Suga koubou / Mbed 2 deprecated madplayer_lpc4088

MP3 Player. You can change fwd/rev speed and skip. see: http://mbed.org/users/okini3939/notebook/lpc4088_madplayer/

Dependencies:   I2SSlave SDFileSystem TLV320 mbed

madplayer/fixed.h@0:8ba6230eefbd, 2014-02-18 (annotated)

Committer:
okini3939
Date:
Tue Feb 18 00:22:50 2014 +0000
Revision:
0:8ba6230eefbd
1st build

Who changed what in which revision?

UserRevisionLine numberNew contents of line
okini3939 0:8ba6230eefbd 1 /*
okini3939 0:8ba6230eefbd 2 * libmad - MPEG audio decoder library
okini3939 0:8ba6230eefbd 3 * Copyright (C) 2000-2004 Underbit Technologies, Inc.
okini3939 0:8ba6230eefbd 4 *
okini3939 0:8ba6230eefbd 5 * This program is free software; you can redistribute it and/or modify
okini3939 0:8ba6230eefbd 6 * it under the terms of the GNU General Public License as published by
okini3939 0:8ba6230eefbd 7 * the Free Software Foundation; either version 2 of the License, or
okini3939 0:8ba6230eefbd 8 * (at your option) any later version.
okini3939 0:8ba6230eefbd 9 *
okini3939 0:8ba6230eefbd 10 * This program is distributed in the hope that it will be useful,
okini3939 0:8ba6230eefbd 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
okini3939 0:8ba6230eefbd 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
okini3939 0:8ba6230eefbd 13 * GNU General Public License for more details.
okini3939 0:8ba6230eefbd 14 *
okini3939 0:8ba6230eefbd 15 * You should have received a copy of the GNU General Public License
okini3939 0:8ba6230eefbd 16 * along with this program; if not, write to the Free Software
okini3939 0:8ba6230eefbd 17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
okini3939 0:8ba6230eefbd 18 *
okini3939 0:8ba6230eefbd 19 * $Id: fixed.h,v 1.1 2010/11/23 20:12:57 andy Exp $
okini3939 0:8ba6230eefbd 20 */
okini3939 0:8ba6230eefbd 21
okini3939 0:8ba6230eefbd 22 # ifndef LIBMAD_FIXED_H
okini3939 0:8ba6230eefbd 23 # define LIBMAD_FIXED_H
okini3939 0:8ba6230eefbd 24
okini3939 0:8ba6230eefbd 25 #include "config.h"
okini3939 0:8ba6230eefbd 26
okini3939 0:8ba6230eefbd 27 # if SIZEOF_INT >= 4
okini3939 0:8ba6230eefbd 28 typedef signed int mad_fixed_t;
okini3939 0:8ba6230eefbd 29
okini3939 0:8ba6230eefbd 30 typedef signed int mad_fixed64hi_t;
okini3939 0:8ba6230eefbd 31 typedef unsigned int mad_fixed64lo_t;
okini3939 0:8ba6230eefbd 32 # else
okini3939 0:8ba6230eefbd 33 typedef signed long mad_fixed_t;
okini3939 0:8ba6230eefbd 34
okini3939 0:8ba6230eefbd 35 typedef signed long mad_fixed64hi_t;
okini3939 0:8ba6230eefbd 36 typedef unsigned long mad_fixed64lo_t;
okini3939 0:8ba6230eefbd 37 # endif
okini3939 0:8ba6230eefbd 38
okini3939 0:8ba6230eefbd 39 # if defined(_MSC_VER)
okini3939 0:8ba6230eefbd 40 # define mad_fixed64_t signed __int64
okini3939 0:8ba6230eefbd 41 # elif 1 || defined(__GNUC__)
okini3939 0:8ba6230eefbd 42 # define mad_fixed64_t signed long long
okini3939 0:8ba6230eefbd 43 # endif
okini3939 0:8ba6230eefbd 44
okini3939 0:8ba6230eefbd 45 # if defined(FPM_FLOAT)
okini3939 0:8ba6230eefbd 46 typedef double mad_sample_t;
okini3939 0:8ba6230eefbd 47 # else
okini3939 0:8ba6230eefbd 48 typedef mad_fixed_t mad_sample_t;
okini3939 0:8ba6230eefbd 49 # endif
okini3939 0:8ba6230eefbd 50
okini3939 0:8ba6230eefbd 51 /*
okini3939 0:8ba6230eefbd 52 * Fixed-point format: 0xABBBBBBB
okini3939 0:8ba6230eefbd 53 * A == whole part (sign + 3 bits)
okini3939 0:8ba6230eefbd 54 * B == fractional part (28 bits)
okini3939 0:8ba6230eefbd 55 *
okini3939 0:8ba6230eefbd 56 * Values are signed two's complement, so the effective range is:
okini3939 0:8ba6230eefbd 57 * 0x80000000 to 0x7fffffff
okini3939 0:8ba6230eefbd 58 * -8.0 to +7.9999999962747097015380859375
okini3939 0:8ba6230eefbd 59 *
okini3939 0:8ba6230eefbd 60 * The smallest representable value is:
okini3939 0:8ba6230eefbd 61 * 0x00000001 == 0.0000000037252902984619140625 (i.e. about 3.725e-9)
okini3939 0:8ba6230eefbd 62 *
okini3939 0:8ba6230eefbd 63 * 28 bits of fractional accuracy represent about
okini3939 0:8ba6230eefbd 64 * 8.6 digits of decimal accuracy.
