uint32_t
dequant_mpeg_inter_altivec_c(int16_t * data,
const int16_t * coeff,
const uint32_t quant,
const uint16_t * mpeg_quant_matrices)
{
register uint32_t sum;
register const uint16_t *inter_matrix = get_inter_matrix(mpeg_quant_matrices);
register vec_sint16_t ox00;
register vec_sint16_t v2048;
register vec_sint16_t level;
register vec_uint16_t vinter;
register vec_uint32_t hi,lo;
register vec_uint32_t sw_hi,sw_lo;
register vec_uint32_t swap;
register vec_uint32_t t,v16;
vec_uint32_t vsum;
vec_uint32_t vquant;
vector bool short zero_eq;
vector bool short zero_less;
vector bool short overflow;
#ifdef DEBUG
if((long)data & 0xf)
fprintf(stderr, "xvidcore: error in dequant_mpeg_inter_altivec_c, incorrect align: %x\n", data);
#endif
/* Initialization */
ox00 = vec_splat_s16(0);
v16 = vec_splat_u32(-16);
v2048 = vec_rl(vec_splat_s16(8),vec_splat_u16(8));
vsum = (vec_uint32_t)ox00;
*((uint32_t*)&vquant) = quant;
vquant = vec_splat(vquant,0);
swap = vec_rl(vquant,v16);
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
DEQUANT_MPEG_INTER();
sum = ((uint32_t*)&vsum)[0];
sum ^= ((uint32_t*)&vsum)[1];
sum ^= ((uint32_t*)&vsum)[2];
sum ^= ((uint32_t*)&vsum)[3];
/* mismatch control */
if((sum & 1) == 0) {
data -= 1;
*data ^= 1;
}
return 0;
}
/* Quantize all blocks -- Inter mode */
static __inline uint8_t
MBQuantInter(const MBParam * pParam,
const FRAMEINFO * const frame,
const MACROBLOCK * pMB,
int16_t data[6 * 64],
int16_t qcoeff[6 * 64],
int bvop,
int limit)
{
int i;
uint8_t cbp = 0;
int sum;
int code_block, mpeg;
quant_interFuncPtr const quant[2] =
{
quant_h263_inter,
quant_mpeg_inter
};
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
for (i = 0; i < 6; i++) {
/* Quantize the block */
start_timer();
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices);
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) {
const uint16_t *matrix;
const static uint16_t h263matrix[] =
{
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16,
16, 16, 16, 16, 16, 16, 16, 16
};
matrix = (mpeg)?get_inter_matrix(pParam->mpeg_quant_matrices):h263matrix;
sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64],
pMB->quant, &scan_tables[0][0],
matrix,
63,
sum,
pMB->lambda[i],
pMB->rel_var8[i],
!!(frame->vop_flags & XVID_VOP_RD_PSNRHVSM));
}
stop_quant_timer();
/*
* We code the block if the sum is higher than the limit and if the first
* two AC coefficients in zig zag order are not zero.
*/
code_block = 0;
if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) {
code_block = 1;
} else {
if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) {
/* dark blocks prevention for direct mode */
if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0))
code_block = 1;
} else {
/* not direct mode */
if (qcoeff[i*64] != 0)
code_block = 1;
}
}
/* Set the corresponding cbp bit */
cbp |= code_block << (5 - i);
}
return(cbp);
}