interpolate8x8.c

用ADI DSP汇编指令编写的全内插函数（包括水平、垂直、交叉内插）

/*****************************************************************************
 *
 *	XVID MPEG-4 VIDEO CODEC
 *	- 8x8 block-based halfpel interpolation -
 *
 *  Copyright(C) 2001-2003 Peter Ross <[e-mail]>
 *
 *  This program is free software ; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation ; either version 2 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 General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program ; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 *
 * $Id$
 *
 ****************************************************************************/

/*#include "../portab.h"
#include "../global.h"
#include "interpolate8x8.h"*/
#include "interpolate.h"

/* function pointers */
/*INTERPOLATE8X8_PTR interpolate8x8_halfpel_h;
INTERPOLATE8X8_PTR interpolate8x8_halfpel_v;
INTERPOLATE8X8_PTR interpolate8x8_halfpel_hv;

INTERPOLATE8X8_AVG2_PTR interpolate8x8_avg2;
INTERPOLATE8X8_AVG4_PTR interpolate8x8_avg4;

INTERPOLATE_LOWPASS_PTR interpolate8x8_lowpass_h;
INTERPOLATE_LOWPASS_PTR interpolate8x8_lowpass_v;

INTERPOLATE_LOWPASS_PTR interpolate16x16_lowpass_h;
INTERPOLATE_LOWPASS_PTR interpolate16x16_lowpass_v;

INTERPOLATE_LOWPASS_HV_PTR interpolate8x8_lowpass_hv;
INTERPOLATE_LOWPASS_HV_PTR interpolate16x16_lowpass_hv;

INTERPOLATE8X8_6TAP_LOWPASS_PTR interpolate8x8_6tap_lowpass_h;
INTERPOLATE8X8_6TAP_LOWPASS_PTR interpolate8x8_6tap_lowpass_v;

void interpolate8x8_avg2_c(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint32_t stride, const uint32_t rounding, const uint32_t height)
{
    uint32_t i;
	const int32_t round = 1 - rounding;

    for(i = 0; i < height; i++)
    {
        dst[0] = (src1[0] + src2[0] + round) >> 1;
        dst[1] = (src1[1] + src2[1] + round) >> 1;
        dst[2] = (src1[2] + src2[2] + round) >> 1;
        dst[3] = (src1[3] + src2[3] + round) >> 1;
        dst[4] = (src1[4] + src2[4] + round) >> 1;
        dst[5] = (src1[5] + src2[5] + round) >> 1;
        dst[6] = (src1[6] + src2[6] + round) >> 1;
        dst[7] = (src1[7] + src2[7] + round) >> 1;

        dst += stride;
        src1 += stride;
        src2 += stride;
    }
}

void interpolate8x8_avg4_c(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, const uint8_t *src3, const uint8_t *src4, const uint32_t stride, const uint32_t rounding)
{
    int32_t i;
	const int32_t round = 2 - rounding;

    for(i = 0; i < 8; i++)
    {
        dst[0] = (src1[0] + src2[0] + src3[0] + src4[0] + round) >> 2;
        dst[1] = (src1[1] + src2[1] + src3[1] + src4[1] + round) >> 2;
        dst[2] = (src1[2] + src2[2] + src3[2] + src4[2] + round) >> 2;
        dst[3] = (src1[3] + src2[3] + src3[3] + src4[3] + round) >> 2;
        dst[4] = (src1[4] + src2[4] + src3[4] + src4[4] + round) >> 2;
        dst[5] = (src1[5] + src2[5] + src3[5] + src4[5] + round) >> 2;
        dst[6] = (src1[6] + src2[6] + src3[6] + src4[6] + round) >> 2;
        dst[7] = (src1[7] + src2[7] + src3[7] + src4[7] + round) >> 2;

		dst += stride;
        src1 += stride;
        src2 += stride;
        src3 += stride;
        src4 += stride;
    }
}

