newlib/libm/machine/amdgcn/amdgcnmach.h

   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 /* Common header file for AMD GCN vector math routines.  */
  16
  17 /*
  18  * ====================================================
  19  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
  20  *
  21  * Developed at SunPro, a Sun Microsystems, Inc. business.
  22  * Permission to use, copy, modify, and distribute this
  23  * software is freely granted, provided that this notice
  24  * is preserved.
  25  * ====================================================
  26  */
  27
  28 /* Copyright (c) 2017-2018 Arm Ltd.  All rights reserved.
  29
  30    SPDX-License-Identifier: BSD-3-Clause
  31
  32    Redistribution and use in source and binary forms, with or without
  33    modification, are permitted provided that the following conditions
  34    are met:
  35    1. Redistributions of source code must retain the above copyright
  36       notice, this list of conditions and the following disclaimer.
  37    2. Redistributions in binary form must reproduce the above copyright
  38       notice, this list of conditions and the following disclaimer in the
  39       documentation and/or other materials provided with the distribution.
  40    3. The name of the company may not be used to endorse or promote
  41       products derived from this software without specific prior written
  42       permission.
  43
  44    THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
  45    WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  46    MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  47    IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  48    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
  49    TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  50    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  51    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  52    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  53    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
  54
  55 /* This header is partially based on:
  56
  57    newlib/libm/common/fdlibm.h
  58    newlib/libm/mathfp/zmath.h
  59    newlib/libm/common/math_errf.c
  60    newlib/libm/common/math_config.h  */
  61
  62 #include <errno.h>
  63 #include <sys/types.h>
  64 #include <machine/ieeefp.h>
  65 #include "amdgcn_veclib.h"
  66
  67 /* Vectorized versions of macros from newlib/libm/common/fdlibm.h  */
  68
  69 #define EXTRACT_WORDS(hi, lo, x) \
  70 do { \
  71   vector_union __tmp; \
  72   __tmp.t_v64df = (x); \
  73   hi = __builtin_convertvector (__tmp.t_v64di >> 32, typeof (hi)); \
  74   lo = __builtin_convertvector (__tmp.t_v64di & 0xffffffff, typeof (lo)); \
  75 } while (0)
  76
  77 #define INSERT_WORDS(x, hi, lo, cond) \
  78 do { \
  79   vector_union __tmp; \
  80   __tmp.t_v64di = __builtin_convertvector (hi, v64di) << 32 | \
  81     __builtin_convertvector (lo, v64di) & 0xffffffff; \
  82   VECTOR_COND_MOVE (x, __tmp.t_v64df, cond); \
  83 } while (0)
  84
  85 #define GET_HIGH_WORD(x, y, cond) \
  86 do { \
  87   vector_union __tmp; \
  88   __tmp.t_v64df = (y); \
  89   VECTOR_COND_MOVE (x, __builtin_convertvector (__tmp.t_v64di >> 32, v64si), \
  90                     (cond)); \
  91 } while (0)
  92
  93 #define GET_LOW_WORD(x, y, cond) \
  94 do { \
  95   vector_union __tmp; \
  96   __tmp.t_v64df = (y); \
  97   VECTOR_COND_MOVE (x, __builtin_convertvector (__tmp.t_v64di & 0xffffffff, \
  98                                                 v64si), (cond)); \
  99 } while (0)
 100
 101 #define SET_HIGH_WORD(x, y, cond) \
 102 do { \
 103   vector_union __tmp; \
 104   __tmp.t_v64df = x; \
 105   __tmp.t_v64di &= 0xffffffff; \
 106   __tmp.t_v64di |= __builtin_convertvector (y, v64di) << 32; \
 107   VECTOR_COND_MOVE (x, __tmp.t_v64df, (cond)); \
 108 } while (0)
 109
 110 #define SET_LOW_WORD(x, y, cond) \
 111 do { \
 112   vector_union __tmp; \
 113   __tmp.t_v64df = x; \
 114   __tmp.t_v64di &= 0xffffffff00000000ULL; \
 115   __tmp.t_v64di |= __builtin_convertvector (y, v64di); \
