motion_est_core_2_optimized.c

基于Linux的ffmepg decoder

			pMB->mv16y_0 = (X<<25)>>25;
			
			pMB->mv16x_1 = pMB->mv16x_0;
			pMB->mv16y_1 = pMB->mv16y_0;
			pMB->mv16x_2 = pMB->mv16x_0;
			pMB->mv16y_2 = pMB->mv16y_0;
			pMB->mv16x_3 = pMB->mv16x_0;
			pMB->mv16y_3 = pMB->mv16y_0;
			pCodec->MVZ = (X == 0);
		} else {
			pMB->mode = MODE_INTER4V;
			X = (pCodec->ME_command_queue0[0+(pCodec->even_odd_1)*4]>>12);
			pMB->mv16x_0 = (X<<18)>>25;
			pMB->mv16y_0 = (X<<25)>>25;
//			*vpe_stop = 0x93000000 | ((pMB->mv16x_0 & 0x7f)<<8) | (pMB->mv16y_0 & 0x7f);
			X = (pCodec->ME_command_queue0[1+(pCodec->even_odd_1)*4]>>12);
			pMB->mv16x_1 = (X<<18)>>25;
			pMB->mv16y_1 = (X<<25)>>25;
//			*vpe_stop = 0x93000000 | ((pMB->mv16x_1 & 0x7f)<<8) | (pMB->mv16y_1 & 0x7f);
			X = (pCodec->ME_command_queue0[2+(pCodec->even_odd_1)*4]>>12);
			pMB->mv16x_2 = (X<<18)>>25;
			pMB->mv16y_2 = (X<<25)>>25;
//			*vpe_stop = 0x93000000 | ((pMB->mv16x_2 & 0x7f)<<8) | (pMB->mv16y_2 & 0x7f);
			X = (pCodec->ME_command_queue0[3+(pCodec->even_odd_1)*4]>>12);
			pMB->mv16x_3 = (X<<18)>>25;
			pMB->mv16y_3 = (X<<25)>>25;
//			*vpe_stop = 0x93000000 | ((pMB->mv16x_3 & 0x7f)<<8) | (pMB->mv16y_3 & 0x7f);
			pCodec->MVZ = 0;
		}
	}
	
	#ifdef DUMP_PMV_RESULT
	switch(pMB->mode)
	  {
	    case MODE_INTRA:
	      fprintf(pmv_result_file,"  Mode is Intra mode\n\n");
	      break;
	    case MODE_INTER:
          fprintf(pmv_result_file,"  (1MV) MVD are : 0x%04x, 0x%04x\n",((pCodec->ME_command_queue0[11]>>16)&0x0ffff), (pCodec->ME_command_queue0[11]&0x0ffff));
          fprintf(pmv_result_file,"  Mode is Inter 1MV mode\n\n");
	      break;
	    case MODE_INTER4V:
          fprintf(pmv_result_file,"  (4MV) MVD are : 0x%04x, 0x%04x, 0x%04x, 0x%04x, 0x%04x, 0x%04x, 0x%04x, 0x%04x\n",((pCodec->ME_command_queue0[8]>>16)&0x0ffff), (pCodec->ME_command_queue0[8]&0x0ffff), ((pCodec->ME_command_queue0[9]>>16)&0x0ffff), (pCodec->ME_command_queue0[9]&0x0ffff), ((pCodec->ME_command_queue0[10]>>16)&0x0ffff), (pCodec->ME_command_queue0[10]&0x0ffff), ((pCodec->ME_command_queue0[11]>>16)&0x0ffff), (pCodec->ME_command_queue0[11]&0x0ffff));
          fprintf(pmv_result_file,"  Mode is Inter 4MV mode\n\n");
	      break;             
	  }
    #endif
	
	tmp1 = pMB->mv16x_1;
	tmp2 = pMB->mv16y_1;
	tmp3 = pMB->mv16x_3;
	tmp4 = pMB->mv16y_3;

	get_pmvdata1_4MV(x,iWcount, &tmp1, &tmp2, &tmp3, &tmp4,pParam,pCodec);

	if (!d_type) {
		MOTION_ACTIVITY=abs(tmp1)+abs(tmp2);
		d_type=(MOTION_ACTIVITY > L1);		//		if (MOTION_ACTIVITY > L1) => Large Diamond else Small Diamond
	}

