1/* Function cosh vectorized with AVX2.
2 Copyright (C) 2021-2024 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
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.
9
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.
14
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/. */
18
19/*
20 * ALGORITHM DESCRIPTION:
21 *
22 * Compute cosh(x) as (exp(x)+exp(-x))/2,
23 * where exp is calculated as
24 * exp(M*ln2 + ln2*(j/2^k) + r) = 2^M * 2^(j/2^k) * exp(r)
25 *
26 * Special cases:
27 *
28 * cosh(NaN) = quiet NaN, and raise invalid exception
29 * cosh(INF) = that INF
30 * cosh(0) = 1
31 * cosh(x) overflows for big x and returns MAXLOG+log(2)
32 *
33 */
34
35/* Offsets for data table __svml_dcosh_data_internal
36 */
37#define _dbT 0
38#define _dbInvLn2 2080
39#define _dbLn2hi 2112
40#define _dbLn2lo 2144
41#define _dbShifter 2176
42#define _iIndexMask 2208
43#define _dPC2 2240
44#define _dPC3 2272
45#define _dPC4 2304
46#define _iMaxIndex 2336
47#define _lExpMask 2368
48#define _dSign 2400
49#define _iDomainRange 2432
50
51#include <sysdep.h>
52
53 .section .text.avx2, "ax", @progbits
54ENTRY(_ZGVdN4v_cosh_avx2)
55 pushq %rbp
56 cfi_def_cfa_offset(16)
57 movq %rsp, %rbp
58 cfi_def_cfa(6, 16)
59 cfi_offset(6, -16)
60 andq $-32, %rsp
61 subq $96, %rsp
62 lea _dbT+__svml_dcosh_data_internal(%rip), %rax
63 vmovupd _dSign+__svml_dcosh_data_internal(%rip), %ymm8
64 vmovupd _dbShifter+__svml_dcosh_data_internal(%rip), %ymm6
65
66 /*
67 * Load argument
68 * dM = x*2^K/log(2) + RShifter
69 */
70 vmovupd _dbInvLn2+__svml_dcosh_data_internal(%rip), %ymm3
71
72 /*
73 * trick
74 * 256=-iIndex
75 */
76 vmovups _iMaxIndex+__svml_dcosh_data_internal(%rip), %xmm14
77
78 /* dXSign=0x001000000000 */
79 vpsrlq $11, %ymm8, %ymm5
80 vmovapd %ymm0, %ymm7
81
82 /* Abs argument */
83 vandnpd %ymm7, %ymm8, %ymm4
84 vfmadd213pd %ymm6, %ymm4, %ymm3
85
86 /* Index and lookup */
87 vextractf128 $1, %ymm3, %xmm12
88 vshufps $136, %xmm12, %xmm3, %xmm13
89 vpand _iIndexMask+__svml_dcosh_data_internal(%rip), %xmm13, %xmm15
90 vpsubd %xmm15, %xmm14, %xmm0
91
92 /* iDomainRange*=3 */
93 vpslld $3, %xmm0, %xmm2
94 vmovd %xmm2, %r9d
95 vpextrd $2, %xmm2, %r11d
96 movslq %r9d, %r9
97 vpextrd $1, %xmm2, %r10d
98 movslq %r11d, %r11
99 movslq %r10d, %r10
100 vmovsd (%rax, %r9), %xmm12
101
102 /*
103 * Check for overflow\underflow
104 *
105 */
106 vextractf128 $1, %ymm4, %xmm9
107 vmovsd (%rax, %r11), %xmm14
108 vmovhpd (%rax, %r10), %xmm12, %xmm13
109 vshufps $221, %xmm9, %xmm4, %xmm10
110
111 /* iIndex*=3 */
112 vpslld $3, %xmm15, %xmm9
113
114 /*
115 * R
116 * dN = dM - RShifter
117 */
118 vsubpd %ymm6, %ymm3, %ymm15
119 vmovd %xmm9, %ecx
120 vpcmpgtd _iDomainRange+__svml_dcosh_data_internal(%rip), %xmm10, %xmm11
121 vmovupd _dbLn2hi+__svml_dcosh_data_internal(%rip), %ymm6
122
123 /*
124 * G1, G2, G3: dTdif, dTn * 2^N, 2^(-N)
125 * NB: copied from sinh_la - to be optimized!!!!!
