/*****************************************************************************
 * i420_rgb16_x86.c : YUV to bitmap RGB conversion module for vlc
 *****************************************************************************
 * Copyright (C) 2000 VLC authors and VideoLAN
 *
 * Authors: Samuel Hocevar <sam@zoy.org>
 *          Damien Fouilleul <damienf@videolan.org>
 *
 * This program 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.
 *
 * This program 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 program; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
 *****************************************************************************/

#ifdef HAVE_CONFIG_H
# include "config.h"
#endif

#include <vlc_common.h>
#include <vlc_filter.h>
#include <vlc_picture.h>
#include <vlc_cpu.h>

#include "i420_rgb.h"
#ifdef PLUGIN_SSE2
# include "i420_rgb_sse2.h"
# define VLC_TARGET VLC_SSE
#endif

/*****************************************************************************
 * SetOffset: build offset array for conversion functions
 *****************************************************************************
 * This function will build an offset array used in later conversion functions.
 * It will also set horizontal and vertical scaling indicators.
 *****************************************************************************/
static void SetOffset( int i_width, int i_height, int i_pic_width,
                       int i_pic_height, bool *pb_hscale,
                       unsigned int *pi_vscale, int *p_offset )
{
    /*
     * Prepare horizontal offset array
     */
    if( i_pic_width - i_width == 0 )
    {   /* No horizontal scaling: YUV conversion is done directly to picture */
        *pb_hscale = 0;
    }
    else if( i_pic_width - i_width > 0 )
    {   /* Prepare scaling array for horizontal extension */
        int i_scale_count = i_pic_width;

        *pb_hscale = 1;
        for( int i_x = i_width; i_x--; )
        {
            while( (i_scale_count -= i_width) > 0 )
            {
                *p_offset++ = 0;
            }
            *p_offset++ = 1;
            i_scale_count += i_pic_width;
        }
    }
    else /* if( i_pic_width - i_width < 0 ) */
    {   /* Prepare scaling array for horizontal reduction */
        int i_scale_count = i_pic_width;

        *pb_hscale = 1;
        for( int i_x = i_pic_width; i_x--; )
        {
            *p_offset = 1;
            while( (i_scale_count -= i_pic_width) > 0 )
            {
                *p_offset += 1;
            }
            p_offset++;
            i_scale_count += i_width;
        }
    }

    /*
     * Set vertical scaling indicator
     */
    if( i_pic_height - i_height == 0 )
        *pi_vscale = 0;
    else if( i_pic_height - i_height > 0 )
        *pi_vscale = 1;
    else /* if( i_pic_height - i_height < 0 ) */
        *pi_vscale = -1;
}

VLC_TARGET
void I420_R5G5B5( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint16_t *p_pic = (uint16_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint16_t *  p_pic_start;       /* beginning of the current line for copy */

    /* Conversion buffer pointer */
    uint16_t *  p_buffer_start;
    uint16_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
               (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint16_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)/16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_16_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_15_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }
            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;

                SSE2_CALL (
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_15_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 2 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)/16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_15_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }
            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;

                SSE2_CALL (
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_15_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 2 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}

VLC_TARGET
void I420_R5G6B5( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint16_t *p_pic = (uint16_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint16_t *  p_pic_start;       /* beginning of the current line for copy */

    /* Conversion buffer pointer */
    uint16_t *  p_buffer_start;
    uint16_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
               (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint16_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)/16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_16_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_16_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }
            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;

                SSE2_CALL (
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_16_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 2 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)/16; i_x--; )
            {
                SSE2_CALL(
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_16_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }
            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;

                SSE2_CALL(
                    SSE2_INIT_16_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_16_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 2 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}

VLC_TARGET
void I420_A8R8G8B8( filter_t *p_filter, picture_t *p_src,
                                            picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint32_t *  p_pic_start;       /* beginning of the current line for copy */
    /* Conversion buffer pointer */
    uint32_t *  p_buffer_start;
    uint32_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width,
               p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height,
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint32_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ARGB_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ARGB_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ARGB_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ARGB_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}

VLC_TARGET
void I420_R8G8B8A8( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint32_t *  p_pic_start;       /* beginning of the current line for copy */
    /* Conversion buffer pointer */
    uint32_t *  p_buffer_start;
    uint32_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
               (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint32_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_RGBA_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_RGBA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_RGBA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_RGBA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}

