static void ff_vp3_idct_add_mxu(uint8_t *src, int stride, DCTELEM *input, uint8_t idct_row)
{
int i;
DCTELEM *blk;
int32_t wf = (int32_t)whirl_idct;
S32LDD(xr5, wf, 0x0); // xr5(w7, w3)
S32LDD(xr6, wf, 0x4); // xr6(w9, w8)
S32LDD(xr7, wf, 0x8); // xr7(w11,w10)
S32LDD(xr8, wf, 0xc); // xr8(w13,w12)
S32LDD(xr9, wf, 0x10); // xr9(w6, w0)
S32LDD(xr10,wf, 0x14);
blk = input - 8;
/* Inverse DCT on the rows now */
for (i=0; i<idct_row; i++) {
S32LDI(xr1, blk, 0x10); // xr1 (x4, x0)
S32LDD(xr2, blk, 0x4); // xr2 (x7, x3)
S32LDD(xr3, blk, 0x8); // xr3 (x6, x1)
S32LDD(xr4, blk, 0xc); // xr4 (x5, x2)
S32OR(xr12, xr2,xr3);
S32OR(xr11,xr12,xr4);
S32OR(xr12,xr11,xr1);
if (S32M2I(xr12) == 0) {
continue; //blk[0]= blk[1]=blk[2]=blk[3]=blk[4]=blk[5]=blk[6]=blk[7]=0
}
S32SFL(xr12,xr0,xr1,xr13,ptn3);
S32OR(xr11,xr11,xr12);
if (S32M2I(xr11) == 0 && S32M2I(xr13) != 0) {
D16MUL_HW(xr0,xr5,xr13,xr13);
D32SAR(xr0,xr0,xr13,xr13,15);
S32SFL(xr0,xr13,xr13,xr13,ptn3);
S32STD(xr13,blk, 0x0);
S32STD(xr13,blk, 0x4);
S32STD(xr13,blk, 0x8);
S32STD(xr13,blk, 0xc);
continue; //blk[0]!=0, and blk[1]=blk[2]=blk[3]=blk[4]=blk[5]=blk[6]=blk[7]=0
}
S32SFL(xr1,xr1,xr2,xr2, ptn3); //xr1:s1, s3, xr2: s0, s2
S32SFL(xr3,xr3,xr4,xr4, ptn3); //xr3:s5, s7, xr4: s4, s6
D16MUL_WW(xr11, xr2, xr5, xr12);//xr11: s0*c4, xr12: s2*c2
D16MAC_AA_WW(xr11,xr4,xr6,xr12);//xr11: s0*c4+s4*c4, xr12: s2*c2+s6*c6
D16MUL_WW(xr13, xr2, xr6, xr14);//xr13: s0*c4, xr14: s2*c6
D16MAC_SS_WW(xr13,xr4,xr5,xr14);//xr13: s0*c4 - s4*c4, xr14: s2*c6-s6*c2
D16MUL_HW(xr2, xr1, xr7, xr4); //xr2: s1*c1, xr4: s1*c3
D16MAC_AS_LW(xr2,xr1,xr9,xr4); //xr2: s1*c1+s3*c3, xr4: s1*c3-s3*c7
D16MAC_AS_HW(xr2,xr3,xr10,xr4); //xr2: s1*c1+s3*c3+s5*c5,
// xr4: s1*c3-s3*c7-s5*c1
D16MAC_AS_LW(xr2,xr3,xr8,xr4); //xr2: s1*c1+s3*c3+s5*c5+s7*c7,
//xr4: s1*c3-s3*c7-s5*c1-s7*c5
D32SAR(xr11, xr11,xr13,xr13,15);
S32SFL(xr0, xr11,xr13,xr11,ptn3);
D32SAR(xr12,xr12,xr14,xr14,15);
S32SFL(xr0, xr12,xr14,xr12,ptn3);
D32SAR(xr2, xr2,xr4,xr4,15);
