newlib/libm/machine/amdgcn/v64sf_fmod.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/sf_fmod.c in Newlib.  */
  27
  28 #include "amdgcnmach.h"
  29
  30 DEF_VS_MATH_FUNC (v64sf, fmodf, v64sf x, v64sf y)
  31 {
  32   FUNCTION_INIT(v64sf);
  33
  34   v64si hx, hy, hz;
  35   GET_FLOAT_WORD (hx, x, NO_COND);
  36   GET_FLOAT_WORD (hy, y, NO_COND);
  37   v64si sx = hx & 0x80000000;   /* sign of x */
  38   hx ^=sx;              /* |x| */
  39   hy &= 0x7fffffff;     /* |y| */
  40
  41   v64sf zeroes = VECTOR_MERGE (VECTOR_INIT (-0.0f),
  42                                VECTOR_INIT (0.0f),
  43                                sx != 0);
  44
  45   /* purge off exception values */
  46   VECTOR_IF ((hy == 0) | (hx >= 0x7f800000)
  47              | (hy > 0x7f800000), cond) // y=0, or x not finite or y is NaN
  48     VECTOR_RETURN ((x * y) / (x * y), cond);
  49   VECTOR_ENDIF
  50   VECTOR_IF (hx < hy, cond)             // |x|<|y| return x
  51     VECTOR_RETURN (x, cond);
  52   VECTOR_ENDIF
  53   VECTOR_IF (hx == hy, cond)
  54     VECTOR_RETURN (zeroes, hx == hy);   // |x|=|y| return x*0
  55   VECTOR_ENDIF
  56
  57   /* determine ix = ilogb(x) */
  58   v64si ix;
  59   VECTOR_IF (hx < 0x00800000, cond)     // subnormal x
  60     ix = VECTOR_INIT (-126);
  61     for (v64si i = (hx << 8);
  62          !ALL_ZEROES_P (cond & (i > 0));
  63          i <<= 1)
  64       VECTOR_COND_MOVE (ix, ix - 1, cond & (i > 0));
  65   VECTOR_ELSE (cond)
  66     VECTOR_COND_MOVE (ix, (hx >> 23) - 127, cond);
  67   VECTOR_ENDIF
  68
  69   /* determine iy = ilogb(y) */
  70   v64si iy;
  71   VECTOR_IF (hy < 0x00800000, cond)     // subnormal y
  72     iy = VECTOR_INIT (-126);
  73     for (v64si i = (hy << 8); !ALL_ZEROES_P (cond & (i >= 0)); i <<= 1)
  74       VECTOR_COND_MOVE (iy, iy - 1, cond & (i >= 0));
  75   VECTOR_ELSE (cond)
  76     VECTOR_COND_MOVE (iy, (hy >> 23) - 127, cond);
  77   VECTOR_ENDIF
  78
  79 /* set up {hx,lx}, {hy,ly} and align y to x */
  80   VECTOR_IF (ix >= -126, cond)
  81     VECTOR_COND_MOVE (hx, 0x00800000 | (0x007fffff & hx), cond);
  82   VECTOR_ELSE (cond)            // subnormal x, shift x to normal
  83     {
  84       v64si n = -126 - ix;
  85       VECTOR_COND_MOVE (hx, hx << n, cond);
  86     }
  87   VECTOR_ENDIF
  88   VECTOR_IF (iy >= -126, cond)
  89     VECTOR_COND_MOVE (hy, 0x00800000 | (0x007fffff & hy), cond);
  90   VECTOR_ELSE (cond)            // subnormal y, shift y to normal
  91     {
  92       v64si n = -126 - iy;
  93       VECTOR_COND_MOVE (hy, hy << n, cond);
  94     }
  95   VECTOR_ENDIF
  96
  97 /* fix point fmod */
  98   v64si n = ix - iy;
  99   v64si cond = n != 0;
 100
 101   while (!ALL_ZEROES_P (cond))
 102     {
 103       hz = hx - hy;
 104       VECTOR_IF2 (hz < 0, cond2, cond)
 105         VECTOR_COND_MOVE (hx, hx + hx, cond2);
 106       VECTOR_ELSE2 (cond2, cond)
 107         VECTOR_IF2 (hz == 0, cond3, cond2)              // return sign(x)*0
 108           VECTOR_RETURN (zeroes, cond3);
 109         VECTOR_ELSE2 (cond3, cond2)
 110           VECTOR_COND_MOVE (hx, hz + hz, cond2);
 111         VECTOR_ENDIF
 112       VECTOR_ENDIF
 113
 114       n += cond;        // Active lanes should be -1
 115       cond &= (n != 0);
 116     }
 117
 118   hz = hx - hy;
 119   VECTOR_COND_MOVE (hx, hz, hz >= 0);
 120
 121   /* convert back to floating value and restore the sign */
 122   VECTOR_RETURN (zeroes, hx == 0);      // return sign(x)*0
 123
 124   cond = hx < 0x00800000;
 125   while (!ALL_ZEROES_P (cond))          // normalize x
 126     {
 127       VECTOR_COND_MOVE (hx, hx + hx, cond);
 128       iy += cond;       // Active lanes should be -1
 129
 130       cond &= (hx < 0x00800000);
 131     }
 132   VECTOR_IF (iy >= -126, cond)          // normalize output
 133     VECTOR_COND_MOVE (hx, (hx - 0x00800000) | ((iy + 127) << 23), cond);
 134     SET_FLOAT_WORD (x, hx | sx, cond);
 135   VECTOR_ELSE (cond)            // subnormal output */
 136     n = -126 - iy;
 137     hx >>= n;
 138     SET_FLOAT_WORD (x, hx | sx, cond);
 139     x *= VECTOR_INIT (1.0f);            /* create necessary signal */
 140   VECTOR_ENDIF
 141
 142   VECTOR_RETURN (x, NO_COND);   /* exact output */
 143
 144   FUNCTION_RETURN;
 145 }
 146
 147 DEF_VARIANTS2 (fmodf, sf, sf)