e7a972e1 |
/*
* Simple IDCT
*
* Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
* |
9096a276 |
* This file is part of FFmpeg. |
e7a972e1 |
* |
9096a276 |
* FFmpeg is free software; you can redistribute it and/or |
e7a972e1 |
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
* |
9096a276 |
* FFmpeg is distributed in the hope that it will be useful, |
e7a972e1 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public |
9096a276 |
* License along with FFmpeg; if not, write to the Free Software |
e7a972e1 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* simpleidct in C.
*/
|
41ed7ab4 |
/* Based upon some commented-out C code from mpeg2dec (idct_mmx.c
* written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>). */ |
e7a972e1 |
|
aba70bb5 |
#include "simple_idct.h"
|
e7a972e1 |
#include "bit_depth_template.c"
#undef W1
#undef W2
#undef W3
#undef W4
#undef W5
#undef W6
#undef W7
#undef ROW_SHIFT
#undef COL_SHIFT
#undef DC_SHIFT
#undef MUL
#undef MAC
#if BIT_DEPTH == 8
#define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
#define ROW_SHIFT 11
#define COL_SHIFT 20
#define DC_SHIFT 3
#define MUL(a, b) MUL16(a, b)
#define MAC(a, b, c) MAC16(a, b, c)
|
ae57e824 |
#elif BIT_DEPTH == 10 || BIT_DEPTH == 12 |
e7a972e1 |
|
2fd14dd8 |
# if BIT_DEPTH == 10
#define W1 22725 // 90901
#define W2 21407 // 85627
#define W3 19265 // 77062
#define W4 16384 // 65535
#define W5 12873 // 51491
#define W6 8867 // 35468
#define W7 4520 // 18081
# ifdef EXTRA_SHIFT
#define ROW_SHIFT 13
#define COL_SHIFT 18
#define DC_SHIFT 1
# else
#define ROW_SHIFT 12
#define COL_SHIFT 19
#define DC_SHIFT 2
# endif
# else |
52954075 |
#define W1 45451
#define W2 42813
#define W3 38531
#define W4 32767
#define W5 25746
#define W6 17734
#define W7 9041
#define ROW_SHIFT 16
#define COL_SHIFT 17 |
ae57e824 |
#define DC_SHIFT -1 |
2fd14dd8 |
# endif |
e7a972e1 |
#define MUL(a, b) ((a) * (b))
#define MAC(a, b, c) ((a) += (b) * (c))
#else
#error "Unsupported bitdepth"
#endif
|
2fd14dd8 |
#ifdef EXTRA_SHIFT
static inline void FUNC(idctRowCondDC_extrashift)(int16_t *row, int extra_shift)
#else |
88bd7fdc |
static inline void FUNC(idctRowCondDC)(int16_t *row, int extra_shift) |
2fd14dd8 |
#endif |
e7a972e1 |
{ |
45198477 |
SUINT a0, a1, a2, a3, b0, b1, b2, b3; |
e7a972e1 |
#if HAVE_FAST_64BIT
#define ROW0_MASK (0xffffLL << 48 * HAVE_BIGENDIAN) |
5df703aa |
if (((AV_RN64A(row) & ~ROW0_MASK) | AV_RN64A(row+4)) == 0) { |
f78cd0c2 |
uint64_t temp; |
4b0cad65 |
if (DC_SHIFT - extra_shift >= 0) { |
e50ae60d |
temp = (row[0] * (1 << (DC_SHIFT - extra_shift))) & 0xffff; |
f78cd0c2 |
} else { |
4b0cad65 |
temp = ((row[0] + (1<<(extra_shift - DC_SHIFT-1))) >> (extra_shift - DC_SHIFT)) & 0xffff; |
f78cd0c2 |
} |
e50ae60d |
temp += temp * (1 << 16);
temp += temp * ((uint64_t) 1 << 32); |
5df703aa |
AV_WN64A(row, temp); |
283226e1 |
AV_WN64A(row + 4, temp); |
2cc4f3b2 |
return;
} |
e7a972e1 |
#else |
