Optimize 1D DCT transform used in MP{1,2,3}. Makes also this function suitable

for being moved to the shared DCT framework in the future.

Originally committed as revision 23514 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Vitor Sessak 2010-06-07 11:43:12 +00:00
parent 77d3f1f792
commit 525ac37fd2

View File

@ -527,46 +527,59 @@ static av_cold int decode_init(AVCodecContext * avctx)
/* butterfly operator */
#define BF(a, b, c, s)\
{\
tmp0 = val##a + val##b;\
tmp1 = val##a - val##b;\
val##a = tmp0;\
val##b = MULH3(tmp1, c, 1<<(s));\
}
#define BF0(a, b, c, s)\
{\
tmp0 = tab[a] + tab[b];\
tmp1 = tab[a] - tab[b];\
tab[a] = tmp0;\
tab[b] = MULH3(tmp1, c, 1<<(s));\
val##a = tmp0;\
val##b = MULH3(tmp1, c, 1<<(s));\
}
#define BF1(a, b, c, d)\
{\
BF(a, b, COS4_0, 1);\
BF(c, d,-COS4_0, 1);\
tab[c] += tab[d];\
val##c += val##d;\
}
#define BF2(a, b, c, d)\
{\
BF(a, b, COS4_0, 1);\
BF(c, d,-COS4_0, 1);\
tab[c] += tab[d];\
tab[a] += tab[c];\
tab[c] += tab[b];\
tab[b] += tab[d];\
val##c += val##d;\
val##a += val##c;\
val##c += val##b;\
val##b += val##d;\
}
#define ADD(a, b) tab[a] += tab[b]
#define ADD(a, b) val##a += val##b
/* DCT32 without 1/sqrt(2) coef zero scaling. */
static void dct32(INTFLOAT *out, INTFLOAT *tab)
static void dct32(INTFLOAT *out, const INTFLOAT *tab)
{
INTFLOAT tmp0, tmp1;
INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 ,
val8 , val9 , val10, val11, val12, val13, val14, val15,
val16, val17, val18, val19, val20, val21, val22, val23,
val24, val25, val26, val27, val28, val29, val30, val31;
/* pass 1 */
BF( 0, 31, COS0_0 , 1);
BF(15, 16, COS0_15, 5);
BF0( 0, 31, COS0_0 , 1);
BF0(15, 16, COS0_15, 5);
/* pass 2 */
BF( 0, 15, COS1_0 , 1);
BF(16, 31,-COS1_0 , 1);
/* pass 1 */
BF( 7, 24, COS0_7 , 1);
BF( 8, 23, COS0_8 , 1);
BF0( 7, 24, COS0_7 , 1);
BF0( 8, 23, COS0_8 , 1);
/* pass 2 */
BF( 7, 8, COS1_7 , 4);
BF(23, 24,-COS1_7 , 4);
@ -576,14 +589,14 @@ static void dct32(INTFLOAT *out, INTFLOAT *tab)
BF(16, 23, COS2_0 , 1);
BF(24, 31,-COS2_0 , 1);
/* pass 1 */
BF( 3, 28, COS0_3 , 1);
BF(12, 19, COS0_12, 2);
BF0( 3, 28, COS0_3 , 1);
BF0(12, 19, COS0_12, 2);
/* pass 2 */
BF( 3, 12, COS1_3 , 1);
BF(19, 28,-COS1_3 , 1);
/* pass 1 */
BF( 4, 27, COS0_4 , 1);
BF(11, 20, COS0_11, 2);
BF0( 4, 27, COS0_4 , 1);
BF0(11, 20, COS0_11, 2);
/* pass 2 */
BF( 4, 11, COS1_4 , 1);
BF(20, 27,-COS1_4 , 1);
@ -605,14 +618,14 @@ static void dct32(INTFLOAT *out, INTFLOAT *tab)
