imgresample.c

mediastreamer2是开源的网络传输媒体流的库

/*
 * High quality image resampling with polyphase filters
 * Copyright (c) 2001 Fabrice Bellard.
 *
 * 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 imgresample.c
 * High quality image resampling with polyphase filters .
 */

#include "avcodec.h"
#include "swscale.h"
#include "dsputil.h"

#ifdef HAVE_ALTIVEC
#include "ppc/imgresample_altivec.h"
#endif

#define NB_COMPONENTS 3

#define PHASE_BITS 4
#define NB_PHASES  (1 << PHASE_BITS)
#define NB_TAPS    4
#define FCENTER    1  /* index of the center of the filter */
//#define TEST    1  /* Test it */

#define POS_FRAC_BITS 16
#define POS_FRAC      (1 << POS_FRAC_BITS)
/* 6 bits precision is needed for MMX */
#define FILTER_BITS   8

#define LINE_BUF_HEIGHT (NB_TAPS * 4)

struct SwsContext {
    AVClass *av_class;
    struct ImgReSampleContext *resampling_ctx;
    enum PixelFormat src_pix_fmt, dst_pix_fmt;
};

struct ImgReSampleContext {
    int iwidth, iheight, owidth, oheight;
    int topBand, bottomBand, leftBand, rightBand;
    int padtop, padbottom, padleft, padright;
    int pad_owidth, pad_oheight;
    int h_incr, v_incr;
    DECLARE_ALIGNED_8(int16_t, h_filters[NB_PHASES][NB_TAPS]); /* horizontal filters */
    DECLARE_ALIGNED_8(int16_t, v_filters[NB_PHASES][NB_TAPS]); /* vertical filters */
    uint8_t *line_buf;
};

void av_build_filter(int16_t *filter, double factor, int tap_count, int phase_count, int scale, int type);

static inline int get_phase(int pos)
{
    return ((pos) >> (POS_FRAC_BITS - PHASE_BITS)) & ((1 << PHASE_BITS) - 1);
}

/* This function must be optimized */
static void h_resample_fast(uint8_t *dst, int dst_width, const uint8_t *src,
                            int src_width, int src_start, int src_incr,
                            int16_t *filters)
{
    int src_pos, phase, sum, i;
    const uint8_t *s;
    int16_t *filter;

    src_pos = src_start;
    for(i=0;i<dst_width;i++) {
#ifdef TEST
        /* test */
        if ((src_pos >> POS_FRAC_BITS) < 0 ||
            (src_pos >> POS_FRAC_BITS) > (src_width - NB_TAPS))
            av_abort();
#endif
        s = src + (src_pos >> POS_FRAC_BITS);
        phase = get_phase(src_pos);
        filter = filters + phase * NB_TAPS;
#if NB_TAPS == 4
        sum = s[0] * filter[0] +
            s[1] * filter[1] +
            s[2] * filter[2] +
            s[3] * filter[3];
#else
        {
            int j;
            sum = 0;
            for(j=0;j<NB_TAPS;j++)
                sum += s[j] * filter[j];
        }
#endif
        sum = sum >> FILTER_BITS;
        if (sum < 0)
            sum = 0;
        else if (sum > 255)
            sum = 255;
        dst[0] = sum;
        src_pos += src_incr;
        dst++;
    }
}

/* This function must be optimized */
static void v_resample(uint8_t *dst, int dst_width, const uint8_t *src,
                       int wrap, int16_t *filter)
{
    int sum, i;
    const uint8_t *s;

    s = src;
    for(i=0;i<dst_width;i++) {
#if NB_TAPS == 4
        sum = s[0 * wrap] * filter[0] +
            s[1 * wrap] * filter[1] +
            s[2 * wrap] * filter[2] +
            s[3 * wrap] * filter[3];
#else
        {
            int j;
            uint8_t *s1 = s;

            sum = 0;
            for(j=0;j<NB_TAPS;j++) {
                sum += s1[0] * filter[j];
                s1 += wrap;
            }
        }
#endif
        sum = sum >> FILTER_BITS;
        if (sum < 0)
            sum = 0;
        else if (sum > 255)
            sum = 255;
        dst[0] = sum;
        dst++;
        s++;
    }
}

#ifdef HAVE_MMX

#include "i386/mmx.h"

