Commit 7e618974 authored by Gaël Hendryckx's avatar Gaël Hendryckx
Browse files

* Le bug des pixels invers�s de la dct mmx � �t� corrig� (et non plus sau-

  vagement kludg�).

* La sparse idct fonctionne.

* Plusieurs versions de la dct ont �t� inclues dans vdec_idct pour pou-
  voir choisir la plus performante quand tout marchera.
parent 57509f75
......@@ -74,11 +74,13 @@ void vdec_InitIDCT (vdec_thread_t * p_vdec)
p_pre[i*64+i] = 1 << SPARSE_SCALE_FACTOR;
vdec_IDCT( p_vdec, &p_pre[i*64], 0) ;
}
return;
}
void vdec_SparseIDCT (vdec_thread_t * p_vdec, dctelem_t * p_block,
int i_sparse_pos)
{
/*debug*/
short int val;
int * dp;
int v;
......@@ -90,7 +92,20 @@ void vdec_SparseIDCT (vdec_thread_t * p_vdec, dctelem_t * p_block,
if ( i_sparse_pos == 0 )
{
dp=(int *)p_block;
val= *p_block >> 6;
// v=*p_block;
/* cuisine a verifier */
/* if (v < 0)
{
val=-v;
val+=4;
val/=8;
val=-val;
}
else
{*/
/* val= (v + 4) /8; */
val=RIGHT_SHIFT((*p_block + 4), 3);
// }
/* Compute int to assign. This speeds things up a bit */
v = ((val & 0xffff) | (val << 16));
dp[0] = v; dp[1] = v; dp[2] = v; dp[3] = v;
......@@ -138,7 +153,227 @@ void vdec_SparseIDCT (vdec_thread_t * p_vdec, dctelem_t * p_block,
void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, int i_idontcare )
{
//IDCT_mmx(p_block);
#if 1
#if 0
/* dct classique: pour tester la meilleure entre la classique et la */
/* no classique */
s32 tmp0, tmp1, tmp2, tmp3;
s32 tmp10, tmp11, tmp12, tmp13;
s32 z1, z2, z3, z4, z5;
dctelem_t * dataptr;
int rowctr;
SHIFT_TEMPS
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = p_block;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--)
{
/* Due to quantization, we will usually find that many of the input
* coefficients are zero, especially the AC terms. We can exploit this
* by short-circuiting the IDCT calculation for any row in which all
* the AC terms are zero. In that case each output is equal to the
* DC coefficient (with scale factor as needed).
* With typical images and quantization tables, half or more of the
* row DCT calculations can be simplified this way.
*/
if ((dataptr[1] | dataptr[2] | dataptr[3] | dataptr[4] |
dataptr[5] | dataptr[6] | dataptr[7]) == 0)
{
/* AC terms all zero */
dctelem_t dcval = (dctelem_t) (dataptr[0] << PASS1_BITS);
dataptr[0] = dcval;
dataptr[1] = dcval;
dataptr[2] = dcval;
dataptr[3] = dcval;
dataptr[4] = dcval;
dataptr[5] = dcval;
dataptr[6] = dcval;
dataptr[7] = dcval;
dataptr += DCTSIZE; /* advance pointer to next row */
continue;
}
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (s32) dataptr[2];
z3 = (s32) dataptr[6];
z1 = MULTIPLY(z2 + z3, FIX(0.541196100));
tmp2 = z1 + MULTIPLY(z3, - FIX(1.847759065));
tmp3 = z1 + MULTIPLY(z2, FIX(0.765366865));
tmp0 = ((s32) dataptr[0] + (s32) dataptr[4]) << CONST_BITS;
tmp1 = ((s32) dataptr[0] - (s32) dataptr[4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (s32) dataptr[7];
tmp1 = (s32) dataptr[5];
tmp2 = (s32) dataptr[3];
tmp3 = (s32) dataptr[1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX(1.175875602)); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX(0.298631336)); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX(2.053119869)); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX(3.072711026)); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX(1.501321110)); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX(0.899976223)); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX(2.562915447)); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX(1.961570560)); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX(0.390180644)); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[0] = (dctelem_t) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
dataptr[7] = (dctelem_t) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
dataptr[1] = (dctelem_t) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
dataptr[6] = (dctelem_t) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
dataptr[2] = (dctelem_t) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
dataptr[5] = (dctelem_t) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
dataptr[3] = (dctelem_t) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
