nengel@2: /*
nengel@2:  * H.264 IDCT
nengel@2:  * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
nengel@2:  *
nengel@2:  * This file is part of FFmpeg.
nengel@2:  *
nengel@2:  * FFmpeg is free software; you can redistribute it and/or
nengel@2:  * modify it under the terms of the GNU Lesser General Public
nengel@2:  * License as published by the Free Software Foundation; either
nengel@2:  * version 2.1 of the License, or (at your option) any later version.
nengel@2:  *
nengel@2:  * FFmpeg is distributed in the hope that it will be useful,
nengel@2:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
nengel@2:  * Lesser General Public License for more details.
nengel@2:  *
nengel@2:  * You should have received a copy of the GNU Lesser General Public
nengel@2:  * License along with FFmpeg; if not, write to the Free Software
nengel@2:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2:  */
nengel@2: 
nengel@2: /**
nengel@2:  * @file
nengel@2:  * H.264 IDCT.
nengel@2:  * @author Michael Niedermayer <michaelni@gmx.at>
nengel@2:  */
nengel@2: 
nengel@2: #include "dsputil.h"
nengel@2: #include "h264_data.h"
nengel@2: 
nengel@2: static av_always_inline void idct_internal(uint8_t *dst, DCTELEM *block, int stride, int block_stride, int shift, int add){
nengel@2:     int i;
nengel@2:     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
nengel@2: 
nengel@2:     block[0] += 1<<(shift-1);
nengel@2: 
nengel@2:     for(i=0; i<4; i++){
nengel@2:         const int z0=  block[0 + block_stride*i]     +  block[2 + block_stride*i];
nengel@2:         const int z1=  block[0 + block_stride*i]     -  block[2 + block_stride*i];
nengel@2:         const int z2= (block[1 + block_stride*i]>>1) -  block[3 + block_stride*i];
nengel@2:         const int z3=  block[1 + block_stride*i]     + (block[3 + block_stride*i]>>1);
nengel@2: 
nengel@2:         block[0 + block_stride*i]= z0 + z3;
nengel@2:         block[1 + block_stride*i]= z1 + z2;
nengel@2:         block[2 + block_stride*i]= z1 - z2;
nengel@2:         block[3 + block_stride*i]= z0 - z3;
nengel@2:     }
nengel@2: 
nengel@2:     for(i=0; i<4; i++){
nengel@2:         const int z0=  block[i + block_stride*0]     +  block[i + block_stride*2];
nengel@2:         const int z1=  block[i + block_stride*0]     -  block[i + block_stride*2];
nengel@2:         const int z2= (block[i + block_stride*1]>>1) -  block[i + block_stride*3];
nengel@2:         const int z3=  block[i + block_stride*1]     + (block[i + block_stride*3]>>1);
nengel@2: 
nengel@2:         dst[i + 0*stride]= cm[ add*dst[i + 0*stride] + ((z0 + z3) >> shift) ];
nengel@2:         dst[i + 1*stride]= cm[ add*dst[i + 1*stride] + ((z1 + z2) >> shift) ];
nengel@2:         dst[i + 2*stride]= cm[ add*dst[i + 2*stride] + ((z1 - z2) >> shift) ];
nengel@2:         dst[i + 3*stride]= cm[ add*dst[i + 3*stride] + ((z0 - z3) >> shift) ];
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride){
nengel@2:     idct_internal(dst, block, stride, 4, 6, 1);
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block){
nengel@2:     idct_internal(dst, block, stride, 8, 3, 1);
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block){
nengel@2:     idct_internal(dst, block, stride, 8, 3, 0);
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride){
nengel@2:     int i;
nengel@2:     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
nengel@2: 
nengel@2:     block[0] += 32;
nengel@2: 
nengel@2:     for( i = 0; i < 8; i++ )
nengel@2:     {
nengel@2:         const int a0 =  block[0+i*8] + block[4+i*8];
nengel@2:         const int a2 =  block[0+i*8] - block[4+i*8];
nengel@2:         const int a4 = (block[2+i*8]>>1) - block[6+i*8];
nengel@2:         const int a6 = (block[6+i*8]>>1) + block[2+i*8];
nengel@2: 
nengel@2:         const int b0 = a0 + a6;
nengel@2:         const int b2 = a2 + a4;
nengel@2:         const int b4 = a2 - a4;
nengel@2:         const int b6 = a0 - a6;
nengel@2: 