126 */
127 vpsllq $44, %ymm3, %ymm3
128 vmovmskps %xmm11, %edx
129
130 /* dR = dX - dN*Log2_hi/2^K */
131 vfnmadd231pd %ymm6, %ymm15, %ymm4
132
133 /* lM now is an EXP(2^N) */
134 vpand _lExpMask+__svml_dcosh_data_internal(%rip), %ymm3, %ymm3
135
136 /* dR = (dX - dN*Log2_hi/2^K) - dN*Log2_lo/2^K */
137 vfnmadd231pd _dbLn2lo+__svml_dcosh_data_internal(%rip), %ymm15, %ymm4
138 movslq %ecx, %rcx
139 vpextrd $2, %xmm9, %edi
140 vpextrd $1, %xmm9, %esi
141 movslq %edi, %rdi
142 vmovsd (%rax, %rcx), %xmm1
143 vpextrd $3, %xmm9, %r8d
144 vpextrd $3, %xmm2, %ecx
145 movslq %esi, %rsi
146 movslq %r8d, %r8
147 movslq %ecx, %rcx
148
149 /* dR2 = dR^2 */
150 vmulpd %ymm4, %ymm4, %ymm0
151 vmovsd (%rax, %rdi), %xmm10
152 vmovhpd (%rax, %rsi), %xmm1, %xmm8
153 vmovhpd (%rax, %r8), %xmm10, %xmm11
154 vmovhpd (%rax, %rcx), %xmm14, %xmm2
155 vinsertf128 $1, %xmm11, %ymm8, %ymm1
156 vinsertf128 $1, %xmm2, %ymm13, %ymm2
157 vpaddq %ymm3, %ymm1, %ymm6
158
159 /* */
160 vpsubq %ymm3, %ymm2, %ymm1
161
162 /*
163 * sinh(r) = r +r*r^2*a3 ....
164 * dSinh_r = r^2*a3
165 */
166 vmulpd _dPC3+__svml_dcosh_data_internal(%rip), %ymm0, %ymm2
167
168 /* lX- = EXP(1/2) */
169 vpsubq %ymm5, %ymm1, %ymm5
170
171 /* dSinh_r = r + r*r^2*a3 */
172 vfmadd213pd %ymm4, %ymm4, %ymm2
173
174 /* poly(r) = dTp + dTn*sinh(dR)+dTp*dR2*(a2 +a4*dR2) */
175 vmovupd _dPC4+__svml_dcosh_data_internal(%rip), %ymm4
176
177 /* dTn = dTn*2^N - dTn*2^-N */
178 vsubpd %ymm5, %ymm6, %ymm1
179
180 /* dTp = dTn*2^N + dTn*2^-N */
181 vaddpd %ymm5, %ymm6, %ymm3
182 vfmadd213pd _dPC2+__svml_dcosh_data_internal(%rip), %ymm0, %ymm4
183 vmulpd %ymm2, %ymm1, %ymm1
184 vmulpd %ymm4, %ymm0, %ymm0
185
186 /* dTn*sinh(dR)+dTp*dR2*(a2 +a4*dR2) */
187 vfmadd213pd %ymm1, %ymm3, %ymm0
188
189 /* _VRES1 = dTp + dTn*sinh(dR)+dTp*dR2*(a2 +a4*dR2) */
190 vaddpd %ymm0, %ymm3, %ymm0
191
192 /* Ret H */
193 testl %edx, %edx
194
195 /* Go to special inputs processing branch */
196 jne L(SPECIAL_VALUES_BRANCH)
197 # LOE rbx r12 r13 r14 r15 edx ymm0 ymm7
198
199 /* Restore registers
200 * and exit the function
201 */
202
203L(EXIT):
204 movq %rbp, %rsp
205 popq %rbp
206 cfi_def_cfa(7, 8)
207 cfi_restore(6)
208 ret
209 cfi_def_cfa(6, 16)
210 cfi_offset(6, -16)
211
212 /* Branch to process
213 * special inputs
214 */
215
216L(SPECIAL_VALUES_BRANCH):
217 vmovupd %ymm7, 32(%rsp)
218 vmovupd %ymm0, 64(%rsp)
219 # LOE rbx r12 r13 r14 r15 edx ymm0
220
221 xorl %eax, %eax
222 # LOE rbx r12 r13 r14 r15 eax edx
223
224 vzeroupper
225 movq %r12, 16(%rsp)