VLC_TARGET
void I420_B8G8R8A8( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint32_t *  p_pic_start;       /* beginning of the current line for copy */
    /* Conversion buffer pointer */
    uint32_t *  p_buffer_start;
    uint32_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
               (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint32_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_BGRA_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_BGRA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_BGRA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_BGRA_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}

VLC_TARGET
void I420_A8B8G8R8( filter_t *p_filter, picture_t *p_src, picture_t *p_dest )
{
    filter_sys_t *p_sys = p_filter->p_sys;

    /* We got this one from the old arguments */
    uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
    uint8_t  *p_y   = p_src->Y_PIXELS;
    uint8_t  *p_u   = p_src->U_PIXELS;
    uint8_t  *p_v   = p_src->V_PIXELS;

    bool  b_hscale;                         /* horizontal scaling type */
    unsigned int i_vscale;                          /* vertical scaling type */
    unsigned int i_x, i_y;                /* horizontal and vertical indexes */

    int         i_right_margin;
    int         i_rewind;
    int         i_scale_count;                       /* scale modulo counter */
    int         i_chroma_width = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 2; /* chroma width */
    uint32_t *  p_pic_start;       /* beginning of the current line for copy */
    /* Conversion buffer pointer */
    uint32_t *  p_buffer_start;
    uint32_t *  p_buffer;

    /* Offset array pointer */
    int *       p_offset_start = p_sys->p_offset;
    int *       p_offset;

    const int i_source_margin = p_src->p[0].i_pitch
                                 - p_src->p[0].i_visible_pitch
                                 - p_filter->fmt_in.video.i_x_offset;
    const int i_source_margin_c = p_src->p[1].i_pitch
                                 - p_src->p[1].i_visible_pitch
                                 - ( p_filter->fmt_in.video.i_x_offset / 2 );

    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;

    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
    SetOffset( (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width),
               (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height),
               (p_filter->fmt_out.video.i_x_offset + p_filter->fmt_out.video.i_visible_width),
               (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height),
               &b_hscale, &i_vscale, p_offset_start );

    if(b_hscale &&
       AllocateOrGrow(&p_sys->p_buffer, &p_sys->i_buffer_size,
                      p_filter->fmt_in.video.i_x_offset +
                      p_filter->fmt_in.video.i_visible_width,
                      p_sys->i_bytespp))
        return;
    else p_buffer_start = (uint32_t*)p_sys->p_buffer;

    /*
     * Perform conversion
     */
    i_scale_count = ( i_vscale == 1 ) ?
                    (p_filter->fmt_out.video.i_y_offset + p_filter->fmt_out.video.i_visible_height) :
                    (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height);

#ifdef PLUGIN_SSE2

    i_rewind = (-(p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width)) & 15;

    /*
    ** SSE2 128 bits fetch/store instructions are faster
    ** if memory access is 16 bytes aligned
    */

    p_buffer = b_hscale ? p_buffer_start : p_pic;

    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
                    p_dest->p->i_pitch|
                    ((intptr_t)p_y)|
                    ((intptr_t)p_buffer))) )
    {
        /* use faster SSE2 aligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_ALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ABGR_ALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ABGR_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }
    else
    {
        /* use slower SSE2 unaligned fetch and store */
        for( i_y = 0; i_y < (p_filter->fmt_in.video.i_y_offset + p_filter->fmt_in.video.i_visible_height); i_y++ )
        {
            p_pic_start = p_pic;

            for ( i_x = (p_filter->fmt_in.video.i_x_offset + p_filter->fmt_in.video.i_visible_width) / 16; i_x--; )
            {
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ABGR_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
                p_buffer += 16;
            }

            /* Here we do some unaligned reads and duplicate conversions, but
             * at least we have all the pixels */
            if( i_rewind )
            {
                p_y -= i_rewind;
                p_u -= i_rewind >> 1;
                p_v -= i_rewind >> 1;
                p_buffer -= i_rewind;
                SSE2_CALL (
                    SSE2_INIT_32_UNALIGNED
                    SSE2_YUV_MUL
                    SSE2_YUV_ADD
                    SSE2_UNPACK_32_ABGR_UNALIGNED
                );
                p_y += 16;
                p_u += 8;
                p_v += 8;
            }
            SCALE_WIDTH;
            SCALE_HEIGHT( 420, 4 );

            p_y += i_source_margin;
            if( i_y % 2 )
            {
                p_u += i_source_margin_c;
                p_v += i_source_margin_c;
            }
            p_buffer = b_hscale ? p_buffer_start : p_pic;
        }
    }

    /* make sure all SSE2 stores are visible thereafter */
    SSE2_END;
#endif
}