sysdeps/x86_64/fpu/multiarch/svml_s_coshf4_core_sse4.S

   1 /* Function coshf vectorized with SSE4.
   2    Copyright (C) 2021 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_scosh_data_internal
  36  */
  37 #define _sInvLn2                        0
  38 #define _sLn2hi                         16
  39 #define _sLn2lo                         32
  40 #define _sSign                          48
  41 #define _sShifter                       64
  42 #define _iDomainRange                   80
  43 #define _sPC1                           96
  44 #define _sPC2                           112
  45 #define _sPC3                           128
  46 #define _sPC4                           144
  47 #define _sPC5                           160
  48 #define _sPC6                           176
  49 #define _iHalf                          192
  50
  51 #include <sysdep.h>
  52
  53         .text
  54         .section .text.sse4,"ax",@progbits
  55 ENTRY(_ZGVbN4v_coshf_sse4)
  56         subq      $72, %rsp
  57         cfi_def_cfa_offset(80)
  58
  59 /*
  60  *  Implementation
  61  *  Abs argument
  62  */
  63         movups    _sSign+__svml_scosh_data_internal(%rip), %xmm1
  64
  65 /*
  66  *  Load argument
  67  * dM = x/log(2) + RShifter
  68  */
  69         movups    _sInvLn2+__svml_scosh_data_internal(%rip), %xmm9
  70         andnps    %xmm0, %xmm1
  71         mulps     %xmm1, %xmm9
  72
  73 /* Check for overflow\underflow  */
  74         movaps    %xmm1, %xmm3
  75         movups    _sShifter+__svml_scosh_data_internal(%rip), %xmm4
  76         movups    _sLn2hi+__svml_scosh_data_internal(%rip), %xmm5
  77         addps     %xmm4, %xmm9
  78
  79 /*
  80  *  R
  81  * sN = sM - RShifter
  82  */
  83         movaps    %xmm9, %xmm6
  84
  85 /*
  86  *  G1,G2 2^N,2^(-N)
  87  * iM now is an EXP(2^N)
  88  */
  89         pslld     $23, %xmm9
  90         movups    _sLn2lo+__svml_scosh_data_internal(%rip), %xmm7
  91         subps     %xmm4, %xmm6
  92
  93 /* sR = sX - sN*Log2_hi */
  94         mulps     %xmm6, %xmm5
  95
  96 /* sR = (sX - sN*Log2_hi) - sN*Log2_lo */
  97         mulps     %xmm6, %xmm7
  98         movdqu    _iDomainRange+__svml_scosh_data_internal(%rip), %xmm2
  99         pcmpgtd   %xmm2, %xmm3
 100         pcmpeqd   %xmm1, %xmm2
 101
 102 /*
 103  * sinh(r) = r*((a1=1)+r^2*(a3+r^2*(a5+{v1 r^2*a7})))) = r + r*(r^2*(a3+r^2*(a5+r^2*a7))) ....
