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
284
285
286
|
; RUN: llc -march=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck --check-prefixes=GCN,GFX8 %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefixes=GCN,GFX9,GFX900 %s
; RUN: llc -march=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck --check-prefixes=GCN,GFX9,GFX906 %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
; GCN-LABEL: {{^}}shl_i16:
; GCN: v_lshlrev_b16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @shl_i16(i16 %x, i16 %y) {
%res = shl i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}lshr_i16:
; GCN: v_lshrrev_b16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @lshr_i16(i16 %x, i16 %y) {
%res = lshr i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}ashr_i16:
; GCN: v_ashrrev_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @ashr_i16(i16 %x, i16 %y) {
%res = ashr i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}add_u16:
; GCN: v_add_{{(nc_)*}}u16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @add_u16(i16 %x, i16 %y) {
%res = add i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}sub_u16:
; GCN: v_sub_{{(nc_)*}}u16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @sub_u16(i16 %x, i16 %y) {
%res = sub i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}mul_lo_u16:
; GCN: v_mul_lo_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @mul_lo_u16(i16 %x, i16 %y) {
%res = mul i16 %x, %y
ret i16 %res
}
; GCN-LABEL: {{^}}min_u16:
; GCN: v_min_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @min_u16(i16 %x, i16 %y) {
%cmp = icmp ule i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
ret i16 %res
}
; GCN-LABEL: {{^}}min_i16:
; GCN: v_min_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @min_i16(i16 %x, i16 %y) {
%cmp = icmp sle i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
ret i16 %res
}
; GCN-LABEL: {{^}}max_u16:
; GCN: v_max_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @max_u16(i16 %x, i16 %y) {
%cmp = icmp uge i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
ret i16 %res
}
; GCN-LABEL: {{^}}max_i16:
; GCN: v_max_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GCN-NEXT: s_setpc_b64
define i16 @max_i16(i16 %x, i16 %y) {
%cmp = icmp sge i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
ret i16 %res
}
; GCN-LABEL: {{^}}shl_i16_zext_i32:
; GCN: v_lshlrev_b16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @shl_i16_zext_i32(i16 %x, i16 %y) {
%res = shl i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}lshr_i16_zext_i32:
; GCN: v_lshrrev_b16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @lshr_i16_zext_i32(i16 %x, i16 %y) {
%res = lshr i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}ashr_i16_zext_i32:
; GCN: v_ashrrev_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @ashr_i16_zext_i32(i16 %x, i16 %y) {
%res = ashr i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}add_u16_zext_i32:
; GCN: v_add_{{(nc_)*}}u16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @add_u16_zext_i32(i16 %x, i16 %y) {
%res = add i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}sub_u16_zext_i32:
; GCN: v_sub_{{(nc_)*}}u16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @sub_u16_zext_i32(i16 %x, i16 %y) {
%res = sub i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}mul_lo_u16_zext_i32:
; GCN: v_mul_lo_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @mul_lo_u16_zext_i32(i16 %x, i16 %y) {
%res = mul i16 %x, %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}min_u16_zext_i32:
; GCN: v_min_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @min_u16_zext_i32(i16 %x, i16 %y) {
%cmp = icmp ule i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}min_i16_zext_i32:
; GCN: v_min_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @min_i16_zext_i32(i16 %x, i16 %y) {
%cmp = icmp sle i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}max_u16_zext_i32:
; GCN: v_max_u16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @max_u16_zext_i32(i16 %x, i16 %y) {
%cmp = icmp uge i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}max_i16_zext_i32:
; GCN: v_max_i16{{[_e32]*}} [[OP:v[0-9]+]],
; GFX10-NEXT: v_and_b32_e32 v{{[0-9]+}}, 0xffff, [[OP]]
; GCN-NEXT: s_setpc_b64
define i32 @max_i16_zext_i32(i16 %x, i16 %y) {
%cmp = icmp sge i16 %x, %y
%res = select i1 %cmp, i16 %x, i16 %y
%zext = zext i16 %res to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}zext_fadd_f16:
; GFX8: v_add_f16_e32 [[ADD:v[0-9]+]], v0, v1
; GFX8-NEXT: s_setpc_b64
; GFX9: v_add_f16_e32 [[ADD:v[0-9]+]], v0, v1
; GFX9-NEXT: s_setpc_b64
; GFX10: v_add_f16_e32 [[ADD:v[0-9]+]], v0, v1
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, [[ADD]]
define i32 @zext_fadd_f16(half %x, half %y) {
%add = fadd half %x, %y
%cast = bitcast half %add to i16
%zext = zext i16 %cast to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}zext_fma_f16:
; GFX8: v_fma_f16 [[FMA:v[0-9]+]], v0, v1, v2
; GFX8-NEXT: s_setpc_b64
; GFX9: v_fma_f16 [[FMA:v[0-9]+]], v0, v1, v2
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, [[FMA]]
; GFX10: v_fmac_f16_e32 [[FMA:v[0-9]+]], v0, v1
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, [[FMA]]
define i32 @zext_fma_f16(half %x, half %y, half %z) {
%fma = call half @llvm.fma.f16(half %x, half %y, half %z)
%cast = bitcast half %fma to i16
%zext = zext i16 %cast to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}zext_div_fixup_f16:
; GFX8: v_div_fixup_f16 v0, v0, v1, v2
; GFX8-NEXT: s_setpc_b64
; GFX9: v_div_fixup_f16 v0, v0, v1, v2
; GFX9-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX10: v_div_fixup_f16 v0, v0, v1, v2
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, v0
define i32 @zext_div_fixup_f16(half %x, half %y, half %z) {
%div.fixup = call half @llvm.amdgcn.div.fixup.f16(half %x, half %y, half %z)
%cast = bitcast half %div.fixup to i16
%zext = zext i16 %cast to i32
ret i32 %zext
}
; We technically could eliminate the and on gfx9 here but we don't try
; to inspect the source of the fptrunc. We're only worried about cases
; that lower to v_fma_mix* instructions.
; GCN-LABEL: {{^}}zext_fptrunc_f16:
; GFX8: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64
; GFX9: v_cvt_f16_f32_e32 v0, v0
; GFX9-NEXT: s_setpc_b64
; GFX10: v_cvt_f16_f32_e32 v0, v0
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, v0
define i32 @zext_fptrunc_f16(float %x) {
%fptrunc = fptrunc float %x to half
%cast = bitcast half %fptrunc to i16
%zext = zext i16 %cast to i32
ret i32 %zext
}
; GCN-LABEL: {{^}}zext_fptrunc_fma_f16:
; GFX8: v_fma_f32 v0, v0, v1, v2
; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX8-NEXT: s_setpc_b64
; GFX900: v_fma_f32 v0, v0, v1, v2
; GFX900-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX900-NEXT: s_setpc_b64
; GFX906: v_fma_mixlo_f16 v0, v0, v1, v2
; GFX906-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX10: v_fma_mixlo_f16 v0, v0, v1, v2
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, v0
define i32 @zext_fptrunc_fma_f16(float %x, float %y, float %z) {
%fma = call float @llvm.fma.f32(float %x, float %y, float %z)
%fptrunc = fptrunc float %fma to half
%cast = bitcast half %fptrunc to i16
%zext = zext i16 %cast to i32
ret i32 %zext
}
declare half @llvm.amdgcn.div.fixup.f16(half, half, half)
declare half @llvm.fma.f16(half, half, half)
declare float @llvm.fma.f32(float, float, float)
|
