1 /* Function pow vectorized with AVX-512. KNL and SKX versions.
2 Copyright (C) 2014-2019 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 <https://www.gnu.org/licenses/>. */
20 #include "svml_d_pow_data.h"
21 #include "svml_d_wrapper_impl.h"
23 /* ALGORITHM DESCRIPTION:
25 1) Calculating log2|x|
26 Here we use the following formula.
27 Let |x|=2^k1*X1, where k1 is integer, 1<=X1<2.
29 Rcp1 ~= 1/X1, X2=Rcp1*X1,
30 Rcp2 ~= 1/X2, X3=Rcp2*X2,
31 Rcp3 ~= 1/X3, Rcp3C ~= C/X3.
33 log2|x| = k1 + log2(1/Rcp1) + log2(1/Rcp2) + log2(C/Rcp3C) +
34 log2(X1*Rcp1*Rcp2*Rcp3C/C),
35 where X1*Rcp1*Rcp2*Rcp3C = C*(1+q), q is very small.
37 The values of Rcp1, log2(1/Rcp1), Rcp2, log2(1/Rcp2),
38 Rcp3C, log2(C/Rcp3C) are taken from tables.
39 Values of Rcp1, Rcp2, Rcp3C are such that RcpC=Rcp1*Rcp2*Rcp3C
40 is exactly represented in target precision.
42 log2(X1*Rcp1*Rcp2*Rcp3C/C) = log2(1+q) = ln(1+q)/ln2 =
43 = 1/(ln2)*q - 1/(2ln2)*q^2 + 1/(3ln2)*q^3 - ... =
44 = 1/(C*ln2)*cq - 1/(2*C^2*ln2)*cq^2 + 1/(3*C^3*ln2)*cq^3 - ... =
45 = (1 + a1)*cq + a2*cq^2 + a3*cq^3 + ...,
46 where cq = X1*Rcp1*Rcp2*Rcp3C-C,
47 a1=1/(C*ln(2))-1 is small,
51 We get 3 parts of log2 result: HH+HL+HLL ~= log2|x|.
53 2) Calculation of y*(HH+HL+HLL).
55 Get high PH and medium PL parts of y*log2|x|.
56 Get low PLL part of y*log2|x|.
57 Now we have PH+PL+PLL ~= y*log2|x|.
59 3) Calculation of 2^(PH+PL+PLL).
60 Mathematical idea of computing 2^(PH+PL+PLL) is the following.
61 Let's represent PH+PL+PLL in the form N + j/2^expK + Z,
62 where expK=7 in this implementation, N and j are integers,
63 0<=j<=2^expK-1, |Z|<2^(-expK-1).
64 Hence 2^(PH+PL+PLL) ~= 2^N * 2^(j/2^expK) * 2^Z,
65 where 2^(j/2^expK) is stored in a table, and
66 2^Z ~= 1 + B1*Z + B2*Z^2 ... + B5*Z^5.
