--- mbtransquant.c 2002/03/26 11:16:08 1.2 +++ mbtransquant.c 2002/04/13 16:30:02 1.5 @@ -42,6 +42,8 @@ * * * Revision history: * * * + * 29.03.2002 interlacing speedup - used transfer strides instead of + * manual field-to-frame conversion * 26.03.2002 interlacing support - moved transfers outside loops * 22.12.2001 get_dc_scaler() moved to common.h * 19.11.2001 introduced coefficient thresholding (Isibaar) * @@ -49,6 +51,8 @@ * * ******************************************************************************/ +#include + #include "../portab.h" #include "mbfunctions.h" @@ -69,30 +73,33 @@ /* this isnt pretty, but its better than 20 ifdefs */ void MBTransQuantIntra(const MBParam *pParam, - MACROBLOCK * pMB, + MACROBLOCK * pMB, const uint32_t x_pos, const uint32_t y_pos, - int16_t data[][64], - int16_t qcoeff[][64], - IMAGE * const pCurrent) + int16_t data[6*64], + int16_t qcoeff[6*64], + IMAGE * const pCurrent) { - const uint32_t stride = pParam->edged_width; + + uint32_t stride = pParam->edged_width; + uint32_t stride2 = stride / 2; + uint32_t next_block = stride * 8; uint32_t i; uint32_t iQuant = pParam->quant; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; - pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); - pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); - pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); + pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); + pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); + pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); start_timer(); - transfer_8to16copy(data[0], pY_Cur, stride); - transfer_8to16copy(data[1], pY_Cur + 8, stride); - transfer_8to16copy(data[2], pY_Cur + 8 * stride, stride); - transfer_8to16copy(data[3], pY_Cur + 8 * stride + 8, stride); - transfer_8to16copy(data[4], pU_Cur, stride / 2); - transfer_8to16copy(data[5], pV_Cur, stride / 2); + transfer_8to16copy(&data[0*64], pY_Cur, stride); + transfer_8to16copy(&data[1*64], pY_Cur + 8, stride); + transfer_8to16copy(&data[2*64], pY_Cur + next_block, stride); + transfer_8to16copy(&data[3*64], pY_Cur + next_block + 8,stride); + transfer_8to16copy(&data[4*64], pU_Cur, stride2); + transfer_8to16copy(&data[5*64], pV_Cur, stride2); stop_transfer_timer(); start_timer(); @@ -108,71 +115,74 @@ uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); start_timer(); - fdct(data[i]); + fdct(&data[i*64]); stop_dct_timer(); if (pParam->quant_type == H263_QUANT) { start_timer(); - quant_intra(qcoeff[i], data[i], iQuant, iDcScaler); + quant_intra(&qcoeff[i*64], &data[i*64], iQuant, iDcScaler); stop_quant_timer(); start_timer(); - dequant_intra(data[i], qcoeff[i], iQuant, iDcScaler); + dequant_intra(&data[i*64], &qcoeff[i*64], iQuant, iDcScaler); stop_iquant_timer(); } else { start_timer(); - quant4_intra(qcoeff[i], data[i], iQuant, iDcScaler); + quant4_intra(&qcoeff[i*64], &data[i*64], iQuant, iDcScaler); stop_quant_timer(); start_timer(); - dequant4_intra(data[i], qcoeff[i], iQuant, iDcScaler); + dequant4_intra(&data[i*64], &qcoeff[i*64], iQuant, iDcScaler); stop_iquant_timer(); } start_timer(); - idct(data[i]); + idct(&data[i*64]); stop_idct_timer(); - } + } - start_timer(); if (pMB->field_dct) { - MBFieldToFrame(data); + next_block = stride; + stride *= 2; } - stop_interlacing_timer(); start_timer(); - transfer_16to8copy(pY_Cur, data[0], stride); - transfer_16to8copy(pY_Cur + 8, data[1], stride); - transfer_16to8copy(pY_Cur + 8 * stride, data[2], stride); - transfer_16to8copy(pY_Cur + 8 + 8 * stride, data[3], stride); - transfer_16to8copy(pU_Cur, data[4], stride / 