S32SFL(xr0, xr2,xr4,xr2,ptn3);
D16MUL_HW(xr4, xr1, xr8, xr15); //xr4: s1*c7, xr15:s1*c5
D16MAC_SS_LW(xr4,xr1,xr10,xr15); //xr4: s1*c7-s3*c5, xr15: s1*c5-s3*c1
D16MAC_AA_HW(xr4,xr3,xr9,xr15); //xr4: s1*c7-s3*c5+s5*c3, xr15: s1*c5-s3*c1+s5*c7
D16MAC_SA_LW(xr4,xr3,xr7,xr15); //xr4: s1*c7-s3*c5+s5*c3-s7*c1
//xr15: s1*c5-s3*c1+s5*c7+s7*c3
Q16ADD_AS_WW(xr11,xr11,xr12,xr12); //xr11: rnd(s0*c4+s4*c4)>>15+rnd(s2*c2+s6*c6)>>15
// rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15
//xr12: rnd(s0*c4+s4*c4)>>15-rnd(s2*c2+s6*c6)>>15
// rnd(s0*c4-s4*c4)>>15-rnd(s2*c6-s6*c2)>>15
D32SAR(xr15,xr15,xr4,xr4,15);
S32SFL(xr0,xr15,xr4,xr15,ptn3);
Q16ADD_AS_WW(xr11, xr11, xr2, xr2);
//xr11: rnd(s0*c4+s4*c4)>>15+rnd(s2*c2+s6*c6)>>15 + rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>15
// : rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15 + rnd(s1*c3-s3*c7-s5*c1-s7*c5)>>15
//xr2: rnd(s0*c4+s4*c4)>>15+rnd(s2*c2+s6*c6)>>15 - rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>15
// : rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15 - rnd(s1*c3-s3*c7-s5*c1-s7*c5)>>15
Q16ADD_AS_XW(xr12, xr12, xr15, xr15);
//xr12: rnd(s0*c4+s4*c4)>>15-rnd(s2*c2+s6*c6)>>15+rnd(s1*c5-s3*c1+s5*c7+s7*c3)>>15
// : rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15+rnd(s1*c7-s3*c5+s5*c3-s7*c1)>>15
//xr15: rnd(s0*c4+s4*c4)>>15-rnd(s2*c2+s6*c6)>>15-rnd(s1*c5-s3*c1+s5*c7+s7*c3)>>15
// : rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15-rnd(s1*c7-s3*c5+s5*c3-s7*c1)>>15
S32SFL(xr11,xr11,xr12,xr12, ptn3);
//xr11: rnd(s0*c4+s4*c4)>>15+rnd(s2*c2+s6*c6)>>15 + rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>15
// : rnd(s0*c4+s4*c4)>>15-rnd(s2*c2+s6*c6)>>15+rnd(s1*c5-s3*c1+s5*c7+s7*c3)>>15
//xr12: rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15 + rnd(s1*c3-s3*c7-s5*c1-s7*c5)>>15
// : rnd(s0*c4-s4*c4)>>15+rnd(s2*c6-s6*c2)>>15+rnd(s1*c7-s3*c5+s5*c3-s7*c1)>>15
S32SFL(xr12,xr12,xr11,xr11, ptn3);