68 We compute 2^(PH+PL+PLL) as follows.
69 Break PH into PHH + PHL, where PHH = N + j/2^expK.
71 Exp2Poly = B1*Z + B2*Z^2 ... + B5*Z^5
72 Get 2^(j/2^expK) from table in the form THI+TLO.
73 Now we have 2^(PH+PL+PLL) ~= 2^N * (THI + TLO) * (1 + Exp2Poly).
75 Get significand of 2^(PH+PL+PLL) in the form ResHi+ResLo:
77 ResLo := THI * Exp2Poly + TLO
79 Get exponent ERes of the result:
81 Result := ex(Res) + N. */
84 ENTRY (_ZGVeN8vv_pow_knl)
85 #ifndef HAVE_AVX512DQ_ASM_SUPPORT
86 WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_pow
89 cfi_adjust_cfa_offset (8)
90 cfi_rel_offset (%rbp, 0)
92 cfi_def_cfa_register (%rbp)
95 vpsrlq $32, %zmm0, %zmm13
97 movq __svml_dpow_data@GOTPCREL(%rip), %rax
99 vpmovqd %zmm13, %ymm10
100 vpsrlq $32, %zmm12, %zmm14
103 vpmovqd %zmm14, %ymm15
105 /* x1 = x; Hi(x1) = (Hi(x1)&0x000fffff)|0x3ff00000 */
106 vmovups _dbOne(%rax), %zmm6
108 /* i = (((Hi(x) & 0x000ffe00) + 0x00000200) >> 10); -> i = (b1..b11 + 1) / 2 */
109 vmovaps %zmm10, %zmm5
111 /* k = Hi(x); k = k - 0x3fe7fe00; k = k >> 20 */
112 vpsubd _i3fe7fe00(%rax), %zmm10, %zmm14{%k1}
113 vpandd _iIndexMask(%rax), %zmm10, %zmm5{%k1}
114 vpsrad $20, %zmm14, %zmm14{%k1}
115 vpxord %zmm9, %zmm9, %zmm9
116 vpaddd _HIDELTA(%rax), %zmm10, %zmm3{%k1}
117 vpaddd _iIndexAdd(%rax), %zmm5, %zmm5{%k1}
118 vpxord %zmm7, %zmm7, %zmm7
119 vpaddd _i2p20_2p19(%rax), %zmm14, %zmm14{%k1}
120 vpcmpd $1, _LORANGE(%rax), %zmm3, %k2{%k1}
121 vpsrld $10, %zmm5, %zmm5{%k1}
122 vpandd _ABSMASK(%rax), %zmm15, %zmm2{%k1}
123 vpbroadcastd %ecx, %zmm1{%k2}{z}
125 /* Index for reciprocal table */
126 vpslld $3, %zmm5, %zmm8{%k1}
128 vgatherdpd 11712(%rax,%ymm8), %zmm9{%k2}
129 vpmovzxdq %ymm14, %zmm10
131 /* Index for log2 table */
132 vpslld $4, %zmm5, %zmm13{%k1}
134 vpsllq $32, %zmm10, %zmm3
135 vpxord %zmm8, %zmm8, %zmm8
136 vpcmpd $5, _INF(%rax), %zmm2, %k3{%k1}
137 vpbroadcastd %ecx, %zmm4{%k3}{z}
138 vpternlogq $248, _iMantissaMask(%rax), %zmm0, %zmm6
140 vpternlogq $168, _iffffffff00000000(%rax), %zmm10, %zmm3
142 /* x1Hi=x1; Lo(x1Hi)&=0xf8000000; x1Lo = x1-x1Hi */
143 vpandq _iHighMask(%rax), %zmm6, %zmm2
144 vgatherdpd 19976(%rax,%ymm13), %zmm8{%k2}
145 vpord %zmm4, %zmm1, %zmm11{%k1}
146 vsubpd _db2p20_2p19(%rax), %zmm3, %zmm1
147 vsubpd %zmm2, %zmm6, %zmm5
150 vmulpd %zmm9, %zmm6, %zmm6
151 vgatherdpd 19968(%rax,%ymm13), %zmm7{%k3}
154 vaddpd _LHN(%rax), %zmm6, %zmm4