okini3939 0:8ba6230eefbd 65 *
okini3939 0:8ba6230eefbd 66 * Fixed-point numbers can be added or subtracted as normal
okini3939 0:8ba6230eefbd 67 * integers, but multiplication requires shifting the 64-bit result
okini3939 0:8ba6230eefbd 68 * from 56 fractional bits back to 28 (and rounding.)
okini3939 0:8ba6230eefbd 69 *
okini3939 0:8ba6230eefbd 70 * Changing the definition of MAD_F_FRACBITS is only partially
okini3939 0:8ba6230eefbd 71 * supported, and must be done with care.
okini3939 0:8ba6230eefbd 72 */
okini3939 0:8ba6230eefbd 73
okini3939 0:8ba6230eefbd 74 # define MAD_F_FRACBITS 28
okini3939 0:8ba6230eefbd 75
okini3939 0:8ba6230eefbd 76 # if MAD_F_FRACBITS == 28
okini3939 0:8ba6230eefbd 77 # define MAD_F(x) ((mad_fixed_t) (x##L))
okini3939 0:8ba6230eefbd 78 # else
okini3939 0:8ba6230eefbd 79 # if MAD_F_FRACBITS < 28
okini3939 0:8ba6230eefbd 80 # warning "MAD_F_FRACBITS < 28"
okini3939 0:8ba6230eefbd 81 # define MAD_F(x) ((mad_fixed_t) \
okini3939 0:8ba6230eefbd 82 (((x##L) + \
okini3939 0:8ba6230eefbd 83 (1L << (28 - MAD_F_FRACBITS - 1))) >> \
okini3939 0:8ba6230eefbd 84 (28 - MAD_F_FRACBITS)))
okini3939 0:8ba6230eefbd 85 # elif MAD_F_FRACBITS > 28
okini3939 0:8ba6230eefbd 86 # error "MAD_F_FRACBITS > 28 not currently supported"
okini3939 0:8ba6230eefbd 87 # define MAD_F(x) ((mad_fixed_t) \
okini3939 0:8ba6230eefbd 88 ((x##L) << (MAD_F_FRACBITS - 28)))
okini3939 0:8ba6230eefbd 89 # endif
okini3939 0:8ba6230eefbd 90 # endif
okini3939 0:8ba6230eefbd 91
okini3939 0:8ba6230eefbd 92 # define MAD_F_MIN ((mad_fixed_t) -0x80000000L)
okini3939 0:8ba6230eefbd 93 # define MAD_F_MAX ((mad_fixed_t) +0x7fffffffL)
okini3939 0:8ba6230eefbd 94
okini3939 0:8ba6230eefbd 95 # define MAD_F_ONE MAD_F(0x10000000)
okini3939 0:8ba6230eefbd 96
okini3939 0:8ba6230eefbd 97 # define mad_f_tofixed(x) ((mad_fixed_t) \
okini3939 0:8ba6230eefbd 98 ((x) * (double) (1L << MAD_F_FRACBITS) + 0.5))
okini3939 0:8ba6230eefbd 99 # define mad_f_todouble(x) ((double) \
okini3939 0:8ba6230eefbd 100 ((x) / (double) (1L << MAD_F_FRACBITS)))
okini3939 0:8ba6230eefbd 101
okini3939 0:8ba6230eefbd 102 # define mad_f_intpart(x) ((x) >> MAD_F_FRACBITS)
okini3939 0:8ba6230eefbd 103 # define mad_f_fracpart(x) ((x) & ((1L << MAD_F_FRACBITS) - 1))
okini3939 0:8ba6230eefbd 104 /* (x should be positive) */
okini3939 0:8ba6230eefbd 105
okini3939 0:8ba6230eefbd 106 # define mad_f_fromint(x) ((x) << MAD_F_FRACBITS)
okini3939 0:8ba6230eefbd 107
okini3939 0:8ba6230eefbd 108 # define mad_f_add(x, y) ((x) + (y))
okini3939 0:8ba6230eefbd 109 # define mad_f_sub(x, y) ((x) - (y))
okini3939 0:8ba6230eefbd 110
okini3939 0:8ba6230eefbd 111 # if defined(FPM_FLOAT)
okini3939 0:8ba6230eefbd 112 # error "FPM_FLOAT not yet supported"
okini3939 0:8ba6230eefbd 113
okini3939 0:8ba6230eefbd 114 # undef MAD_F
okini3939 0:8ba6230eefbd 115 # define MAD_F(x) mad_f_todouble(x)
okini3939 0:8ba6230eefbd 116
okini3939 0:8ba6230eefbd 117 # define mad_f_mul(x, y) ((x) * (y))
okini3939 0:8ba6230eefbd 118 # define mad_f_scale64
okini3939 0:8ba6230eefbd 119
okini3939 0:8ba6230eefbd 120 # undef ASO_ZEROCHECK
okini3939 0:8ba6230eefbd 121
okini3939 0:8ba6230eefbd 122 # elif defined(FPM_64BIT)
okini3939 0:8ba6230eefbd 123
okini3939 0:8ba6230eefbd 124 /*
okini3939 0:8ba6230eefbd 125 * This version should be the most accurate if 64-bit types are supported by
okini3939 0:8ba6230eefbd 126 * the compiler, although it may not be the most efficient.
