1 /* Function cos vectorized with AVX-512, KNL and SKX versions.
2 Copyright (C) 2014-2016 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <http://www.gnu.org/licenses/>. */
20 #include "svml_d_trig_data.h"
21 #include "svml_d_wrapper_impl.h"
24 ENTRY (_ZGVeN8v_cos_knl)
25 #ifndef HAVE_AVX512_ASM_SUPPORT
26 WRAPPER_IMPL_AVX512 _ZGVdN4v_cos
29 ALGORITHM DESCRIPTION:
31 ( low accuracy ( < 4ulp ) or enhanced performance
32 ( half of correct mantissa ) implementation )
34 Argument representation:
35 arg + Pi/2 = (N*Pi + R)
38 cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R)
39 sin(R) is approximated by corresponding polynomial
42 cfi_adjust_cfa_offset (8)
43 cfi_rel_offset (%rbp, 0)
45 cfi_def_cfa_register (%rbp)
48 movq __svml_d_trig_data@GOTPCREL(%rip), %rax
53 /* Check for large arguments path */
57 ARGUMENT RANGE REDUCTION:
58 Add Pi/2 to argument: X' = X+Pi/2
60 vaddpd __dHalfPI(%rax), %zmm0, %zmm5
61 vmovups __dInvPI(%rax), %zmm3
63 /* Get absolute argument value: X' = |X'| */
64 vpandq __dAbsMask(%rax), %zmm5, %zmm1
66 /* Y = X'*InvPi + RS : right shifter add */
67 vfmadd213pd __dRShifter(%rax), %zmm3, %zmm5
68 vmovups __dPI1_FMA(%rax), %zmm6
70 /* N = Y - RS : right shifter sub */
71 vsubpd __dRShifter(%rax), %zmm5, %zmm4
73 /* SignRes = Y<<63 : shift LSB to MSB place for result sign */
74 vpsllq $63, %zmm5, %zmm12
75 vmovups __dC7(%rax), %zmm8
78 vsubpd __dOneHalf(%rax), %zmm4, %zmm10
79 vcmppd $22, __dRangeVal(%rax), %zmm1, %k1
80 vpbroadcastq %rcx, %zmm2{%k1}{z}
81 vfnmadd231pd %zmm10, %zmm6, %zmm7
82 vptestmq %zmm2, %zmm2, %k0
85 vfnmadd231pd __dPI2_FMA(%rax), %zmm10, %zmm7
90 vfnmadd132pd __dPI3_FMA(%rax), %zmm7, %zmm10
93 POLYNOMIAL APPROXIMATION:
96 vmulpd %zmm10, %zmm10, %zmm9
97 vfmadd213pd __dC6(%rax), %zmm9, %zmm8
98 vfmadd213pd __dC5(%rax), %zmm9, %zmm8
99 vfmadd213pd __dC4(%rax), %zmm9, %zmm8
101 /* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */
102 vfmadd213pd __dC3(%rax), %zmm9, %zmm8
104 /* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */
105 vfmadd213pd __dC2(%rax), %zmm9, %zmm8
106 vfmadd213pd __dC1(%rax), %zmm9, %zmm8
107 vmulpd %zmm9, %zmm8, %zmm11
108 vfmadd213pd %zmm10, %zmm10, %zmm11
112 Final sign setting: Res = Poly^SignRes
114 vpxorq %zmm12, %zmm11, %zmm1
122 cfi_def_cfa_register (%rsp)
124 cfi_adjust_cfa_offset (-8)
130 vmovups %zmm0, 1152(%rsp)
131 vmovups %zmm1, 1216(%rsp)
135 kmovw %k4, 1048(%rsp)
137 kmovw %k5, 1040(%rsp)
138 kmovw %k6, 1032(%rsp)
139 kmovw %k7, 1024(%rsp)
140 vmovups %zmm16, 960(%rsp)
141 vmovups %zmm17, 896(%rsp)
142 vmovups %zmm18, 832(%rsp)
143 vmovups %zmm19, 768(%rsp)
144 vmovups %zmm20, 704(%rsp)
145 vmovups %zmm21, 640(%rsp)
146 vmovups %zmm22, 576(%rsp)
147 vmovups %zmm23, 512(%rsp)
148 vmovups %zmm24, 448(%rsp)
149 vmovups %zmm25, 384(%rsp)
150 vmovups %zmm26, 320(%rsp)