226 /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
227 .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
228 movl %eax, %r12d
229 movq %r13, 8(%rsp)
230 /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
231 .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
232 movl %edx, %r13d
233 movq %r14, (%rsp)
234 /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
235 .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
236 # LOE rbx r15 r12d r13d
237
238 /* Range mask
239 * bits check
240 */
241
242L(RANGEMASK_CHECK):
243 btl %r12d, %r13d
244
245 /* Call scalar math function */
246 jc L(SCALAR_MATH_CALL)
247 # LOE rbx r15 r12d r13d
248
249 /* Special inputs
250 * processing loop
251 */
252
253L(SPECIAL_VALUES_LOOP):
254 incl %r12d
255 cmpl $4, %r12d
256
257 /* Check bits in range mask */
258 jl L(RANGEMASK_CHECK)
259 # LOE rbx r15 r12d r13d
260
261 movq 16(%rsp), %r12
262 cfi_restore(12)
263 movq 8(%rsp), %r13
264 cfi_restore(13)
265 movq (%rsp), %r14
266 cfi_restore(14)
267 vmovupd 64(%rsp), %ymm0
268
269 /* Go to exit */
270 jmp L(EXIT)
271 /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
272 .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
273 /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
274 .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
275 /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
276 .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
277 # LOE rbx r12 r13 r14 r15 ymm0
278
279 /* Scalar math function call
280 * to process special input
281 */
282
283L(SCALAR_MATH_CALL):
284 movl %r12d, %r14d
285 vmovsd 32(%rsp, %r14, 8), %xmm0
286 call cosh@PLT
287 # LOE rbx r14 r15 r12d r13d xmm0
288
289 vmovsd %xmm0, 64(%rsp, %r14, 8)
290
291 /* Process special inputs in loop */
292 jmp L(SPECIAL_VALUES_LOOP)
293 # LOE rbx r15 r12d r13d
294END(_ZGVdN4v_cosh_avx2)
295
296 .section .rodata, "a"
297 .align 32
298
299#ifdef __svml_dcosh_data_internal_typedef
300typedef unsigned int VUINT32;
301typedef struct {
302 __declspec(align(32)) VUINT32 _dbT[(1+(1<<8))][2]; // dTpj ONLY!
303 __declspec(align(32)) VUINT32 _dbInvLn2[4][2];
304 __declspec(align(32)) VUINT32 _dbLn2hi[4][2];
305 __declspec(align(32)) VUINT32 _dbLn2lo[4][2];
306 __declspec(align(32)) VUINT32 _dbShifter[4][2];
307 __declspec(align(32)) VUINT32 _iIndexMask[8][1]; // (1<<K)1-
308 __declspec(align(32)) VUINT32 _dPC2[4][2];
309 __declspec(align(32)) VUINT32 _dPC3[4][2];
310 __declspec(align(32)) VUINT32 _dPC4[4][2];
311 __declspec(align(32)) VUINT32 _iMaxIndex[8][1]; // (1<<K)
312 __declspec(align(32)) VUINT32 _lExpMask[4][2];