#define FILTER4(reg) \
{\
        s = src + (src_pos >> POS_FRAC_BITS);\
        phase = get_phase(src_pos);\
        filter = filters + phase * NB_TAPS;\
        movq_m2r(*s, reg);\
        punpcklbw_r2r(mm7, reg);\
        movq_m2r(*filter, mm6);\
        pmaddwd_r2r(reg, mm6);\
        movq_r2r(mm6, reg);\
        psrlq_i2r(32, reg);\
        paddd_r2r(mm6, reg);\
        psrad_i2r(FILTER_BITS, reg);\
        src_pos += src_incr;\
}

#define DUMP(reg) movq_r2m(reg, tmp); printf(#reg "=%016"PRIx64"\n", tmp.uq);

/* XXX: do four pixels at a time */
static void h_resample_fast4_mmx(uint8_t *dst, int dst_width,
                                 const uint8_t *src, int src_width,
                                 int src_start, int src_incr, int16_t *filters)
{
    int src_pos, phase;
    const uint8_t *s;
    int16_t *filter;
    mmx_t tmp;

    src_pos = src_start;
    pxor_r2r(mm7, mm7);

    while (dst_width >= 4) {

        FILTER4(mm0);
        FILTER4(mm1);
        FILTER4(mm2);
        FILTER4(mm3);

        packuswb_r2r(mm7, mm0);
        packuswb_r2r(mm7, mm1);
        packuswb_r2r(mm7, mm3);
        packuswb_r2r(mm7, mm2);
        movq_r2m(mm0, tmp);
        dst[0] = tmp.ub[0];
        movq_r2m(mm1, tmp);
        dst[1] = tmp.ub[0];
        movq_r2m(mm2, tmp);
        dst[2] = tmp.ub[0];
        movq_r2m(mm3, tmp);
        dst[3] = tmp.ub[0];
        dst += 4;
        dst_width -= 4;
    }
    while (dst_width > 0) {
        FILTER4(mm0);
        packuswb_r2r(mm7, mm0);
        movq_r2m(mm0, tmp);
        dst[0] = tmp.ub[0];
        dst++;
        dst_width--;
    }
    emms();
}

static void v_resample4_mmx(uint8_t *dst, int dst_width, const uint8_t *src,
                            int wrap, int16_t *filter)
{
    int sum, i, v;
    const uint8_t *s;
    mmx_t tmp;
    mmx_t coefs[4];

    for(i=0;i<4;i++) {
        v = filter[i];
        coefs[i].uw[0] = v;
        coefs[i].uw[1] = v;
        coefs[i].uw[2] = v;
        coefs[i].uw[3] = v;
    }

    pxor_r2r(mm7, mm7);
    s = src;
    while (dst_width >= 4) {
        movq_m2r(s[0 * wrap], mm0);
        punpcklbw_r2r(mm7, mm0);
        movq_m2r(s[1 * wrap], mm1);
        punpcklbw_r2r(mm7, mm1);
        movq_m2r(s[2 * wrap], mm2);
        punpcklbw_r2r(mm7, mm2);
        movq_m2r(s[3 * wrap], mm3);
        punpcklbw_r2r(mm7, mm3);

        pmullw_m2r(coefs[0], mm0);
        pmullw_m2r(coefs[1], mm1);
        pmullw_m2r(coefs[2], mm2);
        pmullw_m2r(coefs[3], mm3);

        paddw_r2r(mm1, mm0);
        paddw_r2r(mm3, mm2);
        paddw_r2r(mm2, mm0);
        psraw_i2r(FILTER_BITS, mm0);

        packuswb_r2r(mm7, mm0);
        movq_r2m(mm0, tmp);

        *(uint32_t *)dst = tmp.ud[0];
        dst += 4;
        s += 4;
        dst_width -= 4;
    }
    while (dst_width > 0) {
        sum = s[0 * wrap] * filter[0] +
            s[1 * wrap] * filter[1] +
            s[2 * wrap] * filter[2] +
            s[3 * wrap] * filter[3];
        sum = sum >> FILTER_BITS;
        if (sum < 0)
            sum = 0;
        else if (sum > 255)
            sum = 255;
        dst[0] = sum;
        dst++;
        s++;
        dst_width--;
    }
    emms();
}
#endif /* HAVE_MMX */