/* pass 1 */
BF( 1, 30, COS0_1 , 1);
BF(14, 17, COS0_14, 3);
BF0( 1, 30, COS0_1 , 1);
BF0(14, 17, COS0_14, 3);
/* pass 2 */
BF( 1, 14, COS1_1 , 1);
BF(17, 30,-COS1_1 , 1);
/* pass 1 */
BF( 6, 25, COS0_6 , 1);
BF( 9, 22, COS0_9 , 1);
BF0( 6, 25, COS0_6 , 1);
BF0( 9, 22, COS0_9 , 1);
/* pass 2 */
BF( 6, 9, COS1_6 , 2);
BF(22, 25,-COS1_6 , 2);
@ -623,14 +636,14 @@ static void dct32(INTFLOAT *out, INTFLOAT *tab)
BF(25, 30,-COS2_1 , 1);
/* pass 1 */
BF( 2, 29, COS0_2 , 1);
BF(13, 18, COS0_13, 3);
BF0( 2, 29, COS0_2 , 1);
BF0(13, 18, COS0_13, 3);
/* pass 2 */
BF( 2, 13, COS1_2 , 1);
BF(18, 29,-COS1_2 , 1);
/* pass 1 */
BF( 5, 26, COS0_5 , 1);
BF(10, 21, COS0_10, 1);
BF0( 5, 26, COS0_5 , 1);
BF0(10, 21, COS0_10, 1);
/* pass 2 */
BF( 5, 10, COS1_5 , 2);
BF(21, 26,-COS1_5 , 2);
@ -669,22 +682,22 @@ static void dct32(INTFLOAT *out, INTFLOAT *tab)
ADD(13, 11);
ADD(11, 15);
out[ 0] = tab[0];
out[16] = tab[1];
out[ 8] = tab[2];
out[24] = tab[3];
out[ 4] = tab[4];
out[20] = tab[5];
out[12] = tab[6];
out[28] = tab[7];
out[ 2] = tab[8];
out[18] = tab[9];
out[10] = tab[10];
out[26] = tab[11];
out[ 6] = tab[12];
out[22] = tab[13];
out[14] = tab[14];
out[30] = tab[15];
out[ 0] = val0;
out[16] = val1;
out[ 8] = val2;
out[24] = val3;
out[ 4] = val4;
out[20] = val5;
out[12] = val6;
out[28] = val7;
out[ 2] = val8;
out[18] = val9;
out[10] = val10;
out[26] = val11;
out[ 6] = val12;
out[22] = val13;
out[14] = val14;
out[30] = val15;
ADD(24, 28);
ADD(28, 26);
@ -694,22 +707,22 @@ static void dct32(INTFLOAT *out, INTFLOAT *tab)
ADD(29, 27);
ADD(27, 31);
out[ 1] = tab[16] + tab[24];
out[17] = tab[17] + tab[25];
out[ 9] = tab[18] + tab[26];
out[25] = tab[19] + tab[27];
out[ 5] = tab[20] + tab[28];
out[21] = tab[21] + tab[29];
out[13] = tab[22] + tab[30];
out[29] = tab[23] + tab[31];
out[ 3] = tab[24] + tab[20];
out[19] = tab[25] + tab[21];
out[11] = tab[26] + tab[22];
out[27] = tab[27] + tab[23];
out[ 7] = tab[28] + tab[18];
out[23] = tab[29] + tab[19];
out[15] = tab[30] + tab[17];
out[31] = tab[31];
out[ 1] = val16 + val24;
out[17] = val17 + val25;
out[ 9] = val18 + val26;
out[25] = val19 + val27;
out[ 5] = val20 + val28;
out[21] = val21 + val29;
out[13] = val22 + val30;
out[29] = val23 + val31;
out[ 3] = val24 + val20;
out[19] = val25 + val21;
out[11] = val26 + val22;
out[27] = val27 + val23;
out[ 7] = val28 + val18;
out[23] = val29 + val19;
out[15] = val30 + val17;
out[31] = val31;
}
#if CONFIG_FLOAT