5df703aa |
if (!(AV_RN32A(row+2) |
AV_RN32A(row+4) |
AV_RN32A(row+6) | |
2cc4f3b2 |
row[1])) { |
f78cd0c2 |
uint32_t temp; |
4b0cad65 |
if (DC_SHIFT - extra_shift >= 0) { |
e50ae60d |
temp = (row[0] * (1 << (DC_SHIFT - extra_shift))) & 0xffff; |
f78cd0c2 |
} else { |
4b0cad65 |
temp = ((row[0] + (1<<(extra_shift - DC_SHIFT-1))) >> (extra_shift - DC_SHIFT)) & 0xffff; |
f78cd0c2 |
} |
e50ae60d |
temp += temp * (1 << 16); |
5df703aa |
AV_WN32A(row, temp);
AV_WN32A(row+2, temp);
AV_WN32A(row+4, temp);
AV_WN32A(row+6, temp); |
2cc4f3b2 |
return;
} |
e7a972e1 |
#endif
|
a7ea733b |
a0 = (W4 * row[0]) + (1 << (ROW_SHIFT + extra_shift - 1)); |
2cc4f3b2 |
a1 = a0;
a2 = a0;
a3 = a0;
a0 += W2 * row[2];
a1 += W6 * row[2];
a2 -= W6 * row[2];
a3 -= W2 * row[2];
b0 = MUL(W1, row[1]);
MAC(b0, W3, row[3]);
b1 = MUL(W3, row[1]);
MAC(b1, -W7, row[3]);
b2 = MUL(W5, row[1]);
MAC(b2, -W1, row[3]);
b3 = MUL(W7, row[1]);
MAC(b3, -W5, row[3]);
if (AV_RN64A(row + 4)) {
a0 += W4*row[4] + W6*row[6];
a1 += - W4*row[4] - W2*row[6];
a2 += - W4*row[4] + W2*row[6];
a3 += W4*row[4] - W6*row[6];
MAC(b0, W5, row[5]);
MAC(b0, W7, row[7]);
MAC(b1, -W1, row[5]);
MAC(b1, -W5, row[7]);
MAC(b2, W7, row[5]);
MAC(b2, W3, row[7]);
MAC(b3, W3, row[5]);
MAC(b3, -W1, row[7]);
}
|
45198477 |
row[0] = (int)(a0 + b0) >> (ROW_SHIFT + extra_shift);
row[7] = (int)(a0 - b0) >> (ROW_SHIFT + extra_shift);
row[1] = (int)(a1 + b1) >> (ROW_SHIFT + extra_shift);
row[6] = (int)(a1 - b1) >> (ROW_SHIFT + extra_shift);
row[2] = (int)(a2 + b2) >> (ROW_SHIFT + extra_shift);
row[5] = (int)(a2 - b2) >> (ROW_SHIFT + extra_shift);
row[3] = (int)(a3 + b3) >> (ROW_SHIFT + extra_shift);
row[4] = (int)(a3 - b3) >> (ROW_SHIFT + extra_shift); |
e7a972e1 |
}
|
a402f109 |
#define IDCT_COLS do { \
a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4)); \
a1 = a0; \
a2 = a0; \
a3 = a0; \
\
a0 += W2*col[8*2]; \
a1 += W6*col[8*2]; \
a2 += -W6*col[8*2]; \
a3 += -W2*col[8*2]; \
\
b0 = MUL(W1, col[8*1]); \
b1 = MUL(W3, col[8*1]); \
b2 = MUL(W5, col[8*1]); \
b3 = MUL(W7, col[8*1]); \
\
MAC(b0, W3, col[8*3]); \
MAC(b1, -W7, col[8*3]); \
MAC(b2, -W1, col[8*3]); \
MAC(b3, -W5, col[8*3]); \
\
if (col[8*4]) { \
a0 += W4*col[8*4]; \
a1 += -W4*col[8*4]; \
a2 += -W4*col[8*4]; \
a3 += W4*col[8*4]; \
} \
\
if (col[8*5]) { \
MAC(b0, W5, col[8*5]); \
MAC(b1, -W1, col[8*5]); \
MAC(b2, W7, col[8*5]); \
MAC(b3, W3, col[8*5]); \
} \
\
if (col[8*6]) { \
a0 += W6*col[8*6]; \
a1 += -W2*col[8*6]; \
a2 += W2*col[8*6]; \
a3 += -W6*col[8*6]; \
} \
\
if (col[8*7]) { \
MAC(b0, W7, col[8*7]); \
MAC(b1, -W5, col[8*7]); \
MAC(b2, W3, col[8*7]); \
MAC(b3, -W1, col[8*7]); \
} \
} while (0)
|
2fd14dd8 |
#ifdef EXTRA_SHIFT
static inline void FUNC(idctSparseCol_extrashift)(int16_t *col)
#else |
2ec9fa5e |
static inline void FUNC(idctSparseColPut)(pixel *dest, ptrdiff_t line_size, |
88bd7fdc |
int16_t *col) |
e7a972e1 |
{ |
45198477 |
SUINT a0, a1, a2, a3, b0, b1, b2, b3; |
