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
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
define amdgpu_cs float @v_s_exp_f32(float inreg %src) {
; GFX12-LABEL: v_s_exp_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_cmp_lt_f32 s0, 0xc2fc0000
; GFX12-NEXT: s_cselect_b32 s1, 0x42800000, 0
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_2)
; GFX12-NEXT: s_add_f32 s0, s0, s1
; GFX12-NEXT: s_cselect_b32 s1, 0x1f800000, 1.0
; GFX12-NEXT: v_s_exp_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_3)
; GFX12-NEXT: s_mul_f32 s0, s0, s1
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call float @llvm.exp2.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_exp_f16(half inreg %src) {
; GFX12-LABEL: v_s_exp_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_exp_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.exp2.f16(half %src)
ret half %result
}
define amdgpu_cs float @v_s_amdgcn_exp_f32(float inreg %src) {
; GFX12-LABEL: v_s_amdgcn_exp_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_exp_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call float @llvm.amdgcn.exp2.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_amdgcn_exp_f16(half inreg %src) {
; GFX12-LABEL: v_s_amdgcn_exp_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_exp_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.amdgcn.exp2.f16(half %src)
ret half %result
}
define amdgpu_cs float @v_s_log_f32(float inreg %src) {
; GFX12-LABEL: v_s_log_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_cmp_lt_f32 s0, 0x800000
; GFX12-NEXT: s_cselect_b32 s1, 0x4f800000, 1.0
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_2)
; GFX12-NEXT: s_mul_f32 s0, s0, s1
; GFX12-NEXT: s_cselect_b32 s1, 0x42000000, 0
; GFX12-NEXT: v_s_log_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_3)
; GFX12-NEXT: s_sub_f32 s0, s0, s1
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call float @llvm.log2.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_log_f16(half inreg %src) {
; GFX12-LABEL: v_s_log_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_log_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.log2.f16(half %src)
ret half %result
}
define amdgpu_cs float @v_s_amdgcn_log_f32(float inreg %src) {
; GFX12-LABEL: v_s_amdgcn_log_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_log_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call float @llvm.amdgcn.log.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_amdgcn_log_f16(half inreg %src) {
; GFX12-LABEL: v_s_amdgcn_log_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_log_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.amdgcn.log.f16(half %src)
ret half %result
}
define amdgpu_cs float @v_s_rcp_f32(float inreg %src) {
; GFX12-LABEL: v_s_rcp_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_rcp_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call fast float @llvm.amdgcn.rcp.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_rcp_f16(half inreg %src) {
; GFX12-LABEL: v_s_rcp_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_rcp_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call fast half @llvm.amdgcn.rcp.f16(half %src)
ret half %result
}
; TODO-GFX12: GlobalISel should generate v_s_rsq.
define amdgpu_cs float @v_s_rsq_f32(float inreg %src) {
; GFX12-SDAG-LABEL: v_s_rsq_f32:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: v_s_rsq_f32 s0, s0
; GFX12-SDAG-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
; GFX12-SDAG-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: v_s_rsq_f32:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: v_s_sqrt_f32 s0, s0
; GFX12-GISEL-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
; GFX12-GISEL-NEXT: v_s_rcp_f32 s0, s0
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX12-GISEL-NEXT: ; return to shader part epilog
%sqrt = call fast float @llvm.sqrt.f32(float %src)
%fdiv = fdiv fast float 1.0, %sqrt
ret float %fdiv
}
define amdgpu_cs half @v_s_rsq_f16(half inreg %src) {
; GFX12-LABEL: v_s_rsq_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_rsq_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%sqrt = call fast half @llvm.sqrt.f16(half %src)
%result = fdiv fast half 1.0, %sqrt
ret half %result
}
; TODO-GFX12: Should not use any VALU instructions.
