newlib/libm/machine/amdgcn/v64df_asine.c

   1 /*
   2  * Copyright 2023 Siemens
   3  *
   4  * The authors hereby grant permission to use, copy, modify, distribute,
   5  * and license this software and its documentation for any purpose, provided
   6  * that existing copyright notices are retained in all copies and that this
   7  * notice is included verbatim in any distributions.  No written agreement,
   8  * license, or royalty fee is required for any of the authorized uses.
   9  * Modifications to this software may be copyrighted by their authors
  10  * and need not follow the licensing terms described here, provided that
  11  * the new terms are clearly indicated on the first page of each file where
  12  * they apply.
  13  */
  14
  15 /*
  16  * Copyright (c) 1994-2009  Red Hat, Inc. All rights reserved.
  17  *
  18  * This copyrighted material is made available to anyone wishing to use,
  19  * modify, copy, or redistribute it subject to the terms and conditions
  20  * of the BSD License.   This program is distributed in the hope that
  21  * it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
  22  * including the implied warranties of MERCHANTABILITY or FITNESS FOR
  23  * A PARTICULAR PURPOSE.  A copy of this license is available at
  24  * http://www.opensource.org/licenses. Any Red Hat trademarks that are
  25  * incorporated in the source code or documentation are not subject to
  26  * the BSD License and may only be used or replicated with the express
  27  * permission of Red Hat, Inc.
  28  */
  29
  30 /******************************************************************
  31  * The following routines are coded directly from the algorithms
  32  * and coefficients given in "Software Manual for the Elementary
  33  * Functions" by William J. Cody, Jr. and William Waite, Prentice
  34  * Hall, 1980.
  35  ******************************************************************/
  36
  37 /* Based on newlib/libm/mathfp/s_asine.c in Newlib.  */
  38
  39 #include "amdgcnmach.h"
  40
  41 v64si v64df_numtest (v64df);
  42 v64df v64df_sqrt_aux (v64df, v64di);
  43
  44 static const double p[] = { -0.27368494524164255994e+2,
  45                              0.57208227877891731407e+2,
  46                             -0.39688862997404877339e+2,
  47                              0.10152522233806463645e+2,
  48                             -0.69674573447350646411 };
  49 static const double q[] = { -0.16421096714498560795e+3,
  50                              0.41714430248260412556e+3,
  51                             -0.38186303361750149284e+3,
  52                              0.15095270841030604719e+3,
  53                             -0.23823859153670238830e+2 };
  54 static const double a[] = { 0.0, 0.78539816339744830962 };
  55 static const double b[] = { 1.57079632679489661923, 0.78539816339744830962 };
  56
  57 #if defined (__has_builtin) && __has_builtin (__builtin_gcn_fabsv)
  58
  59 DEF_VD_MATH_FUNC (v64df, asine, v64df x, int acosine)
  60 {
  61   FUNCTION_INIT (v64df);
  62
  63   v64si branch = VECTOR_INIT (0);
  64
  65   /* Check for special values. */
  66   v64si i = v64df_numtest (x);
  67   VECTOR_IF ((i == NAN) | (i == INF), cond)
  68     errno = EDOM;
  69     VECTOR_RETURN (VECTOR_MERGE (x, VECTOR_INIT (z_infinity.d),
  70                                  i == NAN),
  71                    cond);
  72   VECTOR_ENDIF
  73
  74   v64df y = __builtin_gcn_fabsv (x);
  75   v64df g, res;
  76
  77   VECTOR_IF (y > 0.5, cond)
  78     VECTOR_COND_MOVE (i, VECTOR_INIT (1 - acosine), cond);
  79
  80     /* Check for range error. */
  81     VECTOR_IF2 (y > 1.0, cond2, cond)
  82       errno = ERANGE;
  83       VECTOR_RETURN (VECTOR_INIT (z_notanum.d), cond2);
  84     VECTOR_ENDIF
  85
  86     VECTOR_COND_MOVE (g, (1.0 - y) / 2.0, cond);
  87     VECTOR_COND_MOVE (y, -2.0 * v64df_sqrt_aux (g, __mask), cond);
  88     VECTOR_COND_MOVE (branch, VECTOR_INIT (-1), cond);
  89   VECTOR_ELSE (cond)
  90     VECTOR_COND_MOVE (i, VECTOR_INIT (acosine), cond);
  91     VECTOR_IF2 (y < z_rooteps, cond2, cond)
  92       VECTOR_COND_MOVE (res, y, cond2);
  93     VECTOR_ELSE2 (cond2, cond)
  94             VECTOR_COND_MOVE (g, y * y, cond2);
  95     VECTOR_ENDIF
  96   VECTOR_ENDIF
  97
  98   VECTOR_IF ((y >= z_rooteps) | __builtin_convertvector(branch, v64di), cond)
  99     {
 100       /* Calculate the Taylor series. */
 101       v64df P = ((((p[4] * g + p[3]) * g + p[2]) * g + p[1]) * g + p[0]) * g;
 102       v64df Q = ((((g + q[4]) * g + q[3]) * g + q[2]) * g + q[1]) * g + q[0];
 103       v64df R = P / Q;
 104
 105       VECTOR_COND_MOVE (res, y + y * R, cond);
 106     }
 107   VECTOR_ENDIF
 108
 109   v64df a_i = VECTOR_MERGE (VECTOR_INIT (a[1]), VECTOR_INIT (a[0]), i != 0);
 110
 111   /* Calculate asine or acose. */
 112   if (acosine == 0)
 113     {
 114       VECTOR_COND_MOVE (res, (a_i + res) + a_i, NO_COND);
 115       VECTOR_IF (x < 0.0, cond)
 116         VECTOR_COND_MOVE (res, -res, cond);
 117       VECTOR_ENDIF
 118     }
 119   else
 120     {
 121       v64df b_i = VECTOR_MERGE (VECTOR_INIT(b[1]), VECTOR_INIT(b[0]), i != 0);
 122
 123       VECTOR_IF (x < 0.0, cond)
 124         VECTOR_COND_MOVE (res, (b_i + res) + b_i, cond);
 125       VECTOR_ELSE (cond)
 126         VECTOR_COND_MOVE (res, (a_i - res) + a_i, cond);
 127       VECTOR_ENDIF
 128     }
 129
 130   VECTOR_RETURN (res, NO_COND);
 131
 132   FUNCTION_RETURN;
 133 }
 134
 135 #endif