//xr12: rnd(s0*c4-s4*c4)>>16+rnd(s2*c6-s6*c2)>>16 + rnd(s1*c3-s3*c7-s5*c1-s7*c5)>>16
// : rnd(s0*c4+s4*c4)>>16+rnd(s2*c2+s6*c6)>>16 + rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>16
//xr11: rnd(s0*c4-s4*c4)>>16+rnd(s2*c6-s6*c2)>>16+rnd(s1*c7-s3*c5+s5*c3-s7*c1)>>16
// : rnd(s0*c4+s4*c4)>>16-rnd(s2*c2+s6*c6)>>16+rnd(s1*c5-s3*c1+s5*c7+s7*c3)>>16
S32STD(xr12, blk, 0x0);
S32STD(xr11, blk, 0x4);
S32STD(xr15, blk, 0x8);
S32STD(xr2, blk, 0xc);
}
blk = input - 2;
for (i=0; i<4; i++) /* idct columns */
{
S32I2M(xr5,wxr5); //xr5: c4 , c2
S32I2M(xr6,wxr6); //xr5: c4 , c2
S32LDI(xr1, blk, 0x4); //xr1: ss0, s0
S32LDD(xr3, blk, 0x20); //xr3: ss2, s2
S32LDD(xr11, blk, 0x40); //xr11: ss4, s4
S32LDD(xr13, blk, 0x60); //xr13: ss6, s6
D16MUL_HW(xr15, xr5, xr1, xr2); //xr15: ss0*c4, xr9: s0*c4
D16MAC_AA_HW(xr15,xr5,xr11,xr2); //xr15: ss0*c4+ss4*c4, xr9: s0*c4+s4*c4
D16MUL_LW(xr10, xr5, xr3, xr9); //xr10: ss2*c2, xr9: s2*c2
D16MAC_AA_LW(xr10,xr6,xr13,xr9); //xr10: ss2*c2+ss6*c6, xr9: s2*c2+s6*c6
D32SAR(xr15,xr15,xr2,xr2,15);
S32SFL(xr0,xr15,xr2,xr15,ptn3); //xr15: (ss0*c4+ss4*c4)>>15
D32SAR(xr10,xr10,xr9,xr9,15);
S32SFL(xr0,xr10,xr9,xr10,ptn3); //xr10: (ss2*c2+ss6*c6)>>15
S32LDD(xr2, blk, 0x10); //xr2: ss1, s1
S32LDD(xr4, blk, 0x30); //xr4: ss3, s3
Q16ADD_AS_WW(xr15,xr15,xr10,xr9); //xr15: rnd(ss0*c4+ss4*c4)>>15+rnd(ss2*c2+ss6*c6)>>15
// :rnd(s0*c4+s4*c4)>>15 + rnd(s2*c2 + s6*c6)>>15
//xr9: rnd(ss0*c4+ss4*c4)>>15 - rnd(ss2*c2+ss6*c6)>>15
// : rnd(s0*c4+s4*c4)>>15 - rnd(s2*c2 + s6*c6)>>15
D16MUL_HW(xr10, xr5, xr1, xr1); //xr10: ss0*c4, xr1: s0*c4
D16MAC_SS_HW(xr10,xr5,xr11,xr1); //xr10: ss0*c4-ss4*c4, xr1: s0*c4 - s4*c4
D16MUL_LW(xr11, xr6, xr3, xr12); //xr11: ss2*c6, xr1: s2*c6
D16MAC_SS_LW(xr11,xr5,xr13,xr12); //xr11: ss2*c6-ss6*c2, xr1: s2*c6-s6*c2
D32SAR(xr10,xr10,xr1,xr1,15);
S32SFL(xr0,xr10,xr1,xr10,ptn3); //xr10: (ss0*c4-ss4*c4)>>15 // : (s0*c4 - s4*c4)>>15
D32SAR(xr11,xr11,xr12,xr12,15);
S32SFL(xr0,xr11,xr12,xr11,ptn3); //xr11:(ss2*c6-ss6*c2)>>15
// :(s2*c6-s6*c2)>>15