156 /* E = -r1+__fence(x1Hi*rcp1) */
157 vfmsub213pd %zmm6, %zmm9, %zmm2
160 vaddpd %zmm7, %zmm1, %zmm7
163 vfmadd213pd %zmm2, %zmm9, %zmm5
166 vaddpd %zmm4, %zmm7, %zmm3
168 /* Rl = T-T_Rh; -> -Rh */
169 vsubpd %zmm3, %zmm7, %zmm9
172 vaddpd %zmm9, %zmm4, %zmm6
174 /* T_Rh_Eh = T_Rh + E */
175 vaddpd %zmm5, %zmm3, %zmm9
177 /* HLL = T_Rh - T_Rh_Eh; -> -Eh */
178 vsubpd %zmm9, %zmm3, %zmm2
181 vaddpd %zmm5, %zmm4, %zmm4
184 vaddpd %zmm2, %zmm5, %zmm1
185 vmovups _clv_2(%rax), %zmm5
187 /* HLL = HLL + (((((((a7)*cq+a6)*cq+a5)*cq+a4)*cq+a3)*cq+a2)*cq+a1)*cq */
188 vfmadd213pd _clv_3(%rax), %zmm4, %zmm5
191 vaddpd %zmm6, %zmm1, %zmm7
193 /* 2^(y*(HH+HL+HLL)) starts here:
194 yH = y; Lo(yH)&=0xf8000000
196 vpandq _iHighMask(%rax), %zmm12, %zmm6
199 vsubpd %zmm6, %zmm12, %zmm2
200 vfmadd213pd _clv_4(%rax), %zmm4, %zmm5
203 vaddpd %zmm8, %zmm7, %zmm8
204 vfmadd213pd _clv_5(%rax), %zmm4, %zmm5
205 vfmadd213pd _clv_6(%rax), %zmm4, %zmm5
206 vfmadd213pd _clv_7(%rax), %zmm4, %zmm5
207 vfmadd213pd %zmm8, %zmm4, %zmm5
209 /* T_Rh_Eh_HLLhi = T_Rh_Eh + HLL */
210 vaddpd %zmm5, %zmm9, %zmm13
212 /* HLLhi = T_Rh_Eh_HLLhi - T_Rh_Eh */
213 vsubpd %zmm9, %zmm13, %zmm10
215 /* HLL = HLL - HLLhi */
216 vsubpd %zmm10, %zmm5, %zmm3
218 /* HH = T_Rh_Eh_HLLhi; Lo(HH)&=0xf8000000 */
219 vpandq _iHighMask(%rax), %zmm13, %zmm5
222 vmulpd %zmm5, %zmm6, %zmm1
224 /* HL = T_Rh_Eh_HLLhi-HH */
225 vsubpd %zmm5, %zmm13, %zmm4
226 vpsrlq $32, %zmm1, %zmm14
229 pHH = pH + *(double*)&db2p45_2p44
231 vaddpd _db2p45_2p44(%rax), %zmm1, %zmm10
232 vpmovqd %zmm14, %ymm15
233 vpandd _ABSMASK(%rax), %zmm15, %zmm14{%k1}
234 vpcmpd $5, _DOMAINRANGE(%rax), %zmm14, %k3{%k1}
236 /* T1 = ((double*)exp2_tbl)[ 2*j ] */
237 vpxord %zmm14, %zmm14, %zmm14
238 vpbroadcastd %ecx, %zmm13{%k3}{z}
239 vpord %zmm13, %zmm11, %zmm11{%k1}
240 vptestmd %zmm11, %zmm11, %k0{%k1}
242 /* pL=yL*HL+yH*HL; pL+=yL*HH */
243 vmulpd %zmm4, %zmm2, %zmm11
245 vfmadd213pd %zmm11, %zmm4, %zmm6
247 /* pHH = pHH - *(double*)&db2p45_2p44 */
248 vsubpd _db2p45_2p44(%rax), %zmm10, %zmm11
249 vpmovqd %zmm10, %ymm4
253 _n = _n & 0xffffff80;
255 Hi(_2n) = (0x3ff+_n)<<20; Lo(_2n) = 0; -> 2^n
257 vpslld $13, %zmm4, %zmm7{%k1}
259 /* j = Lo(pHH)&0x0000007f */
260 vpandd _jIndexMask(%rax), %zmm4, %zmm9{%k1}
261 vfmadd213pd %zmm6, %zmm5, %zmm2
264 vsubpd %zmm11, %zmm1, %zmm1
265 vpaddd _iOne(%rax), %zmm7, %zmm7{%k1}
267 /* t=pL+pLL; t+=pHL */
268 vfmadd231pd %zmm12, %zmm3, %zmm2
