ffmpeg/libavcodec/dsputil.c
Nick Kurshev 1e98dffb7a Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
Originally committed as revision 274 to svn://svn.ffmpeg.org/ffmpeg/trunk
2002-01-20 14:48:02 +00:00

534 lines
21 KiB
C

/*
* DSP utils
* Copyright (c) 2000, 2001 Gerard Lantau.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
#include <stdlib.h>
#include <stdio.h>
#include "avcodec.h"
#include "dsputil.h"
#include "simple_idct.h"
void (*ff_idct)(DCTELEM *block);
void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size);
void (*put_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
void (*add_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);
op_pixels_abs_func pix_abs16x16;
op_pixels_abs_func pix_abs16x16_x2;
op_pixels_abs_func pix_abs16x16_y2;
op_pixels_abs_func pix_abs16x16_xy2;
UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
UINT32 squareTbl[512];
extern UINT16 default_intra_matrix[64];
extern UINT16 default_non_intra_matrix[64];
UINT8 zigzag_direct[64] = {
0, 1, 8, 16, 9, 2, 3, 10,
17, 24, 32, 25, 18, 11, 4, 5,
12, 19, 26, 33, 40, 48, 41, 34,
27, 20, 13, 6, 7, 14, 21, 28,
35, 42, 49, 56, 57, 50, 43, 36,
29, 22, 15, 23, 30, 37, 44, 51,
58, 59, 52, 45, 38, 31, 39, 46,
53, 60, 61, 54, 47, 55, 62, 63
};
UINT8 ff_alternate_horizontal_scan[64] = {
0, 1, 2, 3, 8, 9, 16, 17,
10, 11, 4, 5, 6, 7, 15, 14,
13, 12, 19, 18, 24, 25, 32, 33,
26, 27, 20, 21, 22, 23, 28, 29,
30, 31, 34, 35, 40, 41, 48, 49,
42, 43, 36, 37, 38, 39, 44, 45,
46, 47, 50, 51, 56, 57, 58, 59,
52, 53, 54, 55, 60, 61, 62, 63,
};
UINT8 ff_alternate_vertical_scan[64] = {
0, 8, 16, 24, 1, 9, 2, 10,
17, 25, 32, 40, 48, 56, 57, 49,
41, 33, 26, 18, 3, 11, 4, 12,
19, 27, 34, 42, 50, 58, 35, 43,
51, 59, 20, 28, 5, 13, 6, 14,
21, 29, 36, 44, 52, 60, 37, 45,
53, 61, 22, 30, 7, 15, 23, 31,
38, 46, 54, 62, 39, 47, 55, 63,
};
/* Input permutation for the simple_idct_mmx */
static UINT8 simple_mmx_permutation[64]={
0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,
0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,
0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,
0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,
0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,
0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,
0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,
0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
};
/* used to skip zeros at the end */
UINT8 zigzag_end[64];
UINT8 permutation[64];
//UINT8 invPermutation[64];
static void build_zigzag_end()
{
int lastIndex;
int lastIndexAfterPerm=0;
for(lastIndex=0; lastIndex<64; lastIndex++)
{
if(zigzag_direct[lastIndex] > lastIndexAfterPerm)
lastIndexAfterPerm= zigzag_direct[lastIndex];
zigzag_end[lastIndex]= lastIndexAfterPerm + 1;
}
}
void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size)
{
DCTELEM *p;
const UINT8 *pix;
int i;
/* read the pixels */
p = block;
pix = pixels;
for(i=0;i<8;i++) {
p[0] = pix[0];
p[1] = pix[1];
p[2] = pix[2];
p[3] = pix[3];
p[4] = pix[4];
p[5] = pix[5];
