44cb64ee |
/* |
41ed7ab4 |
* Copyright (C) 2004 The FFmpeg project |
44cb64ee |
* |
b78e7197 |
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or |
44cb64ee |
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either |
b78e7197 |
* version 2.1 of the License, or (at your option) any later version. |
44cb64ee |
* |
b78e7197 |
* FFmpeg is distributed in the hope that it will be useful, |
44cb64ee |
* 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 |
b78e7197 |
* License along with FFmpeg; if not, write to the Free Software |
5509bffa |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
44cb64ee |
*/
/** |
ba87f080 |
* @file |
115329f1 |
* Standard C DSP-oriented functions cribbed from the original VP3 |
44cb64ee |
* source code.
*/
|
28f9ab70 |
#include "libavutil/attributes.h" |
1d9c2dc8 |
#include "libavutil/common.h" |
ed9625eb |
#include "libavutil/intreadwrite.h" |
d2264740 |
|
44cb64ee |
#include "avcodec.h" |
bf6b3ec9 |
#include "rnd_avg.h" |
28f9ab70 |
#include "vp3dsp.h" |
44cb64ee |
#define IdctAdjustBeforeShift 8
#define xC1S7 64277
#define xC2S6 60547
#define xC3S5 54491
#define xC4S4 46341
#define xC5S3 36410
#define xC6S2 25080
#define xC7S1 12785
|
d2264740 |
#define M(a, b) (((a) * (b)) >> 16) |
e003ea06 |
|
d2264740 |
static av_always_inline void idct(uint8_t *dst, int stride,
int16_t *input, int type) |
44cb64ee |
{ |
8b6103da |
int16_t *ip = input; |
44cb64ee |
|
d26a9ea0 |
int A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;
int Ed, Gd, Add, Bdd, Fd, Hd; |
44cb64ee |
|
bf4e3bd2 |
int i; |
115329f1 |
|
44cb64ee |
/* Inverse DCT on the rows now */
for (i = 0; i < 8; i++) {
/* Check for non-zero values */ |
d2264740 |
if (ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8]) { |
d85c9b03 |
A = M(xC1S7, ip[1 * 8]) + M(xC7S1, ip[7 * 8]);
B = M(xC7S1, ip[1 * 8]) - M(xC1S7, ip[7 * 8]);
C = M(xC3S5, ip[3 * 8]) + M(xC5S3, ip[5 * 8]);
D = M(xC3S5, ip[5 * 8]) - M(xC5S3, ip[3 * 8]); |
44cb64ee |
|
d26a9ea0 |
Ad = M(xC4S4, (A - C));
Bd = M(xC4S4, (B - D)); |
44cb64ee |
|
d26a9ea0 |
Cd = A + C;
Dd = B + D; |
44cb64ee |
|
d85c9b03 |
E = M(xC4S4, (ip[0 * 8] + ip[4 * 8]));
F = M(xC4S4, (ip[0 * 8] - ip[4 * 8])); |
44cb64ee |
|
d85c9b03 |
G = M(xC2S6, ip[2 * 8]) + M(xC6S2, ip[6 * 8]);
H = M(xC6S2, ip[2 * 8]) - M(xC2S6, ip[6 * 8]); |
44cb64ee |
|
d26a9ea0 |
Ed = E - G;
Gd = E + G; |
44cb64ee |
|
d26a9ea0 |
Add = F + Ad;
Bdd = Bd - H; |
44cb64ee |
|
d26a9ea0 |
Fd = F - Ad;
Hd = Bd + H; |
44cb64ee |
/* Final sequence of operations over-write original inputs. */ |
d2264740 |
ip[0 * 8] = Gd + Cd;
ip[7 * 8] = Gd - Cd; |
44cb64ee |
|
d85c9b03 |
ip[1 * 8] = Add + Hd;
ip[2 * 8] = Add - Hd; |
44cb64ee |
|
d2264740 |
ip[3 * 8] = Ed + Dd;
ip[4 * 8] = Ed - Dd; |
44cb64ee |
|
d85c9b03 |
ip[5 * 8] = Fd + Bdd;
ip[6 * 8] = Fd - Bdd; |
44cb64ee |
}
|
d85c9b03 |
ip += 1; /* next row */ |
44cb64ee |
} |
115329f1 |
|
8b6103da |
ip = input; |
44cb64ee |
|
d2264740 |
for (i = 0; i < 8; i++) { |
44cb64ee |
/* Check for non-zero values (bitwise or faster than ||) */ |
d2264740 |
if (ip[1] | ip[2] | ip[3] |
ip[4] | ip[5] | ip[6] | ip[7]) { |
d85c9b03 |