269 vpslld $4, %zmm9, %zmm9{%k1}
271 vgatherdpd 36416(%rax,%ymm9), %zmm14{%k1}
272 vpmovzxdq %ymm7, %zmm8
273 vaddpd %zmm1, %zmm2, %zmm2
274 vmovups _cev_1(%rax), %zmm1
275 vpsllq $32, %zmm8, %zmm13
276 vpternlogq $168, _ifff0000000000000(%rax), %zmm8, %zmm13
277 vfmadd213pd _cev_2(%rax), %zmm2, %zmm1
278 vmulpd %zmm14, %zmm13, %zmm15
279 vfmadd213pd _cev_3(%rax), %zmm2, %zmm1
280 vmulpd %zmm2, %zmm15, %zmm3
281 vfmadd213pd _cev_4(%rax), %zmm2, %zmm1
282 vfmadd213pd _cev_5(%rax), %zmm2, %zmm1
283 vfmadd213pd %zmm15, %zmm3, %zmm1
291 cfi_def_cfa_register (%rsp)
293 cfi_adjust_cfa_offset (-8)
299 vmovups %zmm0, 1152(%rsp)
300 vmovups %zmm12, 1216(%rsp)
301 vmovups %zmm1, 1280(%rsp)
305 kmovw %k4, 1048(%rsp)
307 kmovw %k5, 1040(%rsp)
308 kmovw %k6, 1032(%rsp)
309 kmovw %k7, 1024(%rsp)
310 vmovups %zmm16, 960(%rsp)
311 vmovups %zmm17, 896(%rsp)
312 vmovups %zmm18, 832(%rsp)
313 vmovups %zmm19, 768(%rsp)
314 vmovups %zmm20, 704(%rsp)
315 vmovups %zmm21, 640(%rsp)
316 vmovups %zmm22, 576(%rsp)
317 vmovups %zmm23, 512(%rsp)
318 vmovups %zmm24, 448(%rsp)
319 vmovups %zmm25, 384(%rsp)
320 vmovups %zmm26, 320(%rsp)
321 vmovups %zmm27, 256(%rsp)
322 vmovups %zmm28, 192(%rsp)
323 vmovups %zmm29, 128(%rsp)
324 vmovups %zmm30, 64(%rsp)
325 vmovups %zmm31, (%rsp)
326 movq %rsi, 1064(%rsp)
327 movq %rdi, 1056(%rsp)
328 movq %r12, 1096(%rsp)
329 cfi_offset_rel_rsp (12, 1096)
331 movq %r13, 1088(%rsp)
332 cfi_offset_rel_rsp (13, 1088)
334 movq %r14, 1080(%rsp)
335 cfi_offset_rel_rsp (14, 1080)
337 movq %r15, 1072(%rsp)
338 cfi_offset_rel_rsp (15, 1072)
356 kmovw 1048(%rsp), %k4
357 movq 1064(%rsp), %rsi
358 kmovw 1040(%rsp), %k5
359 movq 1056(%rsp), %rdi
360 kmovw 1032(%rsp), %k6
361 movq 1096(%rsp), %r12
363 movq 1088(%rsp), %r13
365 kmovw 1024(%rsp), %k7
366 vmovups 960(%rsp), %zmm16
367 vmovups 896(%rsp), %zmm17
368 vmovups 832(%rsp), %zmm18
369 vmovups 768(%rsp), %zmm19
370 vmovups 704(%rsp), %zmm20
371 vmovups 640(%rsp), %zmm21
372 vmovups 576(%rsp), %zmm22
373 vmovups 512(%rsp), %zmm23
374 vmovups 448(%rsp), %zmm24
375 vmovups 384(%rsp), %zmm25
376 vmovups 320(%rsp), %zmm26
377 vmovups 256(%rsp), %zmm27
378 vmovups 192(%rsp), %zmm28
379 vmovups 128(%rsp), %zmm29
380 vmovups 64(%rsp), %zmm30
381 vmovups (%rsp), %zmm31
382 movq 1080(%rsp), %r14
384 movq 1072(%rsp), %r15
386 vmovups 1280(%rsp), %zmm1
393 vmovsd 1160(%rsp,%r15), %xmm0
394 vmovsd 1224(%rsp,%r15), %xmm1