nengel@2:         const int a1 = -block[3+i*8] + block[5+i*8] - block[7+i*8] - (block[7+i*8]>>1);
nengel@2:         const int a3 =  block[1+i*8] + block[7+i*8] - block[3+i*8] - (block[3+i*8]>>1);
nengel@2:         const int a5 = -block[1+i*8] + block[7+i*8] + block[5+i*8] + (block[5+i*8]>>1);
nengel@2:         const int a7 =  block[3+i*8] + block[5+i*8] + block[1+i*8] + (block[1+i*8]>>1);
nengel@2: 
nengel@2:         const int b1 = (a7>>2) + a1;
nengel@2:         const int b3 =  a3 + (a5>>2);
nengel@2:         const int b5 = (a3>>2) - a5;
nengel@2:         const int b7 =  a7 - (a1>>2);
nengel@2: 
nengel@2:         block[0+i*8] = b0 + b7;
nengel@2:         block[7+i*8] = b0 - b7;
nengel@2:         block[1+i*8] = b2 + b5;
nengel@2:         block[6+i*8] = b2 - b5;
nengel@2:         block[2+i*8] = b4 + b3;
nengel@2:         block[5+i*8] = b4 - b3;
nengel@2:         block[3+i*8] = b6 + b1;
nengel@2:         block[4+i*8] = b6 - b1;
nengel@2:     }
nengel@2:     for( i = 0; i < 8; i++ )
nengel@2:     {
nengel@2:         const int a0 =  block[i+0*8] + block[i+4*8];
nengel@2:         const int a2 =  block[i+0*8] - block[i+4*8];
nengel@2:         const int a4 = (block[i+2*8]>>1) - block[i+6*8];
nengel@2:         const int a6 = (block[i+6*8]>>1) + block[i+2*8];
nengel@2: 
nengel@2:         const int b0 = a0 + a6;
nengel@2:         const int b2 = a2 + a4;
nengel@2:         const int b4 = a2 - a4;
nengel@2:         const int b6 = a0 - a6;
nengel@2: 
nengel@2:         const int a1 = -block[i+3*8] + block[i+5*8] - block[i+7*8] - (block[i+7*8]>>1);
nengel@2:         const int a3 =  block[i+1*8] + block[i+7*8] - block[i+3*8] - (block[i+3*8]>>1);
nengel@2:         const int a5 = -block[i+1*8] + block[i+7*8] + block[i+5*8] + (block[i+5*8]>>1);
nengel@2:         const int a7 =  block[i+3*8] + block[i+5*8] + block[i+1*8] + (block[i+1*8]>>1);
nengel@2: 
nengel@2:         const int b1 = (a7>>2) + a1;
nengel@2:         const int b3 =  a3 + (a5>>2);
nengel@2:         const int b5 = (a3>>2) - a5;
nengel@2:         const int b7 =  a7 - (a1>>2);
nengel@2: 
nengel@2:         dst[i + 0*stride] = cm[ dst[i + 0*stride] + ((b0 + b7) >> 6) ];
nengel@2:         dst[i + 1*stride] = cm[ dst[i + 1*stride] + ((b2 + b5) >> 6) ];
nengel@2:         dst[i + 2*stride] = cm[ dst[i + 2*stride] + ((b4 + b3) >> 6) ];
nengel@2:         dst[i + 3*stride] = cm[ dst[i + 3*stride] + ((b6 + b1) >> 6) ];
nengel@2:         dst[i + 4*stride] = cm[ dst[i + 4*stride] + ((b6 - b1) >> 6) ];
nengel@2:         dst[i + 5*stride] = cm[ dst[i + 5*stride] + ((b4 - b3) >> 6) ];
nengel@2:         dst[i + 6*stride] = cm[ dst[i + 6*stride] + ((b2 - b5) >> 6) ];
nengel@2:         dst[i + 7*stride] = cm[ dst[i + 7*stride] + ((b0 - b7) >> 6) ];
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: // assumes all AC coefs are 0
nengel@2: void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride){
nengel@2:     int i, j;
nengel@2:     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
nengel@2:     int dc = (block[0] + 32) >> 6;
nengel@2:     for( j = 0; j < 4; j++ )
nengel@2:     {
nengel@2:         for( i = 0; i < 4; i++ )
nengel@2:             dst[i] = cm[ dst[i] + dc ];
nengel@2:         dst += stride;
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride){
nengel@2:     int i, j;
nengel@2:     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
nengel@2:     int dc = (block[0] + 32) >> 6;
nengel@2:     for( j = 0; j < 8; j++ )
nengel@2:     {
nengel@2:         for( i = 0; i < 8; i++ )
nengel@2:             dst[i] = cm[ dst[i] + dc ];
nengel@2:         dst += stride;
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct_add16_c(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2:     int i;
nengel@2:     for(i=0; i<16; i++){
nengel@2:         int nnz = nnzc[ scan8[i] ];
nengel@2:         if(nnz){
nengel@2:             if(nnz==1 && block[i*16]) ff_h264_idct_dc_add_c(dst + block_offset[i], block + i*16, stride);