313 __declspec(align(32)) VUINT32 _dSign[4][2]; // 0x8000000000000000
314 __declspec(align(32)) VUINT32 _iDomainRange[8][1];
315} __svml_dcosh_data_internal;
316#endif
317__svml_dcosh_data_internal:
318 /* _dbT */
319 .quad 0x3fe0000000000000, 0x3fe00b1afa5abcbf, 0x3fe0163da9fb3335, 0x3fe02168143b0281
320 .quad 0x3fe02c9a3e778061, 0x3fe037d42e11bbcc, 0x3fe04315e86e7f85, 0x3fe04e5f72f654b1
321 .quad 0x3fe059b0d3158574, 0x3fe0650a0e3c1f89, 0x3fe0706b29ddf6de, 0x3fe07bd42b72a836
322 .quad 0x3fe0874518759bc8, 0x3fe092bdf66607e0, 0x3fe09e3ecac6f383, 0x3fe0a9c79b1f3919
323 .quad 0x3fe0b5586cf9890f, 0x3fe0c0f145e46c85, 0x3fe0cc922b7247f7, 0x3fe0d83b23395dec
324 .quad 0x3fe0e3ec32d3d1a2, 0x3fe0efa55fdfa9c5, 0x3fe0fb66affed31b, 0x3fe1073028d7233e
325 .quad 0x3fe11301d0125b51, 0x3fe11edbab5e2ab6, 0x3fe12abdc06c31cc, 0x3fe136a814f204ab
326 .quad 0x3fe1429aaea92de0, 0x3fe14e95934f312e, 0x3fe15a98c8a58e51, 0x3fe166a45471c3c2
327 .quad 0x3fe172b83c7d517b, 0x3fe17ed48695bbc0, 0x3fe18af9388c8dea, 0x3fe1972658375d2f
328 .quad 0x3fe1a35beb6fcb75, 0x3fe1af99f8138a1c, 0x3fe1bbe084045cd4, 0x3fe1c82f95281c6b
329 .quad 0x3fe1d4873168b9aa, 0x3fe1e0e75eb44027, 0x3fe1ed5022fcd91d, 0x3fe1f9c18438ce4d
330 .quad 0x3fe2063b88628cd6, 0x3fe212be3578a819, 0x3fe21f49917ddc96, 0x3fe22bdda27912d1
331 .quad 0x3fe2387a6e756238, 0x3fe2451ffb82140a, 0x3fe251ce4fb2a63f, 0x3fe25e85711ece75
332 .quad 0x3fe26b4565e27cdd, 0x3fe2780e341ddf29, 0x3fe284dfe1f56381, 0x3fe291ba7591bb70
333 .quad 0x3fe29e9df51fdee1, 0x3fe2ab8a66d10f13, 0x3fe2b87fd0dad990, 0x3fe2c57e39771b2f
334 .quad 0x3fe2d285a6e4030b, 0x3fe2df961f641589, 0x3fe2ecafa93e2f56, 0x3fe2f9d24abd886b
335 .quad 0x3fe306fe0a31b715, 0x3fe31432edeeb2fd, 0x3fe32170fc4cd831, 0x3fe32eb83ba8ea32
336 .quad 0x3fe33c08b26416ff, 0x3fe3496266e3fa2d, 0x3fe356c55f929ff1, 0x3fe36431a2de883b
337 .quad 0x3fe371a7373aa9cb, 0x3fe37f26231e754a, 0x3fe38cae6d05d866, 0x3fe39a401b7140ef
338 .quad 0x3fe3a7db34e59ff7, 0x3fe3b57fbfec6cf4, 0x3fe3c32dc313a8e5, 0x3fe3d0e544ede173
339 .quad 0x3fe3dea64c123422, 0x3fe3ec70df1c5175, 0x3fe3fa4504ac801c, 0x3fe40822c367a024
340 .quad 0x3fe4160a21f72e2a, 0x3fe423fb2709468a, 0x3fe431f5d950a897, 0x3fe43ffa3f84b9d4
341 .quad 0x3fe44e086061892d, 0x3fe45c2042a7d232, 0x3fe46a41ed1d0057, 0x3fe4786d668b3237
342 .quad 0x3fe486a2b5c13cd0, 0x3fe494e1e192aed2, 0x3fe4a32af0d7d3de, 0x3fe4b17dea6db7d7
343 .quad 0x3fe4bfdad5362a27, 0x3fe4ce41b817c114, 0x3fe4dcb299fddd0d, 0x3fe4eb2d81d8abff
344 .quad 0x3fe4f9b2769d2ca7, 0x3fe508417f4531ee, 0x3fe516daa2cf6642, 0x3fe5257de83f4eef