define amdgpu_cs float @v_s_sqrt_f32(float inreg %src) {
; GFX12-SDAG-LABEL: v_s_sqrt_f32:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: s_mul_f32 s1, s0, 0x4f800000
; GFX12-SDAG-NEXT: s_cmp_lt_f32 s0, 0xf800000
; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-SDAG-NEXT: s_cselect_b32 s1, s1, s0
; GFX12-SDAG-NEXT: v_s_sqrt_f32 s2, s1
; GFX12-SDAG-NEXT: s_mov_b32 s4, s1
; GFX12-SDAG-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-SDAG-NEXT: s_add_co_i32 s3, s2, -1
; GFX12-SDAG-NEXT: s_xor_b32 s5, s3, 0x80000000
; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_2)
; GFX12-SDAG-NEXT: s_fmac_f32 s4, s5, s2
; GFX12-SDAG-NEXT: s_mov_b32 s5, s1
; GFX12-SDAG-NEXT: s_cmp_le_f32 s4, 0
; GFX12-SDAG-NEXT: s_cselect_b32 s3, s3, s2
; GFX12-SDAG-NEXT: s_add_co_i32 s4, s2, 1
; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-SDAG-NEXT: s_xor_b32 s6, s4, 0x80000000
; GFX12-SDAG-NEXT: s_fmac_f32 s5, s6, s2
; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(SKIP_4) | instid1(SALU_CYCLE_2)
; GFX12-SDAG-NEXT: s_cmp_gt_f32 s5, 0
; GFX12-SDAG-NEXT: s_cselect_b32 s2, s4, s3
; GFX12-SDAG-NEXT: s_cmp_lt_f32 s0, 0xf800000
; GFX12-SDAG-NEXT: s_mul_f32 s0, s2, 0x37800000
; GFX12-SDAG-NEXT: v_cmp_class_f32_e64 s3, s1, 0x260
; GFX12-SDAG-NEXT: s_cselect_b32 s0, s0, s2
; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12-SDAG-NEXT: s_and_b32 s2, s3, exec_lo
; GFX12-SDAG-NEXT: s_cselect_b32 s0, s1, s0
; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
; GFX12-SDAG-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: v_s_sqrt_f32:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_cmp_gt_f32 0xf800000, s0
; GFX12-GISEL-NEXT: s_mul_f32 s2, s0, 0x4f800000
; GFX12-GISEL-NEXT: s_cselect_b32 s1, 1, 0
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_2) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_cselect_b32 s0, s2, s0
; GFX12-GISEL-NEXT: v_s_sqrt_f32 s2, s0
; GFX12-GISEL-NEXT: s_mov_b32 s4, s0
; GFX12-GISEL-NEXT: s_mov_b32 s6, s0
; GFX12-GISEL-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_add_co_i32 s3, s2, -1
; GFX12-GISEL-NEXT: s_xor_b32 s5, s3, 0x80000000
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_fmac_f32 s4, s5, s2
; GFX12-GISEL-NEXT: s_add_co_i32 s5, s2, 1
; GFX12-GISEL-NEXT: s_xor_b32 s7, s5, 0x80000000
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_2)
; GFX12-GISEL-NEXT: s_cmp_le_f32 s4, 0
; GFX12-GISEL-NEXT: s_fmac_f32 s6, s7, s2
; GFX12-GISEL-NEXT: s_cselect_b32 s2, s3, s2
; GFX12-GISEL-NEXT: s_cmp_gt_f32 s6, 0
; GFX12-GISEL-NEXT: s_cselect_b32 s2, s5, s2
; GFX12-GISEL-NEXT: s_cmp_lg_u32 s1, 0
; GFX12-GISEL-NEXT: s_mul_f32 s3, s2, 0x37800000
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_cselect_b32 s1, s3, s2
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s1
; GFX12-GISEL-NEXT: v_cmp_class_f32_e64 s1, s0, 0x260
; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT: v_cndmask_b32_e64 v0, v0, s0, s1
; GFX12-GISEL-NEXT: ; return to shader part epilog
%result = call float @llvm.sqrt.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_s_sqrt_f16(half inreg %src) {
; GFX12-LABEL: v_s_sqrt_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_sqrt_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.sqrt.f16(half %src)
ret half %result
}
define amdgpu_cs float @v_amdgcn_sqrt_f32(float inreg %src) {
; GFX12-LABEL: v_amdgcn_sqrt_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_sqrt_f32 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call float @llvm.amdgcn.sqrt.f32(float %src)
ret float %result
}
define amdgpu_cs half @v_amdgcn_sqrt_f16(half inreg %src) {