	pCodec->ME_command_queue0[19] = (2<<29) | ((tmp1 & 0x7f) << 19) | ((tmp2 & 0x7f) << 12) | (pCodec->Raddr + (tmp2 >> 1)*16);
	pCodec->ME_command_queue0[20] = (2<<29) | ((pmv[1].x & 0x7f) << 19) | ((pmv[1].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[1].y >> 1)*16);
	pCodec->ME_command_queue0[21] = (2<<29) | ((pmv[3].x & 0x7f) << 19) | ((pmv[3].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[3].y >> 1)*16);
	pCodec->ME_command_queue0[22] = (2<<29) | (1<<28) | ((prevMB->mv16x_3 & 0x7f) << 19) | ((prevMB->mv16y_3 & 0x7f) << 12) | (pCodec->Raddr + (prevMB->mv16y_3 >> 1)*16);
	pCodec->ME_command_queue0[23] = (3<<29) | (d_type << 24) | (Diamond_search_limit << 16) | ThEES;		// dsize is always small

	pCodec->ME_command_queue0[28] = (2<<29) | ((tmp1 & 0x7f) << 19) | ((tmp2 & 0x7f) << 12) | (pCodec->Raddr + (tmp2 >> 1)*16);
	pCodec->ME_command_queue0[29] = (2<<29) | ((pmv[1].x & 0x7f) << 19) | ((pmv[1].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[1].y >> 1)*16);
	pCodec->ME_command_queue0[30] = (2<<29) | ((pmv[3].x & 0x7f) << 19) | ((pmv[3].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[3].y >> 1)*16);
	pCodec->ME_command_queue0[31] = (1<<28) | (2<<29) | ((prevMB->mv16x_0 & 0x7f) << 19) | ((prevMB->mv16y_0 & 0x7f) << 12) | (pCodec->Raddr + (prevMB->mv16y_0 >> 1)*16);

	pCodec->ME_command_queue0[38] = (2<<29) | ((pmv[2].x & 0x7f) << 19) | ((pmv[2].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[2].y >> 1)*16);
	pCodec->ME_command_queue0[39] = (2<<29) | ((pmv[3].x & 0x7f) << 19) | ((pmv[3].y & 0x7f) << 12) | (pCodec->Raddr + (pmv[3].y >> 1)*16);
	pCodec->ME_command_queue0[40] = (1<<28) | (2<<29) | ((prevMB->mv16x_1 & 0x7f) << 19) | ((prevMB->mv16y_1 & 0x7f) << 12) | (pCodec->Raddr + (prevMB->mv16y_1 >> 1)*16);

	pCodec->ME_command_queue0[46] = (2<<29) | ((tmp3 & 0x7f)<<19) | ((tmp4 & 0x7f)<<12) | (pCodec->Raddr23 + (tmp4 >> 1)*16);
	pCodec->ME_command_queue0[49] = (1<<28) | (2<<29) | ((prevMB->mv16x_2 & 0x7f) << 19) | ((prevMB->mv16y_2 & 0x7f) << 12) | (pCodec->Raddr23 + (prevMB->mv16y_2 >> 1)*16);

	pCodec->ME_command_queue0[58] = (1<<28) | (2<<29) | ((prevMB->mv16x_3 & 0x7f) << 19) | ((prevMB->mv16y_3 & 0x7f) << 12) | (pCodec->Raddr23 + (prevMB->mv16y_3 >> 1)*16);
	
	if ((y==(pParam->mb_height-1)) | ((y==(pParam->mb_height-2)) & (x==(iWcount-1)))) {
	
		if (prevMB->mv16y_3 > 15)
			pCodec->ME_command_queue0[58] = pCodec->ME_command_queue0[58] - ((prevMB->mv16y_3 >> 1)*16) + ((15>>1) * 16);
		if (prevMB->mv16y_2 > 15)
			pCodec->ME_command_queue0[49] = pCodec->ME_command_queue0[49] - ((prevMB->mv16y_2  >> 1)*16) + ((15>>1) * 16);
		if (tmp4 > 15)
			pCodec->ME_command_queue0[46] = pCodec->ME_command_queue0[46] - ((tmp4  >> 1)*16) + ((15>>1) * 16);
	}