395 call JUMPTARGET(__pow_finite)
396 vmovsd %xmm0, 1288(%rsp,%r15)
402 vmovsd 1152(%rsp,%r15), %xmm0
403 vmovsd 1216(%rsp,%r15), %xmm1
404 call JUMPTARGET(__pow_finite)
405 vmovsd %xmm0, 1280(%rsp,%r15)
409 END (_ZGVeN8vv_pow_knl)
411 ENTRY (_ZGVeN8vv_pow_skx)
412 #ifndef HAVE_AVX512DQ_ASM_SUPPORT
413 WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_pow
416 cfi_adjust_cfa_offset (8)
417 cfi_rel_offset (%rbp, 0)
419 cfi_def_cfa_register (%rbp)
422 vpsrlq $32, %zmm0, %zmm10
426 vpmovqd %zmm10, %ymm7
427 movq __svml_dpow_data@GOTPCREL(%rip), %rax
429 vpsrlq $32, %zmm6, %zmm13
431 /* i = (((Hi(x) & 0x000ffe00) + 0x00000200) >> 10); -> i = (b1..b11 + 1) / 2 */
432 vpand _iIndexMask(%rax), %ymm7, %ymm15
433 vpaddd _HIDELTA(%rax), %ymm7, %ymm2
435 /* k = Hi(x); k = k - 0x3fe7fe00; k = k >> 20 */
436 vpsubd _i3fe7fe00(%rax), %ymm7, %ymm7
437 vmovdqu _ABSMASK(%rax), %ymm4
438 vmovdqu _LORANGE(%rax), %ymm3
440 /* x1 = x; Hi(x1) = (Hi(x1)&0x000fffff)|0x3ff00000 */
441 vmovups _dbOne(%rax), %zmm11
442 vmovdqu _INF(%rax), %ymm5
443 vpaddd _iIndexAdd(%rax), %ymm15, %ymm12
444 vpmovqd %zmm13, %ymm14
445 vpternlogq $248, _iMantissaMask(%rax), %zmm0, %zmm11
446 vpsrld $10, %ymm12, %ymm10
447 vpsrad $20, %ymm7, %ymm13
449 /* Index for reciprocal table */
450 vpslld $3, %ymm10, %ymm8
452 /* Index for log2 table */
453 vpslld $4, %ymm10, %ymm1
454 vpcmpgtd %ymm2, %ymm3, %ymm3
455 vpand %ymm4, %ymm14, %ymm2
456 vpaddd _i2p20_2p19(%rax), %ymm13, %ymm14
457 vpmovzxdq %ymm14, %zmm15
458 vpsllq $32, %zmm15, %zmm7
459 vpternlogq $168, _iffffffff00000000(%rax), %zmm15, %zmm7
460 vsubpd _db2p20_2p19(%rax), %zmm7, %zmm13
461 vpxord %zmm9, %zmm9, %zmm9
462 vgatherdpd 11712(%rax,%ymm8), %zmm9{%k1}
464 /* T1 = ((double*)exp2_tbl)[ 2*j ] */
466 vpxord %zmm12, %zmm12, %zmm12
467 vpxord %zmm8, %zmm8, %zmm8
468 vgatherdpd 19968(%rax,%ymm1), %zmm12{%k2}
469 vgatherdpd 19976(%rax,%ymm1), %zmm8{%k3}
470 vmovups _iHighMask(%rax), %zmm1
472 /* x1Hi=x1; Lo(x1Hi)&=0xf8000000; x1Lo = x1-x1Hi */
473 vandpd %zmm1, %zmm11, %zmm10
474 vsubpd %zmm10, %zmm11, %zmm15
477 vmulpd %zmm9, %zmm11, %zmm11
479 /* E = -r1+__fence(x1Hi*rcp1) */
480 vfmsub213pd %zmm11, %zmm9, %zmm10
483 vaddpd _LHN(%rax), %zmm11, %zmm14
486 vfmadd213pd %zmm10, %zmm9, %zmm15
489 vaddpd %zmm12, %zmm13, %zmm9
492 vaddpd %zmm14, %zmm9, %zmm11
494 /* T_Rh_Eh = T_Rh + E */
495 vaddpd %zmm15, %zmm11, %zmm13
