view intfloat_readwrite.c @ 728:1fa3820b1a84 libavutil

ARM asm for AV_RN*() ARMv6 and later support unaligned loads and stores for single word/halfword but not double/multiple. GCC is ignorant of this and will always use bytewise accesses for unaligned data. Casting to an int32_t pointer is dangerous since a load/store double or multiple instruction might be used (this happens with some code in FFmpeg). Implementing the AV_[RW]* macros with inline asm using only supported instructions gives fast and safe unaligned accesses. ARM RVCT does the right thing with generic code. This gives an overall speedup of up to 10%.

/*
 * portable IEEE float/double read/write functions
 *
 * Copyright (c) 2005 Michael Niedermayer <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file libavutil/intfloat_readwrite.c
 * portable IEEE float/double read/write functions
 */

#include "common.h"
#include "intfloat_readwrite.h"

double av_int2dbl(int64_t v){
    if(v+v > 0xFFEULL<<52)
        return 0.0/0.0;
    return ldexp(((v&((1LL<<52)-1)) + (1LL<<52)) * (v>>63|1), (v>>52&0x7FF)-1075);
}

float av_int2flt(int32_t v){
    if(v+v > 0xFF000000U)
        return 0.0/0.0;
    return ldexp(((v&0x7FFFFF) + (1<<23)) * (v>>31|1), (v>>23&0xFF)-150);
}

double av_ext2dbl(const AVExtFloat ext){
    uint64_t m = 0;
    int e, i;

    for (i = 0; i < 8; i++)
        m = (m<<8) + ext.mantissa[i];
    e = (((int)ext.exponent[0]&0x7f)<<8) | ext.exponent[1];
    if (e == 0x7fff && m)
        return 0.0/0.0;
    e -= 16383 + 63;        /* In IEEE 80 bits, the whole (i.e. 1.xxxx)
                             * mantissa bit is written as opposed to the
                             * single and double precision formats. */
    if (ext.exponent[0]&0x80)
        m= -m;
    return ldexp(m, e);
}

int64_t av_dbl2int(double d){
    int e;
    if     ( !d) return 0;
    else if(d-d) return 0x7FF0000000000000LL + ((int64_t)(d<0)<<63) + (d!=d);
    d= frexp(d, &e);
    return (int64_t)(d<0)<<63 | (e+1022LL)<<52 | (int64_t)((fabs(d)-0.5)*(1LL<<53));
}

int32_t av_flt2int(float d){
    int e;
    if     ( !d) return 0;
    else if(d-d) return 0x7F800000 + ((d<0)<<31) + (d!=d);
    d= frexp(d, &e);
    return (d<0)<<31 | (e+126)<<23 | (int64_t)((fabs(d)-0.5)*(1<<24));
}

AVExtFloat av_dbl2ext(double d){
    struct AVExtFloat ext= {{0}};
    int e, i; double f; uint64_t m;

    f = fabs(frexp(d, &e));
    if (f >= 0.5 && f < 1) {
        e += 16382;
        ext.exponent[0] = e>>8;
        ext.exponent[1] = e;
        m = (uint64_t)ldexp(f, 64);
        for (i=0; i < 8; i++)
            ext.mantissa[i] = m>>(56-(i<<3));
    } else if (f != 0.0) {
        ext.exponent[0] = 0x7f; ext.exponent[1] = 0xff;
        if (f != 1/0.0)
            ext.mantissa[0] = ~0;
    }
    if (d < 0)
        ext.exponent[0] |= 0x80;
    return ext;
}