 104  * sSinh_r = (a3+r^2*a5)
 105  */
 106         movups    _sPC5+__svml_scosh_data_internal(%rip), %xmm10
 107         por       %xmm2, %xmm3
 108
 109 /*
 110  * sinh(X) = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2)
 111  * sOut = (a4 +a6*sR2)
 112  */
 113         movups    _sPC6+__svml_scosh_data_internal(%rip), %xmm11
 114         subps     %xmm5, %xmm1
 115         movmskps  %xmm3, %edx
 116         movdqu    _iHalf+__svml_scosh_data_internal(%rip), %xmm8
 117         subps     %xmm7, %xmm1
 118
 119 /* sR2 = sR^2,shaffled */
 120         movaps    %xmm1, %xmm13
 121         movdqa    %xmm8, %xmm2
 122         mulps     %xmm1, %xmm13
 123         paddd     %xmm9, %xmm2
 124         mulps     %xmm13, %xmm10
 125         psubd     %xmm9, %xmm8
 126         mulps     %xmm13, %xmm11
 127         addps     _sPC3+__svml_scosh_data_internal(%rip), %xmm10
 128         addps     _sPC4+__svml_scosh_data_internal(%rip), %xmm11
 129
 130 /* sSinh_r = r^2*(a3+r^2*a5) */
 131         mulps     %xmm13, %xmm10
 132
 133 /* sOut = a2+sR2*(a4+a6*sR2) */
 134         mulps     %xmm13, %xmm11
 135
 136 /* sSinh_r = r + r*(r^2*(a3+r^2*a5)) */
 137         mulps     %xmm1, %xmm10
 138         addps     _sPC2+__svml_scosh_data_internal(%rip), %xmm11
 139         addps     %xmm10, %xmm1
 140
 141 /* sOut = sR2*(a2+sR2*(a4+a6*sR2) */
 142         mulps     %xmm11, %xmm13
 143
 144 /* sG1 = 2^(N-1)-2^(-N-1) */
 145         movdqa    %xmm2, %xmm12
 146
 147 /* sG2 = 2^(N-1)+2^(-N-1) */
 148         addps     %xmm8, %xmm2
 149         subps     %xmm8, %xmm12
 150
 151 /* sOut = sG2*sR2*(a2+sR2*(a4+a6*sR2) */
 152         mulps     %xmm2, %xmm13
 153
 154 /* sOut = sG1*sinh(dR)+sG2*sR2*(a2+sR2*(a4+a6*sR2) */
 155         mulps     %xmm1, %xmm12
 156         addps     %xmm12, %xmm13
 157
 158 /* sOut = sG2 + sG1*sinh(dR) + sG2*sR2*(a2+sR2*(a4+a6*sR2) */
 159         addps     %xmm13, %xmm2
 160
 161 /*  Ret H  */
 162         testl     %edx, %edx
 163
 164 /* Go to special inputs processing branch */
 165         jne       L(SPECIAL_VALUES_BRANCH)
 166                                 # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm2
 167
 168 /* Restore registers
 169  * and exit the function
 170  */
 171
 172 L(EXIT):
 173         movaps    %xmm2, %xmm0
 174         addq      $72, %rsp
 175         cfi_def_cfa_offset(8)
 176         ret
 177         cfi_def_cfa_offset(80)
 178
 179 /* Branch to process
 180  * special inputs
 181  */
 182
 183 L(SPECIAL_VALUES_BRANCH):
 184         movups    %xmm0, 32(%rsp)
 185         movups    %xmm2, 48(%rsp)
 186                                 # LOE rbx rbp r12 r13 r14 r15 edx
 187
 188         xorl      %eax, %eax
 189         movq      %r12, 16(%rsp)
 190         cfi_offset(12, -64)
 191         movl      %eax, %r12d
 192         movq      %r13, 8(%rsp)
 193         cfi_offset(13, -72)
 194         movl      %edx, %r13d
 195         movq      %r14, (%rsp)
 196         cfi_offset(14, -80)
 197                                 # LOE rbx rbp r15 r12d r13d
 198
 199 /* Range mask
 200  * bits check
 201  */
 202
 203 L(RANGEMASK_CHECK):
 204         btl       %r12d, %r13d
 205
 206 /* Call scalar math function */
 207         jc        L(SCALAR_MATH_CALL)
 208                                 # LOE rbx rbp r15 r12d r13d
 209
 210 /* Special inputs
 211  * processing loop
 212  */
 213
 214 L(SPECIAL_VALUES_LOOP):
 215         incl      %r12d
 216         cmpl      $4, %r12d
 217
 218 /* Check bits in range mask */
 219         jl        L(RANGEMASK_CHECK)
 220                                 # LOE rbx rbp r15 r12d r13d
 221
 222         movq      16(%rsp), %r12
 223         cfi_restore(12)