151 vmovups %zmm27, 256(%rsp)
152 vmovups %zmm28, 192(%rsp)
153 vmovups %zmm29, 128(%rsp)
154 vmovups %zmm30, 64(%rsp)
155 vmovups %zmm31, (%rsp)
156 movq %rsi, 1064(%rsp)
157 movq %rdi, 1056(%rsp)
158 movq %r12, 1096(%rsp)
159 cfi_offset_rel_rsp (12, 1096)
161 movq %r13, 1088(%rsp)
162 cfi_offset_rel_rsp (13, 1088)
164 movq %r14, 1080(%rsp)
165 cfi_offset_rel_rsp (14, 1080)
167 movq %r15, 1072(%rsp)
168 cfi_offset_rel_rsp (15, 1072)
186 kmovw 1048(%rsp), %k4
187 movq 1064(%rsp), %rsi
188 kmovw 1040(%rsp), %k5
189 movq 1056(%rsp), %rdi
190 kmovw 1032(%rsp), %k6
191 movq 1096(%rsp), %r12
193 movq 1088(%rsp), %r13
195 kmovw 1024(%rsp), %k7
196 vmovups 960(%rsp), %zmm16
197 vmovups 896(%rsp), %zmm17
198 vmovups 832(%rsp), %zmm18
199 vmovups 768(%rsp), %zmm19
200 vmovups 704(%rsp), %zmm20
201 vmovups 640(%rsp), %zmm21
202 vmovups 576(%rsp), %zmm22
203 vmovups 512(%rsp), %zmm23
204 vmovups 448(%rsp), %zmm24
205 vmovups 384(%rsp), %zmm25
206 vmovups 320(%rsp), %zmm26
207 vmovups 256(%rsp), %zmm27
208 vmovups 192(%rsp), %zmm28
209 vmovups 128(%rsp), %zmm29
210 vmovups 64(%rsp), %zmm30
211 vmovups (%rsp), %zmm31
212 movq 1080(%rsp), %r14
214 movq 1072(%rsp), %r15
216 vmovups 1216(%rsp), %zmm1
223 vmovsd 1160(%rsp,%r15), %xmm0
225 vmovsd %xmm0, 1224(%rsp,%r15)
231 vmovsd 1152(%rsp,%r15), %xmm0
233 vmovsd %xmm0, 1216(%rsp,%r15)
236 END (_ZGVeN8v_cos_knl)
238 ENTRY (_ZGVeN8v_cos_skx)
239 #ifndef HAVE_AVX512_ASM_SUPPORT
240 WRAPPER_IMPL_AVX512 _ZGVdN4v_cos
243 ALGORITHM DESCRIPTION:
245 ( low accuracy ( < 4ulp ) or enhanced performance
246 ( half of correct mantissa ) implementation )
248 Argument representation:
249 arg + Pi/2 = (N*Pi + R)
252 cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R)
253 sin(R) is approximated by corresponding polynomial
256 cfi_adjust_cfa_offset (8)
257 cfi_rel_offset (%rbp, 0)
259 cfi_def_cfa_register (%rbp)
262 movq __svml_d_trig_data@GOTPCREL(%rip), %rax
267 /* Check for large arguments path */
268 vpbroadcastq .L_2il0floatpacket.16(%rip), %zmm2
271 ARGUMENT RANGE REDUCTION:
272 Add Pi/2 to argument: X' = X+Pi/2
274 vaddpd __dHalfPI(%rax), %zmm0, %zmm6
275 vmovups __dInvPI(%rax), %zmm3
276 vmovups __dRShifter(%rax), %zmm4
277 vmovups __dPI1_FMA(%rax), %zmm7
278 vmovups __dC7(%rax), %zmm9
280 /* Get absolute argument value: X' = |X'| */
281 vandpd __dAbsMask(%rax), %zmm6, %zmm1
283 /* Y = X'*InvPi + RS : right shifter add */
284 vfmadd213pd %zmm4, %zmm3, %zmm6
285 vcmppd $18, __dRangeVal(%rax), %zmm1, %k1
287 /* SignRes = Y<<63 : shift LSB to MSB place for result sign */
288 vpsllq $63, %zmm6, %zmm13
290 /* N = Y - RS : right shifter sub */
291 vsubpd %zmm4, %zmm6, %zmm5
294 vsubpd __dOneHalf(%rax), %zmm5, %zmm11
295 vfnmadd231pd %zmm11, %zmm7, %zmm8
298 vfnmadd231pd __dPI2_FMA(%rax), %zmm11, %zmm8
301 vfnmadd132pd __dPI3_FMA(%rax), %zmm8, %zmm11