okini3939 0:8ba6230eefbd 127 */
okini3939 0:8ba6230eefbd 128 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 129 # define mad_f_mul(x, y) \
okini3939 0:8ba6230eefbd 130 ((mad_fixed_t) \
okini3939 0:8ba6230eefbd 131 ((((mad_fixed64_t) (x) * (y)) + \
okini3939 0:8ba6230eefbd 132 (1L << (MAD_F_SCALEBITS - 1))) >> MAD_F_SCALEBITS))
okini3939 0:8ba6230eefbd 133 # else
okini3939 0:8ba6230eefbd 134 # define mad_f_mul(x, y) \
okini3939 0:8ba6230eefbd 135 ((mad_fixed_t) (((mad_fixed64_t) (x) * (y)) >> MAD_F_SCALEBITS))
okini3939 0:8ba6230eefbd 136 # endif
okini3939 0:8ba6230eefbd 137
okini3939 0:8ba6230eefbd 138 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 139
okini3939 0:8ba6230eefbd 140 /* --- Intel --------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 141
okini3939 0:8ba6230eefbd 142 # elif defined(FPM_INTEL)
okini3939 0:8ba6230eefbd 143
okini3939 0:8ba6230eefbd 144 # if defined(_MSC_VER)
okini3939 0:8ba6230eefbd 145 # pragma warning(push)
okini3939 0:8ba6230eefbd 146 # pragma warning(disable: 4035) /* no return value */
okini3939 0:8ba6230eefbd 147 static __forceinline
okini3939 0:8ba6230eefbd 148 mad_fixed_t mad_f_mul_inline(mad_fixed_t x, mad_fixed_t y)
okini3939 0:8ba6230eefbd 149 {
okini3939 0:8ba6230eefbd 150 enum {
okini3939 0:8ba6230eefbd 151 fracbits = MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 152 };
okini3939 0:8ba6230eefbd 153
okini3939 0:8ba6230eefbd 154 __asm {
okini3939 0:8ba6230eefbd 155 mov eax, x
okini3939 0:8ba6230eefbd 156 imul y
okini3939 0:8ba6230eefbd 157 shrd eax, edx, fracbits
okini3939 0:8ba6230eefbd 158 }
okini3939 0:8ba6230eefbd 159
okini3939 0:8ba6230eefbd 160 /* implicit return of eax */
okini3939 0:8ba6230eefbd 161 }
okini3939 0:8ba6230eefbd 162 # pragma warning(pop)
okini3939 0:8ba6230eefbd 163
okini3939 0:8ba6230eefbd 164 # define mad_f_mul mad_f_mul_inline
okini3939 0:8ba6230eefbd 165 # define mad_f_scale64
okini3939 0:8ba6230eefbd 166 # else
okini3939 0:8ba6230eefbd 167 /*
okini3939 0:8ba6230eefbd 168 * This Intel version is fast and accurate; the disposition of the least
okini3939 0:8ba6230eefbd 169 * significant bit depends on OPT_ACCURACY via mad_f_scale64().
okini3939 0:8ba6230eefbd 170 */
okini3939 0:8ba6230eefbd 171 # define MAD_F_MLX(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 172 asm ("imull %3" \
okini3939 0:8ba6230eefbd 173 : "=a" (lo), "=d" (hi) \
okini3939 0:8ba6230eefbd 174 : "%a" (x), "rm" (y) \
okini3939 0:8ba6230eefbd 175 : "cc")
okini3939 0:8ba6230eefbd 176
okini3939 0:8ba6230eefbd 177 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 178 /*
okini3939 0:8ba6230eefbd 179 * This gives best accuracy but is not very fast.