2); - transfer_16to8copy(pV_Cur, data[5], stride / 2); + transfer_16to8copy(pY_Cur, &data[0*64], stride); + transfer_16to8copy(pY_Cur + 8, &data[1*64], stride); + transfer_16to8copy(pY_Cur + next_block, &data[2*64], stride); + transfer_16to8copy(pY_Cur + next_block + 8, &data[3*64], stride); + transfer_16to8copy(pU_Cur, &data[4*64], stride2); + transfer_16to8copy(pV_Cur, &data[5*64], stride2); stop_transfer_timer(); + } uint8_t MBTransQuantInter(const MBParam *pParam, - MACROBLOCK * pMB, - const uint32_t x_pos, const uint32_t y_pos, - int16_t data[][64], - int16_t qcoeff[][64], - IMAGE * const pCurrent) + MACROBLOCK * pMB, + const uint32_t x_pos, const uint32_t y_pos, + int16_t data[6*64], + int16_t qcoeff[6*64], + IMAGE * const pCurrent) { - const uint32_t stride = pParam->edged_width; - uint32_t i; - uint32_t iQuant = pParam->quant; + + uint32_t stride = pParam->edged_width; + uint32_t stride2 = stride / 2; + uint32_t next_block = stride * 8; + uint32_t i; + uint32_t iQuant = pParam->quant; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; - uint8_t cbp = 0; + uint8_t cbp = 0; uint32_t sum; - pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); - pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); - pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); + pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); + pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); + pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); start_timer(); pMB->field_dct = 0; @@ -182,26 +192,26 @@ } stop_interlacing_timer(); - for(i = 0; i < 6; i++) + for(i = 0; i < 6; i++) { /* - no need to transfer 8->16-bit - (this is performed already in motion compensation) - */ + * no need to transfer 8->16-bit + * (this is performed already in motion compensation) + */ start_timer(); - fdct(data[i]); + fdct(&data[i*64]); stop_dct_timer(); if (pParam->quant_type == 0) { start_timer(); - sum = quant_inter(qcoeff[i], data[i], iQuant); + sum = quant_inter(&qcoeff[i*64], &data[i*64], iQuant); stop_quant_timer(); } else { start_timer(); - sum = quant4_inter(qcoeff[i], data[i], iQuant); + sum = quant4_inter(&qcoeff[i*64], &data[i*64], iQuant); stop_quant_timer(); } @@ -210,58 +220,58 @@ if (pParam->quant_type == H263_QUANT) { start_timer(); - dequant_inter(data[i], qcoeff[i], iQuant); + dequant_inter(&data[i*64], &qcoeff[i*64], iQuant); stop_iquant_timer(); } else { start_timer(); - dequant4_inter(data[i], qcoeff[i], iQuant); + dequant4_inter(&data[i*64], &qcoeff[i*64], iQuant); stop_iquant_timer(); } cbp |= 1 << (5 - i); start_timer(); - idct(data[i]); + idct(&data[i*64]); stop_idct_timer(); } } - start_timer(); if (pMB->field_dct) { - MBFieldToFrame(data); + next_block = stride; + stride *= 2; } - stop_interlacing_timer(); start_timer(); if (cbp & 32) - transfer_16to8add(pY_Cur, data[0], stride); + transfer_16to8add(pY_Cur, &data[0*64], stride); if (cbp & 16) - transfer_16to8add(pY_Cur + 8, data[1], stride); + transfer_16to8add(pY_Cur + 8, &data[1*64], stride); if (cbp & 8) - transfer_16to8add(pY_Cur + 8 * stride, data[2], stride); + transfer_16to8add(pY_Cur + next_block, &data[2*64], stride); if (cbp & 4) - transfer_16to8add(pY_Cur + 8 + 8 * stride, data[3], stride); + transfer_16to8add(pY_Cur + next_block + 8, &data[3*64], stride); if (cbp & 2) - transfer_16to8add(pU_Cur, data[4], stride / 2); + transfer_16to8add(pU_Cur, &data[4*64], stride2); if (cbp & 1) - transfer_16to8add(pV_Cur, data[5], stride / 2); + transfer_16to8add(pV_Cur, &data[5*64], stride2); stop_transfer_timer(); - return cbp; + return