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
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
|
; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
declare i32 @llvm.amdgcn.workitem.id.y() nounwind readnone
; GCN-LABEL: {{^}}load_i8_to_f32:
; GCN: {{buffer|flat}}_load_ubyte [[LOADREG:v[0-9]+]],
; GCN-NOT: bfe
; GCN-NOT: lshr
; GCN: v_cvt_f32_ubyte0_e32 [[CONV:v[0-9]+]], [[LOADREG]]
; GCN: buffer_store_dword [[CONV]],
define amdgpu_kernel void @load_i8_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i8, i8 addrspace(1)* %in, i32 %tid
%load = load i8, i8 addrspace(1)* %gep, align 1
%cvt = uitofp i8 %load to float
store float %cvt, float addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}load_v2i8_to_v2f32:
; GCN: {{buffer|flat}}_load_ushort [[LD:v[0-9]+]]
; GCN-DAG: v_cvt_f32_ubyte1_e32 v[[HIRESULT:[0-9]+]], [[LD]]
; GCN-DAG: v_cvt_f32_ubyte0_e32 v[[LORESULT:[0-9]+]], [[LD]]
; GCN: buffer_store_dwordx2 v{{\[}}[[LORESULT]]:[[HIRESULT]]{{\]}},
define amdgpu_kernel void @load_v2i8_to_v2f32(<2 x float> addrspace(1)* noalias %out, <2 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x i8>, <2 x i8> addrspace(1)* %in, i32 %tid
%load = load <2 x i8>, <2 x i8> addrspace(1)* %gep, align 2
%cvt = uitofp <2 x i8> %load to <2 x float>
store <2 x float> %cvt, <2 x float> addrspace(1)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}load_v3i8_to_v3f32:
; GCN: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
; GCN-NOT: v_cvt_f32_ubyte3_e32
; GCN-DAG: v_cvt_f32_ubyte2_e32 v{{[0-9]+}}, [[VAL]]
; GCN-DAG: v_cvt_f32_ubyte1_e32 v[[HIRESULT:[0-9]+]], [[VAL]]
; GCN-DAG: v_cvt_f32_ubyte0_e32 v[[LORESULT:[0-9]+]], [[VAL]]
; GCN: buffer_store_dwordx2 v{{\[}}[[LORESULT]]:[[HIRESULT]]{{\]}},
define amdgpu_kernel void @load_v3i8_to_v3f32(<3 x float> addrspace(1)* noalias %out, <3 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <3 x i8>, <3 x i8> addrspace(1)* %in, i32 %tid
%load = load <3 x i8>, <3 x i8> addrspace(1)* %gep, align 4
%cvt = uitofp <3 x i8> %load to <3 x float>
store <3 x float> %cvt, <3 x float> addrspace(1)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}load_v4i8_to_v4f32:
; GCN: {{buffer|flat}}_load_dword [[LOADREG:v[0-9]+]]
; GCN-NOT: bfe
; GCN-NOT: lshr
; GCN-DAG: v_cvt_f32_ubyte3_e32 v[[HIRESULT:[0-9]+]], [[LOADREG]]
; GCN-DAG: v_cvt_f32_ubyte2_e32 v{{[0-9]+}}, [[LOADREG]]
; GCN-DAG: v_cvt_f32_ubyte1_e32 v{{[0-9]+}}, [[LOADREG]]
; GCN-DAG: v_cvt_f32_ubyte0_e32 v[[LORESULT:[0-9]+]], [[LOADREG]]
; GCN: buffer_store_dwordx4 v{{\[}}[[LORESULT]]:[[HIRESULT]]{{\]}},
define amdgpu_kernel void @load_v4i8_to_v4f32(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
%load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 4
%cvt = uitofp <4 x i8> %load to <4 x float>
store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
ret void
}
; This should not be adding instructions to shift into the correct
; position in the word for the component.