okini3939 0:8ba6230eefbd 180 */
okini3939 0:8ba6230eefbd 181 # define MAD_F_MLA(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 182 ({ mad_fixed64hi_t __hi; \
okini3939 0:8ba6230eefbd 183 mad_fixed64lo_t __lo; \
okini3939 0:8ba6230eefbd 184 MAD_F_MLX(__hi, __lo, (x), (y)); \
okini3939 0:8ba6230eefbd 185 asm ("addl %2,%0\n\t" \
okini3939 0:8ba6230eefbd 186 "adcl %3,%1" \
okini3939 0:8ba6230eefbd 187 : "=rm" (lo), "=rm" (hi) \
okini3939 0:8ba6230eefbd 188 : "r" (__lo), "r" (__hi), "0" (lo), "1" (hi) \
okini3939 0:8ba6230eefbd 189 : "cc"); \
okini3939 0:8ba6230eefbd 190 })
okini3939 0:8ba6230eefbd 191 # endif /* OPT_ACCURACY */
okini3939 0:8ba6230eefbd 192
okini3939 0:8ba6230eefbd 193 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 194 /*
okini3939 0:8ba6230eefbd 195 * Surprisingly, this is faster than SHRD followed by ADC.
okini3939 0:8ba6230eefbd 196 */
okini3939 0:8ba6230eefbd 197 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 198 ({ mad_fixed64hi_t __hi_; \
okini3939 0:8ba6230eefbd 199 mad_fixed64lo_t __lo_; \
okini3939 0:8ba6230eefbd 200 mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 201 asm ("addl %4,%2\n\t" \
okini3939 0:8ba6230eefbd 202 "adcl %5,%3" \
okini3939 0:8ba6230eefbd 203 : "=rm" (__lo_), "=rm" (__hi_) \
okini3939 0:8ba6230eefbd 204 : "0" (lo), "1" (hi), \
okini3939 0:8ba6230eefbd 205 "ir" (1L << (MAD_F_SCALEBITS - 1)), "ir" (0) \
okini3939 0:8ba6230eefbd 206 : "cc"); \
okini3939 0:8ba6230eefbd 207 asm ("shrdl %3,%2,%1" \
okini3939 0:8ba6230eefbd 208 : "=rm" (__result) \
okini3939 0:8ba6230eefbd 209 : "0" (__lo_), "r" (__hi_), "I" (MAD_F_SCALEBITS) \
okini3939 0:8ba6230eefbd 210 : "cc"); \
okini3939 0:8ba6230eefbd 211 __result; \
okini3939 0:8ba6230eefbd 212 })
okini3939 0:8ba6230eefbd 213 # elif defined(OPT_INTEL)
okini3939 0:8ba6230eefbd 214 /*
okini3939 0:8ba6230eefbd 215 * Alternate Intel scaling that may or may not perform better.
okini3939 0:8ba6230eefbd 216 */
okini3939 0:8ba6230eefbd 217 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 218 ({ mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 219 asm ("shrl %3,%1\n\t" \
okini3939 0:8ba6230eefbd 220 "shll %4,%2\n\t" \
okini3939 0:8ba6230eefbd 221 "orl %2,%1" \
okini3939 0:8ba6230eefbd 222 : "=rm" (__result) \
okini3939 0:8ba6230eefbd 223 : "0" (lo), "r" (hi), \
okini3939 0:8ba6230eefbd 224 "I" (MAD_F_SCALEBITS), "I" (32 - MAD_F_SCALEBITS) \
okini3939 0:8ba6230eefbd 225 : "cc"); \
okini3939 0:8ba6230eefbd 226 __result; \
okini3939 0:8ba6230eefbd 227 })
okini3939 0:8ba6230eefbd 228 # else
okini3939 0:8ba6230eefbd 229 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 230 ({ mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 231 asm ("shrdl %3,%2,%1" \
okini3939 0:8ba6230eefbd 232 : "=rm" (__result) \
okini3939 0:8ba6230eefbd 233 : "0" (lo), "r" (hi), "I" (MAD_F_SCALEBITS) \
okini3939 0:8ba6230eefbd 234 : "cc"); \
okini3939 0:8ba6230eefbd 235 __result; \
okini3939 0:8ba6230eefbd 236 })
okini3939 0:8ba6230eefbd 237 # endif /* OPT_ACCURACY */
okini3939 0:8ba6230eefbd 238
okini3939 0:8ba6230eefbd 239 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 240 # endif
okini3939 0:8ba6230eefbd 241
okini3939 0:8ba6230eefbd 242 /* --- ARM ----------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 243
okini3939 0:8ba6230eefbd 244 # elif defined(FPM_ARM)
okini3939 0:8ba6230eefbd 245
okini3939 0:8ba6230eefbd 246 /*
okini3939 0:8ba6230eefbd 247 * This ARM V4 version is as accurate as FPM_64BIT but much faster. The
okini3939 0:8ba6230eefbd 248 * least significant bit is properly rounded at no CPU cycle cost!
