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
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
; Test that we use SimplifyDemandedBits on copysign's sign
; operand. These are somewhat simplified extractions from fast pown
; expansions.
define half @test_pown_reduced_fast_f16_known_odd(half %x, i32 %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_f16_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v1, 1, v1
; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX9-NEXT: s_movk_i32 s4, 0x7fff
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_mul_f16_e64 v1, |v0|, v1
; GFX9-NEXT: v_bfi_b32 v0, s4, v1, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or i32 %y.arg, 1
%fabs = call half @llvm.fabs.f16(half %x)
%pownI2F = sitofp i32 %y to half
%ylogx = fmul half %fabs, %pownI2F
%cast_x = bitcast half %x to i16
%pow_sign = and i16 %cast_x, -32768
%cast_sign = bitcast i16 %pow_sign to half
%pow_sign1 = call half @llvm.copysign.f16(half %ylogx, half %cast_sign)
ret half %pow_sign1
}
define <2 x half> @test_pown_reduced_fast_v2f16_known_odd(<2 x half> %x, <2 x i32> %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_v2f16_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v1, 1, v1
; GFX9-NEXT: v_or_b32_e32 v2, 1, v2
; GFX9-NEXT: v_cvt_f32_i32_e32 v2, v2
; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX9-NEXT: v_and_b32_e32 v3, 0x7fff7fff, v0
; GFX9-NEXT: s_movk_i32 s4, 0x7fff
; GFX9-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-NEXT: v_pack_b32_f16 v1, v1, v2
; GFX9-NEXT: v_pk_mul_f16 v1, v3, v1
; GFX9-NEXT: v_bfi_b32 v2, s4, v1, v0
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v0, 16, v0
; GFX9-NEXT: v_bfi_b32 v0, s4, v1, v0
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v0, v0, v2, s4
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or <2 x i32> %y.arg, <i32 1, i32 1>
%fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %x)
%pownI2F = sitofp <2 x i32> %y to <2 x half>
%ylogx = fmul <2 x half> %fabs, %pownI2F
%cast_x = bitcast <2 x half> %x to <2 x i16>
%pow_sign = and <2 x i16> %cast_x, <i16 -32768, i16 -32768>
%cast_sign = bitcast <2 x i16> %pow_sign to <2 x half>
%pow_sign1 = call <2 x half> @llvm.copysign.v2f16(<2 x half> %ylogx, <2 x half> %cast_sign)
ret <2 x half> %pow_sign1
}
define float @test_pown_reduced_fast_f32_known_odd(float %x, i32 %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_f32_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v1, 1, v1
; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_mul_f32_e64 v1, |v0|, v1
; GFX9-NEXT: v_bfi_b32 v0, s4, v1, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or i32 %y.arg, 1
%fabs = call float @llvm.fabs.f32(float %x)
%pownI2F = sitofp i32 %y to float
%ylogx = fmul float %fabs, %pownI2F
%cast_x = bitcast float %x to i32
%pow_sign = and i32 %cast_x, -2147483648
%cast_sign = bitcast i32 %pow_sign to float
%pow_sign1 = call float @llvm.copysign.f32(float %ylogx, float %cast_sign)
ret float %pow_sign1
}
define <2 x float> @test_pown_reduced_fast_v2f32_known_odd(<2 x float> %x, <2 x i32> %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_v2f32_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v3, 1, v3
; GFX9-NEXT: v_or_b32_e32 v2, 1, v2
; GFX9-NEXT: v_cvt_f32_i32_e32 v3, v3
; GFX9-NEXT: v_cvt_f32_i32_e32 v2, v2
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_mul_f32_e64 v3, |v1|, v3
; GFX9-NEXT: v_mul_f32_e64 v2, |v0|, v2
; GFX9-NEXT: v_bfi_b32 v0, s4, v2, v0
; GFX9-NEXT: v_bfi_b32 v1, s4, v3, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or <2 x i32> %y.arg, <i32 1, i32 1>
%fabs = call <2 x float> @llvm.fabs.v2f32(<2 x float> %x)
%pownI2F = sitofp <2 x i32> %y to <2 x float>
%ylogx = fmul <2 x float> %fabs, %pownI2F
%cast_x = bitcast <2 x float> %x to <2 x i32>
%pow_sign = and <2 x i32> %cast_x, <i32 -2147483648, i32 -2147483648>
%cast_sign = bitcast <2 x i32> %pow_sign to <2 x float>
%pow_sign1 = call <2 x float> @llvm.copysign.v2f32(<2 x float> %ylogx, <2 x float> %cast_sign)
ret <2 x float> %pow_sign1
}
define double @test_pown_reduced_fast_f64_known_odd(double %x, i32 %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_f64_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v2, 1, v2
; GFX9-NEXT: v_cvt_f64_i32_e32 v[2:3], v2
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_mul_f64 v[2:3], |v[0:1]|, v[2:3]
; GFX9-NEXT: v_bfi_b32 v1, s4, v3, v1
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or i32 %y.arg, 1