2cc4f3b2 |
IDCT_COLS;
|
45198477 |
dest[0] = av_clip_pixel((int)(a0 + b0) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a1 + b1) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a2 + b2) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a3 + b3) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a3 - b3) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a2 - b2) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a1 - b1) >> COL_SHIFT); |
2cc4f3b2 |
dest += line_size; |
45198477 |
dest[0] = av_clip_pixel((int)(a0 - b0) >> COL_SHIFT); |
e7a972e1 |
}
|
2ec9fa5e |
static inline void FUNC(idctSparseColAdd)(pixel *dest, ptrdiff_t line_size, |
88bd7fdc |
int16_t *col) |
e7a972e1 |
{ |
2cc4f3b2 |
int a0, a1, a2, a3, b0, b1, b2, b3;
IDCT_COLS;
|
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a0 + b0) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a1 + b1) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a2 + b2) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a3 + b3) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a3 - b3) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a2 - b2) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a1 - b1) >> COL_SHIFT)); |
2cc4f3b2 |
dest += line_size; |
c23acbae |
dest[0] = av_clip_pixel(dest[0] + ((a0 - b0) >> COL_SHIFT)); |
e7a972e1 |
}
|
88bd7fdc |
static inline void FUNC(idctSparseCol)(int16_t *col) |
2fd14dd8 |
#endif |
e7a972e1 |
{ |
2cc4f3b2 |
int a0, a1, a2, a3, b0, b1, b2, b3;
IDCT_COLS;
col[0 ] = ((a0 + b0) >> COL_SHIFT);
col[8 ] = ((a1 + b1) >> COL_SHIFT);
col[16] = ((a2 + b2) >> COL_SHIFT);
col[24] = ((a3 + b3) >> COL_SHIFT);
col[32] = ((a3 - b3) >> COL_SHIFT);
col[40] = ((a2 - b2) >> COL_SHIFT);
col[48] = ((a1 - b1) >> COL_SHIFT);
col[56] = ((a0 - b0) >> COL_SHIFT); |
e7a972e1 |
}
|
2fd14dd8 |
#ifndef EXTRA_SHIFT |
2ec9fa5e |
void FUNC(ff_simple_idct_put)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block) |
e7a972e1 |
{
pixel *dest = (pixel *)dest_;
int i; |
e7bcc5ba |
line_size /= sizeof(pixel);
|
2cc4f3b2 |
for (i = 0; i < 8; i++) |
f78cd0c2 |
FUNC(idctRowCondDC)(block + i*8, 0); |
e7a972e1 |
|
2cc4f3b2 |
for (i = 0; i < 8; i++) |
e7a972e1 |
FUNC(idctSparseColPut)(dest + i, line_size, block + i);
}
|
2ec9fa5e |
void FUNC(ff_simple_idct_add)(uint8_t *dest_, ptrdiff_t line_size, int16_t *block) |
e7a972e1 |
{
pixel *dest = (pixel *)dest_;
int i; |
e7bcc5ba |
line_size /= sizeof(pixel);
|
2cc4f3b2 |
for (i = 0; i < 8; i++) |
f78cd0c2 |
FUNC(idctRowCondDC)(block + i*8, 0); |
e7a972e1 |
|
2cc4f3b2 |
for (i = 0; i < 8; i++) |
e7a972e1 |
FUNC(idctSparseColAdd)(dest + i, line_size, block + i);
}
|
88bd7fdc |
void FUNC(ff_simple_idct)(int16_t *block) |
e7a972e1 |
{
int i; |
2cc4f3b2 |
for (i = 0; i < 8; i++) |
f78cd0c2 |
FUNC(idctRowCondDC)(block + i*8, 0); |
e7a972e1 |
|
2cc4f3b2 |
for (i = 0; i < 8; i++) |
e7a972e1 |
FUNC(idctSparseCol)(block + i);
} |
2fd14dd8 |
#endif |