p[6] = pix[6];
p[7] = pix[7];
pix += line_size;
p += 8;
}
}
void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)
{
const DCTELEM *p;
UINT8 *pix;
int i;
UINT8 *cm = cropTbl + MAX_NEG_CROP;
/* read the pixels */
p = block;
pix = pixels;
for(i=0;i<8;i++) {
pix[0] = cm[p[0]];
pix[1] = cm[p[1]];
pix[2] = cm[p[2]];
pix[3] = cm[p[3]];
pix[4] = cm[p[4]];
pix[5] = cm[p[5]];
pix[6] = cm[p[6]];
pix[7] = cm[p[7]];
pix += line_size;
p += 8;
}
}
void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)
{
const DCTELEM *p;
UINT8 *pix;
int i;
UINT8 *cm = cropTbl + MAX_NEG_CROP;
/* read the pixels */
p = block;
pix = pixels;
for(i=0;i<8;i++) {
pix[0] = cm[pix[0] + p[0]];
pix[1] = cm[pix[1] + p[1]];
pix[2] = cm[pix[2] + p[2]];
pix[3] = cm[pix[3] + p[3]];
pix[4] = cm[pix[4] + p[4]];
pix[5] = cm[pix[5] + p[5]];
pix[6] = cm[pix[6] + p[6]];
pix[7] = cm[pix[7] + p[7]];
pix += line_size;
p += 8;
}
}
#define PIXOP(BTYPE, OPNAME, OP, INCR) \
\
static void OPNAME ## _pixels(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
{ \
BTYPE *p; \
const UINT8 *pix; \
\
p = block; \
pix = pixels; \
do { \
OP(p[0], pix[0]); \
OP(p[1], pix[1]); \
OP(p[2], pix[2]); \
OP(p[3], pix[3]); \
OP(p[4], pix[4]); \
OP(p[5], pix[5]); \
OP(p[6], pix[6]); \
OP(p[7], pix[7]); \
pix += line_size; \
p += INCR; \
} while (--h);; \
} \
\
static void OPNAME ## _pixels_x2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
{ \
BTYPE *p; \
const UINT8 *pix; \
\
p = block; \
pix = pixels; \
do { \
OP(p[0], avg2(pix[0], pix[1])); \
OP(p[1], avg2(pix[1], pix[2])); \
OP(p[2], avg2(pix[2], pix[3])); \
OP(p[3], avg2(pix[3], pix[4])); \
OP(p[4], avg2(pix[4], pix[5])); \
OP(p[5], avg2(pix[5], pix[6])); \
OP(p[6], avg2(pix[6], pix[7])); \
OP(p[7], avg2(pix[7], pix[8])); \
pix += line_size; \
p += INCR; \
} while (--h); \
} \
\
static void OPNAME ## _pixels_y2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
{ \
BTYPE *p; \
const UINT8 *pix; \
const UINT8 *pix1; \
\
p = block; \
pix = pixels; \
pix1 = pixels + line_size; \
do { \
OP(p[0], avg2(pix[0], pix1[0])); \
OP(p[1], avg2(pix[1], pix1[1])); \
OP(p[2], avg2(pix[2], pix1[2])); \
OP(p[3], avg2(pix[3], pix1[3])); \
OP(p[4], avg2(pix[4], pix1[4])); \
OP(p[5], avg2(pix[5], pix1[5])); \
OP(p[6], avg2(pix[6], pix1[6])); \
OP(p[7], avg2(pix[7], pix1[7])); \
pix += line_size; \
pix1 += line_size; \
p += INCR; \
} while(--h); \
} \
\
static void OPNAME ## _pixels_xy2(BTYPE *block, const UINT8 *pixels, int line_size, int h) \
{ \
BTYPE *p; \
const UINT8 *pix; \
const UINT8 *pix1; \
\
p = block; \
pix = pixels; \
pix1 = pixels + line_size; \
do { \
OP(p[0], avg4(pix[0], pix[1], pix1[0], pix1[1])); \
OP(p[1], avg4(pix[1], pix[2], pix1[1], pix1[2])); \
OP(p[2], avg4(pix[2], pix[3], pix1[2], pix1[3])); \
OP(p[3], avg4(pix[3], pix[4], pix1[3], pix1[4])); \
OP(p[4], avg4(pix[4], pix[5], pix1[4], pix1[5])); \
OP(p[5], avg4(pix[5], pix[6], pix1[5], pix1[6])); \
OP(p[6], avg4(pix[6], pix[7], pix1[6], pix1[7])); \
OP(p[7], avg4(pix[7], pix[8], pix1[7], pix1[8])); \
