hpel_16x16.asm

adi bf533视频编码程序

    RTS;
    
/*!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!*/

    
/*********************rounding_type ==1 ******************************/ 
/*   
.section L1_code;
.align 8;
.global __hpel_16x16_16_T;

__hpel_16x16_16_T:

    [--SP] = (R7:4, P5:3);
    [--SP] = RETS;
    L0 = 0;
    L1 = 0;
    L2 = 0;
    L3 = 0;
    P5 = R0;                // Address of the best matching block
    R5 = R1;                // SAD corresponding to the best match
    P4 = R2;                // Address of the target block
    P3 = [SP + 44];         // Width of the reference window
    
    
//********************store the best matching block**********************

    I0 = R0;                // Address of best match   
    I1 = R0;                // Address of best match
    P0 = [SP + 64];         //address of MBy_match
    I3 = P0;                //address of MBy_match 
    P2 = 16 (Z);
    R1 = P3;                // WINWIDTH
    R1 += -20;
    M1 = R1;               //M1的值再定一下

    LSETUP(STORE_BEST_ST_T,STORE_BEST_END_T) LC0 = P2;
    DISALGNEXCPT || R0 = [I0++] || R2 = [I1++];
STORE_BEST_ST_T:
        DISALGNEXCPT || R1 = [I0++] || R3  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++] || R2  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(R) || [I3++] = R6 || R1 = [I0++];
        DISALGNEXCPT  || [I3++] = R6 || R3  =[I1++];            
        R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++] || R2  =[I1++];
        R6 = BYTEOP1P(R1:0,R3:2)(R) || [I3++] = R6 || R1 = [I0++M1]; 
        DISALGNEXCPT || [I3++] = R6 || R3  =[I1++M1];               
STORE_BEST_END_T:
        DISALGNEXCPT || R0 = [I0++]|| R2  =[I1++];

//************************************************************************
    
    
    P0 = 344;
    SP -= P0;
    I3 = SP;                // Temporary buffer in stack
    R4 = 0;
//******************** INTERPOLATE DIAGONAL BLOCKS ********************
    M0 = 7;
    M3 = -3 (X);
    R0 = P5;                // Address of the best match
    R0 += -1;
    I0 = R0;
    I1 = R0;
    R2 = P3;                // WINWIDTH
    R1 = R0 - R2 (S);
    I2 = R1;                // Address of best match - (WINWIDTH+1)
    R2 += -17;
    M1 = R2;
    P2 = 17;

    LSETUP(AVG4_ST_T, AVG4_END_T) LC0 = P2;
    DISALGNEXCPT || R0 = [I1++] || R2 = [I2++];
    DISALGNEXCPT || R1 = [I1++M3] || R3 = [I2++M3];
    R7 = BYTEOP2P(R1:0,R3:2) (TL) || R0 = [I1++M0];
AVG4_ST_T:    
        DISALGNEXCPT || I0 += M3 || R2 = [I2++M0];
        R6 = BYTEOP2P(R1:0,R3:2) (TH) || R0 = [I1++M3] || R2 = [I2++M3];
        R7 = R6 + R7 (NS) || I0 -= M3;
    
        R7 = BYTEOP2P(R1:0,R3:2) (TL, R) || [I3++] = R7 || R1 = [I1++M0];
        DISALGNEXCPT || I0 += M3 || R3 = [I2++M0];
        R6 = BYTEOP2P(R1:0,R3:2) (TH, R) || R1 = [I1++M3] || R3 = [I2++M3];
        R7 = R6 + R7 (NS) || I0 -= M3;
    
        R7 = BYTEOP2P(R1:0,R3:2) (TL) || [I3++] = R7 || R0 = [I1++M0];
        DISALGNEXCPT || I0 += M3 || R2 = [I2++M0];
        R6 = BYTEOP2P(R1:0,R3:2) (TH) || R0 = [I1++M3] || R2 = [I2++M3];
        R7 = R6 + R7 (NS) || I0 -= M3;
    
        R7 = BYTEOP2P(R1:0,R3:2) (TL, R) || [I3++] = R7 || R1 = [I1++M0];
        DISALGNEXCPT || I0 += M3 || R3 = [I2++M0];
        R6 = BYTEOP2P(R1:0,R3:2) (TH, R) || R1 = [I1++M3] || R3 = [I2++M3];
        R7 = R6 + R7 (NS) || I0 -= M3;
    
