dword v{{[0-9]}}, off, s[{{[0-9]:[0-9]}}], 0 offset:4 ; encoding: [0x04,0x00,0x70,0xe0
define amdgpu_kernel void @mubuf_store0(ptr addrspace(1) %out) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %out, i64 1
store i32 0, ptr addrspace(1) %0
ret void
}
; MUBUF store with the largest possible immediate offset
; CHECK-LABEL: {{^}}mubuf_store1:
; CHECK: buffer_store_byte v{{[0-9]}}, off, s[{{[0-9]:[0-9]}}], 0 offset:4095 ; encoding: [0xff,0x0f,0x60,0xe0
define amdgpu_kernel void @mubuf_store1(ptr addrspace(1) %out) {
entry:
%0 = getelementptr i8, ptr addrspace(1) %out, i64 4095
store i8 0, ptr addrspace(1) %0
ret void
}
; MUBUF store with an immediate byte offset that doesn't fit into 12-bits
; CHECK-LABEL: {{^}}mubuf_store2:
; CHECK: s_movk_i32 [[SOFFSET:s[0-9]+]], 0x1000
; CHECK: buffer_store_dword v{{[0-9]}}, off, s[{{[0-9]:[0-9]}}], [[SOFFSET]] ; encoding: [0x00,0x00,0x70,0xe0
define amdgpu_kernel void @mubuf_store2(ptr addrspace(1) %out) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %out, i64 1024
store i32 0, ptr addrspace(1) %0
ret void
}
; MUBUF store with a 12-bit immediate offset and a register offset
; CHECK-LABEL: {{^}}mubuf_store3:
; CHECK-NOT: ADD
; CHECK: buffer_store_dword v{{[0-9]}}, v[{{[0-9]:[0-9]}}], s[{{[0-9]:[0-9]}}], 0 addr64 offset:4 ; encoding: [0x04,0x80,0x70,0xe0
define amdgpu_kernel void @mubuf_store3(ptr addrspace(1) %out, i64 %offset) {
entry:
%0 = getelementptr i32, ptr addrspace(1) %out, i64 %offset
%1 = getelementptr i32, ptr addrspace(1) %0, i64 1
store i32 0, ptr addrspace(1) %1
ret void
}
; CHECK-LABEL: {{^}}store_sgpr_ptr:
; CHECK: buffer_store_dword v{{[0-9]+}}, off, s{{\[[0-9]+:[0-9]+\]}}, 0
define amdgpu_kernel void @store_sgpr_ptr(ptr addrspace(1) %out) {
store i32 99, ptr addrspace(1) %out, align 4
ret void
}
; CHECK-LABEL: {{^}}store_sgpr_ptr_offset:
; CHECK: buffer_store_dword v{{[0-9]+}}, off, s{{\[[0-9]+:[0-9]+\]}}, 0 offset:40
define amdgpu_kernel void @store_sgpr_ptr_offset(ptr addrspace(1) %out) {
%out.gep = getelementptr i32, ptr addrspace(1) %out, i32 10
store i32 99, ptr addrspace(1) %out.gep, align 4
ret void
}
; CHECK-LABEL: {{^}}store_sgpr_ptr_large_offset:
; CHECK: s_mov_b32 [[SOFFSET:s[0-9]+]], 0x20000
; CHECK: buffer_store_dword v{{[0-9]+}}, off, s{{\[[0-9]+:[0-9]+\]}}, [[SOFFSET]]
define amdgpu_kernel void @store_sgpr_ptr_large_offset(ptr addrspace(1) %out) {
%out.gep = getelementptr i32, ptr addrspace(1) %out, i32 32768
store i32 99, ptr addrspace(1) %out.gep, align 4
ret void
}
; CHECK-LABEL: {{^}}store_sgpr_ptr_large_offset_atomic:
; CHECK: s_mov_b32 [[SOFFSET:s[0-9]+]], 0x20000
; CHECK: buffer_atomic_add v{{[0-9]+}}, off, s{{\[[0-9]+:[0-9]+\]}}, [[SOFFSET]]
define amdgpu_kernel void @store_sgpr_ptr_large_offset_atomic(ptr addrspace(1) %out) {
%gep = getelementptr i32, ptr addrspace(1) %out, i32 32768
%val = atomicrmw volatile add ptr addrspace(1) %gep, i32 5 seq_cst
ret void
}
; CHECK-LABEL: {{^}}store_vgpr_ptr:
; CHECK: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64
define amdgpu_kernel void @store_vgpr_ptr(ptr addrspace(1) %out) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() readnone
%out.gep = getelementptr i32, ptr addrspace(1) %out, i32 %tid
store i32 99, ptr addrspace(1) %out.gep, align 4
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
declare void @llvm.amdgcn.raw.ptr.tbuffer.store.i32(i32, ptr addrspace(8), i32, i32, i32 immarg, i32 immarg) #2
declare i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8), i32, i32, i32 immarg) #3
attributes #0 = { nounwind readonly }
attributes #1 = { nounwind readnone speculatable willreturn }
attributes #2 = { nounwind willreturn writeonly }
attributes #3 = { nounwind readonly willreturn }
attributes #4 = { readnone }
|