 224         movq      8(%rsp), %r13
 225         cfi_restore(13)
 226         movq      (%rsp), %r14
 227         cfi_restore(14)
 228         movups    48(%rsp), %xmm2
 229
 230 /* Go to exit */
 231         jmp       L(EXIT)
 232         cfi_offset(12, -64)
 233         cfi_offset(13, -72)
 234         cfi_offset(14, -80)
 235                                 # LOE rbx rbp r12 r13 r14 r15 xmm2
 236
 237 /* Scalar math fucntion call
 238  * to process special input
 239  */
 240
 241 L(SCALAR_MATH_CALL):
 242         movl      %r12d, %r14d
 243         movss     32(%rsp,%r14,4), %xmm0
 244         call      coshf@PLT
 245                                 # LOE rbx rbp r14 r15 r12d r13d xmm0
 246
 247         movss     %xmm0, 48(%rsp,%r14,4)
 248
 249 /* Process special inputs in loop */
 250         jmp       L(SPECIAL_VALUES_LOOP)
 251                                 # LOE rbx rbp r15 r12d r13d
 252 END(_ZGVbN4v_coshf_sse4)
 253
 254         .section .rodata, "a"
 255         .align 16
 256
 257 #ifdef __svml_scosh_data_internal_typedef
 258 typedef unsigned int VUINT32;
 259 typedef struct
 260 {
 261         __declspec(align(16)) VUINT32 _sInvLn2[4][1];
 262         __declspec(align(16)) VUINT32 _sLn2hi[4][1];
 263         __declspec(align(16)) VUINT32 _sLn2lo[4][1];
 264         __declspec(align(16)) VUINT32 _sSign[4][1];
 265         __declspec(align(16)) VUINT32 _sShifter[4][1];
 266         __declspec(align(16)) VUINT32 _iDomainRange[4][1];
 267         __declspec(align(16)) VUINT32 _sPC1[4][1];
 268         __declspec(align(16)) VUINT32 _sPC2[4][1];
 269         __declspec(align(16)) VUINT32 _sPC3[4][1];
 270         __declspec(align(16)) VUINT32 _sPC4[4][1];
 271         __declspec(align(16)) VUINT32 _sPC5[4][1];
 272         __declspec(align(16)) VUINT32 _sPC6[4][1];
 273         __declspec(align(16)) VUINT32 _iHalf[4][1];
 274 } __svml_scosh_data_internal;
 275 #endif
 276 __svml_scosh_data_internal:
 277         .long 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B, 0x3FB8AA3B       /* _sInvLn2  */  //k=0
 278         .align 16
 279         .long 0x3F317000, 0x3F317000, 0x3F317000, 0x3F317000       /* _sLn2hi   */
 280         .align 16
 281         .long 0x3805fdf4, 0x3805fdf4, 0x3805fdf4, 0x3805fdf4       /* _sLn2lo   */
 282         .align 16
 283         .long 0x80000000, 0x80000000, 0x80000000, 0x80000000       /* _sSign    */
 284         .align 16
 285         .long 0x4b400000, 0x4b400000, 0x4b400000, 0x4b400000           /* _sShifter */
 286         .align 16
 287         .long 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E, 0x42AEAC4E           /* _iDomainRange */
 288         .align 16
 289         .long 0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000         /* _sPC1=1  */
 290         .align 16
 291         .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000         /* _sPC2  */
 292         .align 16
 293         .long 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57, 0x3e2aaa57         /* _sPC3  */
 294         .align 16
 295         .long 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72, 0x3d2aaa72         /* _sPC4  */
 296         .align 16
 297         .long 0x3c091461, 0x3c091461, 0x3c091461, 0x3c091461         /* _sPC5  */
 298         .align 16
 299         .long 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3, 0x3ab6a8a3         /* _sPC6  */
 300         // Integer constants
 301         .align 16
 302         .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _iHalf*/
 303         .align 16
 304         .type   __svml_scosh_data_internal,@object
 305         .size   __svml_scosh_data_internal,.-__svml_scosh_data_internal