newlib/libm/machine/amdgcn/v64sf_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/sf_asine.c in Newlib.  */
  38
  39 #include "amdgcnmach.h"
  40
  41 v64si v64sf_numtestf (v64sf);
  42 v64sf v64sf_sqrtf (v64sf);
  43
  44 static const float p[] = { 0.933935835, -0.504400557 };
  45 static const float q[] = { 0.560363004e+1, -0.554846723e+1 };
  46 static const float a[] = { 0.0, 0.785398163 };
  47 static const float b[] = { 1.570796326, 0.785398163 };
  48
  49 #if defined (__has_builtin) && __has_builtin (__builtin_gcn_fabsvf)
  50
  51 DEF_VS_MATH_FUNC (v64sf, asinef, v64sf x, int acosine)
  52 {
  53   FUNCTION_INIT (v64sf);
  54
  55   v64si branch = VECTOR_INIT (0);
  56
  57   /* Check for special values. */
  58   v64si i = v64sf_numtestf (x);
  59   VECTOR_IF ((i == NAN) | (i == INF), cond)
  60     errno = EDOM;
  61     VECTOR_RETURN (VECTOR_MERGE (x, VECTOR_INIT (z_infinity_f.f),
  62                                  i == NAN),
  63                    cond);
  64   VECTOR_ENDIF
  65
  66   v64sf y = __builtin_gcn_fabsvf (x);
  67   v64sf g, res;
  68
  69   VECTOR_IF (y > 0.5f, cond)
  70     VECTOR_COND_MOVE (i, VECTOR_INIT (1 - acosine), cond);
  71
  72     /* Check for range error. */
  73     VECTOR_IF2 (y > 1.0f, cond2, cond)
  74       errno = ERANGE;
  75       VECTOR_RETURN (VECTOR_INIT (z_notanum_f.f), cond2);
  76     VECTOR_ENDIF
  77
  78     VECTOR_COND_MOVE (g, (1.0f - y) / 2.0f, cond);
  79     VECTOR_COND_MOVE (y, -2.0f * v64sf_sqrtf (g), cond);
  80     VECTOR_COND_MOVE (branch, VECTOR_INIT (-1), cond);
  81   VECTOR_ELSE (cond)
  82     VECTOR_COND_MOVE (i, VECTOR_INIT (acosine), cond);
  83     VECTOR_IF2 (y < z_rooteps_f, cond2, cond)
  84       VECTOR_COND_MOVE (res, y, cond2);
  85     VECTOR_ELSE2 (cond2, cond)
  86             VECTOR_COND_MOVE (g, y * y, cond2);
  87     VECTOR_ENDIF
  88   VECTOR_ENDIF
  89
  90   VECTOR_IF ((y >= z_rooteps_f) | branch, cond)
  91     {
  92       /* Calculate the Taylor series. */
  93       v64sf P = (p[1] * g + p[0]) * g;
  94       v64sf Q = (g + q[1]) * g + q[0];
  95       v64sf R = P / Q;
  96
  97       VECTOR_COND_MOVE (res, y + y * R, cond);
  98     }
  99   VECTOR_ENDIF
 100
 101   v64sf a_i = VECTOR_MERGE (VECTOR_INIT (a[1]), VECTOR_INIT (a[0]), i != 0);
 102
 103   /* Calculate asine or acose. */
 104   if (acosine == 0)
 105     {
 106       VECTOR_COND_MOVE (res, (a_i + res) + a_i, NO_COND);
 107       VECTOR_IF (x < 0.0f, cond)
 108         VECTOR_COND_MOVE (res, -res, cond);
 109       VECTOR_ENDIF
 110     }
 111   else
 112     {
 113       v64sf b_i = VECTOR_MERGE (VECTOR_INIT(b[1]), VECTOR_INIT(b[0]), i != 0);
 114
 115       VECTOR_IF (x < 0.0f, cond)
 116         VECTOR_COND_MOVE (res, (b_i + res) + b_i, cond);
 117       VECTOR_ELSE (cond)
 118         VECTOR_COND_MOVE (res, (a_i - res) + a_i, cond);
 119       VECTOR_ENDIF
 120     }
 121
 122   VECTOR_RETURN (res, NO_COND);
 123
 124   FUNCTION_RETURN;
 125 }
 126
 127 #endif