A = M(xC1S7, ip[1]) + M(xC7S1, ip[7]);
B = M(xC7S1, ip[1]) - M(xC1S7, ip[7]);
C = M(xC3S5, ip[3]) + M(xC5S3, ip[5]);
D = M(xC3S5, ip[5]) - M(xC5S3, ip[3]); |
44cb64ee |
|
d26a9ea0 |
Ad = M(xC4S4, (A - C));
Bd = M(xC4S4, (B - D)); |
44cb64ee |
|
d26a9ea0 |
Cd = A + C;
Dd = B + D; |
44cb64ee |
|
d85c9b03 |
E = M(xC4S4, (ip[0] + ip[4])) + 8;
F = M(xC4S4, (ip[0] - ip[4])) + 8; |
d9cf32d9 |
|
d2264740 |
if (type == 1) { // HACK
E += 16 * 128;
F += 16 * 128; |
d9cf32d9 |
} |
44cb64ee |
|
d85c9b03 |
G = M(xC2S6, ip[2]) + M(xC6S2, ip[6]);
H = M(xC6S2, ip[2]) - M(xC2S6, ip[6]); |
44cb64ee |
|
d26a9ea0 |
Ed = E - G;
Gd = E + G; |
44cb64ee |
|
d26a9ea0 |
Add = F + Ad;
Bdd = Bd - H; |
44cb64ee |
|
d26a9ea0 |
Fd = F - Ad;
Hd = Bd + H; |
44cb64ee |
/* Final sequence of operations over-write original inputs. */ |
01582122 |
if (type == 1) { |
d2264740 |
dst[0 * stride] = av_clip_uint8((Gd + Cd) >> 4);
dst[7 * stride] = av_clip_uint8((Gd - Cd) >> 4); |
115329f1 |
|
d2264740 |
dst[1 * stride] = av_clip_uint8((Add + Hd) >> 4);
dst[2 * stride] = av_clip_uint8((Add - Hd) >> 4); |
115329f1 |
|
d2264740 |
dst[3 * stride] = av_clip_uint8((Ed + Dd) >> 4);
dst[4 * stride] = av_clip_uint8((Ed - Dd) >> 4); |
115329f1 |
|
d2264740 |
dst[5 * stride] = av_clip_uint8((Fd + Bdd) >> 4);
dst[6 * stride] = av_clip_uint8((Fd - Bdd) >> 4);
} else {
dst[0 * stride] = av_clip_uint8(dst[0 * stride] + ((Gd + Cd) >> 4));
dst[7 * stride] = av_clip_uint8(dst[7 * stride] + ((Gd - Cd) >> 4)); |
115329f1 |
|
d2264740 |
dst[1 * stride] = av_clip_uint8(dst[1 * stride] + ((Add + Hd) >> 4));
dst[2 * stride] = av_clip_uint8(dst[2 * stride] + ((Add - Hd) >> 4)); |
115329f1 |
|
d2264740 |
dst[3 * stride] = av_clip_uint8(dst[3 * stride] + ((Ed + Dd) >> 4));
dst[4 * stride] = av_clip_uint8(dst[4 * stride] + ((Ed - Dd) >> 4)); |
115329f1 |
|
d2264740 |
dst[5 * stride] = av_clip_uint8(dst[5 * stride] + ((Fd + Bdd) >> 4));
dst[6 * stride] = av_clip_uint8(dst[6 * stride] + ((Fd - Bdd) >> 4)); |
8b6103da |
} |
44cb64ee |
} else { |
01582122 |
if (type == 1) { |
d2264740 |
dst[0*stride] =
dst[1*stride] =
dst[2*stride] =
dst[3*stride] =
dst[4*stride] =
dst[5*stride] =
dst[6*stride] =
dst[7*stride] = av_clip_uint8(128 + ((xC4S4 * ip[0] + (IdctAdjustBeforeShift << 16)) >> 20));
} else {
if (ip[0]) {
int v = (xC4S4 * ip[0] + (IdctAdjustBeforeShift << 16)) >> 20;
dst[0 * stride] = av_clip_uint8(dst[0 * stride] + v);
dst[1 * stride] = av_clip_uint8(dst[1 * stride] + v);
dst[2 * stride] = av_clip_uint8(dst[2 * stride] + v);
dst[3 * stride] = av_clip_uint8(dst[3 * stride] + v);
dst[4 * stride] = av_clip_uint8(dst[4 * stride] + v);
dst[5 * stride] = av_clip_uint8(dst[5 * stride] + v);
dst[6 * stride] = av_clip_uint8(dst[6 * stride] + v);
dst[7 * stride] = av_clip_uint8(dst[7 * stride] + v); |
8b6103da |
}
} |
44cb64ee |
}
|
d85c9b03 |
ip += 8; /* next column */ |
8b6103da |
dst++; |
44cb64ee |
}
} |
8b6103da |
|
d2264740 |
static void vp3_idct_put_c(uint8_t *dest /* align 8 */, int line_size,
int16_t *block /* align 16 */) |
88bd7fdc |
{ |
8b6103da |
idct(dest, line_size, block, 1); |
aeaf268e |
memset(block, 0, sizeof(*block) * 64); |