S32LDD(xr12, blk, 0x50); //xr12: ss5, s5
S32LDD(xr14, blk, 0x70); //xr14: ss7, s7
Q16ADD_AS_WW(xr10,xr10,xr11,xr1); //xr10: rnd(ss0*c4-ss4*c4)>>15)+rnd(ss2*c6-ss6*c2)>>15
// : rnd(s0*c4 - s4*c4)>>15 +rnd(s2*c6 - s6*c2)>>15
//xr1 : rnd(ss0*c4-ss4*c4)>>15-rnd(ss2*c6-ss6*c2)>>15
// : rnd(s0*c4 - s4*c4)>>15-rnd(s2*c6 - s6*c2)>>15
D16MUL_HW(xr11, xr7, xr2, xr13); //xr11: ss1*c1, xr13: s1*c1
D16MAC_AA_LW(xr11,xr7,xr4,xr13); //xr11: ss1*c1+ss3*c3, xr13: s1*c1+s3*c3
D16MAC_AA_LW(xr11,xr8,xr12,xr13); //xr11: ss1*c1+ss3*c3+ss5*c5 //xr13: s1*c1+s3*c3+s5*c5
D16MAC_AA_HW(xr11,xr8,xr14,xr13); //xr11: ss1*c1+ss3*c3+ss5*c5+ss7*c7
//xr13: s1*c1+s3*c3+s5*c5+s7*c7
D16MUL_LW(xr3, xr7, xr2, xr5); //xr3: ss1*c3, xr13: s1*c3
D16MAC_SS_HW(xr3,xr8,xr4,xr5); //xr3: ss1*c3-ss3*c7, xr13: s1*c3-s3*c7
D16MAC_SS_HW(xr3,xr7,xr12,xr5); //xr3: ss1*c3-ss3*c7-ss5*c1
//xr13: s1*c3-s3*c7-s5*c1
D16MAC_SS_LW(xr3,xr8,xr14,xr5); //xr3: ss1*c3-ss3*c7-ss5*c1-ss7*c5
//xr13: s1*c3-s3*c7-s7*c5
D32SAR(xr11,xr11,xr13,xr13,15);
S32SFL(xr0,xr11,xr13,xr11,ptn3); //xr11: (ss1*c1+ss3*c3+ss5*c5+ss7*c7)>>15 // : (s1*c1+s3*c3+s5*c5+s7*c7)>>15
D32SAR(xr3,xr3,xr5,xr5,15);
S32SFL(xr0,xr3,xr5,xr3,ptn3); //xr3: (ss1*c3-ss3*c7-ss5*c1-ss7*c5)>>15
// : (s1*c3-s3*c7-s7*c5)>>15
D16MUL_LW(xr5, xr8, xr2, xr13); //xr5: ss1*c5, xr13:s1*c5
D16MAC_SS_HW(xr5,xr7,xr4,xr13); //xr5: ss1*c5-ss3*c1, xr13:s1*c5-s3*c1
D16MAC_AA_HW(xr5,xr8,xr12,xr13); //xr5: ss1*c5-ss3*c1+ss5*c7
// : s1*c5 - s3*c1+ s5*c7
D16MAC_AA_LW(xr5,xr7,xr14,xr13); //xr5: ss1*c5-ss3*c1+ss5*c7+ss7*c1
// : s1*c5 - s3*c1+ s5*c7+ s7*c1
D16MUL_HW(xr2, xr8, xr2, xr6); //xr2: ss1*c7, xr13: s1*c7
D16MAC_SS_LW(xr2,xr8,xr4,xr6); //xr2: ss1*c7-ss3*c5, xr13: s1*c7-s3*c5
D16MAC_AA_LW(xr2,xr7,xr12,xr6); //xr2: ss1*c7-ss3*c5+ss5*c1 //xr13: s1*c7-s3*c5+s5*c1
D16MAC_SS_HW(xr2,xr7,xr14,xr6); //xr2: ss1*c7-ss3*c5+ss5*c1-ss7*c3
//xr13: s1*c7-s3*c5+s5*c1-s7*c3
D32SAR(xr5,xr5,xr13,xr13,15);
S32SFL(xr0,xr5,xr13,xr5,ptn3); //xr5: (ss1*c5-ss3*c1+ss5*c7+ss7*c1)>>15 // :(s1*c5 - s3*c1+ s5*c7+ s7*c1)>>15
