newlib/libm/machine/amdgcn/v64sf_hypot.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/mathfp/ef_hypot.c in Newlib.  */
  27
  28 #include "amdgcnmach.h"
  29
  30 v64sf v64sf_sqrtf_aux (v64sf, v64si);
  31
  32 DEF_VS_MATH_FUNC (v64sf, hypotf, v64sf x, v64sf y)
  33 {
  34   FUNCTION_INIT (v64sf);
  35
  36   v64sf a = x;
  37   v64sf b = y;
  38
  39   v64si ha;
  40   GET_FLOAT_WORD (ha, x, NO_COND);
  41   ha &= 0x7fffffffL;
  42   v64si hb;
  43   GET_FLOAT_WORD (hb, y, NO_COND);
  44   hb &= 0x7fffffffL;
  45
  46   VECTOR_IF (hb > ha, cond)
  47     v64si j = ha;
  48     VECTOR_COND_MOVE (ha, hb, cond);
  49     VECTOR_COND_MOVE (hb, j, cond);
  50   VECTOR_ENDIF
  51   SET_FLOAT_WORD (a, ha, NO_COND);      /* a <- |a| */
  52   SET_FLOAT_WORD (b, hb, NO_COND);      /* b <- |b| */
  53   VECTOR_IF((ha - hb) > 0xf000000L, cond)       // x/y > 2**30 */
  54     VECTOR_RETURN (a + b, cond);
  55   VECTOR_ENDIF
  56
  57   v64si k = VECTOR_INIT (0);
  58
  59   VECTOR_IF (ha > 0x58800000L, cond)            /* a>2**50 */
  60     VECTOR_IF2 (ha >= 0x7f800000L, cond2, cond) /* Inf or NaN */
  61       v64sf w = a + b;                  // for sNaN */
  62       VECTOR_COND_MOVE (w, a, cond2 & (ha == 0x7f800000));
  63       VECTOR_COND_MOVE (w, b, cond2 & (hb == 0x7f800000));
  64       VECTOR_RETURN (w, cond);
  65     VECTOR_ENDIF
  66     /* scale a and b by 2**-60 */
  67     VECTOR_COND_MOVE (ha, ha - 0x5d800000, cond);
  68     VECTOR_COND_MOVE (hb, hb - 0x5d800000, cond);
  69     VECTOR_COND_MOVE (k, k + 60, cond);
  70     SET_FLOAT_WORD (a, ha, cond);
  71     SET_FLOAT_WORD (b, hb, cond);
  72   VECTOR_ENDIF
  73   VECTOR_IF (hb < 0x26800000, cond)             /* b < 2**-50 */
  74     VECTOR_IF2 (hb <= 0x007fffff, cond2, cond)  /* subnormal b or 0 */
  75       VECTOR_RETURN (a, cond2 & (hb == 0));
  76       /* t1=2^126 */
  77       v64sf t1;
  78       SET_FLOAT_WORD (t1, VECTOR_INIT (0x3f000000), cond2);
  79       VECTOR_COND_MOVE (b, b * t1, cond2);
  80       VECTOR_COND_MOVE (a, a * t1, cond2);
  81       VECTOR_COND_MOVE (k, k - 126, cond2);
  82     VECTOR_ELSE2 (cond2, cond)          /* scale a and b by 2^60 */
  83       VECTOR_COND_MOVE (ha, ha + 0x5d800000, cond2);    /* a *= 2^60 */
  84       VECTOR_COND_MOVE (hb, hb + 0x5d800000, cond2);    /* b *= 2^60 */
  85       VECTOR_COND_MOVE (k, k - 60, cond2);
  86       SET_FLOAT_WORD (a, ha, cond2);
  87       SET_FLOAT_WORD (b, hb, cond2);
  88     VECTOR_ENDIF
  89   VECTOR_ENDIF
  90   /* medium size a and b */
  91   v64sf w = a - b;
  92   VECTOR_IF (w > b, cond)
  93     v64sf t1;
  94     SET_FLOAT_WORD (t1, ha & 0xfffff000, cond);
  95     v64sf t2 = a - t1;
  96     VECTOR_COND_MOVE (w, v64sf_sqrtf_aux (t1*t1 - (b*(-b) - t2 * (a + t1)), __mask), cond);
  97   VECTOR_ELSE (cond)
  98     VECTOR_COND_MOVE (a, a+a, cond);
  99     v64sf y1;
 100     SET_FLOAT_WORD (y1, hb & 0xfffff000, cond);
 101     v64sf y2 = b - y1;
 102     v64sf t1;
 103     SET_FLOAT_WORD (t1, ha + 0x00800000, cond);
 104     v64sf t2 = a - t1;
 105     VECTOR_COND_MOVE (w, v64sf_sqrtf_aux (t1*y1 - (w*(-w) - (t1*y2 + t2*b)), __mask), cond);
 106   VECTOR_ENDIF
 107   VECTOR_IF (k != 0, cond)
 108     v64sf t1;
 109     SET_FLOAT_WORD (t1, 0x3f800000 + (k << 23), cond);
 110     VECTOR_RETURN (t1 * w, cond);
 111   VECTOR_ELSE (cond)
 112     VECTOR_RETURN (w, cond);
 113   VECTOR_ENDIF
 114
 115   FUNCTION_RETURN;
 116 }
 117
 118 DEF_VARIANTS2 (hypotf, sf, sf)