nengel@2:             else                      idct_internal        (dst + block_offset[i], block + i*16, stride, 4, 6, 1);
nengel@2:         }
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct_add16intra_c(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2:     int i;
nengel@2:     for(i=0; i<16; i++){
nengel@2:         if(nnzc[ scan8[i] ]) idct_internal        (dst + block_offset[i], block + i*16, stride, 4, 6, 1);
nengel@2:         else if(block[i*16]) ff_h264_idct_dc_add_c(dst + block_offset[i], block + i*16, stride);
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct8_add4_c(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2:     int i;
nengel@2:     for(i=0; i<16; i+=4){
nengel@2:         int nnz = nnzc[ scan8[i] ];
nengel@2:         if(nnz){
nengel@2:             if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_c(dst + block_offset[i], block + i*16, stride);
nengel@2:             else                      ff_h264_idct8_add_c   (dst + block_offset[i], block + i*16, stride);
nengel@2:         }
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: void ff_h264_idct_add8_c(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
nengel@2:     int i;
nengel@2:     for(i=16; i<16+8; i++){
nengel@2:         if(nnzc[ scan8[i] ])
nengel@2:             ff_h264_idct_add_c   (dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2:         else if(block[i*16])
nengel@2:             ff_h264_idct_dc_add_c(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: /**
nengel@2: * IDCT transforms the 16 dc values and dequantizes them.
nengel@2: * @param qp quantization parameter
nengel@2: */
nengel@2: void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qmul){
nengel@2: 	#define stride 16
nengel@2: 	int i;
nengel@2: 	int temp[16]; //FIXME check if this is a good idea
nengel@2: 	static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
nengel@2: 	static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
nengel@2: 
nengel@2: 	//return;
nengel@2: 	for(i=0; i<4; i++){
nengel@2: 		const int offset= y_offset[i];
nengel@2: 		const int z0= block[offset+stride*0] + block[offset+stride*4];
nengel@2: 		const int z1= block[offset+stride*0] - block[offset+stride*4];
nengel@2: 		const int z2= block[offset+stride*1] - block[offset+stride*5];
nengel@2: 		const int z3= block[offset+stride*1] + block[offset+stride*5];
nengel@2: 
nengel@2: 		temp[4*i+0]= z0+z3;
nengel@2: 		temp[4*i+1]= z1+z2;
nengel@2: 		temp[4*i+2]= z1-z2;
nengel@2: 		temp[4*i+3]= z0-z3;
nengel@2: 	}
nengel@2: 
nengel@2: 	for(i=0; i<4; i++){
nengel@2: 		const int offset= x_offset[i];
nengel@2: 		const int z0= temp[4*0+i] + temp[4*2+i];
nengel@2: 		const int z1= temp[4*0+i] - temp[4*2+i];
nengel@2: 		const int z2= temp[4*1+i] - temp[4*3+i];
nengel@2: 		const int z3= temp[4*1+i] + temp[4*3+i];
nengel@2: 
nengel@2: 		block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_residual
nengel@2: 		block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
nengel@2: 		block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
nengel@2: 		block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
nengel@2: 	}
nengel@2: }
nengel@2: 
nengel@2: #undef xStride
nengel@2: #undef stride
nengel@2: 
nengel@2: void chroma_dc_dequant_idct_c(DCTELEM *block, int qmul){
nengel@2: 	const int stride= 16*2;
nengel@2: 	const int xStride= 16;
nengel@2: 	int a,b,c,d,e;
nengel@2: 
nengel@2: 	a= block[stride*0 + xStride*0];
nengel@2: 	b= block[stride*0 + xStride*1];
nengel@2: 	c= block[stride*1 + xStride*0];
nengel@2: 	d= block[stride*1 + xStride*1];
nengel@2: 
nengel@2: 	e= a-b;
nengel@2: 	a= a+b;
nengel@2: 	b= c-d;
nengel@2: 	c= c+d;
nengel@2: 
nengel@2: 	block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
nengel@2: 	block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
nengel@2: 	block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
nengel@2: 	block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
nengel@2: }