okini3939 0:8ba6230eefbd 249 */
okini3939 0:8ba6230eefbd 250 # if 1
okini3939 0:8ba6230eefbd 251 /*
okini3939 0:8ba6230eefbd 252 * This is faster than the default implementation via MAD_F_MLX() and
okini3939 0:8ba6230eefbd 253 * mad_f_scale64().
okini3939 0:8ba6230eefbd 254 */
okini3939 0:8ba6230eefbd 255 # define mad_f_mul(x, y) \
okini3939 0:8ba6230eefbd 256 ({ mad_fixed64hi_t __hi; \
okini3939 0:8ba6230eefbd 257 mad_fixed64lo_t __lo; \
okini3939 0:8ba6230eefbd 258 mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 259 asm ("smull %0, %1, %3, %4\n\t" \
okini3939 0:8ba6230eefbd 260 "movs %0, %0, lsr %5\n\t" \
okini3939 0:8ba6230eefbd 261 "adc %2, %0, %1, lsl %6" \
okini3939 0:8ba6230eefbd 262 : "=&r" (__lo), "=&r" (__hi), "=r" (__result) \
okini3939 0:8ba6230eefbd 263 : "%r" (x), "r" (y), \
okini3939 0:8ba6230eefbd 264 "M" (MAD_F_SCALEBITS), "M" (32 - MAD_F_SCALEBITS) \
okini3939 0:8ba6230eefbd 265 : "cc"); \
okini3939 0:8ba6230eefbd 266 __result; \
okini3939 0:8ba6230eefbd 267 })
okini3939 0:8ba6230eefbd 268 # endif
okini3939 0:8ba6230eefbd 269
okini3939 0:8ba6230eefbd 270 # define MAD_F_MLX(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 271 asm ("smull %0, %1, %2, %3" \
okini3939 0:8ba6230eefbd 272 : "=&r" (lo), "=&r" (hi) \
okini3939 0:8ba6230eefbd 273 : "%r" (x), "r" (y))
okini3939 0:8ba6230eefbd 274
okini3939 0:8ba6230eefbd 275 # define MAD_F_MLA(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 276 asm ("smlal %0, %1, %2, %3" \
okini3939 0:8ba6230eefbd 277 : "+r" (lo), "+r" (hi) \
okini3939 0:8ba6230eefbd 278 : "%r" (x), "r" (y))
okini3939 0:8ba6230eefbd 279
okini3939 0:8ba6230eefbd 280 # define MAD_F_MLN(hi, lo) \
okini3939 0:8ba6230eefbd 281 asm ("rsbs %0, %2, #0\n\t" \
okini3939 0:8ba6230eefbd 282 "rsc %1, %3, #0" \
okini3939 0:8ba6230eefbd 283 : "=r" (lo), "=r" (hi) \
okini3939 0:8ba6230eefbd 284 : "0" (lo), "1" (hi) \
okini3939 0:8ba6230eefbd 285 : "cc")
okini3939 0:8ba6230eefbd 286
okini3939 0:8ba6230eefbd 287 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 288 ({ mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 289 asm ("movs %0, %1, lsr %3\n\t" \
okini3939 0:8ba6230eefbd 290 "adc %0, %0, %2, lsl %4" \
okini3939 0:8ba6230eefbd 291 : "=&r" (__result) \
okini3939 0:8ba6230eefbd 292 : "r" (lo), "r" (hi), \
okini3939 0:8ba6230eefbd 293 "M" (MAD_F_SCALEBITS), "M" (32 - MAD_F_SCALEBITS) \
okini3939 0:8ba6230eefbd 294 : "cc"); \
okini3939 0:8ba6230eefbd 295 __result; \
okini3939 0:8ba6230eefbd 296 })
okini3939 0:8ba6230eefbd 297
okini3939 0:8ba6230eefbd 298 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 299
okini3939 0:8ba6230eefbd 300 /* --- MIPS ---------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 301
okini3939 0:8ba6230eefbd 302 # elif defined(FPM_MIPS)
okini3939 0:8ba6230eefbd 303
okini3939 0:8ba6230eefbd 304 /*
okini3939 0:8ba6230eefbd 305 * This MIPS version is fast and accurate; the disposition of the least
okini3939 0:8ba6230eefbd 306 * significant bit depends on OPT_ACCURACY via mad_f_scale64().