 116   VECTOR_COND_MOVE (x, __tmp.t_v64df, (cond)); \
 117  } while (0)
 118
 119 #define GET_FLOAT_WORD(x, y, cond) \
 120   VECTOR_COND_MOVE(x, CAST_VECTOR(v64si, (y)), (cond))
 121
 122 #define SET_FLOAT_WORD(x, y, cond) \
 123   VECTOR_COND_MOVE(x, CAST_VECTOR(v64sf, (y)), (cond))
 124
 125 /* Definitions from newlib/libm/common/fdlibm.h  */
 126
 127 #ifdef _FLT_LARGEST_EXPONENT_IS_NORMAL
 128 #define FLT_UWORD_IS_FINITE(x) ((x) == (x))
 129 #define FLT_UWORD_IS_NAN(x) ((x) != (x))
 130 #define FLT_UWORD_IS_INFINITE(x) ((x) != (x))
 131 #define FLT_UWORD_MAX 0x7fffffff
 132 #define FLT_UWORD_EXP_MAX 0x43010000
 133 #define FLT_UWORD_LOG_MAX 0x42b2d4fc
 134 #define FLT_UWORD_LOG_2MAX 0x42b437e0
 135 #define HUGE ((float)0X1.FFFFFEP128)
 136 #else
 137 #define FLT_UWORD_IS_FINITE(x) ((x)<0x7f800000)
 138 #define FLT_UWORD_IS_NAN(x) ((x)>0x7f800000)
 139 #define FLT_UWORD_IS_INFINITE(x) ((x)==0x7f800000)
 140 #define FLT_UWORD_MAX 0x7f7fffffL
 141 #define FLT_UWORD_EXP_MAX 0x43000000
 142 #define FLT_UWORD_LOG_MAX 0x42b17217
 143 #define FLT_UWORD_LOG_2MAX 0x42b2d4fc
 144 #define HUGE ((float)3.40282346638528860e+38)
 145 #endif
 146 #define FLT_UWORD_HALF_MAX (FLT_UWORD_MAX-(1L<<23))
 147 #define FLT_LARGEST_EXP (FLT_UWORD_MAX>>23)
 148
 149 #ifdef _FLT_NO_DENORMALS
 150 #define FLT_UWORD_IS_ZERO(x) ((x)<0x00800000)
 151 #define FLT_UWORD_IS_SUBNORMAL(x) ((x) != (x))
 152 #define FLT_UWORD_MIN 0x00800000
 153 #define FLT_UWORD_EXP_MIN 0x42fc0000
 154 #define FLT_UWORD_LOG_MIN 0x42aeac50
 155 #define FLT_SMALLEST_EXP 1
 156 #else
 157 #define FLT_UWORD_IS_ZERO(x) ((x)==0)
 158 #define FLT_UWORD_IS_SUBNORMAL(x) ((x)<0x00800000)
 159 #define FLT_UWORD_MIN 0x00000001
 160 #define FLT_UWORD_EXP_MIN 0x43160000
 161 #define FLT_UWORD_LOG_MIN 0x42cff1b5
 162 #define FLT_SMALLEST_EXP -22
 163 #endif
 164
 165 /* Definitions from newlib/libm/mathfp/zmath.h  */
 166
 167 #define NUM 3
 168 #define NAN 2
 169 #define INF 1
 170
 171 #define __PI 3.14159265358979323846
 172 #define __SQRT_HALF 0.70710678118654752440
 173 #define __PI_OVER_TWO 1.57079632679489661923132
 174 #define __INV_PI_OVER_TWO_2_24 10680707.430881743590348355907974
 175
 176 typedef const union
 177 {
 178   unsigned int l[2];
 179   double d;
 180 } udouble;
 181
 182 typedef const union
 183 {
 184   unsigned int l;
 185   float f;
 186 } ufloat;
 187
 188 extern double BIGX;
 189 extern double SMALLX;
 190
 191 extern udouble z_infinity;
 192 extern udouble z_notanum;
 193 extern double  z_rooteps;
 194
 195 extern ufloat  z_infinity_f;
 196 extern ufloat  z_notanum_f;
 197 extern float   z_rooteps_f;
 198
 199 /* Vectorized versions of functions from newlib/libm/common/math_errf.c  */
 200
 201 static v64sf v64sf_math_oflowf (v64si sign)
 202 {
 203   errno = ERANGE;
 204   return VECTOR_MERGE (VECTOR_INIT (-0x1p97f),
 205                        VECTOR_INIT (0x1p97f), sign) * 0x1p97f;
 206 }
 207
 208 static v64sf v64sf_math_uflowf (v64si sign)
 209 {
 210   errno = ERANGE;
 211   return VECTOR_MERGE (VECTOR_INIT (-0x1p-95f),
 212                        VECTOR_INIT (0x1p-95f), sign) * 0x1p-95f;
 213 }
 214
 215 /* Vectorized versions of functions from newlib/libm/common/math_config.h  */
 216
 217 static v64si v64sf_issignalingf_inline (v64sf x)
 218 {
 219   v64si __mask = VECTOR_INIT (-1);
 220   v64si ix;
 221   GET_FLOAT_WORD (ix, x, NO_COND);
 222   /* Use IEEE-754 2008 encoding - i.e. exponent bits all 1, MSB of
 223      significand is 0 for signalling NaN.  */
 224   return ((ix & 0x7f800000) == 0x7f800000) & ((ix & 0x00400000) == 0);
 225 }
 226
 227 /* Vector extensions to sys/reent.h  */
 228
 229 struct v64_reent {
 230   v64si _v64si_gamma_signgam;
 231 };
 232
 233 extern struct v64_reent *_v64_reent;
 234 #define _V64_REENT _v64_reent
 235
 236 #define _REENT_V64SI_SIGNGAM(ptr)      ((ptr)->_v64si_gamma_signgam)
 237
 238 /* Vector extensions to math.h  */
 239
 240 #define v64si_signgam (*__v64si_signgam())
 241 extern v64si* __v64si_signgam (void);
 242 #define __v64si_signgam_r(ptr) _REENT_V64SI_SIGNGAM(ptr)