	if (counter < iWcount) {
		if (tmp2 < 0)
			pCodec->ME_command_queue0[19] -= (tmp2 >> 1)*16;
		if (pmv[1].y < 0) {
			pCodec->ME_command_queue0[29] -= (pmv[1].y >> 1)*16;
			pCodec->ME_command_queue0[20] -= (pmv[1].y >> 1)*16;
		}
		if (pmv[3].y < 0) {
			pCodec->ME_command_queue0[21] -= (pmv[3].y >> 1)*16;
			pCodec->ME_command_queue0[30] -= (pmv[3].y >> 1)*16;
			pCodec->ME_command_queue0[39] -= (pmv[3].y >> 1)*16;
		}
		if (pmv[2].y < 0)
			pCodec->ME_command_queue0[38] -= (pmv[2].y >> 1)*16;
		if (prevMB->mv16y_0 < 0)
			pCodec->ME_command_queue0[31] -= (prevMB->mv16y_0 >> 1)*16;
		if (prevMB->mv16y_1 < 0)
			pCodec->ME_command_queue0[40] -= (prevMB->mv16y_1 >> 1)*16;
		if (prevMB->mv16y_2 < -16)
			pCodec->ME_command_queue0[49] -= ((prevMB->mv16y_2 + 16) >> 1)*16;
		if (prevMB->mv16y_3 < 0)
			pCodec->ME_command_queue0[22] -= (prevMB->mv16y_3 >> 1)*16;
		if (prevMB->mv16y_3 < -16)
			pCodec->ME_command_queue0[58] -= ((prevMB->mv16y_3 + 16) >> 1)*16;
		if (tmp2 < 0)
			pCodec->ME_command_queue0[28] -= (tmp2 >> 1)*16;
		if (tmp4 < -16)
			pCodec->ME_command_queue0[46] -= ((tmp4 + 16) >> 1)*16;
	}

	return 0;    
}

int32_t
MotionEstimation_blocklast_4MV(MACROBLOCK *const pMB_mc,
				 MBParam * const pParam,
				 Encoder * pEnc)
{
    FTMCP100_CODEC *pCodec=(FTMCP100_CODEC *)pEnc->pCodec;
	uint32_t index;
	int32_t  X;
	volatile MDMA *pmdma = MDMA1;
	MACROBLOCK *pMB_tmp;
    FRAMEINFO * const current1=pEnc->current1;
	unsigned int XDIM=pEnc->mEncParam.u32FrameWidth;
	DECLARE_MP4_PTR

	POLL_MARKER_S	
    // because we have made the DMA chained and sync to MC done.
    // so checking DMA done is equivalent to checking MC done
	while((pmdma->Status & 0x1) == 0) {}
	POLL_MARKER_E

	if (pCodec->MB_mode) {
		index = (MBTransQuantInter(pParam, current1,pCodec)) | (pCodec->MVZ << 16) | (pMB_mc->mode<<6);
	} else {
      if (pParam->resyn==0)
      {
        pMB_tmp = pMB_mc;
        pCodec->acdc_status = 0;
		
		pMB_tmp = pMB_tmp - 1;
		if (pMB_tmp->mode != MODE_INTRA)
			pCodec->acdc_status |= 1;
		pMB_tmp = pMB_tmp - (XDIM/16);
		if (pMB_tmp->mode != MODE_INTRA)
			pCodec->acdc_status |= 4;
		pMB_tmp += 1;
		if (pMB_tmp->mode != MODE_INTRA)
			pCodec->acdc_status |= 2;
	  }
	  else
	  {
		pCodec->acdc_status = 6;
		pMB_tmp = pMB_mc - 1;
		if (pMB_tmp->mode != MODE_INTRA)
			pCodec->acdc_status |= 1;
	  }
		index = MBTransQuantIntra_p(pParam, current1,pCodec);
	}

	SET_MCCTL(index)						//		MC GO	

    {
      unsigned int *pDMA_next; 
	  pDMA_next=pCodec->DMA_COMMAND_local+(pCodec->even_odd_1)*DMA_COMMAND_QUEUE_STRIDE;
      DMA_MOVE(24+(pCodec->even_odd_1^1)*DMA_COMMAND_QUEUE_STRIDE, 0x4B00014)
      pDMA_next[25] = TRANSLATE_LOCAL_MEMORY_BASE_ADDRESS(((pCodec->triple_buffer_selector+2)%3)*384 + INTER_Y0);
      pDMA_next[29] = TRANSLATE_LOCAL_MEMORY_BASE_ADDRESS(((pCodec->triple_buffer_selector+2)%3)*384 + INTER_U0);
	  pDMA_next[33] = TRANSLATE_LOCAL_MEMORY_BASE_ADDRESS(((pCodec->triple_buffer_selector+2)%3)*384 + INTER_V0);
	  pDMA_next[35]=0x00900010;
    }

	/******************       CHECK MC_DONE, INT Y      OK   ****************/
	POLL_MC_DONE_MARKER_START
	// check MC done
	do {
	  READ_CPSTS(X)
	} while(!(X&0x02));
	POLL_MC_DONE_MARKER_END
	
	
	POLL_MARKER_S	
    // because we have made the DMA chained and sync to MC done.
    // so checking DMA done is equivalent to checking MC done
	while((pmdma->Status & 0x1) == 0) {}
	POLL_MARKER_E

