newlib/libm/machine/amdgcn/v64sf_rint.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  * ====================================================
  17  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
  18  *
  19  * Developed at SunPro, a Sun Microsystems, Inc. business.
  20  * Permission to use, copy, modify, and distribute this
  21  * software is freely granted, provided that this notice
  22  * is preserved.
  23  * ====================================================
  24  */
  25
  26 /* Based on newlib/libm/common/sf_rint.c in Newlib.  */
  27
  28 #include "amdgcnmach.h"
  29
  30 static const float TWO23[2] = {
  31   8.3886080000e+06, /* 0x4b000000 */
  32  -8.3886080000e+06, /* 0xcb000000 */
  33 };
  34
  35 DEF_VS_MATH_FUNC (v64sf, rintf, v64sf x)
  36 {
  37   FUNCTION_INIT (v64sf);
  38
  39   v64si i0;
  40   GET_FLOAT_WORD (i0, x, NO_COND);
  41   v64si sx = (i0 >> 31) & 1;
  42   v64sf two23 = VECTOR_MERGE (TWO23[1] + x, TWO23[0] + x, sx != 0);
  43   v64si ix = (i0 & 0x7fffffff);
  44   v64si j0 = (ix >> 23) - 0x7f;
  45   VECTOR_IF (j0 < 23, cond)
  46     VECTOR_RETURN (x, cond & FLT_UWORD_IS_ZERO (ix));
  47     VECTOR_IF2 (j0 < 0, cond2, cond)
  48       v64si i1 = (i0 & 0x07fffff);
  49       VECTOR_COND_MOVE (i0, i0 & 0xfff00000, cond2);
  50       VECTOR_COND_MOVE (i0, i0 | (((i1 | -i1) >> 9) & 0x400000), cond2);
  51       SET_FLOAT_WORD (x, i0, cond2);
  52       v64sf w = two23 + x;
  53       v64sf t = w - two23;
  54       GET_FLOAT_WORD (i0, t, cond2);
  55       SET_FLOAT_WORD (t, (i0&0x7fffffff)|(sx<<31), cond2);
  56       VECTOR_RETURN (t, cond2);
  57     VECTOR_ELSE2 (cond2, cond)
  58       v64si i = (0x007fffff) >> j0;
  59       VECTOR_RETURN (x, cond2 & ((i0 & i) == 0));       /* x is integral */
  60       i >>= 1;
  61       VECTOR_COND_MOVE (i0, (i0 & (~i)) | (0x200000 >> j0),
  62                         cond2 & ((i0 & i) != 0));
  63     VECTOR_ENDIF
  64   VECTOR_ELSE (cond)
  65     VECTOR_RETURN (x + x, cond & ~FLT_UWORD_IS_FINITE (ix));    /* inf or NaN */
  66     VECTOR_RETURN (x, cond); /* x is integral */
  67   VECTOR_ENDIF
  68
  69   SET_FLOAT_WORD (x, i0, NO_COND);
  70   v64sf w = two23 + x;
  71   VECTOR_RETURN (w - two23, NO_COND);
  72
  73   FUNCTION_RETURN;
  74 }
  75
  76 DEF_VARIANTS (rintf, sf, sf)