8b6103da |
}
|
d2264740 |
static void vp3_idct_add_c(uint8_t *dest /* align 8 */, int line_size,
int16_t *block /* align 16 */) |
88bd7fdc |
{ |
8b6103da |
idct(dest, line_size, block, 2); |
aeaf268e |
memset(block, 0, sizeof(*block) * 64); |
8b6103da |
} |
9971331d |
|
d2264740 |
static void vp3_idct_dc_add_c(uint8_t *dest /* align 8 */, int line_size,
int16_t *block /* align 16 */) |
88bd7fdc |
{ |
c0fda017 |
int i, dc = (block[0] + 15) >> 5; |
eb6a6cd7 |
|
d2264740 |
for (i = 0; i < 8; i++) { |
c23acbae |
dest[0] = av_clip_uint8(dest[0] + dc);
dest[1] = av_clip_uint8(dest[1] + dc);
dest[2] = av_clip_uint8(dest[2] + dc);
dest[3] = av_clip_uint8(dest[3] + dc);
dest[4] = av_clip_uint8(dest[4] + dc);
dest[5] = av_clip_uint8(dest[5] + dc);
dest[6] = av_clip_uint8(dest[6] + dc);
dest[7] = av_clip_uint8(dest[7] + dc); |
d2264740 |
dest += line_size; |
eb6a6cd7 |
} |
aeaf268e |
block[0] = 0; |
eb6a6cd7 |
}
|
28f9ab70 |
static void vp3_v_loop_filter_c(uint8_t *first_pixel, int stride,
int *bounding_values) |
9971331d |
{
unsigned char *end;
int filter_value; |
d2264740 |
const int nstride = -stride; |
9971331d |
|
d2264740 |
for (end = first_pixel + 8; first_pixel < end; first_pixel++) {
filter_value = (first_pixel[2 * nstride] - first_pixel[stride]) +
(first_pixel[0] - first_pixel[nstride]) * 3; |
9971331d |
filter_value = bounding_values[(filter_value + 4) >> 3]; |
d2264740 |
|
9971331d |
first_pixel[nstride] = av_clip_uint8(first_pixel[nstride] + filter_value); |
d2264740 |
first_pixel[0] = av_clip_uint8(first_pixel[0] - filter_value); |
9971331d |
}
}
|
28f9ab70 |
static void vp3_h_loop_filter_c(uint8_t *first_pixel, int stride,
int *bounding_values) |
9971331d |
{
unsigned char *end;
int filter_value;
|
d2264740 |
for (end = first_pixel + 8 * stride; first_pixel != end; first_pixel += stride) {
filter_value = (first_pixel[-2] - first_pixel[1]) +
(first_pixel[ 0] - first_pixel[-1]) * 3; |
9971331d |
filter_value = bounding_values[(filter_value + 4) >> 3]; |
d2264740 |
|
9971331d |
first_pixel[-1] = av_clip_uint8(first_pixel[-1] + filter_value);
first_pixel[ 0] = av_clip_uint8(first_pixel[ 0] - filter_value);
}
} |
28f9ab70 |
|
4a73fbd9 |
static void put_no_rnd_pixels_l2(uint8_t *dst, const uint8_t *src1,
const uint8_t *src2, ptrdiff_t stride, int h)
{
int i;
for (i = 0; i < h; i++) {
uint32_t a, b;
|
0ee8293a |
a = AV_RN32(&src1[i * stride]);
b = AV_RN32(&src2[i * stride]); |
4a73fbd9 |
AV_WN32A(&dst[i * stride], no_rnd_avg32(a, b)); |
0ee8293a |
a = AV_RN32(&src1[i * stride + 4]);
b = AV_RN32(&src2[i * stride + 4]); |
4a73fbd9 |
AV_WN32A(&dst[i * stride + 4], no_rnd_avg32(a, b));
}
}
|
28f9ab70 |
av_cold void ff_vp3dsp_init(VP3DSPContext *c, int flags)
{ |
4a73fbd9 |
c->put_no_rnd_pixels_l2 = put_no_rnd_pixels_l2;
|
28f9ab70 |
c->idct_put = vp3_idct_put_c;
c->idct_add = vp3_idct_add_c;
c->idct_dc_add = vp3_idct_dc_add_c;
c->v_loop_filter = vp3_v_loop_filter_c;
c->h_loop_filter = vp3_h_loop_filter_c;
if (ARCH_ARM)
ff_vp3dsp_init_arm(c, flags);
if (ARCH_PPC)
ff_vp3dsp_init_ppc(c, flags);
if (ARCH_X86)
ff_vp3dsp_init_x86(c, flags);
} |