/* dst = interpolate(src) */

void
interpolate8x8_halfpel_h_c(uint8_t * const dst,
						   const uint8_t * const src,
						   const uint32_t stride,
						   const uint32_t rounding)
{
	uintptr_t j;

	if (rounding)
		for (j = 0; j < 8*stride; j+=stride)
		{
				dst[j + 0] = (uint8_t)((src[j + 0] + src[j + 1] )>>1);
				dst[j + 1] = (uint8_t)((src[j + 1] + src[j + 2] )>>1);
				dst[j + 2] = (uint8_t)((src[j + 2] + src[j + 3] )>>1);
				dst[j + 3] = (uint8_t)((src[j + 3] + src[j + 4] )>>1);
				dst[j + 4] = (uint8_t)((src[j + 4] + src[j + 5] )>>1);
				dst[j + 5] = (uint8_t)((src[j + 5] + src[j + 6] )>>1);
				dst[j + 6] = (uint8_t)((src[j + 6] + src[j + 7] )>>1);
				dst[j + 7] = (uint8_t)((src[j + 7] + src[j + 8] )>>1);
		}
	else
		for (j = 0; j < 8*stride; j+=stride)		/* forward or backwards? Who knows ... */
		{
				dst[j + 0] = (uint8_t)((src[j + 0] + src[j + 1] + 1)>>1);
				dst[j + 1] = (uint8_t)((src[j + 1] + src[j + 2] + 1)>>1);
				dst[j + 2] = (uint8_t)((src[j + 2] + src[j + 3] + 1)>>1);
				dst[j + 3] = (uint8_t)((src[j + 3] + src[j + 4] + 1)>>1);
				dst[j + 4] = (uint8_t)((src[j + 4] + src[j + 5] + 1)>>1);
				dst[j + 5] = (uint8_t)((src[j + 5] + src[j + 6] + 1)>>1);
				dst[j + 6] = (uint8_t)((src[j + 6] + src[j + 7] + 1)>>1);
				dst[j + 7] = (uint8_t)((src[j + 7] + src[j + 8] + 1)>>1);
		}
}



void
interpolate8x8_halfpel_v_c(uint8_t * const dst,
						   const uint8_t * const src,
						   const uint32_t stride,
						   const uint32_t rounding)
{
	uintptr_t j;


	if (rounding)
		for (j = 0; j < 8*stride; j+=stride)		/* forward is better. Some automatic prefetch perhaps. */
		{
				dst[j + 0] = (uint8_t)((src[j + 0] + src[j + stride + 0] )>>1);
				dst[j + 1] = (uint8_t)((src[j + 1] + src[j + stride + 1] )>>1);
				dst[j + 2] = (uint8_t)((src[j + 2] + src[j + stride + 2] )>>1);
				dst[j + 3] = (uint8_t)((src[j + 3] + src[j + stride + 3] )>>1);
				dst[j + 4] = (uint8_t)((src[j + 4] + src[j + stride + 4] )>>1);
				dst[j + 5] = (uint8_t)((src[j + 5] + src[j + stride + 5] )>>1);
				dst[j + 6] = (uint8_t)((src[j + 6] + src[j + stride + 6] )>>1);
				dst[j + 7] = (uint8_t)((src[j + 7] + src[j + stride + 7] )>>1);
		}
	else
		for (j = 0; j < 8*stride; j+=stride)
		{
				dst[j + 0] = (uint8_t)((src[j + 0] + src[j + stride + 0] + 1)>>1);
				dst[j + 1] = (uint8_t)((src[j + 1] + src[j + stride + 1] + 1)>>1);
				dst[j + 2] = (uint8_t)((src[j + 2] + src[j + stride + 2] + 1)>>1);
				dst[j + 3] = (uint8_t)((src[j + 3] + src[j + stride + 3] + 1)>>1);
				dst[j + 4] = (uint8_t)((src[j + 4] + src[j + stride + 4] + 1)>>1);
				dst[j + 5] = (uint8_t)((src[j + 5] + src[j + stride + 5] + 1)>>1);
				dst[j + 6] = (uint8_t)((src[j + 6] + src[j + stride + 6] + 1)>>1);
				dst[j + 7] = (uint8_t)((src[j + 7] + src[j + stride + 7] + 1)>>1);
		}
}


void
interpolate8x8_halfpel_hv_c(uint8_t * const dst,
							const uint8_t * const src,
							const uint32_t stride,
							const uint32_t rounding)
{
	uintptr_t j;

