imgresample.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.7平台上编译为嵌入式图形界面操作系统。

/*
 * High quality image resampling with polyphase filters 
 * Copyright (c) 2001 Fabrice Bellard.
 *
 * This library 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 of the License, or (at your option) any later version.
 *
 * This library 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 this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
#include "avcodec.h"
#include "dsputil.h"

#ifdef USE_FASTMEMCPY
#include "fastmemcpy.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 ImgReSampleContext {
    int iwidth, iheight, owidth, oheight, topBand, bottomBand, leftBand, rightBand;
    int h_incr, v_incr;
    INT16 h_filters[NB_PHASES][NB_TAPS] __align8; /* horizontal filters */
    INT16 v_filters[NB_PHASES][NB_TAPS] __align8; /* vertical filters */
    UINT8 *line_buf;
};

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 *dst, int dst_width, UINT8 *src, int src_width,
                            int src_start, int src_incr, INT16 *filters)
{
    int src_pos, phase, sum, i;
    UINT8 *s;
    INT16 *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 *dst, int dst_width, UINT8 *src, int wrap, 
                       INT16 *filter)
{
    int sum, i;
    UINT8 *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 *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 "=%016Lx\n", tmp.uq);

/* XXX: do four pixels at a time */
static void h_resample_fast4_mmx(UINT8 *dst, int dst_width, UINT8 *src, int src_width,
                                 int src_start, int src_incr, INT16 *filters)
{
    int src_pos, phase;
    UINT8 *s;
    INT16 *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 *dst, int dst_width, UINT8 *src, int wrap, 
                            INT16 *filter)
{
    int sum, i, v;
    UINT8 *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 *)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

/* slow version to handle limit cases. Does not need optimisation */
static void h_resample_slow(UINT8 *dst, int dst_width, UINT8 *src, int src_width,
                            int src_start, int src_incr, INT16 *filters)
{
    int src_pos, phase, sum, j, v, i;
    UINT8 *s, *src_end;
    INT16 *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 *dst, int dst_width, UINT8 *src, int src_width,
                       int src_start, int src_incr, INT16 *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 {