497 /* Rl = T-T_Rh; -> -Rh */
498 vsubpd %zmm11, %zmm9, %zmm12
500 /* HLL = T_Rh - T_Rh_Eh; -> -Eh */
501 vsubpd %zmm13, %zmm11, %zmm9
504 vaddpd %zmm12, %zmm14, %zmm10
507 vaddpd %zmm9, %zmm15, %zmm7
510 vaddpd %zmm10, %zmm7, %zmm12
512 /* 2^(y*(HH+HL+HLL)) starts here:
513 yH = y; Lo(yH)&=0xf8000000
515 vandpd %zmm1, %zmm6, %zmm7
518 vaddpd %zmm8, %zmm12, %zmm12
521 vaddpd %zmm15, %zmm14, %zmm8
522 vmovups _clv_2(%rax), %zmm14
524 /* HLL = HLL + (((((((a7)*cq+a6)*cq+a5)*cq+a4)*cq+a3)*cq+a2)*cq+a1)*cq */
525 vfmadd213pd _clv_3(%rax), %zmm8, %zmm14
526 vfmadd213pd _clv_4(%rax), %zmm8, %zmm14
527 vfmadd213pd _clv_5(%rax), %zmm8, %zmm14
528 vfmadd213pd _clv_6(%rax), %zmm8, %zmm14
529 vfmadd213pd _clv_7(%rax), %zmm8, %zmm14
530 vfmadd213pd %zmm12, %zmm8, %zmm14
533 vsubpd %zmm7, %zmm6, %zmm8
535 /* T_Rh_Eh_HLLhi = T_Rh_Eh + HLL */
536 vaddpd %zmm14, %zmm13, %zmm15
538 /* HH = T_Rh_Eh_HLLhi; Lo(HH)&=0xf8000000 */
539 vandpd %zmm1, %zmm15, %zmm11
541 /* HLLhi = T_Rh_Eh_HLLhi - T_Rh_Eh */
542 vsubpd %zmm13, %zmm15, %zmm13
545 vmulpd %zmm11, %zmm7, %zmm9
547 /* HLL = HLL - HLLhi */
548 vsubpd %zmm13, %zmm14, %zmm12
550 /* HL = T_Rh_Eh_HLLhi-HH */
551 vsubpd %zmm11, %zmm15, %zmm10
552 vpsrlq $32, %zmm9, %zmm1
553 vmovdqu _DOMAINRANGE(%rax), %ymm13
555 vpand %ymm4, %ymm1, %ymm1
556 vpcmpgtd %ymm5, %ymm2, %ymm4
557 vpcmpeqd %ymm5, %ymm2, %ymm5
558 vpternlogd $254, %ymm5, %ymm4, %ymm3
559 vpcmpgtd %ymm13, %ymm1, %ymm2
560 vpcmpeqd %ymm13, %ymm1, %ymm4
561 vpternlogd $254, %ymm4, %ymm2, %ymm3
564 vmovups _db2p45_2p44(%rax), %zmm2
566 /* pHH = pH + *(double*)&db2p45_2p44 */
567 vaddpd %zmm2, %zmm9, %zmm1
570 /* j = Lo(pHH)&0x0000007f */
571 vpand _jIndexMask(%rax), %ymm5, %ymm14
572 vpslld $4, %ymm14, %ymm15
573 vmovmskps %ymm3, %ecx
575 /* pL=yL*HL+yH*HL; pL+=yL*HH */
576 vmulpd %zmm10, %zmm8, %zmm3
577 vfmadd213pd %zmm3, %zmm10, %zmm7
578 vfmadd213pd %zmm7, %zmm11, %zmm8
583 Hi(_2n) = (0x3ff+_n)<<20; Lo(_2n) = 0; -> 2^n
585 vpslld $13, %ymm5, %ymm7
587 /* t=pL+pLL; t+=pHL */
588 vfmadd231pd %zmm6, %zmm12, %zmm8
589 vpaddd _iOne(%rax), %ymm7, %ymm10
590 vpmovzxdq %ymm10, %zmm11
591 vpsllq $32, %zmm11, %zmm3
592 vpternlogq $168, _ifff0000000000000(%rax), %zmm11, %zmm3
594 /* pHH = pHH - *(double*)&db2p45_2p44 */
595 vsubpd %zmm2, %zmm1, %zmm11
596 vmovups _cev_1(%rax), %zmm2
599 vsubpd %zmm11, %zmm9, %zmm9
600 vaddpd %zmm9, %zmm8, %zmm8
601 vfmadd213pd _cev_2(%rax), %zmm8, %zmm2