	DMA_MOVE(24+(pCodec->even_odd_1)*DMA_COMMAND_QUEUE_STRIDE, 0x4B0000c)

	return 0;
}

int32_t
MotionEstimation_block0_1MV(MACROBLOCK *const pMB,
				 Encoder *pEnc)
{
    FTMCP100_CODEC *pCodec=(FTMCP100_CODEC *)pEnc->pCodec;
    FRAMEINFO * const current1=pEnc->current1;
    FRAMEINFO * const reference=pEnc->reference;
    
	MACROBLOCK *prevMB = &reference->mbs[0];
	VECTOR pmv[4];
	int32_t MOTION_ACTIVITY;
	int32_t  X, d_type;
	uint32_t lam_16;
	int32_t pmvsx, pmvsy;
	volatile MDMA *pmdma = MDMA1;
	DECLARE_MP4_PTR

    unsigned int *pDMA1,*pDMA2;
	unsigned int XDIM=pEnc->mEncParam.u32FrameWidth;
	pDMA1=pCodec->DMA_COMMAND_local;
	pDMA2=pCodec->DMA_COMMAND_local+DMA_COMMAND_QUEUE_STRIDE;

    SET_VOPSIZE((pEnc->mEncParam.u32FrameWidth)<<16 | (pEnc->mEncParam.u32FrameHeight))
	
	SET_PMVBUF(PMV_BUF)
	SET_HOFFSET(0)

	pCodec->ME_command_queue0[16] = (0<<29) | (1<<25) | (0<<23);
	pCodec->ME_command_queue0[17] = (1<<29) | (1<<25) | (0<<23);
	pCodec->ME_command_queue0[18] = (2<<29) | (1<<27) | pCodec->Raddr;				//	median result
	pCodec->ME_command_queue0[19] = (2<<29) | pCodec->Raddr;
	pCodec->ME_command_queue0[20] = (2<<29) | pCodec->Raddr;
	pCodec->ME_command_queue0[21] = (2<<29) | pCodec->Raddr;
	pCodec->ME_command_queue0[22] = (1<<28) | (2<<29) | ((prevMB->mv16x_3 & 0x7f) << 19) | ((prevMB->mv16y_3 & 0x7f) << 12) 
							| pCodec->Raddr + (prevMB->mv16y_3 < 0 ? 0 : (prevMB->mv16y_3 >> 1)*16);
	pCodec->ME_command_queue0[23] = (3<<29) | (Diamond_search_limit << 16) | ThEES;		// dsize is always small
	pCodec->ME_command_queue0[24] = ((unsigned int)4<<29) | ThEES1; // to add unsigned int to avoid warning ... half pixel refine
	
	pCodec->ME_command_queue0[25] = ((unsigned int)5<<29) | (1<<28) | MV16_INTER_BIAS; // to add unsigned int to avoid warning ... P-I switch
	pCodec->ME_command_queue0[26] = ((unsigned int)6<<29); // to add unsigned int to avoid warning ... Y interpolation
	pCodec->ME_command_queue0[27] = ((unsigned int)6<<29) | (((uint32_t) (REF_U + 32*8)) >> 2) | (1<<27); // to add unsigned int to avoid warning
	pCodec->ME_command_queue0[28] = ((unsigned int)6<<29) | (((uint32_t) (REF_V + 32*8)) >> 2) | (1<<27) | (1<<28); // to add unsigned int to avoid warning

	/************	for next block	********************************/

 	lam_16 = NEIGH_TEND_16X16*lambda_vec16[current1->quant];
	SET_MECR(lam_16)
	SET_CMDADDR(ME_COMMAND_QUEUE_ADDR)
	SET_MECADDR(CUR_Y0)
	SET_MEIADDR(INTER_Y0)

	X = current1->rounding_type << 2 | 1 << 11 | 1 | 1<<3 | 1<<4;		//	1 << 11 => fcode

	POLL_MARKER_S
	while((pmdma->Status & 0x1) == 0) {}			//	check REFyuv, CURyuv OK
	POLL_MARKER_E

	SET_MECTL(X)									//		ME GO  ************************

	// since we have already prepared the secondary part of DMA commands right in the FrameCodeP(),
    // so we can directly begin to execute the secondary part of DMA command buffer (from 0th to 0+40(0x28)-1=39th)
	DMA_MOVE(DMA_COMMAND_QUEUE_STRIDE, 0x4B00028)  // enable Transfer Done flag mask
                                                   // Enable DMA start transferring
                                                   // Enable chain transfer
                                                   // From sequential Local memory to sequential System memory                            
                                                   // transfer 0x28(40) words