304 POLYNOMIAL APPROXIMATION:
307 vmulpd %zmm11, %zmm11, %zmm10
308 vfmadd213pd __dC6(%rax), %zmm10, %zmm9
309 vfmadd213pd __dC5(%rax), %zmm10, %zmm9
310 vfmadd213pd __dC4(%rax), %zmm10, %zmm9
312 /* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */
313 vfmadd213pd __dC3(%rax), %zmm10, %zmm9
315 /* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */
316 vfmadd213pd __dC2(%rax), %zmm10, %zmm9
317 vfmadd213pd __dC1(%rax), %zmm10, %zmm9
318 vmulpd %zmm10, %zmm9, %zmm12
319 vfmadd213pd %zmm11, %zmm11, %zmm12
320 vpandnq %zmm1, %zmm1, %zmm2{%k1}
321 vcmppd $3, %zmm2, %zmm2, %k0
325 Final sign setting: Res = Poly^SignRes
327 vxorpd %zmm13, %zmm12, %zmm1
336 cfi_def_cfa_register (%rsp)
338 cfi_adjust_cfa_offset (-8)
344 vmovups %zmm0, 1152(%rsp)
345 vmovups %zmm1, 1216(%rsp)
350 kmovw %k4, 1048(%rsp)
351 kmovw %k5, 1040(%rsp)
352 kmovw %k6, 1032(%rsp)
353 kmovw %k7, 1024(%rsp)
354 vmovups %zmm16, 960(%rsp)
355 vmovups %zmm17, 896(%rsp)
356 vmovups %zmm18, 832(%rsp)
357 vmovups %zmm19, 768(%rsp)
358 vmovups %zmm20, 704(%rsp)
359 vmovups %zmm21, 640(%rsp)
360 vmovups %zmm22, 576(%rsp)
361 vmovups %zmm23, 512(%rsp)
362 vmovups %zmm24, 448(%rsp)
363 vmovups %zmm25, 384(%rsp)
364 vmovups %zmm26, 320(%rsp)
365 vmovups %zmm27, 256(%rsp)
366 vmovups %zmm28, 192(%rsp)
367 vmovups %zmm29, 128(%rsp)
368 vmovups %zmm30, 64(%rsp)
369 vmovups %zmm31, (%rsp)
370 movq %rsi, 1064(%rsp)
371 movq %rdi, 1056(%rsp)
372 movq %r12, 1096(%rsp)
373 cfi_offset_rel_rsp (12, 1096)
375 movq %r13, 1088(%rsp)
376 cfi_offset_rel_rsp (13, 1088)
378 movq %r14, 1080(%rsp)
379 cfi_offset_rel_rsp (14, 1080)
381 movq %r15, 1072(%rsp)
382 cfi_offset_rel_rsp (15, 1072)
400 kmovw 1048(%rsp), %k4
401 kmovw 1040(%rsp), %k5
402 kmovw 1032(%rsp), %k6
403 kmovw 1024(%rsp), %k7
404 vmovups 960(%rsp), %zmm16
405 vmovups 896(%rsp), %zmm17
406 vmovups 832(%rsp), %zmm18
407 vmovups 768(%rsp), %zmm19
408 vmovups 704(%rsp), %zmm20
409 vmovups 640(%rsp), %zmm21
410 vmovups 576(%rsp), %zmm22
411 vmovups 512(%rsp), %zmm23
412 vmovups 448(%rsp), %zmm24
413 vmovups 384(%rsp), %zmm25
414 vmovups 320(%rsp), %zmm26
415 vmovups 256(%rsp), %zmm27
416 vmovups 192(%rsp), %zmm28
417 vmovups 128(%rsp), %zmm29
418 vmovups 64(%rsp), %zmm30
419 vmovups (%rsp), %zmm31
420 vmovups 1216(%rsp), %zmm1
421 movq 1064(%rsp), %rsi
422 movq 1056(%rsp), %rdi
423 movq 1096(%rsp), %r12
425 movq 1088(%rsp), %r13
427 movq 1080(%rsp), %r14
429 movq 1072(%rsp), %r15
437 vmovsd 1160(%rsp,%r15), %xmm0
439 vmovsd 1160(%rsp,%r15), %xmm0
443 vmovsd %xmm0, 1224(%rsp,%r15)
449 vmovsd 1152(%rsp,%r15), %xmm0
451 vmovsd 1152(%rsp,%r15), %xmm0
455 vmovsd %xmm0, 1216(%rsp,%r15)
458 END (_ZGVeN8v_cos_skx)
460 .section .rodata, "a"
461 .L_2il0floatpacket.16:
462 .long 0xffffffff,0xffffffff
463 .type .L_2il0floatpacket.16,@object