sysdeps/x86_64/fpu/multiarch/svml_s_sinf4_core_sse4.S

   1 /* Function sinf vectorized with SSE4.
   2    Copyright (C) 2014-2019 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 #include <sysdep.h>
  21 #include "svml_s_trig_data.h"
  22
  23         .text
  24 ENTRY(_ZGVbN4v_sinf_sse4)
  25 /*
  26    ALGORITHM DESCRIPTION:
  27
  28    1) Range reduction to [-Pi/2; +Pi/2] interval
  29       a) Grab sign from source argument and save it.
  30       b) Remove sign using AND operation
  31       c) Getting octant Y by 1/Pi multiplication
  32       d) Add "Right Shifter" value
  33       e) Treat obtained value as integer for destination sign setting.
  34          Shift first bit of this value to the last (sign) position
  35       f) Change destination sign if source sign is negative
  36          using XOR operation.
  37       g) Subtract "Right Shifter" value
  38       h) Subtract Y*PI from X argument, where PI divided to 4 parts:
  39          X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4;
  40    2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval)
  41       a) Calculate X^2 = X * X
  42       b) Calculate polynomial:
  43          R = X + X * X^2 * (A3 + x^2 * (A5 + ......
  44    3) Destination sign setting
  45       a) Set shifted destination sign using XOR operation:
  46          R = XOR( R, S );
  47  */
  48         pushq     %rbp
  49         cfi_adjust_cfa_offset (8)
  50         cfi_rel_offset (%rbp, 0)
  51         movq      %rsp, %rbp
  52         cfi_def_cfa_register (%rbp)
  53         andq      $-64, %rsp
  54         subq      $320, %rsp
  55         movaps    %xmm0, %xmm5
  56         movq      __svml_s_trig_data@GOTPCREL(%rip), %rax
  57         movups    __sAbsMask(%rax), %xmm2
  58
  59 /* b) Remove sign using AND operation */
  60         movaps    %xmm2, %xmm4
  61
  62 /*
  63   f) Change destination sign if source sign is negative
  64   using XOR operation.
  65  */
  66         andnps    %xmm5, %xmm2
  67         movups    __sInvPI(%rax), %xmm1
  68         andps     %xmm5, %xmm4
  69
  70 /* c) Getting octant Y by 1/Pi multiplication
  71    d) Add "Right Shifter" value  */
  72         mulps     %xmm4, %xmm1
  73
  74 /* h) Subtract Y*PI from X argument, where PI divided to 4 parts:
  75    X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4  */
  76         movaps    %xmm4, %xmm0
  77
  78 /* Check for large and special values */
  79         cmpnleps  __sRangeReductionVal(%rax), %xmm4
  80         movups    __sRShifter(%rax), %xmm6
  81         movups    __sPI1(%rax), %xmm7
  82         addps     %xmm6, %xmm1
  83         movmskps  %xmm4, %ecx
  84
  85 /* e) Treat obtained value as integer for destination sign setting.
  86    Shift first bit of this value to the last (sign) position  */
  87         movaps    %xmm1, %xmm3
  88
  89 /* g) Subtract "Right Shifter" value */
  90         subps     %xmm6, %xmm1
  91         mulps     %xmm1, %xmm7
  92         pslld     $31, %xmm3
  93         movups    __sPI2(%rax), %xmm6
  94         subps     %xmm7, %xmm0
  95         mulps     %xmm1, %xmm6
  96         movups    __sPI3(%rax), %xmm7
  97         subps     %xmm6, %xmm0
  98         mulps     %xmm1, %xmm7
  99         movups    __sPI4(%rax), %xmm6
 100         subps     %xmm7, %xmm0
 101         mulps     %xmm6, %xmm1
 102         subps     %xmm1, %xmm0
 103
 104 /* 2) Polynomial (minimax for sin within [-Pi/2; +Pi/2] interval)
 105    a) Calculate X^2 = X * X
 106    b) Calculate polynomial:
 107    R = X + X * X^2 * (A3 + x^2 * (A5 + ...... */
 108         movaps    %xmm0, %xmm1
 109         mulps     %xmm0, %xmm1