okini3939 0:8ba6230eefbd 307 */
okini3939 0:8ba6230eefbd 308 # define MAD_F_MLX(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 309 asm ("mult %2,%3" \
okini3939 0:8ba6230eefbd 310 : "=l" (lo), "=h" (hi) \
okini3939 0:8ba6230eefbd 311 : "%r" (x), "r" (y))
okini3939 0:8ba6230eefbd 312
okini3939 0:8ba6230eefbd 313 # if defined(HAVE_MADD_ASM)
okini3939 0:8ba6230eefbd 314 # define MAD_F_MLA(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 315 asm ("madd %2,%3" \
okini3939 0:8ba6230eefbd 316 : "+l" (lo), "+h" (hi) \
okini3939 0:8ba6230eefbd 317 : "%r" (x), "r" (y))
okini3939 0:8ba6230eefbd 318 # elif defined(HAVE_MADD16_ASM)
okini3939 0:8ba6230eefbd 319 /*
okini3939 0:8ba6230eefbd 320 * This loses significant accuracy due to the 16-bit integer limit in the
okini3939 0:8ba6230eefbd 321 * multiply/accumulate instruction.
okini3939 0:8ba6230eefbd 322 */
okini3939 0:8ba6230eefbd 323 # define MAD_F_ML0(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 324 asm ("mult %2,%3" \
okini3939 0:8ba6230eefbd 325 : "=l" (lo), "=h" (hi) \
okini3939 0:8ba6230eefbd 326 : "%r" ((x) >> 12), "r" ((y) >> 16))
okini3939 0:8ba6230eefbd 327 # define MAD_F_MLA(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 328 asm ("madd16 %2,%3" \
okini3939 0:8ba6230eefbd 329 : "+l" (lo), "+h" (hi) \
okini3939 0:8ba6230eefbd 330 : "%r" ((x) >> 12), "r" ((y) >> 16))
okini3939 0:8ba6230eefbd 331 # define MAD_F_MLZ(hi, lo) ((mad_fixed_t) (lo))
okini3939 0:8ba6230eefbd 332 # endif
okini3939 0:8ba6230eefbd 333
okini3939 0:8ba6230eefbd 334 # if defined(OPT_SPEED)
okini3939 0:8ba6230eefbd 335 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 336 ((mad_fixed_t) ((hi) << (32 - MAD_F_SCALEBITS)))
okini3939 0:8ba6230eefbd 337 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 338 # endif
okini3939 0:8ba6230eefbd 339
okini3939 0:8ba6230eefbd 340 /* --- SPARC --------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 341
okini3939 0:8ba6230eefbd 342 # elif defined(FPM_SPARC)
okini3939 0:8ba6230eefbd 343
okini3939 0:8ba6230eefbd 344 /*
okini3939 0:8ba6230eefbd 345 * This SPARC V8 version is fast and accurate; the disposition of the least
okini3939 0:8ba6230eefbd 346 * significant bit depends on OPT_ACCURACY via mad_f_scale64().
okini3939 0:8ba6230eefbd 347 */
okini3939 0:8ba6230eefbd 348 # define MAD_F_MLX(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 349 asm ("smul %2, %3, %0\n\t" \
okini3939 0:8ba6230eefbd 350 "rd %%y, %1" \
okini3939 0:8ba6230eefbd 351 : "=r" (lo), "=r" (hi) \
okini3939 0:8ba6230eefbd 352 : "%r" (x), "rI" (y))
okini3939 0:8ba6230eefbd 353
okini3939 0:8ba6230eefbd 354 /* --- PowerPC ------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 355
okini3939 0:8ba6230eefbd 356 # elif defined(FPM_PPC)
okini3939 0:8ba6230eefbd 357
okini3939 0:8ba6230eefbd 358 /*
okini3939 0:8ba6230eefbd 359 * This PowerPC version is fast and accurate; the disposition of the least
okini3939 0:8ba6230eefbd 360 * significant bit depends on OPT_ACCURACY via mad_f_scale64().
okini3939 0:8ba6230eefbd 361 */
okini3939 0:8ba6230eefbd 362 # define MAD_F_MLX(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 363 do { \
okini3939 0:8ba6230eefbd 364 asm ("mullw %0,%1,%2" \
okini3939 0:8ba6230eefbd 365 : "=r" (lo) \
okini3939 0:8ba6230eefbd 366 : "%r" (x), "r" (y)); \
okini3939 0:8ba6230eefbd 367 asm ("mulhw %0,%1,%2" \
okini3939 0:8ba6230eefbd 368 : "=r" (hi) \
okini3939 0:8ba6230eefbd 369 : "%r" (x), "r" (y)); \
okini3939 0:8ba6230eefbd 370 } \
okini3939 0:8ba6230eefbd 371 while (0)
okini3939 0:8ba6230eefbd 372
okini3939 0:8ba6230eefbd 373 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 374 /*
okini3939 0:8ba6230eefbd 375 * This gives best accuracy but is not very fast.