%fabs = call double @llvm.fabs.f64(double %x)
%pownI2F = sitofp i32 %y to double
%ylogx = fmul double %fabs, %pownI2F
%cast_x = bitcast double %x to i64
%pow_sign = and i64 %cast_x, -9223372036854775808
%cast_sign = bitcast i64 %pow_sign to double
%pow_sign1 = call double @llvm.copysign.f64(double %ylogx, double %cast_sign)
ret double %pow_sign1
}
define <2 x double> @test_pown_reduced_fast_v2f64_known_odd(<2 x double> %x, <2 x i32> %y.arg) #0 {
; GFX9-LABEL: test_pown_reduced_fast_v2f64_known_odd:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_or_b32_e32 v6, 1, v5
; GFX9-NEXT: v_or_b32_e32 v4, 1, v4
; GFX9-NEXT: v_cvt_f64_i32_e32 v[4:5], v4
; GFX9-NEXT: v_cvt_f64_i32_e32 v[6:7], v6
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_mul_f64 v[4:5], |v[0:1]|, v[4:5]
; GFX9-NEXT: v_mul_f64 v[6:7], |v[2:3]|, v[6:7]
; GFX9-NEXT: v_bfi_b32 v1, s4, v5, v1
; GFX9-NEXT: v_bfi_b32 v3, s4, v7, v3
; GFX9-NEXT: v_mov_b32_e32 v0, v4
; GFX9-NEXT: v_mov_b32_e32 v2, v6
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = or <2 x i32> %y.arg, <i32 1, i32 1>
%fabs = call <2 x double> @llvm.fabs.v2f64(<2 x double> %x)
%pownI2F = sitofp <2 x i32> %y to <2 x double>
%ylogx = fmul <2 x double> %fabs, %pownI2F
%cast_x = bitcast <2 x double> %x to <2 x i64>
%pow_sign = and <2 x i64> %cast_x, <i64 -9223372036854775808, i64 -9223372036854775808>
%cast_sign = bitcast <2 x i64> %pow_sign to <2 x double>
%pow_sign1 = call <2 x double> @llvm.copysign.f64(<2 x double> %ylogx, <2 x double> %cast_sign)
ret <2 x double> %pow_sign1
}
define float @copysign_f32_f32_sign_known_p0_or_n0(float %x, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define double @copysign_f64_f32_sign_known_p0_or_n0(double %x, i32 %y.i) {
; GFX9-LABEL: copysign_f64_f32_sign_known_p0_or_n0:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 31, v2
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v1, s4, v1, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%y.even.as.f32.fpext = fpext float %y.even.as.f32 to double
%copysign = call double @llvm.copysign.f64(double %x, double %y.even.as.f32.fpext)
ret double %copysign
}
define half @copysign_f16_f32_sign_known_p0_or_n0(half %x, i32 %y.i) {
; GFX9-LABEL: copysign_f16_f32_sign_known_p0_or_n0:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: v_lshrrev_b32_e32 v1, 16, v1
; GFX9-NEXT: s_movk_i32 s4, 0x7fff
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%y.even.as.f32.fptrunc = fptrunc float %y.even.as.f32 to half
%copysign = call half @llvm.copysign.f16(half %x, half %y.even.as.f32.fptrunc)
ret half %copysign
}
define float @copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_fabs(float %x.arg, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_fabs:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%x = call float @llvm.fabs.f32(float %x.arg)
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define float @copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_select(float %x.arg, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_select:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_cmp_lt_f32_e32 vcc, 0, v0
; GFX9-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%x.ule.0 = fcmp ule float %x.arg, 0.0
%x = select i1 %x.ule.0, float 0.0, float %x.arg
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define float @copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_nnan_nsz_sqrt(float %x.arg, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0__mag_known_positive_nnan_nsz_sqrt:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0xf800000
; GFX9-NEXT: v_mul_f32_e32 v2, 0x4f800000, v0
; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0
; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; GFX9-NEXT: v_sqrt_f32_e32 v2, v0
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: v_add_u32_e32 v3, -1, v2
; GFX9-NEXT: v_fma_f32 v4, -v3, v2, v0
; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], 0, v4
; GFX9-NEXT: v_add_u32_e32 v4, 1, v2
; GFX9-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[4:5]
; GFX9-NEXT: v_fma_f32 v2, -v4, v2, v0
; GFX9-NEXT: v_cmp_lt_f32_e64 s[4:5], 0, v2
; GFX9-NEXT: v_cndmask_b32_e64 v2, v3, v4, s[4:5]
; GFX9-NEXT: v_mul_f32_e32 v3, 0x37800000, v2
; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
; GFX9-NEXT: v_mov_b32_e32 v3, 0x260
; GFX9-NEXT: v_cmp_class_f32_e32 vcc, v0, v3
; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%x = call nnan nsz float @llvm.sqrt.f32(float %x.arg)