345 .quad 0x3fe5342b569d4f82, 0x3fe542e2f4f6ad27, 0x3fe551a4ca5d920f, 0x3fe56070dde910d2
346 .quad 0x3fe56f4736b527da, 0x3fe57e27dbe2c4cf, 0x3fe58d12d497c7fd, 0x3fe59c0827ff07cc
347 .quad 0x3fe5ab07dd485429, 0x3fe5ba11fba87a03, 0x3fe5c9268a5946b7, 0x3fe5d84590998b93
348 .quad 0x3fe5e76f15ad2148, 0x3fe5f6a320dceb71, 0x3fe605e1b976dc09, 0x3fe6152ae6cdf6f4
349 .quad 0x3fe6247eb03a5585, 0x3fe633dd1d1929fd, 0x3fe6434634ccc320, 0x3fe652b9febc8fb7
350 .quad 0x3fe6623882552225, 0x3fe671c1c70833f6, 0x3fe68155d44ca973, 0x3fe690f4b19e9538
351 .quad 0x3fe6a09e667f3bcd, 0x3fe6b052fa75173e, 0x3fe6c012750bdabf, 0x3fe6cfdcddd47645
352 .quad 0x3fe6dfb23c651a2f, 0x3fe6ef9298593ae5, 0x3fe6ff7df9519484, 0x3fe70f7466f42e87
353 .quad 0x3fe71f75e8ec5f74, 0x3fe72f8286ead08a, 0x3fe73f9a48a58174, 0x3fe74fbd35d7cbfd
354 .quad 0x3fe75feb564267c9, 0x3fe77024b1ab6e09, 0x3fe780694fde5d3f, 0x3fe790b938ac1cf6
355 .quad 0x3fe7a11473eb0187, 0x3fe7b17b0976cfdb, 0x3fe7c1ed0130c132, 0x3fe7d26a62ff86f0
356 .quad 0x3fe7e2f336cf4e62, 0x3fe7f3878491c491, 0x3fe80427543e1a12, 0x3fe814d2add106d9
357 .quad 0x3fe82589994cce13, 0x3fe8364c1eb941f7, 0x3fe8471a4623c7ad, 0x3fe857f4179f5b21
358 .quad 0x3fe868d99b4492ed, 0x3fe879cad931a436, 0x3fe88ac7d98a6699, 0x3fe89bd0a478580f
359 .quad 0x3fe8ace5422aa0db, 0x3fe8be05bad61778, 0x3fe8cf3216b5448c, 0x3fe8e06a5e0866d9
360 .quad 0x3fe8f1ae99157736, 0x3fe902fed0282c8a, 0x3fe9145b0b91ffc6, 0x3fe925c353aa2fe2
361 .quad 0x3fe93737b0cdc5e5, 0x3fe948b82b5f98e5, 0x3fe95a44cbc8520f, 0x3fe96bdd9a7670b3
362 .quad 0x3fe97d829fde4e50, 0x3fe98f33e47a22a2, 0x3fe9a0f170ca07ba, 0x3fe9b2bb4d53fe0d
363 .quad 0x3fe9c49182a3f090, 0x3fe9d674194bb8d5, 0x3fe9e86319e32323, 0x3fe9fa5e8d07f29e
364 .quad 0x3fea0c667b5de565, 0x3fea1e7aed8eb8bb, 0x3fea309bec4a2d33, 0x3fea42c980460ad8
365 .quad 0x3fea5503b23e255d, 0x3fea674a8af46052, 0x3fea799e1330b358, 0x3fea8bfe53c12e59
366 .quad 0x3fea9e6b5579fdbf, 0x3feab0e521356eba, 0x3feac36bbfd3f37a, 0x3fead5ff3a3c2774
367 .quad 0x3feae89f995ad3ad, 0x3feafb4ce622f2ff, 0x3feb0e07298db666, 0x3feb20ce6c9a8952
368 .quad 0x3feb33a2b84f15fb, 0x3feb468415b749b1, 0x3feb59728de5593a, 0x3feb6c6e29f1c52a
369 .quad 0x3feb7f76f2fb5e47, 0x3feb928cf22749e4, 0x3feba5b030a1064a, 0x3febb8e0b79a6f1f
370 .quad 0x3febcc1e904bc1d2, 0x3febdf69c3f3a207, 0x3febf2c25bd71e09, 0x3fec06286141b33d
371 .quad 0x3fec199bdd85529c, 0x3fec2d1cd9fa652c, 0x3fec40ab5fffd07a, 0x3fec544778fafb22
372 .quad 0x3fec67f12e57d14b, 0x3fec7ba88988c933, 0x3fec8f6d9406e7b5, 0x3feca3405751c4db