	if (rounding)
		for (j = 0; j < 8*stride; j+=stride)
		{
				dst[j + 0] = (uint8_t)((src[j+0] + src[j+1] + src[j+stride+0] + src[j+stride+1] +1)>>2);
				dst[j + 1] = (uint8_t)((src[j+1] + src[j+2] + src[j+stride+1] + src[j+stride+2] +1)>>2);
				dst[j + 2] = (uint8_t)((src[j+2] + src[j+3] + src[j+stride+2] + src[j+stride+3] +1)>>2);
				dst[j + 3] = (uint8_t)((src[j+3] + src[j+4] + src[j+stride+3] + src[j+stride+4] +1)>>2);
				dst[j + 4] = (uint8_t)((src[j+4] + src[j+5] + src[j+stride+4] + src[j+stride+5] +1)>>2);
				dst[j + 5] = (uint8_t)((src[j+5] + src[j+6] + src[j+stride+5] + src[j+stride+6] +1)>>2);
				dst[j + 6] = (uint8_t)((src[j+6] + src[j+7] + src[j+stride+6] + src[j+stride+7] +1)>>2);
				dst[j + 7] = (uint8_t)((src[j+7] + src[j+8] + src[j+stride+7] + src[j+stride+8] +1)>>2);
		}
	else
		for (j = 0; j < 8*stride; j+=stride)
		{
				dst[j + 0] = (uint8_t)((src[j+0] + src[j+1] + src[j+stride+0] + src[j+stride+1] +2)>>2);
				dst[j + 1] = (uint8_t)((src[j+1] + src[j+2] + src[j+stride+1] + src[j+stride+2] +2)>>2);
				dst[j + 2] = (uint8_t)((src[j+2] + src[j+3] + src[j+stride+2] + src[j+stride+3] +2)>>2);
				dst[j + 3] = (uint8_t)((src[j+3] + src[j+4] + src[j+stride+3] + src[j+stride+4] +2)>>2);
				dst[j + 4] = (uint8_t)((src[j+4] + src[j+5] + src[j+stride+4] + src[j+stride+5] +2)>>2);
				dst[j + 5] = (uint8_t)((src[j+5] + src[j+6] + src[j+stride+5] + src[j+stride+6] +2)>>2);
				dst[j + 6] = (uint8_t)((src[j+6] + src[j+7] + src[j+stride+6] + src[j+stride+7] +2)>>2);
				dst[j + 7] = (uint8_t)((src[j+7] + src[j+8] + src[j+stride+7] + src[j+stride+8] +2)>>2);
		}
}

/*************************************************************
 * QPEL STUFF STARTS HERE                                    *
 *************************************************************/

/*void interpolate8x8_6tap_lowpass_h_c(uint8_t *dst, uint8_t *src, int32_t stride, int32_t rounding)
{
    int32_t i;
	uint8_t round_add = 16 - rounding;

    for(i = 0; i < 8; i++)
    {

        dst[0] = CLIP((((src[-2] + src[3]) + 5 * (((src[0] + src[1])<<2) - (src[-1] + src[2])) + round_add) >> 5), 0, 255);
        dst[1] = CLIP((((src[-1] + src[4]) + 5 * (((src[1] + src[2])<<2) - (src[0] + src[3])) + round_add) >> 5), 0, 255);
        dst[2] = CLIP((((src[0] + src[5]) + 5 * (((src[2] + src[3])<<2) - (src[1] + src[4])) + round_add) >> 5), 0, 255);
        dst[3] = CLIP((((src[1] + src[6]) + 5 * (((src[3] + src[4])<<2) - (src[2] + src[5])) + round_add) >> 5), 0, 255);
        dst[4] = CLIP((((src[2] + src[7]) + 5 * (((src[4] + src[5])<<2) - (src[3] + src[6])) + round_add) >> 5), 0, 255);
        dst[5] = CLIP((((src[3] + src[8]) + 5 * (((src[5] + src[6])<<2) - (src[4] + src[7])) + round_add) >> 5), 0, 255);
        dst[6] = CLIP((((src[4] + src[9]) + 5 * (((src[6] + src[7])<<2) - (src[5] + src[8])) + round_add) >> 5), 0, 255);
        dst[7] = CLIP((((src[5] + src[10]) + 5 * (((src[7] + src[8])<<2) - (src[6] + src[9])) + round_add) >> 5), 0, 255);

        dst += stride;
        src += stride;
    }
}

void interpolate16x16_lowpass_h_c(uint8_t *dst, uint8_t *src, int32_t stride, int32_t rounding)