D32SAR(xr2,xr2,xr6,xr6,15);
S32SFL(xr0,xr2,xr6,xr2,ptn3); //xr2:(ss1*c7-ss3*c5+ss5*c1-ss7*c3)>>15
// :(s1*c7-s3*c5+s5*c1-s7*c3)>>15
S32I2M(xr4, 0x00080008);//round value 8;
Q16ADD_AS_WW(xr15,xr15,xr11,xr11); //xr15:rnd(ss0*c4+ss4*c4)>>16+rnd(ss2*c2+ss6*c6)>>16+
// rnd(ss1*c1+ss3*c3+ss5*c5+ss7*c7)>>16
// rnd(s0*c4+s4*c4)>>16 + rnd(s2*c2 + s6*c6)>>16+
// rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>16
//xr11:rnd(ss0*c4+ss4*c4)>>16+rnd(ss2*c2+ss6*c6)>>16-
// rnd(ss1*c1+ss3*c3+ss5*c5+ss7*c7)>>16
// rnd(s0*c4+s4*c4)>>16 + rnd(s2*c2 + s6*c6)>>16-
// rnd(s1*c1+s3*c3+s5*c5+s7*c7)>>16
Q16ADD_AS_WW(xr10,xr10,xr3,xr3); //xr10:rnd(ss0*c4-ss4*c4)>>16)+rnd(ss2*c6-ss6*c2)>>16+
// rnd(ss1*c3-ss3*c7-ss5*c1-ss7*c5)>>16
// rnd(s0*c4 - s4*c4)>>16 +rnd(s2*c6 - s6*c2)>>16+
// rnd(s1*c3-s3*c7-s7*c5)>>16
//xr10:rnd(ss0*c4-ss4*c4)>>16)+rnd(ss2*c6-ss6*c2)>>16-
// rnd(ss1*c3-ss3*c7-ss5*c1-ss7*c5)>>16
// rnd(s0*c4 - s4*c4)>>16 +rnd(s2*c6 - s6*c2)>>16-
// rnd(s1*c3-s3*c7-s7*c5)>>16
Q16ADD_AS_WW(xr1,xr1,xr5,xr5); //xr1: rnd(ss0*c4-ss4*c4)>>16-rnd(ss2*c6-ss6*c2)>>16+
// rnd(ss1*c5-ss3*c1+ss5*c7+ss7*c1)>>16
// rnd(s0*c4 - s4*c4)>>16 +rnd(s2*c6 - s6*c2)>>16+
// rnd(s1*c5 - s3*c1+ s5*c7+ s7*c1)>>16
//xr1: rnd(ss0*c4-ss4*c4)>>16-rnd(ss2*c6-ss6*c2)>>16-
// rnd(ss1*c5-ss3*c1+ss5*c7+ss7*c1)>>16
// rnd(s0*c4 - s4*c4)>>16 +rnd(s2*c6 - s6*c2)>>16-
// rnd(s1*c5 - s3*c1+ s5*c7+ s7*c1)>>16
Q16ADD_AS_WW(xr9,xr9,xr2,xr2); //xr9: rnd(ss0*c4+ss4*c4)>>16 - rnd(ss2*c2+ss6*c6)>>16+
// rnd(ss1*c7-ss3*c5+ss5*c1-ss7*c3)>>16
// rnd(s0*c4+s4*c4)>>16 - rnd(s2*c2 + s6*c6)>>16+
// rnd(s1*c7-s3*c5+s5*c1-s7*c3)>>16
//xr9: rnd(ss0*c4+ss4*c4)>>16 - rnd(ss2*c2+ss6*c6)>>16-
// rnd(ss1*c7-ss3*c5+ss5*c1-ss7*c3)>>16
// rnd(s0*c4+s4*c4)>>16 - rnd(s2*c2 + s6*c6)>>16-
// rnd(s1*c7-s3*c5+s5*c1-s7*c3)>>16
Q16ACCM_AA(xr15,xr4,xr4,xr10);
Q16ACCM_AA(xr11,xr4,xr4,xr1);
Q16ACCM_AA(xr9,xr4,xr4,xr2);
Q16ACCM_AA(xr5,xr4,xr4,xr3);
Q16SAR(xr15,xr15,xr10,xr10,4);