/* slow version to handle limit cases. Does not need optimization */
static void h_resample_slow(uint8_t *dst, int dst_width,
                            const uint8_t *src, int src_width,
                            int src_start, int src_incr, int16_t *filters)
{
    int src_pos, phase, sum, j, v, i;
    const uint8_t *s, *src_end;
    int16_t *filter;

    src_end = src + src_width;
    src_pos = src_start;
    for(i=0;i<dst_width;i++) {
        s = src + (src_pos >> POS_FRAC_BITS);
        phase = get_phase(src_pos);
        filter = filters + phase * NB_TAPS;
        sum = 0;
        for(j=0;j<NB_TAPS;j++) {
            if (s < src)
                v = src[0];
            else if (s >= src_end)
                v = src_end[-1];
            else
                v = s[0];
            sum += v * filter[j];
            s++;
        }
        sum = sum >> FILTER_BITS;
        if (sum < 0)
            sum = 0;
        else if (sum > 255)
            sum = 255;
        dst[0] = sum;
        src_pos += src_incr;
        dst++;
    }
}

static void h_resample(uint8_t *dst, int dst_width, const uint8_t *src,
                       int src_width, int src_start, int src_incr,
                       int16_t *filters)
{
    int n, src_end;

    if (src_start < 0) {
        n = (0 - src_start + src_incr - 1) / src_incr;
        h_resample_slow(dst, n, src, src_width, src_start, src_incr, filters);
        dst += n;
        dst_width -= n;
        src_start += n * src_incr;
    }
    src_end = src_start + dst_width * src_incr;
    if (src_end > ((src_width - NB_TAPS) << POS_FRAC_BITS)) {
        n = (((src_width - NB_TAPS + 1) << POS_FRAC_BITS) - 1 - src_start) /
            src_incr;
    } else {
        n = dst_width;
    }
#ifdef HAVE_MMX
    if ((mm_flags & MM_MMX) && NB_TAPS == 4)
        h_resample_fast4_mmx(dst, n,
                             src, src_width, src_start, src_incr, filters);
    else
#endif
        h_resample_fast(dst, n,
                        src, src_width, src_start, src_incr, filters);
    if (n < dst_width) {
        dst += n;
        dst_width -= n;
        src_start += n * src_incr;
        h_resample_slow(dst, dst_width,
                        src, src_width, src_start, src_incr, filters);
    }
}

static void component_resample(ImgReSampleContext *s,
                               uint8_t *output, int owrap, int owidth, int oheight,
                               uint8_t *input, int iwrap, int iwidth, int iheight)
{
    int src_y, src_y1, last_src_y, ring_y, phase_y, y1, y;
    uint8_t *new_line, *src_line;

    last_src_y = - FCENTER - 1;
    /* position of the bottom of the filter in the source image */
    src_y = (last_src_y + NB_TAPS) * POS_FRAC;
    ring_y = NB_TAPS; /* position in ring buffer */
    for(y=0;y<oheight;y++) {
        /* apply horizontal filter on new lines from input if needed */
        src_y1 = src_y >> POS_FRAC_BITS;
        while (last_src_y < src_y1) {
            if (++ring_y >= LINE_BUF_HEIGHT + NB_TAPS)
                ring_y = NB_TAPS;
            last_src_y++;
            /* handle limit conditions : replicate line (slightly
               inefficient because we filter multiple times) */
            y1 = last_src_y;
            if (y1 < 0) {
                y1 = 0;
            } else if (y1 >= iheight) {
                y1 = iheight - 1;
            }
            src_line = input + y1 * iwrap;
            new_line = s->line_buf + ring_y * owidth;
            /* apply filter and handle limit cases correctly */
            h_resample(new_line, owidth,
                       src_line, iwidth, - FCENTER * POS_FRAC, s->h_incr,
                       &s->h_filters[0][0]);
            /* handle ring buffer wrapping */
            if (ring_y >= LINE_BUF_HEIGHT) {
                memcpy(s->line_buf + (ring_y - LINE_BUF_HEIGHT) * owidth,
                       new_line, owidth);
            }
        }
        /* apply vertical filter */
        phase_y = get_phase(src_y);
#ifdef HAVE_MMX
        /* desactivated MMX because loss of precision */
        if ((mm_flags & MM_MMX) && NB_TAPS == 4 && 0)
            v_resample4_mmx(output, owidth,
                            s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth,
                            &s->v_filters[phase_y][0]);
        else
#endif
#ifdef HAVE_ALTIVEC
            if ((mm_flags & MM_ALTIVEC) && NB_TAPS == 4 && FILTER_BITS <= 6)
                v_resample16_altivec(output, owidth,
                                s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth,
                                &s->v_filters[phase_y][0]);