okini3939 0:8ba6230eefbd 376 */
okini3939 0:8ba6230eefbd 377 # define MAD_F_MLA(hi, lo, x, y) \
okini3939 0:8ba6230eefbd 378 ({ mad_fixed64hi_t __hi; \
okini3939 0:8ba6230eefbd 379 mad_fixed64lo_t __lo; \
okini3939 0:8ba6230eefbd 380 MAD_F_MLX(__hi, __lo, (x), (y)); \
okini3939 0:8ba6230eefbd 381 asm ("addc %0,%2,%3\n\t" \
okini3939 0:8ba6230eefbd 382 "adde %1,%4,%5" \
okini3939 0:8ba6230eefbd 383 : "=r" (lo), "=r" (hi) \
okini3939 0:8ba6230eefbd 384 : "%r" (lo), "r" (__lo), \
okini3939 0:8ba6230eefbd 385 "%r" (hi), "r" (__hi) \
okini3939 0:8ba6230eefbd 386 : "xer"); \
okini3939 0:8ba6230eefbd 387 })
okini3939 0:8ba6230eefbd 388 # endif
okini3939 0:8ba6230eefbd 389
okini3939 0:8ba6230eefbd 390 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 391 /*
okini3939 0:8ba6230eefbd 392 * This is slower than the truncating version below it.
okini3939 0:8ba6230eefbd 393 */
okini3939 0:8ba6230eefbd 394 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 395 ({ mad_fixed_t __result, __round; \
okini3939 0:8ba6230eefbd 396 asm ("rotrwi %0,%1,%2" \
okini3939 0:8ba6230eefbd 397 : "=r" (__result) \
okini3939 0:8ba6230eefbd 398 : "r" (lo), "i" (MAD_F_SCALEBITS)); \
okini3939 0:8ba6230eefbd 399 asm ("extrwi %0,%1,1,0" \
okini3939 0:8ba6230eefbd 400 : "=r" (__round) \
okini3939 0:8ba6230eefbd 401 : "r" (__result)); \
okini3939 0:8ba6230eefbd 402 asm ("insrwi %0,%1,%2,0" \
okini3939 0:8ba6230eefbd 403 : "+r" (__result) \
okini3939 0:8ba6230eefbd 404 : "r" (hi), "i" (MAD_F_SCALEBITS)); \
okini3939 0:8ba6230eefbd 405 asm ("add %0,%1,%2" \
okini3939 0:8ba6230eefbd 406 : "=r" (__result) \
okini3939 0:8ba6230eefbd 407 : "%r" (__result), "r" (__round)); \
okini3939 0:8ba6230eefbd 408 __result; \
okini3939 0:8ba6230eefbd 409 })
okini3939 0:8ba6230eefbd 410 # else
okini3939 0:8ba6230eefbd 411 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 412 ({ mad_fixed_t __result; \
okini3939 0:8ba6230eefbd 413 asm ("rotrwi %0,%1,%2" \
okini3939 0:8ba6230eefbd 414 : "=r" (__result) \
okini3939 0:8ba6230eefbd 415 : "r" (lo), "i" (MAD_F_SCALEBITS)); \
okini3939 0:8ba6230eefbd 416 asm ("insrwi %0,%1,%2,0" \
okini3939 0:8ba6230eefbd 417 : "+r" (__result) \
okini3939 0:8ba6230eefbd 418 : "r" (hi), "i" (MAD_F_SCALEBITS)); \
okini3939 0:8ba6230eefbd 419 __result; \
okini3939 0:8ba6230eefbd 420 })
okini3939 0:8ba6230eefbd 421 # endif
okini3939 0:8ba6230eefbd 422
okini3939 0:8ba6230eefbd 423 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 424
okini3939 0:8ba6230eefbd 425 /* --- Default ------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 426
okini3939 0:8ba6230eefbd 427 # elif defined(FPM_DEFAULT)
okini3939 0:8ba6230eefbd 428
okini3939 0:8ba6230eefbd 429 /*
okini3939 0:8ba6230eefbd 430 * This version is the most portable but it loses significant accuracy.
okini3939 0:8ba6230eefbd 431 * Furthermore, accuracy is biased against the second argument, so care
okini3939 0:8ba6230eefbd 432 * should be taken when ordering operands.
okini3939 0:8ba6230eefbd 433 *
okini3939 0:8ba6230eefbd 434 * The scale factors are constant as this is not used with SSO.
okini3939 0:8ba6230eefbd 435 *
okini3939 0:8ba6230eefbd 436 * Pre-rounding is required to stay within the limits of compliance.