602 vfmadd213pd _cev_3(%rax), %zmm8, %zmm2
603 vfmadd213pd _cev_4(%rax), %zmm8, %zmm2
604 vfmadd213pd _cev_5(%rax), %zmm8, %zmm2
605 vpxord %zmm4, %zmm4, %zmm4
606 vgatherdpd 36416(%rax,%ymm15), %zmm4{%k1}
607 vmulpd %zmm4, %zmm3, %zmm1
608 vmulpd %zmm8, %zmm1, %zmm12
609 vfmadd213pd %zmm1, %zmm12, %zmm2
617 cfi_def_cfa_register (%rsp)
619 cfi_adjust_cfa_offset (-8)
625 vmovups %zmm0, 1152(%rsp)
626 vmovups %zmm6, 1216(%rsp)
627 vmovups %zmm2, 1280(%rsp)
632 kmovw %k4, 1048(%rsp)
633 kmovw %k5, 1040(%rsp)
634 kmovw %k6, 1032(%rsp)
635 kmovw %k7, 1024(%rsp)
636 vmovups %zmm16, 960(%rsp)
637 vmovups %zmm17, 896(%rsp)
638 vmovups %zmm18, 832(%rsp)
639 vmovups %zmm19, 768(%rsp)
640 vmovups %zmm20, 704(%rsp)
641 vmovups %zmm21, 640(%rsp)
642 vmovups %zmm22, 576(%rsp)
643 vmovups %zmm23, 512(%rsp)
644 vmovups %zmm24, 448(%rsp)
645 vmovups %zmm25, 384(%rsp)
646 vmovups %zmm26, 320(%rsp)
647 vmovups %zmm27, 256(%rsp)
648 vmovups %zmm28, 192(%rsp)
649 vmovups %zmm29, 128(%rsp)
650 vmovups %zmm30, 64(%rsp)
651 vmovups %zmm31, (%rsp)
652 movq %rsi, 1064(%rsp)
653 movq %rdi, 1056(%rsp)
654 movq %r12, 1096(%rsp)
655 cfi_offset_rel_rsp (12, 1096)
657 movq %r13, 1088(%rsp)
658 cfi_offset_rel_rsp (13, 1088)
660 movq %r14, 1080(%rsp)
661 cfi_offset_rel_rsp (14, 1080)
663 movq %r15, 1072(%rsp)
664 cfi_offset_rel_rsp (15, 1072)
682 kmovw 1048(%rsp), %k4
683 kmovw 1040(%rsp), %k5
684 kmovw 1032(%rsp), %k6
685 kmovw 1024(%rsp), %k7
686 vmovups 960(%rsp), %zmm16
687 vmovups 896(%rsp), %zmm17
688 vmovups 832(%rsp), %zmm18
689 vmovups 768(%rsp), %zmm19
690 vmovups 704(%rsp), %zmm20
691 vmovups 640(%rsp), %zmm21
692 vmovups 576(%rsp), %zmm22
693 vmovups 512(%rsp), %zmm23
694 vmovups 448(%rsp), %zmm24
695 vmovups 384(%rsp), %zmm25
696 vmovups 320(%rsp), %zmm26
697 vmovups 256(%rsp), %zmm27
698 vmovups 192(%rsp), %zmm28
699 vmovups 128(%rsp), %zmm29
700 vmovups 64(%rsp), %zmm30
701 vmovups (%rsp), %zmm31
702 vmovups 1280(%rsp), %zmm2
703 movq 1064(%rsp), %rsi
704 movq 1056(%rsp), %rdi
705 movq 1096(%rsp), %r12
707 movq 1088(%rsp), %r13
709 movq 1080(%rsp), %r14
711 movq 1072(%rsp), %r15
719 vmovsd 1224(%rsp,%r15), %xmm1
721 vmovsd 1160(%rsp,%r15), %xmm0
723 call JUMPTARGET(__pow_finite)
725 vmovsd %xmm0, 1288(%rsp,%r15)
731 vmovsd 1216(%rsp,%r15), %xmm1
733 vmovsd 1152(%rsp,%r15), %xmm0
735 call JUMPTARGET(__pow_finite)
737 vmovsd %xmm0, 1280(%rsp,%r15)
741 END (_ZGVeN8vv_pow_skx)