        R7 = BYTEOP2P(R1:0,R3:2) (TL) || [I3++] = R7 || R0 = [I1++M1];
        DISALGNEXCPT || I0 += M3 || R2 = [I2++M1];
        R6 = BYTEOP2P(R1:0,R3:2) (TH) || R0 = [I1++] || R2 = [I2++];
        R7 = R6 + R7 (NS) || I0 -= M3;
    
        DISALGNEXCPT || R1 = [I1++M3] || R3 = [I2++M3];
AVG4_END_T:
        R7 = BYTEOP2P(R1:0,R3:2) (TL) || [I3++] = R7 || R0 = [I1++M0];       
    
//**************** CALCULATE SAD FOR DIAGONAL BLOCKS *******************
	R6=0(Z);                   //the offset from SP
    R7 = -1;                // (V,H) -> R7.H = -1, R7.L = -1
    R0 = P4;
    R1 = SP;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    
    R7.L = 1;               // (V,H) -> R7.H = -1, R7.L = 1
    R0 = P4;
    R1 = SP;
    R1 += 1;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    R1 = 1(Z);
    IF CC R6 = R1;
    
    R7.H = 1;               // (V,H) -> R7.H = 1, R7.L = 1
    R0 = P4;
    R1 = SP;
    R1 += 21;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    R1 = 21(Z);
    IF CC R6 = R1;
    
    R7.L = -1;              // (V,H) -> R7.H = 1, R7.L = -1
    R0 = P4;
    R1 = SP;
    R1 += 20;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    R1 = 20(Z);
    IF CC R6 = R1;
    
    CC = R4==0;
    IF CC JUMP STORE_BYPASS0_T; 
    //if half pel match better, store the interpolated block
 //********************store the best matching block**********************
	R0 = SP;
	R0 = R0 + R6;
    I0 = R0;                // Address of half pel best match   
    I1 = R0;                
    P0 = [SP + 408];         //address of MBy_match :SP + 344 + 64
    I3 = P0;                //address of MBy_match 
    P2 = 16 (Z);
    
    
    LSETUP(STORE_BEST_0ST,STORE_BEST_0END) LC0 = P2;
    DISALGNEXCPT || R0 = [I0++] || R2 = [I1++];
STORE_BEST_0ST:
        DISALGNEXCPT || R1 = [I0++] || R3  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++] || R2  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(R) || [I3++] = R6 || R1 = [I0++];
        DISALGNEXCPT  || [I3++] = R6 || R3  =[I1++];            
        R6 = BYTEOP1P(R1:0,R3:2) || R0 = [I0++] || R2  =[I1++];
        R6 = BYTEOP1P(R1:0,R3:2)(R) || [I3++] = R6 || R0 = [I0++]; 
STORE_BEST_0END:       
		DISALGNEXCPT || [I3++] = R6 || R2  =[I1++];               

//************************************************************************
   
STORE_BYPASS0_T:    
//******************** INTERPOLATE LEFT/RIGHT BLOCKS ********************
    R0 = P5;                // Address of the best match
    I0 = R0;                // Address of best match
    R0 += -1;
    I1 = R0;                // Address of best match - 1
    I3 = SP;                // Output buffer
    P2 = 16 (Z);
    R1 = P3;                // WINWIDTH
    R1 += -20;
    M1 = R1;
    MNOP;

    LSETUP(AVG2_LR_ST_T, AVG2_LR_END_T) LC0 = P2;
    DISALGNEXCPT || R0 = [I0++] || R2 = [I1++];
AVG2_LR_ST_T:
        DISALGNEXCPT || R1 = [I0++] || R3  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++] || R2  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(T,R) || [I3++] = R6 || R1 = [I0++];
        DISALGNEXCPT  || [I3++] = R6 || R3  =[I1++];            
        R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++] || R2  =[I1++];
        R6 = BYTEOP1P(R1:0,R3:2)(T,R) || [I3++] = R6 || R1 = [I0++M1]; 
        DISALGNEXCPT || [I3++] = R6 || R3  =[I1++M1];               
        R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++];
AVG2_LR_END_T:
        DISALGNEXCPT || [I3++] = R6 || R2  =[I1++];
    