okini3939 0:8ba6230eefbd 437 */
okini3939 0:8ba6230eefbd 438 # if defined(OPT_SPEED)
okini3939 0:8ba6230eefbd 439 # define mad_f_mul(x, y) (((x) >> 12) * ((y) >> 16))
okini3939 0:8ba6230eefbd 440 # else
okini3939 0:8ba6230eefbd 441 # define mad_f_mul(x, y) ((((x) + (1L << 11)) >> 12) * \
okini3939 0:8ba6230eefbd 442 (((y) + (1L << 15)) >> 16))
okini3939 0:8ba6230eefbd 443 # endif
okini3939 0:8ba6230eefbd 444
okini3939 0:8ba6230eefbd 445 /* ------------------------------------------------------------------------- */
okini3939 0:8ba6230eefbd 446
okini3939 0:8ba6230eefbd 447 # else
okini3939 0:8ba6230eefbd 448 # error "no FPM selected"
okini3939 0:8ba6230eefbd 449 # endif
okini3939 0:8ba6230eefbd 450
okini3939 0:8ba6230eefbd 451 /* default implementations */
okini3939 0:8ba6230eefbd 452
okini3939 0:8ba6230eefbd 453 # if !defined(mad_f_mul)
okini3939 0:8ba6230eefbd 454 # define mad_f_mul(x, y) \
okini3939 0:8ba6230eefbd 455 ({ register mad_fixed64hi_t __hi; \
okini3939 0:8ba6230eefbd 456 register mad_fixed64lo_t __lo; \
okini3939 0:8ba6230eefbd 457 MAD_F_MLX(__hi, __lo, (x), (y)); \
okini3939 0:8ba6230eefbd 458 mad_f_scale64(__hi, __lo); \
okini3939 0:8ba6230eefbd 459 })
okini3939 0:8ba6230eefbd 460 # endif
okini3939 0:8ba6230eefbd 461
okini3939 0:8ba6230eefbd 462 # if !defined(MAD_F_MLA)
okini3939 0:8ba6230eefbd 463 # define MAD_F_ML0(hi, lo, x, y) ((lo) = mad_f_mul((x), (y)))
okini3939 0:8ba6230eefbd 464 # define MAD_F_MLA(hi, lo, x, y) ((lo) += mad_f_mul((x), (y)))
okini3939 0:8ba6230eefbd 465 # define MAD_F_MLN(hi, lo) ((lo) = -(lo))
okini3939 0:8ba6230eefbd 466 # define MAD_F_MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo))
okini3939 0:8ba6230eefbd 467 # endif
okini3939 0:8ba6230eefbd 468
okini3939 0:8ba6230eefbd 469 # if !defined(MAD_F_ML0)
okini3939 0:8ba6230eefbd 470 # define MAD_F_ML0(hi, lo, x, y) MAD_F_MLX((hi), (lo), (x), (y))
okini3939 0:8ba6230eefbd 471 # endif
okini3939 0:8ba6230eefbd 472
okini3939 0:8ba6230eefbd 473 # if !defined(MAD_F_MLN)
okini3939 0:8ba6230eefbd 474 # define MAD_F_MLN(hi, lo) ((hi) = ((lo) = -(lo)) ? ~(hi) : -(hi))
okini3939 0:8ba6230eefbd 475 # endif
okini3939 0:8ba6230eefbd 476
okini3939 0:8ba6230eefbd 477 # if !defined(MAD_F_MLZ)
okini3939 0:8ba6230eefbd 478 # define MAD_F_MLZ(hi, lo) mad_f_scale64((hi), (lo))
okini3939 0:8ba6230eefbd 479 # endif
okini3939 0:8ba6230eefbd 480
okini3939 0:8ba6230eefbd 481 # if !defined(mad_f_scale64)
okini3939 0:8ba6230eefbd 482 # if defined(OPT_ACCURACY)
okini3939 0:8ba6230eefbd 483 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 484 ((((mad_fixed_t) \
okini3939 0:8ba6230eefbd 485 (((hi) << (32 - (MAD_F_SCALEBITS - 1))) | \
okini3939 0:8ba6230eefbd 486 ((lo) >> (MAD_F_SCALEBITS - 1)))) + 1) >> 1)
okini3939 0:8ba6230eefbd 487 # else
okini3939 0:8ba6230eefbd 488 # define mad_f_scale64(hi, lo) \
okini3939 0:8ba6230eefbd 489 ((mad_fixed_t) \
okini3939 0:8ba6230eefbd 490 (((hi) << (32 - MAD_F_SCALEBITS)) | \
okini3939 0:8ba6230eefbd 491 ((lo) >> MAD_F_SCALEBITS)))
okini3939 0:8ba6230eefbd 492 # endif
okini3939 0:8ba6230eefbd 493 # define MAD_F_SCALEBITS MAD_F_FRACBITS
okini3939 0:8ba6230eefbd 494 # endif
okini3939 0:8ba6230eefbd 495
okini3939 0:8ba6230eefbd 496 /* C routines */
okini3939 0:8ba6230eefbd 497
okini3939 0:8ba6230eefbd 498 mad_fixed_t mad_f_abs(mad_fixed_t);
okini3939 0:8ba6230eefbd 499 mad_fixed_t mad_f_div(mad_fixed_t, mad_fixed_t);
okini3939 0:8ba6230eefbd 500
okini3939 0:8ba6230eefbd 501 # endif