Q16SAR(xr11,xr11,xr1,xr1,4);
Q16SAR(xr9,xr9,xr2,xr2,4);
Q16SAR(xr5,xr5,xr3,xr3,4);
S32STD(xr15, blk, 0x00);
S32STD(xr10, blk, 0x10);
S32STD(xr1, blk, 0x20);
S32STD(xr9, blk, 0x30);
S32STD(xr2, blk, 0x40);
S32STD(xr5, blk, 0x50);
S32STD(xr3, blk, 0x60);
S32STD(xr11, blk, 0x70);
}
blk = input - 8;
src -= stride;
for (i=0; i<8; i++) {
S32LDIV(xr1, src, stride, 0x0);
S32LDI(xr3, blk, 0x10);
S32LDD(xr4, blk, 0x4);
Q8ACCE_AA(xr4, xr1, xr0, xr3);
S32LDD(xr2, src, 0x4);
S32LDD(xr5, blk, 0x8);
S32LDD(xr6, blk, 0xc);
Q8ACCE_AA(xr6, xr2, xr0, xr5);
Q16SAT(xr1, xr4, xr3);
S32STD(xr1, src, 0x0);
Q16SAT(xr2, xr6, xr5);
S32STD(xr2, src, 0x4);
}
}
void Predict_16x16_C(const NEW_GMC_DATA * const This,
uint8_t *dst,
const uint8_t *src,
int dststride,
int srcstride,
int x,
int y,
int rounding)
{
const int W = This->sW;
const int H = This->sH;
const int rho = 3 - This->accuracy;
const int Rounder = ( (1<<7) - (rounding<<(2*rho)) ) << 16;
const int dUx = This->dU[0];
const int dVx = This->dV[0];
const int dUy = This->dU[1];
const int dVy = This->dV[1];
int Uo = This->Uo + 16*(dUy*y + dUx*x);
int Vo = This->Vo + 16*(dVy*y + dVx*x);
int i, j;
dst += 16;
{
unsigned int ri = 16;
unsigned int rj = 16;
int Offset;
int u,v;
uint8_t *srctmp;
uint32_t tmpf = 0;
S32I2M(xr15,dUx);
S32I2M(xr14,dVx);
S32I2M(xr13,dUy);
S32I2M(xr12,dVy);
S32I2M(xr11,Uo); // Uo 11
S32I2M(xr10,Vo); // Vo 10
S32I2M(xr5, Rounder);
for (j = 16; j>0; --j)
{
D32SLL(xr9,xr11,xr10,xr8, 0x0); // U 9 ,V 8
D32ASUM_AA(xr11,xr13,xr12,xr10); // += dUy; +=dVy;
for (i = -16; i<0; ++i)
{
ri = 16;
rj = 16;
// ( U >> 16 ) ,( V >> 16 )
D32SAR(xr7,xr9,xr8,xr6, 0x8);
D32SAR(xr7,xr7,xr6,xr6, 0x8);
D32SLLV(xr7,xr6, rho); // << rho
u = S32M2I(xr7);
v = S32M2I(xr6);
D32ASUM_AA(xr9,xr15,xr14,xr8); // U += dUx; V += dVx;
if (u > 0 && u <= W)
{
ri = MTab[u&15];
Offset = u>>4;
}
else
{
if (u > W)
Offset = W>>4;
else
Offset = 0;
ri = MTab[0];
}
if (v > 0 && v <= H)
{
rj = MTab[v&15];
Offset += (v>>4)*srcstride;
}