373 .quad 0x3fecb720dcef9069, 0x3feccb0f2e6d1675, 0x3fecdf0b555dc3fa, 0x3fecf3155b5bab74
374 .quad 0x3fed072d4a07897c, 0x3fed1b532b08c968, 0x3fed2f87080d89f2, 0x3fed43c8eacaa1d6
375 .quad 0x3fed5818dcfba487, 0x3fed6c76e862e6d3, 0x3fed80e316c98398, 0x3fed955d71ff6075
376 .quad 0x3feda9e603db3285, 0x3fedbe7cd63a8315, 0x3fedd321f301b460, 0x3fede7d5641c0658
377 .quad 0x3fedfc97337b9b5f, 0x3fee11676b197d17, 0x3fee264614f5a129, 0x3fee3b333b16ee12
378 .quad 0x3fee502ee78b3ff6, 0x3fee653924676d76, 0x3fee7a51fbc74c83, 0x3fee8f7977cdb740
379 .quad 0x3feea4afa2a490da, 0x3feeb9f4867cca6e, 0x3feecf482d8e67f1, 0x3feee4aaa2188510
380 .quad 0x3feefa1bee615a27, 0x3fef0f9c1cb6412a, 0x3fef252b376bba97, 0x3fef3ac948dd7274
381 .quad 0x3fef50765b6e4540, 0x3fef6632798844f8, 0x3fef7bfdad9cbe14, 0x3fef91d802243c89
382 .quad 0x3fefa7c1819e90d8, 0x3fefbdba3692d514, 0x3fefd3c22b8f71f1, 0x3fefe9d96b2a23d9
383 .quad 0x3ff0000000000000
384 .align 32
385 .quad 0x3ff71547652b82fe, 0x3ff71547652b82fe, 0x3ff71547652b82fe, 0x3ff71547652b82fe /* _dbInvLn2 = 1/log(2) */
386 .align 32
387 .quad 0x3FE62E42FEFC0000, 0x3FE62E42FEFC0000, 0x3FE62E42FEFC0000, 0x3FE62E42FEFC0000 /* _dbLn2hi = log(2) hi */
388 .align 32
389 .quad 0xBDAC610CA86C3899, 0xBDAC610CA86C3899, 0xBDAC610CA86C3899, 0xBDAC610CA86C3899 /* _dbLn2lo = log(2) lo */
390 .align 32
391 .quad 0x42B8000000000000, 0x42B8000000000000, 0x42B8000000000000, 0x42B8000000000000 /* _dbShifter */
392 .align 32
393 .long 0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF /* _iIndexMask */
394 .align 32
395 .quad 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD, 0x3FDFFFFFFFFFFDBD /* _dPC2 */
396 .align 32
397 .quad 0x3FC5555570813E14, 0x3FC5555570813E14, 0x3FC5555570813E14, 0x3FC5555570813E14 /* _dPC3 */
398 .align 32
399 .quad 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299, 0x3FA55555CF16D299 /* _dPC4 */
400 .align 32
401 .long 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100 /* _iMaxIndex */
402 .align 32
403 .quad 0x7ff0000000000000, 0x7ff0000000000000, 0x7ff0000000000000, 0x7ff0000000000000 /* _lExpMask */
404 .align 32
405 .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 /* _dSign */
406 .align 32
407 .long 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99, 0x40861d99 /* _iDomainRange 0x40861d9ac12a3e85 =(1021*2^K-0.5)*log(2)/2^K -needed for quick exp */
408 .align 32
409 .type __svml_dcosh_data_internal, @object
410 .size __svml_dcosh_data_internal, .-__svml_dcosh_data_internal
411

source code of glibc/sysdeps/x86_64/fpu/multiarch/svml_d_cosh4_core_avx2.S