; FIXME: Packing bytes
; GCN-LABEL: {{^}}load_v4i8_to_v4f32_unaligned:
; GCN: {{buffer|flat}}_load_ubyte [[LOADREG3:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ubyte [[LOADREG2:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ubyte [[LOADREG1:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ubyte [[LOADREG0:v[0-9]+]]
; GCN-DAG: v_lshlrev_b32
; GCN-DAG: v_or_b32
; GCN-DAG: v_cvt_f32_ubyte0_e32 v[[LORESULT:[0-9]+]],
; GCN-DAG: v_cvt_f32_ubyte0_e32 v{{[0-9]+}},
; GCN-DAG: v_cvt_f32_ubyte0_e32 v{{[0-9]+}},
; GCN-DAG: v_cvt_f32_ubyte0_e32 v[[HIRESULT:[0-9]+]]
; GCN: buffer_store_dwordx4
define amdgpu_kernel void @load_v4i8_to_v4f32_unaligned(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
%load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
%cvt = uitofp <4 x i8> %load to <4 x float>
store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
ret void
}
; FIXME: Need to handle non-uniform case for function below (load without gep).
; Instructions still emitted to repack bytes for add use.
; GCN-LABEL: {{^}}load_v4i8_to_v4f32_2_uses:
; GCN: {{buffer|flat}}_load_dword
; GCN-DAG: v_cvt_f32_ubyte0_e32
; GCN-DAG: v_cvt_f32_ubyte1_e32
; GCN-DAG: v_cvt_f32_ubyte2_e32
; GCN-DAG: v_cvt_f32_ubyte3_e32
; GCN-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 24
; SI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16
; SI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 8
; SI-DAG: v_and_b32_e32 v{{[0-9]+}}, 0xffff,
; SI-DAG: v_and_b32_e32 v{{[0-9]+}}, 0xff00,
; SI-DAG: v_add_i32
; VI-DAG: v_and_b32_e32 v{{[0-9]+}}, 0xffffff00,
; VI-DAG: v_add_u16_e32
; VI-DAG: v_add_u16_e32
; GCN: {{buffer|flat}}_store_dwordx4
; GCN: {{buffer|flat}}_store_dword
; GCN: s_endpgm
define amdgpu_kernel void @load_v4i8_to_v4f32_2_uses(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %out2, <4 x i8> addrspace(1)* noalias %in) nounwind {
%tid.x = call i32 @llvm.amdgcn.workitem.id.x()
%in.ptr = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid.x
%load = load <4 x i8>, <4 x i8> addrspace(1)* %in.ptr, align 4
%cvt = uitofp <4 x i8> %load to <4 x float>
store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
%add = add <4 x i8> %load, <i8 9, i8 9, i8 9, i8 9> ; Second use of %load
store <4 x i8> %add, <4 x i8> addrspace(1)* %out2, align 4
ret void
}
; Make sure this doesn't crash.
; GCN-LABEL: {{^}}load_v7i8_to_v7f32:
; GCN: s_endpgm
define amdgpu_kernel void @load_v7i8_to_v7f32(<7 x float> addrspace(1)* noalias %out, <7 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <7 x i8>, <7 x i8> addrspace(1)* %in, i32 %tid
%load = load <7 x i8>, <7 x i8> addrspace(1)* %gep, align 1
%cvt = uitofp <7 x i8> %load to <7 x float>
store <7 x float> %cvt, <7 x float> addrspace(1)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}load_v8i8_to_v8f32:
; GCN: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LOLOAD:[0-9]+]]:[[HILOAD:[0-9]+]]{{\]}},
; GCN-NOT: bfe
; GCN-NOT: lshr
; GCN-DAG: v_cvt_f32_ubyte3_e32 v{{[0-9]+}}, v[[LOLOAD]]
; GCN-DAG: v_cvt_f32_ubyte2_e32 v{{[0-9]+}}, v[[LOLOAD]]
; GCN-DAG: v_cvt_f32_ubyte1_e32 v{{[0-9]+}}, v[[LOLOAD]]
; GCN-DAG: v_cvt_f32_ubyte0_e32 v{{[0-9]+}}, v[[LOLOAD]]
; GCN-DAG: v_cvt_f32_ubyte3_e32 v{{[0-9]+}}, v[[HILOAD]]
; GCN-DAG: v_cvt_f32_ubyte2_e32 v{{[0-9]+}}, v[[HILOAD]]
; GCN-DAG: v_cvt_f32_ubyte1_e32 v{{[0-9]+}}, v[[HILOAD]]
; GCN-DAG: v_cvt_f32_ubyte0_e32 v{{[0-9]+}}, v[[HILOAD]]
; GCN-NOT: bfe
; GCN-NOT: lshr
; GCN: buffer_store_dwordx4
; GCN: buffer_store_dwordx4
define amdgpu_kernel void @load_v8i8_to_v8f32(<8 x float> addrspace(1)* noalias %out, <8 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <8 x i8>, <8 x i8> addrspace(1)* %in, i32 %tid
%load = load <8 x i8>, <8 x i8> addrspace(1)* %gep, align 8