; GFX12-LABEL: v_amdgcn_sqrt_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_sqrt_f16 s0, s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%result = call half @llvm.amdgcn.sqrt.f16(half %src)
ret half %result
}
define amdgpu_cs float @srcmods_abs_f32(float inreg %src) {
; GFX12-LABEL: srcmods_abs_f32:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_bitset0_b32 s0, 31
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12-NEXT: s_cmp_lt_f32 s0, 0x800000
; GFX12-NEXT: s_cselect_b32 s1, 0x4f800000, 1.0
; GFX12-NEXT: s_mul_f32 s0, s0, s1
; GFX12-NEXT: s_cselect_b32 s1, 0x42000000, 0
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
; GFX12-NEXT: v_s_log_f32 s0, s0
; GFX12-NEXT: s_sub_f32 s0, s0, s1
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%abs = call float @llvm.fabs.f32(float %src)
%result = call float @llvm.log2.f32(float %abs)
ret float %result
}
define amdgpu_cs float @srcmods_neg_f32(float inreg %src) {
; GFX12-SDAG-LABEL: srcmods_neg_f32:
; GFX12-SDAG: ; %bb.0:
; GFX12-SDAG-NEXT: s_xor_b32 s1, s0, 0x80000000
; GFX12-SDAG-NEXT: s_cmp_gt_f32 s0, 0x80800000
; GFX12-SDAG-NEXT: s_cselect_b32 s0, 0x4f800000, 1.0
; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_2)
; GFX12-SDAG-NEXT: s_mul_f32 s0, s1, s0
; GFX12-SDAG-NEXT: s_cselect_b32 s1, 0x42000000, 0
; GFX12-SDAG-NEXT: v_s_log_f32 s0, s0
; GFX12-SDAG-NEXT: s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_3)
; GFX12-SDAG-NEXT: s_sub_f32 s0, s0, s1
; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
; GFX12-SDAG-NEXT: ; return to shader part epilog
;
; GFX12-GISEL-LABEL: srcmods_neg_f32:
; GFX12-GISEL: ; %bb.0:
; GFX12-GISEL-NEXT: s_xor_b32 s0, s0, 0x80000000
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
; GFX12-GISEL-NEXT: s_cmp_lt_f32 s0, 0x800000
; GFX12-GISEL-NEXT: s_cselect_b32 s1, 0x4f800000, 1.0
; GFX12-GISEL-NEXT: s_mul_f32 s0, s0, s1
; GFX12-GISEL-NEXT: s_cselect_b32 s1, 0x42000000, 0
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
; GFX12-GISEL-NEXT: v_s_log_f32 s0, s0
; GFX12-GISEL-NEXT: s_sub_f32 s0, s0, s1
; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
; GFX12-GISEL-NEXT: ; return to shader part epilog
%neg = fneg float %src
%result = call float @llvm.log2.f32(float %neg)
ret float %result
}
define amdgpu_cs half @srcmods_abs_f16(half inreg %src) {
; GFX12-LABEL: srcmods_abs_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_log_f16 s0, |s0|
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%abs = call half @llvm.fabs.f16(half %src)
%result = call half @llvm.log2.f16(half %abs)
ret half %result
}
define amdgpu_cs half @srcmods_neg_f16(half inreg %src) {
; GFX12-LABEL: srcmods_neg_f16:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_s_log_f16 s0, -s0
; GFX12-NEXT: s_delay_alu instid0(TRANS32_DEP_1)
; GFX12-NEXT: v_mov_b32_e32 v0, s0
; GFX12-NEXT: ; return to shader part epilog
%neg = fneg half %src
%result = call half @llvm.log2.f16(half %neg)
ret half %result
}
declare half @llvm.exp2.f16(half)
declare float @llvm.exp2.f32(float)
declare half @llvm.amdgcn.exp2.f16(half)
declare float @llvm.amdgcn.exp2.f32(float)
declare half @llvm.log2.f16(half)
declare float @llvm.log2.f32(float)
declare half @llvm.amdgcn.log.f16(half)
declare float @llvm.amdgcn.log.f32(float)
declare half @llvm.amdgcn.rcp.f16(half)
declare float @llvm.amdgcn.rcp.f32(float)
declare half @llvm.sqrt.f16(half)
declare float @llvm.sqrt.f32(float)
declare half @llvm.amdgcn.sqrt.f16(half)
declare float @llvm.amdgcn.sqrt.f32(float)
declare half @llvm.fabs.f16(half)
declare float @llvm.fabs.f32(float)
|