cbp; + } /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ -#define ABS(X) (X)<0 ? -(X) : (X) -uint32_t MBDecideFieldDCT(int16_t data[][64]) +uint32_t MBDecideFieldDCT(int16_t data[6*64]) { - const uint8_t blocks[] = {0, 0, 0, 0, 2, 2, 2, 2}; - const uint8_t lines[] = {0, 16, 32, 48, 0, 16, 32, 48}; + + const uint8_t blocks[] = {0*64, 0*64, 0*64, 0*64, 2*64, 2*64, 2*64, 2*64}; + const uint8_t lines[] = {0, 16, 32, 48, 0, 16, 32, 48}; int frame = 0, field = 0; int i, j; @@ -270,15 +280,19 @@ { for (j=0 ; j<8 ; ++j) { - frame += ABS(data[0][(i+1)*8 + j] - data[0][i*8 + j]); - frame += ABS(data[1][(i+1)*8 + j] - data[1][i*8 + j]); - frame += ABS(data[2][(i+1)*8 + j] - data[2][i*8 + j]); - frame += ABS(data[3][(i+1)*8 + j] - data[3][i*8 + j]); - - field += ABS(data[blocks[i+1]][lines[i+1] + j] - data[blocks[i]][lines[i] + j]); - field += ABS(data[blocks[i+1]][lines[i+1] + 8 + j] - data[blocks[i]][lines[i] + 8 + j]); - field += ABS(data[blocks[i+1]+1][lines[i+1] + j] - data[blocks[i]+1][lines[i] + j]); - field += ABS(data[blocks[i+1]+1][lines[i+1] + 8 + j] - data[blocks[i]+1][lines[i] + 8 + j]); + frame += ABS(data[0*64 + (i+1)*8 + j] - data[0*64 + i*8 + j]); + frame += ABS(data[1*64 + (i+1)*8 + j] - data[1*64 + i*8 + j]); + frame += ABS(data[2*64 + (i+1)*8 + j] - data[2*64 + i*8 + j]); + frame += ABS(data[3*64 + (i+1)*8 + j] - data[3*64 + i*8 + j]); + + field += ABS(data[blocks[i+1] + lines[i+1] + j] -\ + data[blocks[i ] + lines[i ] + j]); + field += ABS(data[blocks[i+1] + lines[i+1] + 8 + j] -\ + data[blocks[i ] + lines[i ] + 8 + j]); + field += ABS(data[blocks[i+1] + 64 + lines[i+1] + j] -\ + data[blocks[i ] + 64 + lines[i ] + j]); + field += ABS(data[blocks[i+1] + 64 + lines[i+1] + 8 + j] -\ + data[blocks[i ] + 64 + lines[i ] + 8 + j]); } } @@ -294,9 +308,9 @@ /* deinterlace Y blocks vertically */ #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp)) -#define LINE(X,Y) &data[X][Y*8] +#define LINE(X,Y) &data[X*64 + Y*8] -void MBFrameToField(int16_t data[][64]) +void MBFrameToField(int16_t data[6*64]) { int16_t tmp[8]; @@ -356,67 +370,3 @@ MOVLINE(LINE(3,5), LINE(3,3)); MOVLINE(LINE(3,3), tmp); } - - -/* interlace Y blocks vertically */ - -void MBFieldToFrame(int16_t data[][64]) -{ - uint16_t tmp[8]; - - /* left blocks */ - - // 1=8, 8=4, 4=2, 2=1 - MOVLINE(tmp, LINE(0,1)); - MOVLINE(LINE(0,1), LINE(2,0)); - MOVLINE(LINE(2,0), LINE(0,4)); - MOVLINE(LINE(0,4), LINE(0,2)); - MOVLINE(LINE(0,2), tmp); - - // 3=9, 9=12, 12=6, 6=3 - MOVLINE(tmp, LINE(0,3)); - MOVLINE(LINE(0,3), LINE(2,1)); - MOVLINE(LINE(2,1), LINE(2,4)); - MOVLINE(LINE(2,4), LINE(0,6)); - MOVLINE(LINE(0,6), tmp); - - // 5=10, 10=5 - MOVLINE(tmp, LINE(0,5)); - MOVLINE(LINE(0,5), LINE(2,2)); - MOVLINE(LINE(2,2), tmp); - - // 7=11, 11=13, 13=14, 14=7 - MOVLINE(tmp, LINE(0,7)); - MOVLINE(LINE(0,7), LINE(2,3)); - MOVLINE(LINE(2,3), LINE(2,5)); - MOVLINE(LINE(2,5), LINE(2,6)); - MOVLINE(LINE(2,6), tmp); - - /* right blocks */ - - // 1=8, 8=4, 4=2, 2=1 - MOVLINE(tmp, LINE(1,1)); - MOVLINE(LINE(1,1), LINE(3,0)); - MOVLINE(LINE(3,0), LINE(1,4)); - MOVLINE(LINE(1,4), LINE(1,2)); - MOVLINE(LINE(1,2), tmp); - - // 3=9, 9=12, 12=6, 6=3 - MOVLINE(tmp, LINE(1,3)); - MOVLINE(LINE(1,3), LINE(3,1)); - MOVLINE(LINE(3,1), LINE(3,4)); - MOVLINE(LINE(3,4), LINE(1,6)); - MOVLINE(LINE(1,6), tmp); - - // 5=10, 10=5 - MOVLINE(tmp, LINE(1,5)); - MOVLINE(LINE(1,5), LINE(3,2)); - MOVLINE(LINE(3,2), tmp); - - // 7=11, 11=13, 13=14, 14=7 - MOVLINE(tmp, LINE(1,7)); - MOVLINE(LINE(1,7), LINE(3,3)); - MOVLINE(LINE(3,3), LINE(3,5)); - MOVLINE(LINE(3,5), LINE(3,6)); - MOVLINE(LINE(3,6), tmp); -}