//**************** CALCULATE SAD FOR LEFT/RIGHT BLOCKS *******************
    P2 = R4;                //backup R4
	R7 = 1;                 // (V,H) -> R7.H = 0, R7.L = 1
    R0 = P4;
    R1 = SP;
    R1 += 1;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    
    R7.L = -1;              // (V,H) -> R7.H = 0, R7.L = -1
    R0 = P4;
    R1 = SP;
    CALL _compute_sad;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    
    R3 = P2;
    CC = R4==R3;         //R4 has been modified?
    IF CC JUMP STORE_BYPASS1_T;
 
    //if half pel match better, store the 
//********************store the best matching block**********************
	R0 = SP;
    I0 = R0;                // Address of half pel interpolated block                
    P0 = [SP + 408];         //address of MBy_match :SP + 344 + 64
    I3 = P0;                //address of MBy_match 
    P2 = 16 (Z);
    CC = R4==R7;
    
    IF !CC JUMP STORE_DISALIGN_T;

    LSETUP(STORE_BEST_1ST_ALIGN,STORE_BEST_1END_ALIGN) LC0 = P2;
    M1 = 8(Z);
    R0 = [I0++];
STORE_BEST_1ST_ALIGN: // (V,H)=(0,-1)
    [I3++]=R0||R0 = [I0++];
    [I3++]=R0||R0 = [I0++];
    [I3++]=R0||R0 = [I0 ++ M1];
STORE_BEST_1END_ALIGN:    
	[I3++]=R0||R0 = [I0++];
	
	JUMP STORE_BYPASS1_T;
	
STORE_DISALIGN_T:	   // (V,H)=(0,1)
    LSETUP(STORE_BEST_1ST,STORE_BEST_1END) LC0 = P2;
    DISALGNEXCPT || R0 = [I0++];
    DISALGNEXCPT || R1 = [I0++];
STORE_BEST_1ST:
		R6 = ALIGN8(R1,R0) || R0 = [I0++];
		R6 = ALIGN8(R0,R1) || [I3++] = R6 || R1 = [I0++];		
        R6 = ALIGN8(R1,R0) || [I3++] = R6 || R0  =[I0++]; 
        R6 = ALIGN8(R0,R1) || [I3++] = R6 || R0 = [I0++];                
		
STORE_BEST_1END:		
		DISALGNEXCPT || [I3++] = R6 || R1 = [I0++];
							
    
STORE_BYPASS1_T:   
//******************** INTERPOLATE TOP/BOTTOM BLOCKS ********************
    R1 = P3;                // WINWIDTH
    R0 = P5;                // Address of the best match
    I0 = R0;                // Address of best match
    R0 = R0 - R1(S) || NOP;
    I1 = R0;                // Address of best match - WINWIDTH
    I3 = SP;                // Output buffer
    P2 = 17;
    R1 += -16;
    M1 = R1;

    LSETUP(AVG2_TB_ST_T, AVG2_TB_END_T) LC0 = P2;
    DISALGNEXCPT || R0 = [I0++] || R2 = [I1++];
AVG2_TB_ST_T:
        DISALGNEXCPT || R1 = [I0++] || R3  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++] || R2  =[I1++]; 
        R6 = BYTEOP1P(R1:0,R3:2)(T,R) || [I3++] = R6 || R1 = [I0++];
        DISALGNEXCPT  || [I3++] = R6 || R3  =[I1++];            
        R6 = BYTEOP1P(R1:0,R3:2)(T) || R0 = [I0++M1] || R2  =[I1++M1];
        R6 = BYTEOP1P(R1:0,R3:2)(T,R) || [I3++] = R6 || R0 = [I0++]; 
AVG2_TB_END_T:
        DISALGNEXCPT || [I3++] = R6 || R2  =[I1++];             
    
//**************** CALCULATE SAD FOR TOP/BOTTOM BLOCKS *******************
    
	P2 = R4;                //backup R4
	R7.H = -1;              // (V,H) -> R7.H = -1, R7.L = 0
    R7.L = 0;
    R0 = P4;
    R1 = SP;
    CALL _compute_sad_aligned;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
    