pix += line_size; \
pix1 += line_size; \
p += INCR; \
} while(--h); \
} \
\
void (*OPNAME ## _pixels_tab[4])(BTYPE *block, const UINT8 *pixels, int line_size, int h) = { \
OPNAME ## _pixels, \
OPNAME ## _pixels_x2, \
OPNAME ## _pixels_y2, \
OPNAME ## _pixels_xy2, \
};
/* rounding primitives */
#define avg2(a,b) ((a+b+1)>>1)
#define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
#define op_put(a, b) a = b
#define op_avg(a, b) a = avg2(a, b)
#define op_sub(a, b) a -= b
PIXOP(UINT8, put, op_put, line_size)
PIXOP(UINT8, avg, op_avg, line_size)
PIXOP(DCTELEM, sub, op_sub, 8)
/* not rounding primitives */
#undef avg2
#undef avg4
#define avg2(a,b) ((a+b)>>1)
#define avg4(a,b,c,d) ((a+b+c+d+1)>>2)
PIXOP(UINT8, put_no_rnd, op_put, line_size)
PIXOP(UINT8, avg_no_rnd, op_avg, line_size)
/* motion estimation */
#undef avg2
#undef avg4
#define avg2(a,b) ((a+b+1)>>1)
#define avg4(a,b,c,d) ((a+b+c+d+2)>>2)
int pix_abs16x16_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
{
int s, i;
s = 0;
for(i=0;i<h;i++) {
s += abs(pix1[0] - pix2[0]);
s += abs(pix1[1] - pix2[1]);
s += abs(pix1[2] - pix2[2]);
s += abs(pix1[3] - pix2[3]);
s += abs(pix1[4] - pix2[4]);
s += abs(pix1[5] - pix2[5]);
s += abs(pix1[6] - pix2[6]);
s += abs(pix1[7] - pix2[7]);
s += abs(pix1[8] - pix2[8]);
s += abs(pix1[9] - pix2[9]);
s += abs(pix1[10] - pix2[10]);
s += abs(pix1[11] - pix2[11]);
s += abs(pix1[12] - pix2[12]);
s += abs(pix1[13] - pix2[13]);
s += abs(pix1[14] - pix2[14]);
s += abs(pix1[15] - pix2[15]);
pix1 += line_size;
pix2 += line_size;
}
return s;
}
int pix_abs16x16_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
{
int s, i;
s = 0;
for(i=0;i<h;i++) {
s += abs(pix1[0] - avg2(pix2[0], pix2[1]));
s += abs(pix1[1] - avg2(pix2[1], pix2[2]));
s += abs(pix1[2] - avg2(pix2[2], pix2[3]));
s += abs(pix1[3] - avg2(pix2[3], pix2[4]));
s += abs(pix1[4] - avg2(pix2[4], pix2[5]));
s += abs(pix1[5] - avg2(pix2[5], pix2[6]));
s += abs(pix1[6] - avg2(pix2[6], pix2[7]));
s += abs(pix1[7] - avg2(pix2[7], pix2[8]));
s += abs(pix1[8] - avg2(pix2[8], pix2[9]));
s += abs(pix1[9] - avg2(pix2[9], pix2[10]));
s += abs(pix1[10] - avg2(pix2[10], pix2[11]));
s += abs(pix1[11] - avg2(pix2[11], pix2[12]));
s += abs(pix1[12] - avg2(pix2[12], pix2[13]));
s += abs(pix1[13] - avg2(pix2[13], pix2[14]));
s += abs(pix1[14] - avg2(pix2[14], pix2[15]));
s += abs(pix1[15] - avg2(pix2[15], pix2[16]));
pix1 += line_size;
pix2 += line_size;
}
return s;
}
int pix_abs16x16_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
{
int s, i;
UINT8 *pix3 = pix2 + line_size;
s = 0;
for(i=0;i<h;i++) {
s += abs(pix1[0] - avg2(pix2[0], pix3[0]));
s += abs(pix1[1] - avg2(pix2[1], pix3[1]));
s += abs(pix1[2] - avg2(pix2[2], pix3[2]));
s += abs(pix1[3] - avg2(pix2[3], pix3[3]));
s += abs(pix1[4] - avg2(pix2[4], pix3[4]));
s += abs(pix1[5] - avg2(pix2[5], pix3[5]));
s += abs(pix1[6] - avg2(pix2[6], pix3[6]));
s += abs(pix1[7] - avg2(pix2[7], pix3[7]));
s += abs(pix1[8] - avg2(pix2[8], pix3[8]));
s += abs(pix1[9] - avg2(pix2[9], pix3[9]));