 110         xorps     %xmm3, %xmm0
 111         movups    __sA9(%rax), %xmm3
 112         mulps     %xmm1, %xmm3
 113         addps     __sA7(%rax), %xmm3
 114         mulps     %xmm1, %xmm3
 115         addps     __sA5(%rax), %xmm3
 116         mulps     %xmm1, %xmm3
 117         addps     __sA3(%rax), %xmm3
 118         mulps     %xmm3, %xmm1
 119         mulps     %xmm0, %xmm1
 120         addps     %xmm1, %xmm0
 121
 122 /* 3) Destination sign setting
 123    a) Set shifted destination sign using XOR operation:
 124    R = XOR( R, S ); */
 125         xorps     %xmm2, %xmm0
 126         testl     %ecx, %ecx
 127         jne       .LBL_1_3
 128
 129 .LBL_1_2:
 130         cfi_remember_state
 131         movq      %rbp, %rsp
 132         cfi_def_cfa_register (%rsp)
 133         popq      %rbp
 134         cfi_adjust_cfa_offset (-8)
 135         cfi_restore (%rbp)
 136         ret
 137
 138 .LBL_1_3:
 139         cfi_restore_state
 140         movups    %xmm5, 192(%rsp)
 141         movups    %xmm0, 256(%rsp)
 142         je        .LBL_1_2
 143
 144         xorb      %dl, %dl
 145         xorl      %eax, %eax
 146         movups    %xmm8, 112(%rsp)
 147         movups    %xmm9, 96(%rsp)
 148         movups    %xmm10, 80(%rsp)
 149         movups    %xmm11, 64(%rsp)
 150         movups    %xmm12, 48(%rsp)
 151         movups    %xmm13, 32(%rsp)
 152         movups    %xmm14, 16(%rsp)
 153         movups    %xmm15, (%rsp)
 154         movq      %rsi, 136(%rsp)
 155         movq      %rdi, 128(%rsp)
 156         movq      %r12, 168(%rsp)
 157         cfi_offset_rel_rsp (12, 168)
 158         movb      %dl, %r12b
 159         movq      %r13, 160(%rsp)
 160         cfi_offset_rel_rsp (13, 160)
 161         movl      %ecx, %r13d
 162         movq      %r14, 152(%rsp)
 163         cfi_offset_rel_rsp (14, 152)
 164         movl      %eax, %r14d
 165         movq      %r15, 144(%rsp)
 166         cfi_offset_rel_rsp (15, 144)
 167         cfi_remember_state
 168
 169 .LBL_1_6:
 170         btl       %r14d, %r13d
 171         jc        .LBL_1_12
 172
 173 .LBL_1_7:
 174         lea       1(%r14), %esi
 175         btl       %esi, %r13d
 176         jc        .LBL_1_10
 177
 178 .LBL_1_8:
 179         incb      %r12b
 180         addl      $2, %r14d
 181         cmpb      $16, %r12b
 182         jb        .LBL_1_6
 183
 184         movups    112(%rsp), %xmm8
 185         movups    96(%rsp), %xmm9
 186         movups    80(%rsp), %xmm10
 187         movups    64(%rsp), %xmm11
 188         movups    48(%rsp), %xmm12
 189         movups    32(%rsp), %xmm13
 190         movups    16(%rsp), %xmm14
 191         movups    (%rsp), %xmm15
 192         movq      136(%rsp), %rsi
 193         movq      128(%rsp), %rdi
 194         movq      168(%rsp), %r12
 195         cfi_restore (%r12)
 196         movq      160(%rsp), %r13
 197         cfi_restore (%r13)
 198         movq      152(%rsp), %r14
 199         cfi_restore (%r14)
 200         movq      144(%rsp), %r15
 201         cfi_restore (%r15)
 202         movups    256(%rsp), %xmm0
 203         jmp       .LBL_1_2
 204
 205 .LBL_1_10:
 206         cfi_restore_state
 207         movzbl    %r12b, %r15d
 208         movss     196(%rsp,%r15,8), %xmm0
 209
 210         call      JUMPTARGET(sinf)
 211
 212         movss     %xmm0, 260(%rsp,%r15,8)
 213         jmp       .LBL_1_8
 214
 215 .LBL_1_12:
 216         movzbl    %r12b, %r15d
 217         movss     192(%rsp,%r15,8), %xmm0
 218
 219         call      JUMPTARGET(sinf)
 220
 221         movss     %xmm0, 256(%rsp,%r15,8)
 222         jmp       .LBL_1_7
 223
 224 END(_ZGVbN4v_sinf_sse4)