%cvt = uitofp <8 x i8> %load to <8 x float>
store <8 x float> %cvt, <8 x float> addrspace(1)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}i8_zext_inreg_i32_to_f32:
; GCN: {{buffer|flat}}_load_dword [[LOADREG:v[0-9]+]],
; GCN: v_add_{{[iu]}}32_e32 [[ADD:v[0-9]+]], vcc, 2, [[LOADREG]]
; GCN-NEXT: v_cvt_f32_ubyte0_e32 [[CONV:v[0-9]+]], [[ADD]]
; GCN: buffer_store_dword [[CONV]],
define amdgpu_kernel void @i8_zext_inreg_i32_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%load = load i32, i32 addrspace(1)* %gep, align 4
%add = add i32 %load, 2
%inreg = and i32 %add, 255
%cvt = uitofp i32 %inreg to float
store float %cvt, float addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}i8_zext_inreg_hi1_to_f32:
define amdgpu_kernel void @i8_zext_inreg_hi1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%load = load i32, i32 addrspace(1)* %gep, align 4
%inreg = and i32 %load, 65280
%shr = lshr i32 %inreg, 8
%cvt = uitofp i32 %shr to float
store float %cvt, float addrspace(1)* %out, align 4
ret void
}
; We don't get these ones because of the zext, but instcombine removes
; them so it shouldn't really matter.
; GCN-LABEL: {{^}}i8_zext_i32_to_f32:
define amdgpu_kernel void @i8_zext_i32_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i8, i8 addrspace(1)* %in, i32 %tid
%load = load i8, i8 addrspace(1)* %gep, align 1
%ext = zext i8 %load to i32
%cvt = uitofp i32 %ext to float
store float %cvt, float addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}v4i8_zext_v4i32_to_v4f32:
define amdgpu_kernel void @v4i8_zext_v4i32_to_v4f32(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
%load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
%ext = zext <4 x i8> %load to <4 x i32>
%cvt = uitofp <4 x i32> %ext to <4 x float>
store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}extract_byte0_to_f32:
; GCN: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
; GCN-NOT: [[VAL]]
; GCN: v_cvt_f32_ubyte0_e32 [[CONV:v[0-9]+]], [[VAL]]
; GCN: buffer_store_dword [[CONV]]
define amdgpu_kernel void @extract_byte0_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%val = load i32, i32 addrspace(1)* %gep
%and = and i32 %val, 255
%cvt = uitofp i32 %and to float
store float %cvt, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}extract_byte1_to_f32:
; GCN: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
; GCN-NOT: [[VAL]]
; GCN: v_cvt_f32_ubyte1_e32 [[CONV:v[0-9]+]], [[VAL]]
; GCN: buffer_store_dword [[CONV]]
define amdgpu_kernel void @extract_byte1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%val = load i32, i32 addrspace(1)* %gep
%srl = lshr i32 %val, 8
%and = and i32 %srl, 255
%cvt = uitofp i32 %and to float
store float %cvt, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}extract_byte2_to_f32:
; GCN: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
; GCN-NOT: [[VAL]]
; GCN: v_cvt_f32_ubyte2_e32 [[CONV:v[0-9]+]], [[VAL]]
; GCN: buffer_store_dword [[CONV]]
define amdgpu_kernel void @extract_byte2_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%val = load i32, i32 addrspace(1)* %gep
%srl = lshr i32 %val, 16
%and = and i32 %srl, 255
%cvt = uitofp i32 %and to float
store float %cvt, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}extract_byte3_to_f32:
; GCN: {{buffer|flat}}_load_dword [[VAL:v[0-9]+]]
; GCN-NOT: [[VAL]]
; GCN: v_cvt_f32_ubyte3_e32 [[CONV:v[0-9]+]], [[VAL]]
; GCN: buffer_store_dword [[CONV]]
define amdgpu_kernel void @extract_byte3_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
%val = load i32, i32 addrspace(1)* %gep
%srl = lshr i32 %val, 24
%and = and i32 %srl, 255
%cvt = uitofp i32 %and to float
store float %cvt, float addrspace(1)* %out
ret void
}
|