s += abs(pix1[10] - avg2(pix2[10], pix3[10]));
s += abs(pix1[11] - avg2(pix2[11], pix3[11]));
s += abs(pix1[12] - avg2(pix2[12], pix3[12]));
s += abs(pix1[13] - avg2(pix2[13], pix3[13]));
s += abs(pix1[14] - avg2(pix2[14], pix3[14]));
s += abs(pix1[15] - avg2(pix2[15], pix3[15]));
pix1 += line_size;
pix2 += line_size;
pix3 += line_size;
}
return s;
}
int pix_abs16x16_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size, int h)
{
int s, i;
UINT8 *pix3 = pix2 + line_size;
s = 0;
for(i=0;i<h;i++) {
s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));
s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));
s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
pix1 += line_size;
pix2 += line_size;
pix3 += line_size;
}
return s;
}
/* permute block according so that it corresponds to the MMX idct
order */
#ifdef SIMPLE_IDCT
/* general permutation, but perhaps slightly slower */
void block_permute(INT16 *block)
{
int i;
INT16 temp[64];
for(i=0; i<64; i++) temp[ block_permute_op(i) ] = block[i];
for(i=0; i<64; i++) block[i] = temp[i];
}
#else
void block_permute(INT16 *block)
{
int tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
int i;
for(i=0;i<8;i++) {
tmp1 = block[1];
tmp2 = block[2];
tmp3 = block[3];
tmp4 = block[4];
tmp5 = block[5];
tmp6 = block[6];
block[1] = tmp2;
block[2] = tmp4;
block[3] = tmp6;
block[4] = tmp1;
block[5] = tmp3;
block[6] = tmp5;
block += 8;
}
}
#endif
void dsputil_init(void)
{
int i, j;
int use_permuted_idct;
for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;
for(i=0;i<MAX_NEG_CROP;i++) {
cropTbl[i] = 0;
cropTbl[i + MAX_NEG_CROP + 256] = 255;
}
for(i=0;i<512;i++) {
squareTbl[i] = (i - 256) * (i - 256);
}
#ifdef SIMPLE_IDCT
ff_idct = simple_idct;
#else
ff_idct = j_rev_dct;
#endif
get_pixels = get_pixels_c;
put_pixels_clamped = put_pixels_clamped_c;
add_pixels_clamped = add_pixels_clamped_c;
pix_abs16x16 = pix_abs16x16_c;
pix_abs16x16_x2 = pix_abs16x16_x2_c;
pix_abs16x16_y2 = pix_abs16x16_y2_c;
pix_abs16x16_xy2 = pix_abs16x16_xy2_c;
av_fdct = jpeg_fdct_ifast;
use_permuted_idct = 1;
#ifdef HAVE_MMX
dsputil_init_mmx();
#endif
#ifdef ARCH_ARMV4L
dsputil_init_armv4l();
#endif
#ifdef HAVE_MLIB
dsputil_init_mlib();
use_permuted_idct = 0;
#endif
#ifdef ARCH_ALPHA
dsputil_init_alpha();
use_permuted_idct = 0;
#endif
#ifdef SIMPLE_IDCT
if(ff_idct == simple_idct) use_permuted_idct=0;
#endif
if(use_permuted_idct)
#ifdef SIMPLE_IDCT
for(i=0; i<64; i++) permutation[i]= simple_mmx_permutation[i];
#else
for(i=0; i<64; i++) permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
#endif
else
for(i=0; i<64; i++) permutation[i]=i;
if (use_permuted_idct) {
/* permute for IDCT */
for(i=0;i<64;i++) {
j = zigzag_direct[i];
zigzag_direct[i] = block_permute_op(j);
j = ff_alternate_horizontal_scan[i];
ff_alternate_horizontal_scan[i] = block_permute_op(j);
j = ff_alternate_vertical_scan[i];
ff_alternate_vertical_scan[i] = block_permute_op(j);
}
block_permute(default_intra_matrix);
block_permute(default_non_intra_matrix);
}
build_zigzag_end();
}