    R7.H = 1;               // (V,H) -> R7.H = 1, R7.L = 0
    R0 = P4;
    R1 = SP;
    R1 += 16;
    CALL _compute_sad_aligned;
    CC = R0 < R5;
    IF CC R5 = R0;
    IF CC R4 = R7;
 
//********************store the best matching block**********************   
    R3 = P2;
    CC = R4==R3;
    IF CC JUMP STORE_BYPASS2_T;
    
    R0 = 0(Z);
    R1 = 16(Z);
    CC = R4==R7; 
    IF !CC R1=R0;
    R0 = SP;
    R0 = R0+R1; // Address of half pel best match
    I0 = R0;
    P0 = [SP + 408];         //address of MBy_match :SP + 344 + 64
    I3 = P0;                //address of MBy_match     
    P2 = 16 (Z);    
    
    LSETUP(STORE_BEST_2ST,STORE_BEST_2END) LC0 = P2;
    R0 = [I0++];
STORE_BEST_2ST:
    [I3++]=R0||R0 = [I0++];
    [I3++]=R0||R0 = [I0++];
    [I3++]=R0||R0 = [I0++];
STORE_BEST_2END:    
	[I3++]=R0||R0 = [I0++];
		 
    
STORE_BYPASS2_T:    
//********************************************************************
    R0 = R4;                // Return horizontal and vertical half pel
    P0 = 344;
    SP = SP + P0;
    [SP + 48] = R5;        //store SAD corresponding to the best match
    RETS = [SP++];
    (R7:4, P5:3) = [SP++];
    RTS;

///////////////////////////////////////////////////////////////////////////   
*/        
.align 8;
_compute_sad:
    
    [--SP] = (R7:4, P5:3);
    [--SP] = RETS;
    I0 = R0;                // Address of the target
    I1 = R1;                // Address of the interpolated block
    P0 = 16 (Z);
    A1 = A0 = 0;
    //R2=372;//Step-12
    //M0=R2;
    LSETUP (MAD_START1, MAD_END1) LC0=P0;
    DISALGNEXCPT || R0 = [I0++] || R2 = [I1++]; 
MAD_START1:
        DISALGNEXCPT || R3 = [I1++];
        SAA (R1:0,R3:2) || R1 = [I0++]  || R2 = [I1++];
                            // Compute absolute difference and acc 
        SAA (R1:0,R3:2) (R) || R0 = [I0++] || R3 = [I1++];
        SAA (R1:0,R3:2) || R1 = [I0++] || R2 = [I1++];
MAD_END1:
        SAA (R1:0,R3:2) (R) || R0 = [I0++] || R2 = [I1++];
    R3=A1.L+A1.H,R2=A0.L+A0.H;    
    R0 = R2 + R3 (S);       // Add the accumulated values in both MACs
   
    RETS = [SP++];
    (R7:4, P5:3) = [SP++];
_compute_sad.end:
    RTS;
    
.align 8;
_compute_sad_aligned:
    [--SP] = (R7:4, P5:3);
    [--SP] = RETS;

    I0 = R0;                // Address of the target
    I1 = R1;                // Address of the interpolated block
    P0 = 16;
    //R2=372;//Step-12
    //M0=R2;
    
    
    LSETUP (MAD_START, MAD_END) LC0=P0;
    A1=A0=0 || R0 = [I0++] || R2 = [I1++];
                            // Initialize accumulators 
MAD_START:
        SAA (R1:0,R3:2) || R1 = [I0++]  || R3 = [I1++];
                            // Compute absolute difference and acc. 
        SAA (R1:0,R3:2) (R) || R0 = [I0++] || R2 = [I1++];
        SAA (R1:0,R3:2) || R1 = [I0++] || R3 = [I1++];
MAD_END:SAA (R1:0,R3:2) (R) || R0 = [I0++] || R2 = [I1++];
    R3=A1.L+A1.H,R2=A0.L+A0.H;    
    R0 = R2 + R3 (S);       // Add the accumulated values in both MACs
   
    RETS = [SP++];
    (R7:4, P5:3) = [SP++];   
_compute_sad_aligned.end:    
    
    RTS;