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define float @copysign_f32_f32_sign_known_p0_or_n0__mag_almost_positive_nsz_sqrt(float %x.arg, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0__mag_almost_positive_nsz_sqrt:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0xf800000
; GFX9-NEXT: v_mul_f32_e32 v2, 0x4f800000, v0
; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0
; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; GFX9-NEXT: v_sqrt_f32_e32 v2, v0
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: v_add_u32_e32 v3, -1, v2
; GFX9-NEXT: v_fma_f32 v4, -v3, v2, v0
; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], 0, v4
; GFX9-NEXT: v_add_u32_e32 v4, 1, v2
; GFX9-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[4:5]
; GFX9-NEXT: v_fma_f32 v2, -v4, v2, v0
; GFX9-NEXT: v_cmp_lt_f32_e64 s[4:5], 0, v2
; GFX9-NEXT: v_cndmask_b32_e64 v2, v3, v4, s[4:5]
; GFX9-NEXT: v_mul_f32_e32 v3, 0x37800000, v2
; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
; GFX9-NEXT: v_mov_b32_e32 v3, 0x260
; GFX9-NEXT: v_cmp_class_f32_e32 vcc, v0, v3
; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%x = call nsz float @llvm.sqrt.f32(float %x.arg)
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define float @copysign_f32_f32_sign_known_p0_or_n0__mag_almost_positive_nnan_sqrt(float %x.arg, i32 %y.i) {
; GFX9-LABEL: copysign_f32_f32_sign_known_p0_or_n0__mag_almost_positive_nnan_sqrt:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0xf800000
; GFX9-NEXT: v_mul_f32_e32 v2, 0x4f800000, v0
; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0
; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
; GFX9-NEXT: v_sqrt_f32_e32 v2, v0
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: v_add_u32_e32 v3, -1, v2
; GFX9-NEXT: v_fma_f32 v4, -v3, v2, v0
; GFX9-NEXT: v_cmp_ge_f32_e64 s[4:5], 0, v4
; GFX9-NEXT: v_add_u32_e32 v4, 1, v2
; GFX9-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[4:5]
; GFX9-NEXT: v_fma_f32 v2, -v4, v2, v0
; GFX9-NEXT: v_cmp_lt_f32_e64 s[4:5], 0, v2
; GFX9-NEXT: v_cndmask_b32_e64 v2, v3, v4, s[4:5]
; GFX9-NEXT: v_mul_f32_e32 v3, 0x37800000, v2
; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc
; GFX9-NEXT: v_mov_b32_e32 v3, 0x260
; GFX9-NEXT: v_cmp_class_f32_e32 vcc, v0, v3
; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
%x = call nnan float @llvm.sqrt.f32(float %x.arg)
%y.even = shl i32 %y.i, 31
%y.even.as.f32 = bitcast i32 %y.even to float
%copysign = call float @llvm.copysign.f32(float %x, float %y.even.as.f32)
ret float %copysign
}
define float @test_copysign_pow_fast_f32__integral_y(float %x, i32 %y.i) {
; GFX9-LABEL: test_copysign_pow_fast_f32__integral_y:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0x800000
; GFX9-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
; GFX9-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
; GFX9-NEXT: v_lshlrev_b32_e32 v3, 5, v3
; GFX9-NEXT: v_ldexp_f32 v3, |v0|, v3
; GFX9-NEXT: v_log_f32_e32 v3, v3
; GFX9-NEXT: v_cvt_f32_i32_e32 v1, v1
; GFX9-NEXT: v_mov_b32_e32 v2, 0x42000000
; GFX9-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc
; GFX9-NEXT: v_sub_f32_e32 v2, v3, v2
; GFX9-NEXT: v_mul_f32_e32 v3, v2, v1
; GFX9-NEXT: s_mov_b32 s4, 0xc2fc0000
; GFX9-NEXT: v_mov_b32_e32 v4, 0x42800000
; GFX9-NEXT: v_cmp_gt_f32_e32 vcc, s4, v3
; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v4, vcc
; GFX9-NEXT: v_fma_f32 v2, v2, v1, v3
; GFX9-NEXT: v_cvt_i32_f32_e32 v1, v1
; GFX9-NEXT: v_exp_f32_e32 v2, v2
; GFX9-NEXT: v_not_b32_e32 v3, 63
; GFX9-NEXT: v_cndmask_b32_e32 v3, 0, v3, vcc
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 31, v1
; GFX9-NEXT: v_ldexp_f32 v2, v2, v3
; GFX9-NEXT: v_and_b32_e32 v0, v1, v0
; GFX9-NEXT: s_brev_b32 s4, -2
; GFX9-NEXT: v_bfi_b32 v0, s4, v2, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
%y = sitofp i32 %y.i to float
%y.fptosi = fptosi float %y to i32
%fabs = call fast float @llvm.fabs.f32(float %x)
%log2 = call fast float @llvm.log2.f32(float %fabs)
%pownI2F = sitofp i32 %y.i to float
%ylogx = fmul fast float %log2, %pownI2F
%exp2 = call fast float @llvm.exp2.f32(float %ylogx)
%yeven = shl i32 %y.fptosi, 31
%x.i32 = bitcast float %x to i32
%pow_sign = and i32 %yeven, %x.i32
%pow_sign.f32 = bitcast i32 %pow_sign to float
%pow_sign1 = call fast float @llvm.copysign.f32(float %exp2, float %pow_sign.f32)
ret float %pow_sign1
}
attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
|