dataptr[4] = (dctelem_t) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
dataptr += DCTSIZE; /* advance pointer to next row */
}
/* Pass 2: process columns. */
/* Note that we must descale the results by a factor of 8 == 2**3, */
/* and also undo the PASS1_BITS scaling. */
dataptr = p_block;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--)
{
/* Columns of zeroes can be exploited in the same way as we did with rows.
* However, the row calculation has created many nonzero AC terms, so the
* simplification applies less often (typically 5% to 10% of the time).
* On machines with very fast multiplication, it's possible that the
* test takes more time than it's worth. In that case this section
* may be commented out.
*/
#ifndef NO_ZERO_COLUMN_TEST /*ajoute un test mais evite des calculs */
if ((dataptr[DCTSIZE*1] | dataptr[DCTSIZE*2] | dataptr[DCTSIZE*3] |
dataptr[DCTSIZE*4] | dataptr[DCTSIZE*5] | dataptr[DCTSIZE*6] |
dataptr[DCTSIZE*7]) == 0)
{
/* AC terms all zero */
dctelem_t dcval = (dctelem_t) DESCALE((s32) dataptr[0], PASS1_BITS+3);
dataptr[DCTSIZE*0] = dcval;
dataptr[DCTSIZE*1] = dcval;
dataptr[DCTSIZE*2] = dcval;
dataptr[DCTSIZE*3] = dcval;
dataptr[DCTSIZE*4] = dcval;
dataptr[DCTSIZE*5] = dcval;
dataptr[DCTSIZE*6] = dcval;
dataptr[DCTSIZE*7] = dcval;
dataptr++; /* advance pointer to next column */
continue;
}
#endif
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (s32) dataptr[DCTSIZE*2];
z3 = (s32) dataptr[DCTSIZE*6];
z1 = MULTIPLY(z2 + z3, FIX(0.541196100));
tmp2 = z1 + MULTIPLY(z3, - FIX(1.847759065));
tmp3 = z1 + MULTIPLY(z2, FIX(0.765366865));
tmp0 = ((s32) dataptr[DCTSIZE*0] + (s32) dataptr[DCTSIZE*4]) << CONST_BITS;
tmp1 = ((s32) dataptr[DCTSIZE*0] - (s32) dataptr[DCTSIZE*4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (s32) dataptr[DCTSIZE*7];
tmp1 = (s32) dataptr[DCTSIZE*5];
tmp2 = (s32) dataptr[DCTSIZE*3];
tmp3 = (s32) dataptr[DCTSIZE*1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX(1.175875602)); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX(0.298631336)); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX(2.053119869)); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX(3.072711026)); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX(1.501321110)); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX(0.899976223)); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX(2.562915447)); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX(1.961570560)); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX(0.390180644)); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[DCTSIZE*0] = (dctelem_t) DESCALE(tmp10 + tmp3,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*7] = (dctelem_t) DESCALE(tmp10 - tmp3,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*1] = (dctelem_t) DESCALE(tmp11 + tmp2,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*6] = (dctelem_t) DESCALE(tmp11 - tmp2,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*2] = (dctelem_t) DESCALE(tmp12 + tmp1,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*5] = (dctelem_t) DESCALE(tmp12 - tmp1,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*3] = (dctelem_t) DESCALE(tmp13 + tmp0,
CONST_BITS+PASS1_BITS+3);
dataptr[DCTSIZE*4] = (dctelem_t) DESCALE(tmp13 - tmp0,
CONST_BITS+PASS1_BITS+3);
dataptr++; /* advance pointer to next column */
}
#endif
#if 1 /*dct avec no classique*/
s32 tmp0, tmp1, tmp2, tmp3;
s32 tmp10, tmp11, tmp12, tmp13;
s32 z1, z2, z3, z4, z5;
......
......@@ -1460,7 +1460,7 @@ static void vpar_DecodeMPEG2Intra( vpar_thread_t * p_vpar, macroblock_t * p_mb,
: i_level;
break;
case DCT_EOB:
if( i_nc <= 0 )
if( i_nc <= 1 )
{
p_mb->pf_idct[i_b] = vdec_SparseIDCT;
p_mb->pi_sparse_pos[i_b] = i_coef;
......
......@@ -149,8 +149,8 @@ boolean_t vpar_SynchroChoose( vpar_thread_t * p_vpar, int i_coding_type,
int i_structure )
{
// return( 1 );
return( i_coding_type == I_CODING_TYPE || i_coding_type == P_CODING_TYPE );
//return( i_coding_type == I_CODING_TYPE );
// return( i_coding_type == I_CODING_TYPE || i_coding_type == P_CODING_TYPE );
return( i_coding_type == I_CODING_TYPE );
}
/*****************************************************************************
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment