Commit f0a149e5 authored by Paul B Mahol's avatar Paul B Mahol

lavfi: remove bad inverse telecine filters

Signed-off-by: 's avatarPaul B Mahol <onemda@gmail.com>
parent a5ebd2d9
......@@ -4303,16 +4303,13 @@ the named filter.
The list of the currently supported filters follows:
@table @var
@item detc
@item dint
@item divtc
@item down3dright
@item eq2
@item eq
@item fil
@item fspp
@item ilpack
@item ivtc
@item mcdeint
@item ow
@item perspective
......
......@@ -192,17 +192,13 @@ OBJS-$(CONFIG_NULLSINK_FILTER) += vsink_nullsink.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/mp_image.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/img_format.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_detc.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_dint.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_divtc.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_down3dright.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_eq2.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_eq.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_fil.o
#OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_filmdint.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_fspp.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_ilpack.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_ivtc.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_mcdeint.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_ow.o
OBJS-$(CONFIG_MP_FILTER) += libmpcodecs/vf_perspective.o
......
/*
* This file is part of MPlayer.
*
* MPlayer 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.
*
* MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "config.h"
#include "mp_msg.h"
#include "img_format.h"
#include "mp_image.h"
#include "vf.h"
#include "libvo/fastmemcpy.h"
struct metrics {
int even;
int odd;
int noise;
int temp;
};
struct vf_priv_s {
int frame;
int drop, lastdrop;
struct metrics pm;
int thres[5];
int inframes, outframes;
int mode;
int (*analyze)(struct vf_priv_s *, mp_image_t *, mp_image_t *);
int needread;
};
#define COMPE(a,b,e) (abs((a)-(b)) < (((a)+(b))>>(e)))
#define COMPARABLE(a,b) COMPE((a),(b),2)
#define VERYCLOSE(a,b) COMPE((a),(b),3)
#define OUTER_TC_NBHD(s) ( \
COMPARABLE((s)[-1].m.even,(s)[-1].m.odd) && \
COMPARABLE((s)[1].m.even,(s)[0].m.odd) && \
COMPARABLE((s)[2].m.even,(s)[1].m.odd) && \
COMPARABLE((s)[-1].m.noise,(s)[0].m.temp) && \
COMPARABLE((s)[2].m.noise,(s)[2].m.temp) )
#define INNER_TC_NBHD(s,l,h) ( \
COMPARABLE((s)[0].m.even,(l)) && \
COMPARABLE((s)[2].m.odd,(l)) && ( \
COMPARABLE((s)[0].m.noise,(h)) || \
COMPARABLE((s)[1].m.noise,(h)) ) )
enum {
TC_DROP,
TC_PROG,
TC_IL1,
TC_IL2
};
static void block_diffs(struct metrics *m, unsigned char *old, unsigned char *new, int os, int ns)
{
int x, y, even=0, odd=0, noise, temp;
unsigned char *oldp, *newp;
m->noise = m->temp = 0;
for (x = 8; x; x--) {
oldp = old++;
newp = new++;
noise = temp = 0;
for (y = 4; y; y--) {
even += abs(newp[0]-oldp[0]);
odd += abs(newp[ns]-oldp[os]);
noise += newp[ns]-newp[0];
temp += oldp[os]-newp[0];
oldp += os<<1;
newp += ns<<1;
}
m->noise += abs(noise);
m->temp += abs(temp);
}
m->even = even;
m->odd = odd;
}
static void diff_planes(struct metrics *m, unsigned char *old, unsigned char *new, int w, int h, int os, int ns)
{
int x, y, me=0, mo=0, mn=0, mt=0;
struct metrics l;
for (y = 0; y < h-7; y += 8) {
for (x = 0; x < w-7; x += 8) {
block_diffs(&l, old+x+y*os, new+x+y*ns, os, ns);
if (l.even > me) me = l.even;
if (l.odd > mo) mo = l.odd;
if (l.noise > mn) mn = l.noise;
if (l.temp > mt) mt = l.temp;
}
}
m->even = me;
m->odd = mo;
m->noise = mn;
m->temp = mt;
}
static void diff_fields(struct metrics *metr, mp_image_t *old, mp_image_t *new)
{
struct metrics m, mu, mv;
diff_planes(&m, old->planes[0], new->planes[0],
new->w, new->h, old->stride[0], new->stride[0]);
if (new->flags & MP_IMGFLAG_PLANAR) {
diff_planes(&mu, old->planes[1], new->planes[1],
new->chroma_width, new->chroma_height,
old->stride[1], new->stride[1]);
diff_planes(&mv, old->planes[2], new->planes[2],
new->chroma_width, new->chroma_height,
old->stride[2], new->stride[2]);
if (mu.even > m.even) m.even = mu.even;
if (mu.odd > m.odd) m.odd = mu.odd;
if (mu.noise > m.noise) m.noise = mu.noise;
if (mu.temp > m.temp) m.temp = mu.temp;
if (mv.even > m.even) m.even = mv.even;
if (mv.odd > m.odd) m.odd = mv.odd;
if (mv.noise > m.noise) m.noise = mv.noise;
if (mv.temp > m.temp) m.temp = mv.temp;
}
*metr = m;
}
static void status(int f, struct metrics *m)
{
ff_mp_msg(MSGT_VFILTER, MSGL_V, "frame %d: e=%d o=%d n=%d t=%d\n",
f, m->even, m->odd, m->noise, m->temp);
}
static int analyze_fixed_pattern(struct vf_priv_s *p, mp_image_t *new, mp_image_t *old)
{
if (p->frame >= 0) p->frame = (p->frame+1)%5;
ff_mp_msg(MSGT_VFILTER, MSGL_V, "frame %d\n", p->frame);
switch (p->frame) {
case -1: case 0: case 1: case 2:
return TC_PROG;
case 3:
return TC_IL1;
case 4:
return TC_IL2;
}
return 0;
}
static int analyze_aggressive(struct vf_priv_s *p, mp_image_t *new, mp_image_t *old)
{
struct metrics m, pm;
if (p->frame >= 0) p->frame = (p->frame+1)%5;
diff_fields(&m, old, new);
status(p->frame, &m);
pm = p->pm;
p->pm = m;
if (p->frame == 4) {
/* We need to break at scene changes, but is this a valid test? */
if ((m.even > p->thres[2]) && (m.odd > p->thres[2]) && (m.temp > p->thres[3])
&& (m.temp > 5*pm.temp) && (m.temp*2 > m.noise)) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "scene change breaking telecine!\n");
p->frame = -1;
return TC_DROP;
}
/* Thres. is to compensate for quantization errors when noise is low */
if (m.noise - m.temp > -p->thres[4]) {
if (COMPARABLE(m.even, pm.odd)) {
//ff_mp_msg(MSGT_VFILTER, MSGL_V, "confirmed field match!\n");
return TC_IL2;
} else if ((m.even < p->thres[0]) && (m.odd < p->thres[0]) && VERYCLOSE(m.even, m.odd)
&& VERYCLOSE(m.noise,m.temp) && VERYCLOSE(m.noise,pm.noise)) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "interlaced frame appears in duplicate!!!\n");
p->pm = pm; /* hack :) */
p->frame = 3;
return TC_IL1;
}
} else {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "mismatched telecine fields!\n");
p->frame = -1;
}
}
if (2*m.even*m.temp < m.odd*m.noise) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "caught telecine sync!\n");
p->frame = 3;
return TC_IL1;
}
if (p->frame < 3) {
if (m.noise > p->thres[3]) {
if (m.noise > 2*m.temp) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "merging fields out of sequence!\n");
return TC_IL2;
}
if ((m.noise > 2*pm.noise) && (m.even > p->thres[2]) && (m.odd > p->thres[2])) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "dropping horrible interlaced frame!\n");
return TC_DROP;
}
}
}
switch (p->frame) {
case -1:
if (4*m.noise > 5*m.temp) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "merging fields out of sequence!\n");
return TC_IL2;
}
case 0:
case 1:
case 2:
return TC_PROG;
case 3:
if ((m.even > p->thres[1]) && (m.even > m.odd) && (m.temp > m.noise)) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "lost telecine tracking!\n");
p->frame = -1;
return TC_PROG;
}
return TC_IL1;
case 4:
return TC_IL2;
}
return 0;
}
static void copy_image(mp_image_t *dmpi, mp_image_t *mpi, int field)
{
switch (field) {
case 0:
my_memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h/2,
dmpi->stride[0]*2, mpi->stride[0]*2);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(dmpi->planes[1], mpi->planes[1],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[1]*2, mpi->stride[1]*2);
my_memcpy_pic(dmpi->planes[2], mpi->planes[2],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[2]*2, mpi->stride[2]*2);
}
break;
case 1:
my_memcpy_pic(dmpi->planes[0]+dmpi->stride[0],
mpi->planes[0]+mpi->stride[0], mpi->w, mpi->h/2,
dmpi->stride[0]*2, mpi->stride[0]*2);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(dmpi->planes[1]+dmpi->stride[1],
mpi->planes[1]+mpi->stride[1],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[1]*2, mpi->stride[1]*2);
my_memcpy_pic(dmpi->planes[2]+dmpi->stride[2],
mpi->planes[2]+mpi->stride[2],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[2]*2, mpi->stride[2]*2);
}
break;
case 2:
memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h,
dmpi->stride[0], mpi->stride[0]);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
memcpy_pic(dmpi->planes[1], mpi->planes[1],
mpi->chroma_width, mpi->chroma_height,
dmpi->stride[1], mpi->stride[1]);
memcpy_pic(dmpi->planes[2], mpi->planes[2],
mpi->chroma_width, mpi->chroma_height,
dmpi->stride[2], mpi->stride[2]);
}
break;
}
}
static int do_put_image(struct vf_instance *vf, mp_image_t *dmpi)
{
struct vf_priv_s *p = vf->priv;
int dropflag;
switch (p->drop) {
default:
dropflag = 0;
break;
case 1:
dropflag = (++p->lastdrop >= 5);
break;
case 2:
dropflag = (++p->lastdrop >= 5) && (4*p->inframes <= 5*p->outframes);
break;
}
if (dropflag) {
ff_mp_msg(MSGT_VFILTER, MSGL_V, "drop! [%d/%d=%g]\n",
p->outframes, p->inframes, (float)p->outframes/p->inframes);
p->lastdrop = 0;
return 0;
}
p->outframes++;
return ff_vf_next_put_image(vf, dmpi, MP_NOPTS_VALUE);
}
static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts)
{
int ret=0;
mp_image_t *dmpi;
struct vf_priv_s *p = vf->priv;
p->inframes++;
if (p->needread) dmpi = ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_STATIC, MP_IMGFLAG_ACCEPT_STRIDE |
MP_IMGFLAG_PRESERVE | MP_IMGFLAG_READABLE,
mpi->width, mpi->height);
/* FIXME: is there a good way to get rid of static type? */
else dmpi = ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_STATIC, MP_IMGFLAG_ACCEPT_STRIDE |
MP_IMGFLAG_PRESERVE, mpi->width, mpi->height);
switch (p->analyze(p, mpi, dmpi)) {
case TC_DROP:
/* Don't copy anything unless we'll need to read it. */
if (p->needread) copy_image(dmpi, mpi, 2);
p->lastdrop = 0;
break;
case TC_PROG:
/* Copy and display the whole frame. */
copy_image(dmpi, mpi, 2);
ret = do_put_image(vf, dmpi);
break;
case TC_IL1:
/* Only copy bottom field unless we need to read. */
if (p->needread) copy_image(dmpi, mpi, 2);
else copy_image(dmpi, mpi, 1);
p->lastdrop = 0;
break;
case TC_IL2:
/* Copy top field and show frame, then copy bottom if needed. */
copy_image(dmpi, mpi, 0);
ret = do_put_image(vf, dmpi);
if (p->needread) copy_image(dmpi, mpi, 1);
break;
}
return ret;
}
static int query_format(struct vf_instance *vf, unsigned int fmt)
{
/* FIXME - figure out which other formats work */
switch (fmt) {
case IMGFMT_YV12:
case IMGFMT_IYUV:
case IMGFMT_I420:
return ff_vf_next_query_format(vf, fmt);
}
return 0;
}
static int config(struct vf_instance *vf,
int width, int height, int d_width, int d_height,
unsigned int flags, unsigned int outfmt)
{
return ff_vf_next_config(vf,width,height,d_width,d_height,flags,outfmt);
}
static void uninit(struct vf_instance *vf)
{
free(vf->priv);
}
static struct {
const char *name;
int (*func)(struct vf_priv_s *p, mp_image_t *new, mp_image_t *old);
int needread;
} anal_funcs[] = {
{ "fixed", analyze_fixed_pattern, 0 },
{ "aggressive", analyze_aggressive, 1 },
{ NULL, NULL, 0 }
};
#define STARTVARS if (0)
#define GETVAR(str, name, out, func) \
else if (!strncmp((str), name "=", sizeof(name))) \
(out) = (func)((str) + sizeof(name))
static void parse_var(struct vf_priv_s *p, char *var)
{
STARTVARS;
GETVAR(var, "dr", p->drop, atoi);
GETVAR(var, "t0", p->thres[0], atoi);
GETVAR(var, "t1", p->thres[1], atoi);
GETVAR(var, "t2", p->thres[2], atoi);
GETVAR(var, "t3", p->thres[3], atoi);
GETVAR(var, "t4", p->thres[4], atoi);
GETVAR(var, "fr", p->frame, atoi);
GETVAR(var, "am", p->mode, atoi);
}
static void parse_args(struct vf_priv_s *p, char *args)
{
char *next, *orig;
for (args=orig=strdup(args); args; args=next) {
next = strchr(args, ':');
if (next) *next++ = 0;
parse_var(p, args);
}
free(orig);
}
static int vf_open(vf_instance_t *vf, char *args)
{
struct vf_priv_s *p;
vf->config = config;
vf->put_image = put_image;
vf->query_format = query_format;
vf->uninit = uninit;
vf->default_reqs = VFCAP_ACCEPT_STRIDE;
vf->priv = p = calloc(1, sizeof(struct vf_priv_s));
p->frame = -1;
p->thres[0] = 440;
p->thres[1] = 720;
p->thres[2] = 2500;
p->thres[3] = 2500;
p->thres[4] = 800;
p->drop = 0;
p->mode = 1;
if (args) parse_args(p, args);
p->analyze = anal_funcs[p->mode].func;
p->needread = anal_funcs[p->mode].needread;
return 1;
}
const vf_info_t ff_vf_info_detc = {
"de-telecine filter",
"detc",
"Rich Felker",
"",
vf_open,
NULL
};
/*
* This file is part of MPlayer.
*
* MPlayer 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.
*
* MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include <math.h>
#include "config.h"
#include "mp_msg.h"
#include "cpudetect.h"
#include "libavutil/common.h"
#include "libavutil/x86/asm.h"
#include "mpbswap.h"
#include "img_format.h"
#include "mp_image.h"
#include "vf.h"
#include "libvo/fastmemcpy.h"
const vf_info_t ff_vf_info_divtc;
struct vf_priv_s
{
int deghost, pass, phase, window, fcount, bcount, frameno, misscount,
ocount, sum[5];
double threshold;
FILE *file;
int8_t *bdata;
unsigned int *csdata;
int *history;
};
/*
* diff_MMX and diff_C stolen from vf_decimate.c
*/
#if HAVE_MMX && HAVE_EBX_AVAILABLE
static int diff_MMX(unsigned char *old, unsigned char *new, int os, int ns)
{
volatile short out[4];
__asm__ (
"movl $8, %%ecx \n\t"
"pxor %%mm4, %%mm4 \n\t"
"pxor %%mm7, %%mm7 \n\t"
ASMALIGN(4)
"1: \n\t"
"movq (%%"REG_S"), %%mm0 \n\t"
"movq (%%"REG_S"), %%mm2 \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"movq (%%"REG_D"), %%mm1 \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"psubusb %%mm1, %%mm2 \n\t"
"psubusb %%mm0, %%mm1 \n\t"
"movq %%mm2, %%mm0 \n\t"
"movq %%mm1, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpckhbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm0, %%mm4 \n\t"
"paddw %%mm1, %%mm4 \n\t"
"paddw %%mm2, %%mm4 \n\t"
"paddw %%mm3, %%mm4 \n\t"
"decl %%ecx \n\t"
"jnz 1b \n\t"
"movq %%mm4, (%%"REG_d") \n\t"
"emms \n\t"
:
: "S" (old), "D" (new), "a" ((long)os), "b" ((long)ns), "d" (out)
: "%ecx", "memory"
);
return out[0]+out[1]+out[2]+out[3];
}
#endif
static int diff_C(unsigned char *old, unsigned char *new, int os, int ns)
{
int x, y, d=0;
for(y=8; y; y--, new+=ns, old+=os)
for(x=8; x; x--)
d+=abs(new[x]-old[x]);
return d;
}
static int (*diff)(unsigned char *, unsigned char *, int, int);
static int diff_plane(unsigned char *old, unsigned char *new,
int w, int h, int os, int ns, int arg)
{
int x, y, d, max=0, sum=0, n=0;
for(y=0; y<h-7; y+=8)
{
for(x=0; x<w-7; x+=8)
{
d=diff(old+x+y*os, new+x+y*ns, os, ns);
if(d>max) max=d;
sum+=d;
n++;
}
}
return (sum+n*max)/2;
}
/*
static unsigned int checksum_plane(unsigned char *p, unsigned char *z,
int w, int h, int s, int zs, int arg)
{
unsigned int shift, sum;
unsigned char *e;
for(sum=0; h; h--, p+=s-w)
for(e=p+w, shift=32; p<e;)
sum^=(*p++)<<(shift=(shift-8)&31);
return sum;
}
*/
static unsigned int checksum_plane(unsigned char *p, unsigned char *z,
int w, int h, int s, int zs, int arg)
{
unsigned int shift;
uint32_t sum, t;
unsigned char *e, *e2;
#if HAVE_FAST_64BIT
typedef uint64_t wsum_t;
#else
typedef uint32_t wsum_t;
#endif
wsum_t wsum;
for(sum=0; h; h--, p+=s-w)
{
for(shift=0, e=p+w; (int)p&(sizeof(wsum_t)-1) && p<e;)
sum^=*p++<<(shift=(shift-8)&31);
for(wsum=0, e2=e-sizeof(wsum_t)+1; p<e2; p+=sizeof(wsum_t))
wsum^=*(wsum_t *)p;
#if HAVE_FAST_64BIT
t=be2me_32((uint32_t)(wsum>>32^wsum));
#else
t=be2me_32(wsum);
#endif
for(sum^=(t<<shift|t>>(32-shift)); p<e;)
sum^=*p++<<(shift=(shift-8)&31);
}
return sum;
}
static int deghost_plane(unsigned char *d, unsigned char *s,
int w, int h, int ds, int ss, int threshold)
{
int t;
unsigned char *e;
for(; h; h--, s+=ss-w, d+=ds-w)
for(e=d+w; d<e; d++, s++)
if(abs(*d-*s)>=threshold)
*d=(t=(*d<<1)-*s)<0?0:t>255?255:t;
return 0;
}
static int copyop(unsigned char *d, unsigned char *s, int bpl, int h, int dstride, int sstride, int dummy) {
memcpy_pic(d, s, bpl, h, dstride, sstride);
return 0;
}
static int imgop(int(*planeop)(unsigned char *, unsigned char *,
int, int, int, int, int),
mp_image_t *dst, mp_image_t *src, int arg)
{
if(dst->flags&MP_IMGFLAG_PLANAR)
return planeop(dst->planes[0], src?src->planes[0]:0,
dst->w, dst->h,
dst->stride[0], src?src->stride[0]:0, arg)+
planeop(dst->planes[1], src?src->planes[1]:0,
dst->chroma_width, dst->chroma_height,
dst->stride[1], src?src->stride[1]:0, arg)+
planeop(dst->planes[2], src?src->planes[2]:0,
dst->chroma_width, dst->chroma_height,
dst->stride[2], src?src->stride[2]:0, arg);
return planeop(dst->planes[0], src?src->planes[0]:0,
dst->w*(dst->bpp/8), dst->h,
dst->stride[0], src?src->stride[0]:0, arg);
}
/*
* Find the phase in which the telecine pattern fits best to the
* given 5 frame slice of frame difference measurements.
*
* If phase1 and phase2 are not negative, only the two specified
* phases are tested.
*/
static int match(struct vf_priv_s *p, int *diffs,
int phase1, int phase2, double *strength)
{
static const int pattern1[]={ -4, 1, 1, 1, 1 },
pattern2[]={ -2, -3, 4, 4, -3 }, *pattern;
int f, m, n, t[5];
pattern=p->deghost>0?pattern2:pattern1;
for(f=0; f<5; f++)
{
if(phase1<0 || phase2<0 || f==phase1 || f==phase2)
{
for(n=t[f]=0; n<5; n++)
t[f]+=diffs[n]*pattern[(n-f+5)%5];
}
else
t[f]=INT_MIN;
}
/* find the best match */
for(m=0, n=1; n<5; n++)
if(t[n]>t[m]) m=n;
if(strength)
{
/* the second best match */
for(f=m?0:1, n=f+1; n<5; n++)
if(n!=m && t[n]>t[f]) f=n;
*strength=(t[m]>0?(double)(t[m]-t[f])/t[m]:0.0);
}
return m;
}
static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts)
{
mp_image_t *dmpi, *tmpi=0;
int n, m, f, newphase;
struct vf_priv_s *p=vf->priv;
unsigned int checksum;
double d;
dmpi=ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_STATIC, MP_IMGFLAG_ACCEPT_STRIDE |
MP_IMGFLAG_PRESERVE | MP_IMGFLAG_READABLE,
mpi->width, mpi->height);
ff_vf_clone_mpi_attributes(dmpi, mpi);
newphase=p->phase;
switch(p->pass)
{
case 1:
fprintf(p->file, "%08x %d\n",
(unsigned int)imgop((void *)checksum_plane, mpi, 0, 0),
p->frameno?imgop(diff_plane, dmpi, mpi, 0):0);
break;
case 2:
if(p->frameno/5>p->bcount)
{
ff_mp_msg(MSGT_VFILTER, MSGL_ERR,
"\n%s: Log file ends prematurely! "
"Switching to one pass mode.\n", vf->info->name);
p->pass=0;
break;
}
checksum=(unsigned int)imgop((void *)checksum_plane, mpi, 0, 0);
if(checksum!=p->csdata[p->frameno])
{
for(f=0; f<100; f++)
if(p->frameno+f<p->fcount && p->csdata[p->frameno+f]==checksum)
break;
else if(p->frameno-f>=0 && p->csdata[p->frameno-f]==checksum)
{
f=-f;
break;
}
if(f<100)
{
ff_mp_msg(MSGT_VFILTER, MSGL_INFO,
"\n%s: Mismatch with pass-1: %+d frame(s).\n",
vf->info->name, f);
p->frameno+=f;
p->misscount=0;
}
else if(p->misscount++>=30)
{
ff_mp_msg(MSGT_VFILTER, MSGL_ERR,
"\n%s: Sync with pass-1 lost! "
"Switching to one pass mode.\n", vf->info->name);
p->pass=0;
break;
}
}
n=(p->frameno)/5;
if(n>=p->bcount) n=p->bcount-1;
newphase=p->bdata[n];
break;
default:
if(p->frameno)
{
int *sump=p->sum+p->frameno%5,
*histp=p->history+p->frameno%p->window;
*sump-=*histp;
*sump+=(*histp=imgop(diff_plane, dmpi, mpi, 0));
}
m=match(p, p->sum, -1, -1, &d);
if(d>=p->threshold)
newphase=m;
}
n=p->ocount++%5;
if(newphase!=p->phase && ((p->phase+4)%5<n)==((newphase+4)%5<n))
{
p->phase=newphase;
ff_mp_msg(MSGT_VFILTER, MSGL_STATUS,
"\n%s: Telecine phase %d.\n", vf->info->name, p->phase);
}
switch((p->frameno++-p->phase+10)%5)
{
case 0:
imgop(copyop, dmpi, mpi, 0);
return 0;
case 4:
if(p->deghost>0)
{
tmpi=ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE |
MP_IMGFLAG_READABLE,
mpi->width, mpi->height);
ff_vf_clone_mpi_attributes(tmpi, mpi);
imgop(copyop, tmpi, mpi, 0);
imgop(deghost_plane, tmpi, dmpi, p->deghost);
imgop(copyop, dmpi, mpi, 0);
return ff_vf_next_put_image(vf, tmpi, MP_NOPTS_VALUE);
}
}
imgop(copyop, dmpi, mpi, 0);
return ff_vf_next_put_image(vf, dmpi, MP_NOPTS_VALUE);
}
static int analyze(struct vf_priv_s *p)
{
int *buf=0, *bp, bufsize=0, n, b, f, i, j, m, s;
unsigned int *cbuf=0, *cp;
int8_t *pbuf;
int8_t lbuf[256];
int sum[5];
double d;
/* read the file */
n=15;
while(fgets(lbuf, 256, p->file))
{
if(n>=bufsize-19)
{
bufsize=bufsize?bufsize*2:30000;
if((bp=realloc(buf, bufsize*sizeof *buf))) buf=bp;
if((cp=realloc(cbuf, bufsize*sizeof *cbuf))) cbuf=cp;
if(!bp || !cp)
{
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "%s: Not enough memory.\n",
ff_vf_info_divtc.name);
free(buf);
free(cbuf);
return 0;
}
}
sscanf(lbuf, "%x %d", cbuf+n, buf+n);
n++;
}
if(n <= 15)
{
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "%s: Empty 2-pass log file.\n",
ff_vf_info_divtc.name);
free(buf);
free(cbuf);
return 0;
}
/* generate some dummy data past the beginning and end of the array */
buf+=15, cbuf+=15;
n-=15;
memcpy(buf-15, buf, 15*sizeof *buf);
memset(cbuf-15, 0, 15*sizeof *cbuf);
while(n%5)
buf[n]=buf[n-5], cbuf[n]=0, n++;
memcpy(buf+n, buf+n-15, 15*sizeof *buf);
memset(cbuf+n, 0, 15*sizeof *cbuf);
p->csdata=cbuf;
p->fcount=n;
/* array with one slot for each slice of 5 frames */
p->bdata=pbuf=malloc(p->bcount=b=(n/5));
memset(pbuf, 255, b);
/* resolve the automatic mode */
if(p->deghost<0)
{
int deghost=-p->deghost;
double s0=0.0, s1=0.0;
for(f=0; f<n; f+=5)
{
p->deghost=0; match(p, buf+f, -1, -1, &d); s0+=d;
p->deghost=1; match(p, buf+f, -1, -1, &d); s1+=d;
}
p->deghost=s1>s0?deghost:0;
ff_mp_msg(MSGT_VFILTER, MSGL_INFO,
"%s: Deghosting %-3s (relative pattern strength %+.2fdB).\n",
ff_vf_info_divtc.name,
p->deghost?"ON":"OFF",
10.0*log10(s1/s0));
}
/* analyze the data */
for(f=0; f<5; f++)
for(sum[f]=0, n=-15; n<20; n+=5)
sum[f]+=buf[n+f];
for(f=0; f<b; f++)
{
m=match(p, sum, -1, -1, &d);
if(d>=p->threshold)
pbuf[f]=m;
if(f<b-1)
for(n=0; n<5; n++)
sum[n]=sum[n]-buf[5*(f-3)+n]+buf[5*(f+4)+n];
}
/* fill in the gaps */
/* the beginning */
for(f=0; f<b && pbuf[f]==-1; f++);
if(f==b)
{
free(buf-15);
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "%s: No telecine pattern found!\n",
ff_vf_info_divtc.name);
return 0;
}
for(n=0; n<f; pbuf[n++]=pbuf[f]);
/* the end */
for(f=b-1; pbuf[f]==-1; f--);
for(n=f+1; n<b; pbuf[n++]=pbuf[f]);
/* the rest */
for(f=0;;)
{
while(f<b && pbuf[f]!=-1) f++;
if(f==b) break;
for(n=f; pbuf[n]==-1; n++);
if(pbuf[f-1]==pbuf[n])
{
/* just a gap */
while(f<n) pbuf[f++]=pbuf[n];
}
else
{
/* phase change, reanalyze the original data in the gap with zero
threshold for only the two phases that appear at the ends */
for(i=0; i<5; i++)
for(sum[i]=0, j=5*f-15; j<5*f; j+=5)
sum[i]+=buf[i+j];
for(i=f; i<n; i++)
{
pbuf[i]=match(p, sum, pbuf[f-1], pbuf[n], 0);
for(j=0; j<5; j++)
sum[j]=sum[j]-buf[5*(i-3)+j]+buf[5*(i+4)+j];
}
/* estimate the transition point by dividing the gap
in the same proportion as the number of matches of each kind */
for(i=f, m=f; i<n; i++)
if(pbuf[i]==pbuf[f-1]) m++;
/* find the transition of the right direction nearest to the
estimated point */
if(m>f && m<n)
{
for(j=m; j>f; j--)
if(pbuf[j-1]==pbuf[f-1] && pbuf[j]==pbuf[n]) break;
for(s=m; s<n; s++)
if(pbuf[s-1]==pbuf[f-1] && pbuf[s]==pbuf[n]) break;
m=(s-m<m-j)?s:j;
}
/* and rewrite the data to allow only this one transition */
for(i=f; i<m; i++)
pbuf[i]=pbuf[f-1];
for(; i<n; i++)
pbuf[i]=pbuf[n];
f=n;
}
}
free(buf-15);
return 1;
}
static int query_format(struct vf_instance *vf, unsigned int fmt)
{
switch(fmt)
{
case IMGFMT_444P: case IMGFMT_IYUV: case IMGFMT_RGB24:
case IMGFMT_422P: case IMGFMT_UYVY: case IMGFMT_BGR24:
case IMGFMT_411P: case IMGFMT_YUY2: case IMGFMT_IF09:
case IMGFMT_YV12: case IMGFMT_I420: case IMGFMT_YVU9:
case IMGFMT_IUYV: case IMGFMT_Y800: case IMGFMT_Y8:
return ff_vf_next_query_format(vf,fmt);
}
return 0;
}
static void uninit(struct vf_instance *vf)
{
if(vf->priv)
{
if(vf->priv->file) fclose(vf->priv->file);
if(vf->priv->csdata) free(vf->priv->csdata-15);
free(vf->priv->bdata);
free(vf->priv->history);
free(vf->priv);
}
}
static int vf_open(vf_instance_t *vf, char *args)
{
struct vf_priv_s *p;
const char *filename="framediff.log";
char *ap, *q, *a;
if(args && !(args=strdup(args)))
{
nomem:
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"%s: Not enough memory.\n", vf->info->name);
fail:
uninit(vf);
free(args);
return 0;
}
vf->put_image=put_image;
vf->uninit=uninit;
vf->query_format=query_format;
vf->default_reqs=VFCAP_ACCEPT_STRIDE;
if(!(vf->priv=p=calloc(1, sizeof(struct vf_priv_s))))
goto nomem;
p->phase=5;
p->threshold=0.5;
p->window=30;
if((ap=args))
while(*ap)
{
q=ap;
if((ap=strchr(q, ':'))) *ap++=0; else ap=q+strlen(q);
if((a=strchr(q, '='))) *a++=0; else a=q+strlen(q);
switch(*q)
{
case 0: break;
case 'f': filename=a; break;
case 't': p->threshold=atof(a); break;
case 'w': p->window=5*(atoi(a)+4)/5; break;
case 'd': p->deghost=atoi(a); break;
case 'p':
if(q[1]=='h') p->phase=atoi(a);
else p->pass=atoi(a);
break;
case 'h':
ff_mp_msg(MSGT_VFILTER, MSGL_INFO,
"\n%s options:\n\n"
"pass=1|2 - Use 2-pass mode.\n"
"file=filename - Set the 2-pass log file name "
"(default %s).\n"
"threshold=value - Set the pattern recognition "
"sensitivity (default %g).\n"
"deghost=value - Select deghosting threshold "
"(default %d).\n"
"window=numframes - Set the statistics window "
"for 1-pass mode (default %d).\n"
"phase=0|1|2|3|4 - Set the initial phase "
"for 1-pass mode (default %d).\n\n"
"The option names can be abbreviated to the shortest "
"unique prefix.\n\n",
vf->info->name, filename, p->threshold, p->deghost,
p->window, p->phase%5);
break;
default:
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"%s: Unknown argument %s.\n", vf->info->name, q);
goto fail;
}
}
switch(p->pass)
{
case 1:
if(!(p->file=fopen(filename, "w")))
{
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"%s: Can't create file %s.\n", vf->info->name, filename);
goto fail;
}
break;
case 2:
if(!(p->file=fopen(filename, "r")))
{
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"%s: Can't open file %s.\n", vf->info->name, filename);
goto fail;
}
if(!analyze(p))
goto fail;
fclose(p->file);
p->file=0;
break;
}
if(p->window<5) p->window=5;
if(!(p->history=calloc(sizeof *p->history, p->window)))
goto nomem;
diff = diff_C;
#if HAVE_MMX && HAVE_EBX_AVAILABLE
if(ff_gCpuCaps.hasMMX) diff = diff_MMX;
#endif
free(args);
return 1;
}
const vf_info_t ff_vf_info_divtc =
{
"inverse telecine for deinterlaced video",
"divtc",
"Ville Saari",
"",
vf_open,
NULL
};
/*
* This file is part of MPlayer.
*
* MPlayer 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.
*
* MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>
#include "config.h"
#include "mp_msg.h"
#include "cpudetect.h"
#include "img_format.h"
#include "mp_image.h"
#include "vd.h"
#include "vf.h"
#include "cmmx.h"
#include "libavutil/x86/asm.h"
#include "libvo/fastmemcpy.h"
#define NUM_STORED 4
enum pu_field_type_t {
PU_1ST_OF_3,
PU_2ND_OF_3,
PU_3RD_OF_3,
PU_1ST_OF_2,
PU_2ND_OF_2,
PU_INTERLACED
};
struct metrics {
/* This struct maps to a packed word 64-bit MMX register */
unsigned short int even;
unsigned short int odd;
unsigned short int noise;
unsigned short int temp;
} __attribute__ ((aligned (8)));
struct frame_stats {
struct metrics tiny, low, high, bigger, twox, max;
struct { unsigned int even, odd, noise, temp; } sad;
unsigned short interlaced_high;
unsigned short interlaced_low;
unsigned short num_blocks;
};
struct vf_priv_s {
unsigned long inframes;
unsigned long outframes;
enum pu_field_type_t prev_type;
unsigned swapped, chroma_swapped;
unsigned luma_only;
unsigned verbose;
unsigned fast;
unsigned long w, h, cw, ch, stride, chroma_stride, nplanes;
unsigned long sad_thres;
unsigned long dint_thres;
unsigned char *memory_allocated;
unsigned char *planes[2*NUM_STORED][4];
unsigned char **old_planes;
unsigned long static_idx;
unsigned long temp_idx;
unsigned long crop_x, crop_y, crop_cx, crop_cy;
unsigned long export_count, merge_count;
unsigned long num_breaks;
unsigned long num_copies;
long in_inc, out_dec, iosync;
long num_fields;
long prev_fields;
long notout;
long mmx2;
unsigned small_bytes[2];
unsigned mmx_temp[2];
struct frame_stats stats[2];
struct metrics thres;
char chflag;
double diff_time, merge_time, decode_time, vo_time, filter_time;
};
#define PPZ { 2000, 2000, 0, 2000 }
#define PPR { 2000, 2000, 0, 2000 }
static const struct frame_stats ppzs = {PPZ,PPZ,PPZ,PPZ,PPZ,PPZ,PPZ,0,0,9999};
static const struct frame_stats pprs = {PPR,PPR,PPR,PPR,PPR,PPR,PPR,0,0,9999};
#ifndef MIN
#define MIN(a,b) (((a)<(b))?(a):(b))
#endif
#ifndef MAX
#define MAX(a,b) (((a)>(b))?(a):(b))
#endif
#define PDIFFUB(X,Y,T) "movq " #X "," #T "\n\t" \
"psubusb " #Y "," #T "\n\t" \
"psubusb " #X "," #Y "\n\t" \
"paddusb " #Y "," #T "\n\t"
#define PDIFFUBT(X,Y,T) "movq " #X "," #T "\n\t" \
"psubusb " #Y "," #T "\n\t" \
"psubusb " #X "," #Y "\n\t" \
"paddusb " #T "," #Y "\n\t"
#define PSUMBW(X,T,Z) "movq " #X "," #T "\n\t" \
"punpcklbw " #Z "," #X "\n\t" \
"punpckhbw " #Z "," #T "\n\t" \
"paddw " #T "," #X "\n\t" \
"movq " #X "," #T "\n\t" \
"psllq $32, " #T "\n\t" \
"paddw " #T "," #X "\n\t" \
"movq " #X "," #T "\n\t" \
"psllq $16, " #T "\n\t" \
"paddw " #T "," #X "\n\t" \
"psrlq $48, " #X "\n\t"
#define PSADBW(X,Y,T,Z) PDIFFUBT(X,Y,T) PSUMBW(Y,T,Z)
#define PMAXUB(X,Y) "psubusb " #X "," #Y "\n\tpaddusb " #X "," #Y "\n\t"
#define PMAXUW(X,Y) "psubusw " #X "," #Y "\n\tpaddusw " #X "," #Y "\n\t"
#define PMINUBT(X,Y,T) "movq " #Y "," #T "\n\t" \
"psubusb " #X "," #T "\n\t" \
"psubusb " #T "," #Y "\n\t"
#define PAVGB(X,Y) "pavgusb " #X "," #Y "\n\t"
static inline void
get_metrics_c(unsigned char *a, unsigned char *b, int as, int bs, int lines,
struct metrics *m)
{
a -= as;
b -= bs;
do {
cmmx_t old_po = *(cmmx_t*)(a );
cmmx_t po = *(cmmx_t*)(b );
cmmx_t e = *(cmmx_t*)(b + bs);
cmmx_t old_o = *(cmmx_t*)(a + 2*as);
cmmx_t o = *(cmmx_t*)(b + 2*bs);
cmmx_t ne = *(cmmx_t*)(b + 3*bs);
cmmx_t old_no = *(cmmx_t*)(a + 4*as);
cmmx_t no = *(cmmx_t*)(b + 4*bs);
cmmx_t qup_old_odd = p31avgb(old_o, old_po);
cmmx_t qup_odd = p31avgb( o, po);
cmmx_t qdown_old_odd = p31avgb(old_o, old_no);
cmmx_t qdown_odd = p31avgb( o, no);
cmmx_t qup_even = p31avgb(ne, e);
cmmx_t qdown_even = p31avgb(e, ne);
cmmx_t temp_up_diff = pdiffub(qdown_even, qup_old_odd);
cmmx_t noise_up_diff = pdiffub(qdown_even, qup_odd);
cmmx_t temp_down_diff = pdiffub(qup_even, qdown_old_odd);
cmmx_t noise_down_diff = pdiffub(qup_even, qdown_odd);
cmmx_t odd_diff = pdiffub(o, old_o);
m->odd += psumbw(odd_diff);
m->even += psadbw(e, *(cmmx_t*)(a+as));
temp_up_diff = pminub(temp_up_diff, temp_down_diff);
temp_up_diff = pminub(temp_up_diff, odd_diff);
m->temp += psumbw(temp_up_diff);
noise_up_diff = pminub(noise_up_diff, odd_diff);
noise_up_diff = pminub(noise_up_diff, noise_down_diff);
m->noise += psumbw(noise_up_diff);
a += 2*as;
b += 2*bs;
} while (--lines);
}
static inline void
get_metrics_fast_c(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct metrics *m)
{
a -= as;
b -= bs;
do {
cmmx_t old_po = (*(cmmx_t*)(a ) >> 1) & ~SIGN_BITS;
cmmx_t po = (*(cmmx_t*)(b ) >> 1) & ~SIGN_BITS;
cmmx_t old_e = (*(cmmx_t*)(a + as) >> 1) & ~SIGN_BITS;
cmmx_t e = (*(cmmx_t*)(b + bs) >> 1) & ~SIGN_BITS;
cmmx_t old_o = (*(cmmx_t*)(a + 2*as) >> 1) & ~SIGN_BITS;
cmmx_t o = (*(cmmx_t*)(b + 2*bs) >> 1) & ~SIGN_BITS;
cmmx_t ne = (*(cmmx_t*)(b + 3*bs) >> 1) & ~SIGN_BITS;
cmmx_t old_no = (*(cmmx_t*)(a + 4*as) >> 1) & ~SIGN_BITS;
cmmx_t no = (*(cmmx_t*)(b + 4*bs) >> 1) & ~SIGN_BITS;
cmmx_t qup_old_odd = p31avgb_s(old_o, old_po);
cmmx_t qup_odd = p31avgb_s( o, po);
cmmx_t qdown_old_odd = p31avgb_s(old_o, old_no);
cmmx_t qdown_odd = p31avgb_s( o, no);
cmmx_t qup_even = p31avgb_s(ne, e);
cmmx_t qdown_even = p31avgb_s(e, ne);
cmmx_t temp_up_diff = pdiffub_s(qdown_even, qup_old_odd);
cmmx_t noise_up_diff = pdiffub_s(qdown_even, qup_odd);
cmmx_t temp_down_diff = pdiffub_s(qup_even, qdown_old_odd);
cmmx_t noise_down_diff = pdiffub_s(qup_even, qdown_odd);
cmmx_t odd_diff = pdiffub_s(o, old_o);
m->odd += psumbw_s(odd_diff) << 1;
m->even += psadbw_s(e, old_e) << 1;
temp_up_diff = pminub_s(temp_up_diff, temp_down_diff);
temp_up_diff = pminub_s(temp_up_diff, odd_diff);
m->temp += psumbw_s(temp_up_diff) << 1;
noise_up_diff = pminub_s(noise_up_diff, odd_diff);
noise_up_diff = pminub_s(noise_up_diff, noise_down_diff);
m->noise += psumbw_s(noise_up_diff) << 1;
a += 2*as;
b += 2*bs;
} while (--lines);
}
static inline void
get_metrics_faster_c(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct metrics *m)
{
a -= as;
b -= bs;
do {
cmmx_t old_po = (*(cmmx_t*)(a )>>1) & ~SIGN_BITS;
cmmx_t po = (*(cmmx_t*)(b )>>1) & ~SIGN_BITS;
cmmx_t old_e = (*(cmmx_t*)(a + as)>>1) & ~SIGN_BITS;
cmmx_t e = (*(cmmx_t*)(b + bs)>>1) & ~SIGN_BITS;
cmmx_t old_o = (*(cmmx_t*)(a + 2*as)>>1) & ~SIGN_BITS;
cmmx_t o = (*(cmmx_t*)(b + 2*bs)>>1) & ~SIGN_BITS;
cmmx_t ne = (*(cmmx_t*)(b + 3*bs)>>1) & ~SIGN_BITS;
cmmx_t down_even = p31avgb_s(e, ne);
cmmx_t up_odd = p31avgb_s(o, po);
cmmx_t up_old_odd = p31avgb_s(old_o, old_po);
cmmx_t odd_diff = pdiffub_s(o, old_o);
cmmx_t temp_diff = pdiffub_s(down_even, up_old_odd);
cmmx_t noise_diff = pdiffub_s(down_even, up_odd);
m->even += psadbw_s(e, old_e) << 1;
m->odd += psumbw_s(odd_diff) << 1;
temp_diff = pminub_s(temp_diff, odd_diff);
noise_diff = pminub_s(noise_diff, odd_diff);
m->noise += psumbw_s(noise_diff) << 1;
m->temp += psumbw_s(temp_diff) << 1;
a += 2*as;
b += 2*bs;
} while (--lines);
}
static inline void
get_block_stats(struct metrics *m, struct vf_priv_s *p, struct frame_stats *s)
{
unsigned two_e = m->even + MAX(m->even , p->thres.even );
unsigned two_o = m->odd + MAX(m->odd , p->thres.odd );
unsigned two_n = m->noise + MAX(m->noise, p->thres.noise);
unsigned two_t = m->temp + MAX(m->temp , p->thres.temp );
unsigned e_big = m->even >= (m->odd + two_o + 1)/2;
unsigned o_big = m->odd >= (m->even + two_e + 1)/2;
unsigned n_big = m->noise >= (m->temp + two_t + 1)/2;
unsigned t_big = m->temp >= (m->noise + two_n + 1)/2;
unsigned e2x = m->even >= two_o;
unsigned o2x = m->odd >= two_e;
unsigned n2x = m->noise >= two_t;
unsigned t2x = m->temp >= two_n;
unsigned ntiny_e = m->even > p->thres.even ;
unsigned ntiny_o = m->odd > p->thres.odd ;
unsigned ntiny_n = m->noise > p->thres.noise;
unsigned ntiny_t = m->temp > p->thres.temp ;
unsigned nlow_e = m->even > 2*p->thres.even ;
unsigned nlow_o = m->odd > 2*p->thres.odd ;
unsigned nlow_n = m->noise > 2*p->thres.noise;
unsigned nlow_t = m->temp > 2*p->thres.temp ;
unsigned high_e = m->even > 4*p->thres.even ;
unsigned high_o = m->odd > 4*p->thres.odd ;
unsigned high_n = m->noise > 4*p->thres.noise;
unsigned high_t = m->temp > 4*p->thres.temp ;
unsigned low_il = !n_big && !t_big && ntiny_n && ntiny_t;
unsigned high_il = !n_big && !t_big && nlow_n && nlow_t;
if (low_il | high_il) {
s->interlaced_low += low_il;
s->interlaced_high += high_il;
} else {
s->tiny.even += ntiny_e;
s->tiny.odd += ntiny_o;
s->tiny.noise += ntiny_n;
s->tiny.temp += ntiny_t;
s->low .even += nlow_e ;
s->low .odd += nlow_o ;
s->low .noise += nlow_n ;
s->low .temp += nlow_t ;
s->high.even += high_e ;
s->high.odd += high_o ;
s->high.noise += high_n ;
s->high.temp += high_t ;
if (m->even >= p->sad_thres) s->sad.even += m->even ;
if (m->odd >= p->sad_thres) s->sad.odd += m->odd ;
if (m->noise >= p->sad_thres) s->sad.noise += m->noise;
if (m->temp >= p->sad_thres) s->sad.temp += m->temp ;
}
s->num_blocks++;
s->max.even = MAX(s->max.even , m->even );
s->max.odd = MAX(s->max.odd , m->odd );
s->max.noise = MAX(s->max.noise, m->noise);
s->max.temp = MAX(s->max.temp , m->temp );
s->bigger.even += e_big ;
s->bigger.odd += o_big ;
s->bigger.noise += n_big ;
s->bigger.temp += t_big ;
s->twox.even += e2x ;
s->twox.odd += o2x ;
s->twox.noise += n2x ;
s->twox.temp += t2x ;
}
static inline struct metrics
block_metrics_c(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct vf_priv_s *p, struct frame_stats *s)
{
struct metrics tm;
tm.even = tm.odd = tm.noise = tm.temp = 0;
get_metrics_c(a, b, as, bs, lines, &tm);
if (sizeof(cmmx_t) < 8)
get_metrics_c(a+4, b+4, as, bs, lines, &tm);
get_block_stats(&tm, p, s);
return tm;
}
static inline struct metrics
block_metrics_fast_c(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct vf_priv_s *p, struct frame_stats *s)
{
struct metrics tm;
tm.even = tm.odd = tm.noise = tm.temp = 0;
get_metrics_fast_c(a, b, as, bs, lines, &tm);
if (sizeof(cmmx_t) < 8)
get_metrics_fast_c(a+4, b+4, as, bs, lines, &tm);
get_block_stats(&tm, p, s);
return tm;
}
static inline struct metrics
block_metrics_faster_c(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct vf_priv_s *p, struct frame_stats *s)
{
struct metrics tm;
tm.even = tm.odd = tm.noise = tm.temp = 0;
get_metrics_faster_c(a, b, as, bs, lines, &tm);
if (sizeof(cmmx_t) < 8)
get_metrics_faster_c(a+4, b+4, as, bs, lines, &tm);
get_block_stats(&tm, p, s);
return tm;
}
#define MEQ(X,Y) ((X).even == (Y).even && (X).odd == (Y).odd && (X).temp == (Y).temp && (X).noise == (Y).noise)
#define BLOCK_METRICS_TEMPLATE() \
__asm__ volatile("pxor %mm7, %mm7\n\t" /* The result is colleted in mm7 */ \
"pxor %mm6, %mm6\n\t" /* Temp to stay at 0 */ \
); \
a -= as; \
b -= bs; \
do { \
__asm__ volatile( \
"movq (%0,%2), %%mm0\n\t" \
"movq (%1,%3), %%mm1\n\t" /* mm1 = even */ \
PSADBW(%%mm1, %%mm0, %%mm4, %%mm6) \
"paddusw %%mm0, %%mm7\n\t" /* even diff */ \
"movq (%0,%2,2), %%mm0\n\t" /* mm0 = old odd */ \
"movq (%1,%3,2), %%mm2\n\t" /* mm2 = odd */ \
"movq (%0), %%mm3\n\t" \
"psubusb %4, %%mm3\n\t" \
PAVGB(%%mm0, %%mm3) \
PAVGB(%%mm0, %%mm3) /* mm3 = qup old odd */ \
"movq %%mm0, %%mm5\n\t" \
PSADBW(%%mm2, %%mm0, %%mm4, %%mm6) \
"psllq $16, %%mm0\n\t" \
"paddusw %%mm0, %%mm7\n\t" \
"movq (%1), %%mm4\n\t" \
"lea (%0,%2,2), %0\n\t" \
"lea (%1,%3,2), %1\n\t" \
"psubusb %4, %%mm4\n\t" \
PAVGB(%%mm2, %%mm4) \
PAVGB(%%mm2, %%mm4) /* mm4 = qup odd */ \
PDIFFUBT(%%mm5, %%mm2, %%mm0) /* mm2 =abs(oldodd-odd) */ \
"movq (%1,%3), %%mm5\n\t" \
"psubusb %4, %%mm5\n\t" \
PAVGB(%%mm1, %%mm5) \
PAVGB(%%mm5, %%mm1) /* mm1 = qdown even */ \
PAVGB((%1,%3), %%mm5) /* mm5 = qup next even */ \
PDIFFUBT(%%mm1, %%mm3, %%mm0) /* mm3 = abs(qupoldo-qde) */ \
PDIFFUBT(%%mm1, %%mm4, %%mm0) /* mm4 = abs(qupodd-qde) */ \
PMINUBT(%%mm2, %%mm3, %%mm0) /* limit temp to odd diff */ \
PMINUBT(%%mm2, %%mm4, %%mm0) /* limit noise to odd diff */ \
"movq (%1,%3,2), %%mm2\n\t" \
"psubusb %4, %%mm2\n\t" \
PAVGB((%1), %%mm2) \
PAVGB((%1), %%mm2) /* mm2 = qdown odd */ \
"movq (%0,%2,2), %%mm1\n\t" \
"psubusb %4, %%mm1\n\t" \
PAVGB((%0), %%mm1) \
PAVGB((%0), %%mm1) /* mm1 = qdown old odd */ \
PDIFFUBT(%%mm5, %%mm2, %%mm0) /* mm2 = abs(qdo-qune) */ \
PDIFFUBT(%%mm5, %%mm1, %%mm0) /* mm1 = abs(qdoo-qune) */ \
PMINUBT(%%mm4, %%mm2, %%mm0) /* current */ \
PMINUBT(%%mm3, %%mm1, %%mm0) /* old */ \
PSUMBW(%%mm2, %%mm0, %%mm6) \
PSUMBW(%%mm1, %%mm0, %%mm6) \
"psllq $32, %%mm2\n\t" \
"psllq $48, %%mm1\n\t" \
"paddusw %%mm2, %%mm7\n\t" \
"paddusw %%mm1, %%mm7\n\t" \
: "=r" (a), "=r" (b) \
: "r"((x86_reg)as), "r"((x86_reg)bs), "m" (ones), "0"(a), "1"(b), "X"(*a), "X"(*b) \
); \
} while (--lines);
static inline struct metrics
block_metrics_3dnow(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct vf_priv_s *p, struct frame_stats *s)
{
struct metrics tm;
#if !HAVE_AMD3DNOW
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "block_metrics_3dnow: internal error\n");
#else
static const unsigned long long ones = 0x0101010101010101ull;
BLOCK_METRICS_TEMPLATE();
__asm__ volatile("movq %%mm7, %0\n\temms" : "=m" (tm));
get_block_stats(&tm, p, s);
#endif
return tm;
}
#undef PSUMBW
#undef PSADBW
#undef PMAXUB
#undef PMINUBT
#undef PAVGB
#define PSUMBW(X,T,Z) "psadbw " #Z "," #X "\n\t"
#define PSADBW(X,Y,T,Z) "psadbw " #X "," #Y "\n\t"
#define PMAXUB(X,Y) "pmaxub " #X "," #Y "\n\t"
#define PMINUBT(X,Y,T) "pminub " #X "," #Y "\n\t"
#define PAVGB(X,Y) "pavgb " #X "," #Y "\n\t"
static inline struct metrics
block_metrics_mmx2(unsigned char *a, unsigned char *b, int as, int bs,
int lines, struct vf_priv_s *p, struct frame_stats *s)
{
struct metrics tm;
#if !HAVE_MMX
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "block_metrics_mmx2: internal error\n");
#else
static const unsigned long long ones = 0x0101010101010101ull;
x86_reg interlaced;
x86_reg prefetch_line = (((long)a>>3) & 7) + 10;
#ifdef DEBUG
struct frame_stats ts = *s;
#endif
__asm__ volatile("prefetcht0 (%0,%2)\n\t"
"prefetcht0 (%1,%3)\n\t" :
: "r" (a), "r" (b),
"r" (prefetch_line * as), "r" (prefetch_line * bs));
BLOCK_METRICS_TEMPLATE();
s->num_blocks++;
__asm__ volatile(
"movq %3, %%mm0\n\t"
"movq %%mm7, %%mm1\n\t"
"psubusw %%mm0, %%mm1\n\t"
"movq %%mm1, %%mm2\n\t"
"paddusw %%mm0, %%mm2\n\t"
"paddusw %%mm7, %%mm2\n\t"
"pshufw $0xb1, %%mm2, %%mm3\n\t"
"pavgw %%mm7, %%mm2\n\t"
"pshufw $0xb1, %%mm2, %%mm2\n\t"
"psubusw %%mm7, %%mm2\n\t"
"pcmpeqw %%mm6, %%mm2\n\t" /* 1 if >= 1.5x */
"psubusw %%mm7, %%mm3\n\t"
"pcmpeqw %%mm6, %%mm3\n\t" /* 1 if >= 2x */
"movq %1, %%mm4\n\t"
"movq %2, %%mm5\n\t"
"psubw %%mm2, %%mm4\n\t"
"psubw %%mm3, %%mm5\n\t"
"movq %%mm4, %1\n\t"
"movq %%mm5, %2\n\t"
"pxor %%mm4, %%mm4\n\t"
"pcmpeqw %%mm1, %%mm4\n\t" /* 1 if <= t */
"psubusw %%mm0, %%mm1\n\t"
"pxor %%mm5, %%mm5\n\t"
"pcmpeqw %%mm1, %%mm5\n\t" /* 1 if <= 2t */
"psubusw %%mm0, %%mm1\n\t"
"psubusw %%mm0, %%mm1\n\t"
"pcmpeqw %%mm6, %%mm1\n\t" /* 1 if <= 4t */
"pshufw $0xb1, %%mm2, %%mm0\n\t"
"por %%mm2, %%mm0\n\t" /* 1 if not close */
"punpckhdq %%mm0, %%mm0\n\t"
"movq %%mm4, %%mm2\n\t" /* tttt */
"punpckhdq %%mm5, %%mm2\n\t" /* ttll */
"por %%mm2, %%mm0\n\t"
"pcmpeqd %%mm6, %%mm0\n\t" /* close && big */
"psrlq $16, %%mm0\n\t"
"psrlw $15, %%mm0\n\t"
"movd %%mm0, %0\n\t"
: "=r" (interlaced), "=m" (s->bigger), "=m" (s->twox)
: "m" (p->thres)
);
if (interlaced) {
s->interlaced_high += interlaced >> 16;
s->interlaced_low += interlaced;
} else {
__asm__ volatile(
"pcmpeqw %%mm0, %%mm0\n\t" /* -1 */
"psubw %%mm0, %%mm4\n\t"
"psubw %%mm0, %%mm5\n\t"
"psubw %%mm0, %%mm1\n\t"
"paddw %0, %%mm4\n\t"
"paddw %1, %%mm5\n\t"
"paddw %2, %%mm1\n\t"
"movq %%mm4, %0\n\t"
"movq %%mm5, %1\n\t"
"movq %%mm1, %2\n\t"
: "=m" (s->tiny), "=m" (s->low), "=m" (s->high)
);
__asm__ volatile(
"pshufw $0, %2, %%mm0\n\t"
"psubusw %%mm7, %%mm0\n\t"
"pcmpeqw %%mm6, %%mm0\n\t" /* 0 if below sad_thres */
"pand %%mm7, %%mm0\n\t"
"movq %%mm0, %%mm1\n\t"
"punpcklwd %%mm6, %%mm0\n\t" /* sad even, odd */
"punpckhwd %%mm6, %%mm1\n\t" /* sad noise, temp */
"paddd %0, %%mm0\n\t"
"paddd %1, %%mm1\n\t"
"movq %%mm0, %0\n\t"
"movq %%mm1, %1\n\t"
: "=m" (s->sad.even), "=m" (s->sad.noise)
: "m" (p->sad_thres)
);
}
__asm__ volatile(
"movq %%mm7, (%1)\n\t"
PMAXUW((%0), %%mm7)
"movq %%mm7, (%0)\n\t"
"emms"
: : "r" (&s->max), "r" (&tm), "X" (s->max)
: "memory"
);
#ifdef DEBUG
if (1) {
struct metrics cm;
a -= 7*as;
b -= 7*bs;
cm = block_metrics_c(a, b, as, bs, 4, p, &ts);
if (!MEQ(tm, cm))
ff_mp_msg(MSGT_VFILTER, MSGL_WARN, "Bad metrics\n");
if (s) {
# define CHECK(X) if (!MEQ(s->X, ts.X)) \
ff_mp_msg(MSGT_VFILTER, MSGL_WARN, "Bad " #X "\n");
CHECK(tiny);
CHECK(low);
CHECK(high);
CHECK(sad);
CHECK(max);
}
}
#endif
#endif
return tm;
}
static inline int
dint_copy_line_mmx2(unsigned char *dst, unsigned char *a, long bos,
long cos, int ds, int ss, int w, int t)
{
#if !HAVE_MMX
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL, "dint_copy_line_mmx2: internal error\n");
return 0;
#else
unsigned long len = (w+7) >> 3;
int ret;
__asm__ volatile (
"pxor %%mm6, %%mm6 \n\t" /* deinterlaced pixel counter */
"movd %0, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t"
"punpcklwd %%mm7, %%mm7 \n\t"
"punpckldq %%mm7, %%mm7 \n\t" /* mm7 = threshold */
: /* no output */
: "rm" (t)
);
do {
__asm__ volatile (
"movq (%0), %%mm0\n\t"
"movq (%0,%3,2), %%mm1\n\t"
"movq %%mm0, (%2)\n\t"
"pmaxub %%mm1, %%mm0\n\t"
"pavgb (%0), %%mm1\n\t"
"psubusb %%mm1, %%mm0\n\t"
"paddusb %%mm7, %%mm0\n\t" /* mm0 = max-avg+thr */
"movq (%0,%1), %%mm2\n\t"
"movq (%0,%5), %%mm3\n\t"
"movq %%mm2, %%mm4\n\t"
PDIFFUBT(%%mm1, %%mm2, %%mm5)
PDIFFUBT(%%mm1, %%mm3, %%mm5)
"pminub %%mm2, %%mm3\n\t"
"pcmpeqb %%mm3, %%mm2\n\t" /* b = min */
"pand %%mm2, %%mm4\n\t"
"pandn (%0,%5), %%mm2\n\t"
"por %%mm4, %%mm2\n\t"
"pminub %%mm0, %%mm3\n\t"
"pcmpeqb %%mm0, %%mm3\n\t" /* set to 1s if >= threshold */
"psubb %%mm3, %%mm6\n\t" /* count pixels above thr. */
"pand %%mm3, %%mm1 \n\t"
"pandn %%mm2, %%mm3 \n\t"
"por %%mm3, %%mm1 \n\t" /* avg if >= threshold */
"movq %%mm1, (%2,%4) \n\t"
: /* no output */
: "r" (a), "r" ((x86_reg)bos), "r" ((x86_reg)dst), "r" ((x86_reg)ss), "r" ((x86_reg)ds), "r" ((x86_reg)cos)
);
a += 8;
dst += 8;
} while (--len);
__asm__ volatile ("pxor %%mm7, %%mm7 \n\t"
"psadbw %%mm6, %%mm7 \n\t"
"movd %%mm7, %0 \n\t"
"emms \n\t"
: "=r" (ret)
);
return ret;
#endif
}
static inline int
dint_copy_line(unsigned char *dst, unsigned char *a, long bos,
long cos, int ds, int ss, int w, int t)
{
unsigned long len = ((unsigned long)w+sizeof(cmmx_t)-1) / sizeof(cmmx_t);
cmmx_t dint_count = 0;
cmmx_t thr;
t |= t << 8;
thr = t | (t << 16);
if (sizeof(cmmx_t) > 4)
thr |= thr << (sizeof(cmmx_t)*4);
do {
cmmx_t e = *(cmmx_t*)a;
cmmx_t ne = *(cmmx_t*)(a+2*ss);
cmmx_t o = *(cmmx_t*)(a+bos);
cmmx_t oo = *(cmmx_t*)(a+cos);
cmmx_t maxe = pmaxub(e, ne);
cmmx_t avge = pavgb(e, ne);
cmmx_t max_diff = maxe - avge + thr; /* 0<=max-avg<128, thr<128 */
cmmx_t diffo = pdiffub(avge, o);
cmmx_t diffoo = pdiffub(avge, oo);
cmmx_t diffcmp = pcmpgtub(diffo, diffoo);
cmmx_t bo = ((oo ^ o) & diffcmp) ^ o;
cmmx_t diffbo = ((diffoo ^ diffo) & diffcmp) ^ diffo;
cmmx_t above_thr = ~pcmpgtub(max_diff, diffbo);
cmmx_t bo_or_avg = ((avge ^ bo) & above_thr) ^ bo;
dint_count += above_thr & ONE_BYTES;
*(cmmx_t*)(dst) = e;
*(cmmx_t*)(dst+ds) = bo_or_avg;
a += sizeof(cmmx_t);
dst += sizeof(cmmx_t);
} while (--len);
return psumbw(dint_count);
}
static int
dint_copy_plane(unsigned char *d, unsigned char *a, unsigned char *b,
unsigned char *c, unsigned long w, unsigned long h,
unsigned long ds, unsigned long ss, unsigned long threshold,
long field, long mmx2)
{
unsigned long ret = 0;
long bos = b - a;
long cos = c - a;
if (field) {
fast_memcpy(d, b, w);
h--;
d += ds;
a += ss;
}
bos += ss;
cos += ss;
while (h > 2) {
if (threshold >= 128) {
fast_memcpy(d, a, w);
fast_memcpy(d+ds, a+bos, w);
} else if (mmx2 == 1) {
ret += dint_copy_line_mmx2(d, a, bos, cos, ds, ss, w, threshold);
} else
ret += dint_copy_line(d, a, bos, cos, ds, ss, w, threshold);
h -= 2;
d += 2*ds;
a += 2*ss;
}
fast_memcpy(d, a, w);
if (h == 2)
fast_memcpy(d+ds, a+bos, w);
return ret;
}
static void
copy_merge_fields(struct vf_priv_s *p, mp_image_t *dmpi,
unsigned char **old, unsigned char **new, unsigned long show)
{
unsigned long threshold = 256;
unsigned long field = p->swapped;
unsigned long dint_pixels = 0;
unsigned char **other = old;
if (show >= 12 || !(show & 3))
show >>= 2, other = new, new = old;
if (show <= 2) { /* Single field: de-interlace */
threshold = p->dint_thres;
field ^= show & 1;
old = new;
} else if (show == 3)
old = new;
else
field ^= 1;
dint_pixels +=dint_copy_plane(dmpi->planes[0], old[0], new[0],
other[0], p->w, p->h, dmpi->stride[0],
p->stride, threshold, field, p->mmx2);
if (dmpi->flags & MP_IMGFLAG_PLANAR) {
if (p->luma_only)
old = new, other = new;
else
threshold = threshold/2 + 1;
field ^= p->chroma_swapped;
dint_copy_plane(dmpi->planes[1], old[1], new[1],
other[1], p->cw, p->ch, dmpi->stride[1],
p->chroma_stride, threshold, field, p->mmx2);
dint_copy_plane(dmpi->planes[2], old[2], new[2],
other[2], p->cw, p->ch, dmpi->stride[2],
p->chroma_stride, threshold, field, p->mmx2);
}
if (dint_pixels > 0 && p->verbose)
ff_mp_msg(MSGT_VFILTER,MSGL_INFO,"Deinterlaced %lu pixels\n",dint_pixels);
}
static void diff_planes(struct vf_priv_s *p, struct frame_stats *s,
unsigned char *of, unsigned char *nf,
int w, int h, int os, int ns, int swapped)
{
int i, y;
int align = -(long)nf & 7;
of += align;
nf += align;
w -= align;
if (swapped)
of -= os, nf -= ns;
i = (h*3 >> 7) & ~1;
of += i*os + 8;
nf += i*ns + 8;
h -= i;
w -= 16;
memset(s, 0, sizeof(*s));
for (y = (h-8) >> 3; y; y--) {
if (p->mmx2 == 1) {
for (i = 0; i < w; i += 8)
block_metrics_mmx2(of+i, nf+i, os, ns, 4, p, s);
} else if (p->mmx2 == 2) {
for (i = 0; i < w; i += 8)
block_metrics_3dnow(of+i, nf+i, os, ns, 4, p, s);
} else if (p->fast > 3) {
for (i = 0; i < w; i += 8)
block_metrics_faster_c(of+i, nf+i, os, ns, 4, p, s);
} else if (p->fast > 1) {
for (i = 0; i < w; i += 8)
block_metrics_fast_c(of+i, nf+i, os, ns, 4, p, s);
} else {
for (i = 0; i < w; i += 8)
block_metrics_c(of+i, nf+i, os, ns, 4, p, s);
}
of += 8*os;
nf += 8*ns;
}
}
#define METRICS(X) (X).even, (X).odd, (X).noise, (X).temp
static void diff_fields(struct vf_priv_s *p, struct frame_stats *s,
unsigned char **old, unsigned char **new)
{
diff_planes(p, s, old[0], new[0], p->w, p->h,
p->stride, p->stride, p->swapped);
s->sad.even = (s->sad.even * 16ul) / s->num_blocks;
s->sad.odd = (s->sad.odd * 16ul) / s->num_blocks;
s->sad.noise = (s->sad.noise * 16ul) / s->num_blocks;
s->sad.temp = (s->sad.temp * 16ul) / s->num_blocks;
if (p->verbose)
ff_mp_msg(MSGT_VFILTER, MSGL_INFO, "%lu%c M:%d/%d/%d/%d - %d, "
"t:%d/%d/%d/%d, l:%d/%d/%d/%d, h:%d/%d/%d/%d, bg:%d/%d/%d/%d, "
"2x:%d/%d/%d/%d, sad:%d/%d/%d/%d, lil:%d, hil:%d, ios:%.1f\n",
p->inframes, p->chflag, METRICS(s->max), s->num_blocks,
METRICS(s->tiny), METRICS(s->low), METRICS(s->high),
METRICS(s->bigger), METRICS(s->twox), METRICS(s->sad),
s->interlaced_low, s->interlaced_high,
p->iosync / (double) p->in_inc);
}
static const char *parse_args(struct vf_priv_s *p, const char *args)
{
args--;
while (args && *++args &&
(sscanf(args, "io=%lu:%lu", &p->out_dec, &p->in_inc) == 2 ||
sscanf(args, "diff_thres=%hu", &p->thres.even ) == 1 ||
sscanf(args, "comb_thres=%hu", &p->thres.noise) == 1 ||
sscanf(args, "sad_thres=%lu", &p->sad_thres ) == 1 ||
sscanf(args, "dint_thres=%lu", &p->dint_thres ) == 1 ||
sscanf(args, "fast=%u", &p->fast ) == 1 ||
sscanf(args, "mmx2=%lu", &p->mmx2 ) == 1 ||
sscanf(args, "luma_only=%u", &p->luma_only ) == 1 ||
sscanf(args, "verbose=%u", &p->verbose ) == 1 ||
sscanf(args, "crop=%lu:%lu:%lu:%lu", &p->w,
&p->h, &p->crop_x, &p->crop_y) == 4))
args = strchr(args, '/');
return args;
}
static unsigned long gcd(unsigned long x, unsigned long y)
{
unsigned long t;
if (x > y)
t = x, x = y, y = t;
while (x) {
t = y % x;
y = x;
x = t;
}
return y;
}
static void init(struct vf_priv_s *p, mp_image_t *mpi)
{
unsigned long i;
unsigned long plane_size, chroma_plane_size;
unsigned char *plane;
unsigned long cos, los;
p->crop_cx = p->crop_x >> mpi->chroma_x_shift;
p->crop_cy = p->crop_y >> mpi->chroma_y_shift;
if (mpi->flags & MP_IMGFLAG_ACCEPT_STRIDE) {
p->stride = (mpi->w + 15) & ~15;
p->chroma_stride = p->stride >> mpi->chroma_x_shift;
} else {
p->stride = mpi->width;
p->chroma_stride = mpi->chroma_width;
}
p->cw = p->w >> mpi->chroma_x_shift;
p->ch = p->h >> mpi->chroma_y_shift;
p->nplanes = 1;
p->static_idx = 0;
p->temp_idx = 0;
p->old_planes = p->planes[0];
plane_size = mpi->h * p->stride;
chroma_plane_size = mpi->flags & MP_IMGFLAG_PLANAR ?
mpi->chroma_height * p->chroma_stride : 0;
p->memory_allocated =
malloc(NUM_STORED * (plane_size+2*chroma_plane_size) +
8*p->chroma_stride + 4096);
/* align to page boundary */
plane = p->memory_allocated + (-(long)p->memory_allocated & 4095);
memset(plane, 0, NUM_STORED * plane_size);
los = p->crop_x + p->crop_y * p->stride;
cos = p->crop_cx + p->crop_cy * p->chroma_stride;
for (i = 0; i != NUM_STORED; i++, plane += plane_size) {
p->planes[i][0] = plane;
p->planes[NUM_STORED + i][0] = plane + los;
}
if (mpi->flags & MP_IMGFLAG_PLANAR) {
p->nplanes = 3;
memset(plane, 0x80, NUM_STORED * 2 * chroma_plane_size);
for (i = 0; i != NUM_STORED; i++) {
p->planes[i][1] = plane;
p->planes[NUM_STORED + i][1] = plane + cos;
plane += chroma_plane_size;
p->planes[i][2] = plane;
p->planes[NUM_STORED + i][2] = plane + cos;
plane += chroma_plane_size;
}
}
p->out_dec <<= 2;
i = gcd(p->in_inc, p->out_dec);
p->in_inc /= i;
p->out_dec /= i;
p->iosync = 0;
p->num_fields = 3;
}
static inline double get_time(void)
{
struct timeval tv;
gettimeofday(&tv, 0);
return tv.tv_sec + tv.tv_usec * 1e-6;
}
static void get_image(struct vf_instance *vf, mp_image_t *mpi)
{
struct vf_priv_s *p = vf->priv;
static unsigned char **planes, planes_idx;
if (mpi->type == MP_IMGTYPE_STATIC) return;
if (!p->planes[0][0]) init(p, mpi);
if (mpi->type == MP_IMGTYPE_TEMP ||
(mpi->type == MP_IMGTYPE_IPB && !(mpi->flags & MP_IMGFLAG_READABLE)))
planes_idx = NUM_STORED/2 + (++p->temp_idx % (NUM_STORED/2));
else
planes_idx = ++p->static_idx % (NUM_STORED/2);
planes = p->planes[planes_idx];
mpi->priv = p->planes[NUM_STORED + planes_idx];
if (mpi->priv == p->old_planes) {
unsigned char **old_planes =
p->planes[NUM_STORED + 2 + (++p->temp_idx & 1)];
my_memcpy_pic(old_planes[0], p->old_planes[0],
p->w, p->h, p->stride, p->stride);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(old_planes[1], p->old_planes[1],
p->cw, p->ch, p->chroma_stride, p->chroma_stride);
my_memcpy_pic(old_planes[2], p->old_planes[2],
p->cw, p->ch, p->chroma_stride, p->chroma_stride);
}
p->old_planes = old_planes;
p->num_copies++;
}
mpi->planes[0] = planes[0];
mpi->stride[0] = p->stride;
if (mpi->flags & MP_IMGFLAG_PLANAR) {
mpi->planes[1] = planes[1];
mpi->planes[2] = planes[2];
mpi->stride[1] = mpi->stride[2] = p->chroma_stride;
}
mpi->width = p->stride;
mpi->flags |= MP_IMGFLAG_DIRECT;
mpi->flags &= ~MP_IMGFLAG_DRAW_CALLBACK;
}
static inline long
cmpe(unsigned long x, unsigned long y, unsigned long err, unsigned long e)
{
long diff = x-y;
long unit = ((x+y+err) >> e);
long ret = (diff > unit) - (diff < -unit);
unit >>= 1;
return ret + (diff > unit) - (diff < -unit);
}
static unsigned long
find_breaks(struct vf_priv_s *p, struct frame_stats *s)
{
struct frame_stats *ps = &p->stats[(p->inframes-1) & 1];
long notfilm = 5*p->in_inc - p->out_dec;
unsigned long n = s->num_blocks >> 8;
unsigned long sad_comb_cmp = cmpe(s->sad.temp, s->sad.noise, 512, 1);
unsigned long ret = 8;
if (cmpe(s->sad.temp, s->sad.even, 512, 1) > 0)
ff_mp_msg(MSGT_VFILTER, MSGL_WARN,
"@@@@@@@@ Bottom-first field??? @@@@@@@@\n");
if (s->sad.temp > 1000 && s->sad.noise > 1000)
return 3;
if (s->interlaced_high >= 2*n && s->sad.temp > 256 && s->sad.noise > 256)
return 3;
if (s->high.noise > s->num_blocks/4 && s->sad.noise > 10000 &&
s->sad.noise > 2*s->sad.even && s->sad.noise > 2*ps->sad.odd) {
// Mid-frame scene change
if (s->tiny.temp + s->interlaced_low < n ||
s->low.temp + s->interlaced_high < n/4 ||
s->high.temp + s->interlaced_high < n/8 ||
s->sad.temp < 160)
return 1;
return 3;
}
if (s->high.temp > s->num_blocks/4 && s->sad.temp > 10000 &&
s->sad.temp > 2*ps->sad.odd && s->sad.temp > 2*ps->sad.even) {
// Start frame scene change
if (s->tiny.noise + s->interlaced_low < n ||
s->low.noise + s->interlaced_high < n/4 ||
s->high.noise + s->interlaced_high < n/8 ||
s->sad.noise < 160)
return 2;
return 3;
}
if (sad_comb_cmp == 2)
return 2;
if (sad_comb_cmp == -2)
return 1;
if (s->tiny.odd > 3*MAX(n,s->tiny.even) + s->interlaced_low)
return 1;
if (s->tiny.even > 3*MAX(n,s->tiny.odd)+s->interlaced_low &&
(!sad_comb_cmp || (s->low.noise <= n/4 && s->low.temp <= n/4)))
return 4;
if (s->sad.noise < 64 && s->sad.temp < 64 &&
s->low.noise <= n/2 && s->high.noise <= n/4 &&
s->low.temp <= n/2 && s->high.temp <= n/4)
goto still;
if (s->tiny.temp > 3*MAX(n,s->tiny.noise) + s->interlaced_low)
return 2;
if (s->tiny.noise > 3*MAX(n,s->tiny.temp) + s->interlaced_low)
return 1;
if (s->low.odd > 3*MAX(n/4,s->low.even) + s->interlaced_high)
return 1;
if (s->low.even > 3*MAX(n/4,s->low.odd)+s->interlaced_high &&
s->sad.even > 2*s->sad.odd &&
(!sad_comb_cmp || (s->low.noise <= n/4 && s->low.temp <= n/4)))
return 4;
if (s->low.temp > 3*MAX(n/4,s->low.noise) + s->interlaced_high)
return 2;
if (s->low.noise > 3*MAX(n/4,s->low.temp) + s->interlaced_high)
return 1;
if (sad_comb_cmp == 1 && s->sad.noise < 64)
return 2;
if (sad_comb_cmp == -1 && s->sad.temp < 64)
return 1;
if (s->tiny.odd <= n || (s->tiny.noise <= n/2 && s->tiny.temp <= n/2)) {
if (s->interlaced_low <= n) {
if (p->num_fields == 1)
goto still;
if (s->tiny.even <= n || ps->tiny.noise <= n/2)
/* Still frame */
goto still;
if (s->bigger.even >= 2*MAX(n,s->bigger.odd) + s->interlaced_low)
return 4;
if (s->low.even >= 2*n + s->interlaced_low)
return 4;
goto still;
}
}
if (s->low.odd <= n/4) {
if (s->interlaced_high <= n/4) {
if (p->num_fields == 1)
goto still;
if (s->low.even <= n/4)
/* Still frame */
goto still;
if (s->bigger.even >= 2*MAX(n/4,s->bigger.odd)+s->interlaced_high)
return 4;
if (s->low.even >= n/2 + s->interlaced_high)
return 4;
goto still;
}
}
if (s->bigger.temp > 2*MAX(n,s->bigger.noise) + s->interlaced_low)
return 2;
if (s->bigger.noise > 2*MAX(n,s->bigger.temp) + s->interlaced_low)
return 1;
if (s->bigger.temp > 2*MAX(n,s->bigger.noise) + s->interlaced_high)
return 2;
if (s->bigger.noise > 2*MAX(n,s->bigger.temp) + s->interlaced_high)
return 1;
if (s->twox.temp > 2*MAX(n,s->twox.noise) + s->interlaced_high)
return 2;
if (s->twox.noise > 2*MAX(n,s->twox.temp) + s->interlaced_high)
return 1;
if (s->bigger.even > 2*MAX(n,s->bigger.odd) + s->interlaced_low &&
s->bigger.temp < n && s->bigger.noise < n)
return 4;
if (s->interlaced_low > MIN(2*n, s->tiny.odd))
return 3;
ret = 8 + (1 << (s->sad.temp > s->sad.noise));
still:
if (p->num_fields == 1 && p->prev_fields == 3 && notfilm >= 0 &&
(s->tiny.temp <= s->tiny.noise || s->sad.temp < s->sad.noise+16))
return 1;
if (p->notout < p->num_fields && p->iosync > 2*p->in_inc && notfilm < 0)
notfilm = 0;
if (p->num_fields < 2 ||
(p->num_fields == 2 && p->prev_fields == 2 && notfilm < 0))
return ret;
if (!notfilm && (p->prev_fields&~1) == 2) {
if (p->prev_fields + p->num_fields == 5) {
if (s->tiny.noise <= s->tiny.temp ||
s->low.noise == 0 || s->low.noise < s->low.temp ||
s->sad.noise < s->sad.temp+16)
return 2;
}
if (p->prev_fields + p->num_fields == 4) {
if (s->tiny.temp <= s->tiny.noise ||
s->low.temp == 0 || s->low.temp < s->low.noise ||
s->sad.temp < s->sad.noise+16)
return 1;
}
}
if (p->num_fields > 2 &&
ps->sad.noise > s->sad.noise && ps->sad.noise > s->sad.temp)
return 4;
return 2 >> (s->sad.noise > s->sad.temp);
}
#define ITOC(X) (!(X) ? ' ' : (X) + ((X)>9 ? 'a'-10 : '0'))
static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts)
{
mp_image_t *dmpi;
struct vf_priv_s *p = vf->priv;
unsigned char **planes, **old_planes;
struct frame_stats *s = &p->stats[p->inframes & 1];
struct frame_stats *ps = &p->stats[(p->inframes-1) & 1];
int swapped = 0;
const int flags = mpi->fields;
int breaks, prev;
int show_fields = 0;
int dropped_fields = 0;
double start_time, diff_time;
char prev_chflag = p->chflag;
int keep_rate;
if (!p->planes[0][0]) init(p, mpi);
old_planes = p->old_planes;
if ((mpi->flags & MP_IMGFLAG_DIRECT) && mpi->priv) {
planes = mpi->priv;
mpi->priv = 0;
} else {
planes = p->planes[2 + (++p->temp_idx & 1)];
my_memcpy_pic(planes[0],
mpi->planes[0] + p->crop_x + p->crop_y * mpi->stride[0],
p->w, p->h, p->stride, mpi->stride[0]);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(planes[1],
mpi->planes[1] + p->crop_cx + p->crop_cy * mpi->stride[1],
p->cw, p->ch, p->chroma_stride, mpi->stride[1]);
my_memcpy_pic(planes[2],
mpi->planes[2] + p->crop_cx + p->crop_cy * mpi->stride[2],
p->cw, p->ch, p->chroma_stride, mpi->stride[2]);
p->num_copies++;
}
}
p->old_planes = planes;
p->chflag = ';';
if (flags & MP_IMGFIELD_ORDERED) {
swapped = !(flags & MP_IMGFIELD_TOP_FIRST);
p->chflag = (flags & MP_IMGFIELD_REPEAT_FIRST ? '|' :
flags & MP_IMGFIELD_TOP_FIRST ? ':' : '.');
}
p->swapped = swapped;
start_time = get_time();
if (p->chflag == '|') {
*s = ppzs;
p->iosync += p->in_inc;
} else if ((p->fast & 1) && prev_chflag == '|')
*s = pprs;
else
diff_fields(p, s, old_planes, planes);
diff_time = get_time();
p->diff_time += diff_time - start_time;
breaks = p->inframes ? find_breaks(p, s) : 2;
p->inframes++;
keep_rate = 4*p->in_inc == p->out_dec;
switch (breaks) {
case 0:
case 8:
case 9:
case 10:
if (!keep_rate && p->notout < p->num_fields && p->iosync < 2*p->in_inc)
break;
if (p->notout < p->num_fields)
dropped_fields = -2;
case 4:
if (keep_rate || p->iosync >= -2*p->in_inc)
show_fields = (4<<p->num_fields)-1;
break;
case 3:
if (keep_rate)
show_fields = 2;
else if (p->iosync > 0) {
if (p->notout >= p->num_fields && p->iosync > 2*p->in_inc) {
show_fields = 4; /* prev odd only */
if (p->num_fields > 1)
show_fields |= 8; /* + prev even */
} else {
show_fields = 2; /* even only */
if (p->notout >= p->num_fields)
dropped_fields += p->num_fields;
}
}
break;
case 2:
if (p->iosync <= -3*p->in_inc) {
if (p->notout >= p->num_fields)
dropped_fields = p->num_fields;
break;
}
if (p->num_fields == 1) {
int prevbreak = ps->sad.noise >= 128;
if (p->iosync < 4*p->in_inc) {
show_fields = 3;
dropped_fields = prevbreak;
} else {
show_fields = 4 | (!prevbreak << 3);
if (p->notout < 1 + p->prev_fields)
dropped_fields = -!prevbreak;
}
break;
}
default:
if (keep_rate)
show_fields = 3 << (breaks & 1);
else if (p->notout >= p->num_fields &&
p->iosync >= (breaks == 1 ? -p->in_inc :
p->in_inc << (p->num_fields == 1))) {
show_fields = (1 << (2 + p->num_fields)) - (1<<breaks);
} else {
if (p->notout >= p->num_fields)
dropped_fields += p->num_fields + 2 - breaks;
if (breaks == 1) {
if (p->iosync >= 4*p->in_inc)
show_fields = 6;
} else if (p->iosync > -3*p->in_inc)
show_fields = 3; /* odd+even */
}
break;
}
show_fields &= 15;
prev = p->prev_fields;
if (breaks < 8) {
if (p->num_fields == 1)
breaks &= ~4;
if (breaks)
p->num_breaks++;
if (breaks == 3)
p->prev_fields = p->num_fields = 1;
else if (breaks) {
p->prev_fields = p->num_fields + (breaks==1) - (breaks==4);
p->num_fields = breaks - (breaks == 4) + (p->chflag == '|');
} else
p->num_fields += 2;
} else
p->num_fields += 2;
p->iosync += 4 * p->in_inc;
if (p->chflag == '|')
p->iosync += p->in_inc;
if (show_fields) {
p->iosync -= p->out_dec;
p->notout = !(show_fields & 1) + !(show_fields & 3);
if (((show_fields & 3) == 3 &&
(s->low.noise + s->interlaced_low < (s->num_blocks>>8) ||
s->sad.noise < 160)) ||
((show_fields & 12) == 12 &&
(ps->low.noise + ps->interlaced_low < (s->num_blocks>>8) ||
ps->sad.noise < 160))) {
p->export_count++;
dmpi = ff_vf_get_image(vf->next, mpi->imgfmt, MP_IMGTYPE_EXPORT,
MP_IMGFLAG_PRESERVE|MP_IMGFLAG_READABLE,
p->w, p->h);
if ((show_fields & 3) != 3) planes = old_planes;
dmpi->planes[0] = planes[0];
dmpi->stride[0] = p->stride;
dmpi->width = mpi->width;
if (mpi->flags & MP_IMGFLAG_PLANAR) {
dmpi->planes[1] = planes[1];
dmpi->planes[2] = planes[2];
dmpi->stride[1] = p->chroma_stride;
dmpi->stride[2] = p->chroma_stride;
}
} else {
p->merge_count++;
dmpi = ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE,
p->w, p->h);
copy_merge_fields(p, dmpi, old_planes, planes, show_fields);
}
p->outframes++;
} else
p->notout += 2;
if (p->verbose)
ff_mp_msg(MSGT_VFILTER, MSGL_INFO, "%lu %lu: %x %c %c %lu%s%s%c%s\n",
p->inframes, p->outframes,
breaks, breaks<8 && breaks>0 ? (int) p->prev_fields+'0' : ' ',
ITOC(show_fields),
p->num_breaks, 5*p->in_inc == p->out_dec && breaks<8 &&
breaks>0 && ((prev&~1)!=2 || prev+p->prev_fields!=5) ?
" ######## bad telecine ########" : "",
dropped_fields ? " ======== dropped ":"", ITOC(dropped_fields),
!show_fields || (show_fields & (show_fields-1)) ?
"" : " @@@@@@@@@@@@@@@@@");
p->merge_time += get_time() - diff_time;
return show_fields ? ff_vf_next_put_image(vf, dmpi, MP_NOPTS_VALUE) : 0;
}
static int query_format(struct vf_instance *vf, unsigned int fmt)
{
/* FIXME - support more formats */
switch (fmt) {
case IMGFMT_YV12:
case IMGFMT_IYUV:
case IMGFMT_I420:
case IMGFMT_411P:
case IMGFMT_422P:
case IMGFMT_444P:
return ff_vf_next_query_format(vf, fmt);
}
return 0;
}
static int config(struct vf_instance *vf,
int width, int height, int d_width, int d_height,
unsigned int flags, unsigned int outfmt)
{
unsigned long cxm = 0;
unsigned long cym = 0;
struct vf_priv_s *p = vf->priv;
// rounding:
if(!IMGFMT_IS_RGB(outfmt) && !IMGFMT_IS_BGR(outfmt)){
switch(outfmt){
case IMGFMT_444P:
case IMGFMT_Y800:
case IMGFMT_Y8:
break;
case IMGFMT_YVU9:
case IMGFMT_IF09:
cym = 3;
case IMGFMT_411P:
cxm = 3;
break;
case IMGFMT_YV12:
case IMGFMT_I420:
case IMGFMT_IYUV:
cym = 1;
default:
cxm = 1;
}
}
p->chroma_swapped = !!(p->crop_y & (cym+1));
if (p->w) p->w += p->crop_x & cxm;
if (p->h) p->h += p->crop_y & cym;
p->crop_x &= ~cxm;
p->crop_y &= ~cym;
if (!p->w || p->w > width ) p->w = width;
if (!p->h || p->h > height) p->h = height;
if (p->crop_x + p->w > width ) p->crop_x = 0;
if (p->crop_y + p->h > height) p->crop_y = 0;
if(!opt_screen_size_x && !opt_screen_size_y){
d_width = d_width * p->w/width;
d_height = d_height * p->h/height;
}
return ff_vf_next_config(vf, p->w, p->h, d_width, d_height, flags, outfmt);
}
static void uninit(struct vf_instance *vf)
{
struct vf_priv_s *p = vf->priv;
ff_mp_msg(MSGT_VFILTER, MSGL_INFO, "diff_time: %.3f, merge_time: %.3f, "
"export: %lu, merge: %lu, copy: %lu\n", p->diff_time, p->merge_time,
p->export_count, p->merge_count, p->num_copies);
free(p->memory_allocated);
free(p);
}
static int vf_open(vf_instance_t *vf, char *args)
{
struct vf_priv_s *p;
vf->get_image = get_image;
vf->put_image = put_image;
vf->config = config;
vf->query_format = query_format;
vf->uninit = uninit;
vf->default_reqs = VFCAP_ACCEPT_STRIDE;
vf->priv = p = calloc(1, sizeof(struct vf_priv_s));
p->out_dec = 5;
p->in_inc = 4;
p->thres.noise = 128;
p->thres.even = 128;
p->sad_thres = 64;
p->dint_thres = 4;
p->luma_only = 0;
p->fast = 3;
p->mmx2 = ff_gCpuCaps.hasMMX2 ? 1 : ff_gCpuCaps.has3DNow ? 2 : 0;
if (args) {
const char *args_remain = parse_args(p, args);
if (args_remain) {
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"filmdint: unknown suboption: %s\n", args_remain);
return 0;
}
if (p->out_dec < p->in_inc) {
ff_mp_msg(MSGT_VFILTER, MSGL_FATAL,
"filmdint: increasing the frame rate is not supported\n");
return 0;
}
}
if (p->mmx2 > 2)
p->mmx2 = 0;
#if !HAVE_MMX
p->mmx2 = 0;
#endif
#if !HAVE_AMD3DNOW
p->mmx2 &= 1;
#endif
p->thres.odd = p->thres.even;
p->thres.temp = p->thres.noise;
p->diff_time = 0;
p->merge_time = 0;
return 1;
}
const vf_info_t ff_vf_info_filmdint = {
"Advanced inverse telecine filer",
"filmdint",
"Zoltan Hidvegi",
"",
vf_open,
NULL
};
/*
* This file is part of MPlayer.
*
* MPlayer 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.
*
* MPlayer 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 MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "config.h"
#include "mp_msg.h"
#include "cpudetect.h"
#include "img_format.h"
#include "mp_image.h"
#include "vf.h"
#include "libavutil/x86/asm.h"
#include "libvo/fastmemcpy.h"
struct metrics {
/* difference: total, even lines, odd lines */
int d, e, o;
/* noise: temporal, spacial (current), spacial (past) */
int t, s, p;
};
struct frameinfo {
/* peak, relative, mean */
struct metrics p, r, m;
};
struct vf_priv_s {
struct frameinfo fi[2];
mp_image_t *dmpi;
int first;
int drop, lastdrop, dropnext;
int inframes, outframes;
};
enum {
F_DROP,
F_MERGE,
F_NEXT,
F_SHOW
};
#if HAVE_MMX && HAVE_EBX_AVAILABLE
static void block_diffs_MMX(struct metrics *m, unsigned char *old, unsigned char *new, int os, int ns)
{
int i;
short out[24]; // output buffer for the partial metrics from the mmx code
__asm__ (
"movl $4, %%ecx \n\t"
"pxor %%mm4, %%mm4 \n\t" // 4 even difference sums
"pxor %%mm5, %%mm5 \n\t" // 4 odd difference sums
"pxor %%mm7, %%mm7 \n\t" // all zeros
ASMALIGN(4)
"1: \n\t"
// Even difference
"movq (%%"REG_S"), %%mm0 \n\t"
"movq (%%"REG_S"), %%mm2 \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"movq (%%"REG_D"), %%mm1 \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"psubusb %%mm1, %%mm2 \n\t"
"psubusb %%mm0, %%mm1 \n\t"
"movq %%mm2, %%mm0 \n\t"
"movq %%mm1, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpckhbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm0, %%mm4 \n\t"
"paddw %%mm1, %%mm4 \n\t"
"paddw %%mm2, %%mm4 \n\t"
"paddw %%mm3, %%mm4 \n\t"
// Odd difference
"movq (%%"REG_S"), %%mm0 \n\t"
"movq (%%"REG_S"), %%mm2 \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"movq (%%"REG_D"), %%mm1 \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"psubusb %%mm1, %%mm2 \n\t"
"psubusb %%mm0, %%mm1 \n\t"
"movq %%mm2, %%mm0 \n\t"
"movq %%mm1, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpckhbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm0, %%mm5 \n\t"
"paddw %%mm1, %%mm5 \n\t"
"paddw %%mm2, %%mm5 \n\t"
"paddw %%mm3, %%mm5 \n\t"
"decl %%ecx \n\t"
"jnz 1b \n\t"
"movq %%mm4, (%%"REG_d") \n\t"
"movq %%mm5, 8(%%"REG_d") \n\t"
:
: "S" (old), "D" (new), "a" (os), "b" (ns), "d" (out)
: "memory"
);
m->e = out[0]+out[1]+out[2]+out[3];
m->o = out[4]+out[5]+out[6]+out[7];
m->d = m->e + m->o;
__asm__ (
// First loop to measure first four columns
"movl $4, %%ecx \n\t"
"pxor %%mm4, %%mm4 \n\t" // Past spacial noise
"pxor %%mm5, %%mm5 \n\t" // Temporal noise
"pxor %%mm6, %%mm6 \n\t" // Current spacial noise
ASMALIGN(4)
"2: \n\t"
"movq (%%"REG_S"), %%mm0 \n\t"
"movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"movq (%%"REG_D"), %%mm2 \n\t"
"movq (%%"REG_D",%%"REG_b"), %%mm3 \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpcklbw %%mm7, %%mm3 \n\t"
"paddw %%mm1, %%mm4 \n\t"
"paddw %%mm1, %%mm5 \n\t"
"paddw %%mm3, %%mm6 \n\t"
"psubw %%mm0, %%mm4 \n\t"
"psubw %%mm2, %%mm5 \n\t"
"psubw %%mm2, %%mm6 \n\t"
"decl %%ecx \n\t"
"jnz 2b \n\t"
"movq %%mm0, %%mm1 \n\t"
"movq %%mm0, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"pcmpgtw %%mm4, %%mm1 \n\t"
"pcmpgtw %%mm5, %%mm2 \n\t"
"pcmpgtw %%mm6, %%mm3 \n\t"
"pxor %%mm1, %%mm4 \n\t"
"pxor %%mm2, %%mm5 \n\t"
"pxor %%mm3, %%mm6 \n\t"
"psubw %%mm1, %%mm4 \n\t"
"psubw %%mm2, %%mm5 \n\t"
"psubw %%mm3, %%mm6 \n\t"
"movq %%mm4, (%%"REG_d") \n\t"
"movq %%mm5, 16(%%"REG_d") \n\t"
"movq %%mm6, 32(%%"REG_d") \n\t"
"mov %%"REG_a", %%"REG_c" \n\t"
"shl $3, %%"REG_c" \n\t"
"sub %%"REG_c", %%"REG_S" \n\t"
"mov %%"REG_b", %%"REG_c" \n\t"
"shl $3, %%"REG_c" \n\t"
"sub %%"REG_c", %%"REG_D" \n\t"
// Second loop for the last four columns
"movl $4, %%ecx \n\t"
"pxor %%mm4, %%mm4 \n\t"
"pxor %%mm5, %%mm5 \n\t"
"pxor %%mm6, %%mm6 \n\t"
ASMALIGN(4)
"3: \n\t"
"movq (%%"REG_S"), %%mm0 \n\t"
"movq (%%"REG_S",%%"REG_a"), %%mm1 \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"add %%"REG_a", %%"REG_S" \n\t"
"movq (%%"REG_D"), %%mm2 \n\t"
"movq (%%"REG_D",%%"REG_b"), %%mm3 \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"add %%"REG_b", %%"REG_D" \n\t"
"punpckhbw %%mm7, %%mm0 \n\t"
"punpckhbw %%mm7, %%mm1 \n\t"
"punpckhbw %%mm7, %%mm2 \n\t"
"punpckhbw %%mm7, %%mm3 \n\t"
"paddw %%mm1, %%mm4 \n\t"
"paddw %%mm1, %%mm5 \n\t"
"paddw %%mm3, %%mm6 \n\t"
"psubw %%mm0, %%mm4 \n\t"
"psubw %%mm2, %%mm5 \n\t"
"psubw %%mm2, %%mm6 \n\t"
"decl %%ecx \n\t"
"jnz 3b \n\t"
"movq %%mm0, %%mm1 \n\t"
"movq %%mm0, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"pcmpgtw %%mm4, %%mm1 \n\t"
"pcmpgtw %%mm5, %%mm2 \n\t"
"pcmpgtw %%mm6, %%mm3 \n\t"
"pxor %%mm1, %%mm4 \n\t"
"pxor %%mm2, %%mm5 \n\t"
"pxor %%mm3, %%mm6 \n\t"
"psubw %%mm1, %%mm4 \n\t"
"psubw %%mm2, %%mm5 \n\t"
"psubw %%mm3, %%mm6 \n\t"
"movq %%mm4, 8(%%"REG_d") \n\t"
"movq %%mm5, 24(%%"REG_d") \n\t"
"movq %%mm6, 40(%%"REG_d") \n\t"
"emms \n\t"
:
: "S" (old), "D" (new), "a" ((long)os), "b" ((long)ns), "d" (out)
: "memory"
);
m->p = m->t = m->s = 0;
for (i=0; i<8; i++) {
m->p += out[i];
m->t += out[8+i];
m->s += out[16+i];
}
//printf("e=%d o=%d d=%d p=%d t=%d s=%d\n", m->e, m->o, m->d, m->p, m->t, m->s);
}
#endif
//#define MAG(a) ((a)*(a))
//#define MAG(a) (abs(a))
#define MAG(a) (((a)^((a)>>31))-((a)>>31))
//#define LOWPASS(s) (((s)[-2] + 4*(s)[-1] + 6*(s)[0] + 4*(s)[1] + (s)[2])>>4)
//#define LOWPASS(s) (((s)[-1] + 2*(s)[0] + (s)[1])>>2)
#define LOWPASS(s) ((s)[0])
static void block_diffs_C(struct metrics *m, unsigned char *old, unsigned char *new, int os, int ns)
{
int x, y, e=0, o=0, s=0, p=0, t=0;
unsigned char *oldp, *newp;
m->s = m->p = m->t = 0;
for (x = 8; x; x--) {
oldp = old++;
newp = new++;
s = p = t = 0;
for (y = 4; y; y--) {
e += MAG(newp[0]-oldp[0]);
o += MAG(newp[ns]-oldp[os]);
s += newp[ns]-newp[0];
p += oldp[os]-oldp[0];
t += oldp[os]-newp[0];
oldp += os<<1;
newp += ns<<1;
}
m->s += MAG(s);
m->p += MAG(p);
m->t += MAG(t);
}
m->e = e;
m->o = o;
m->d = e+o;
}
static void (*block_diffs)(struct metrics *, unsigned char *, unsigned char *, int, int);
#define MAXUP(a,b) ((a) = ((a)>(b)) ? (a) : (b))
static void diff_planes(struct frameinfo *fi,
unsigned char *old, unsigned char *new, int w, int h, int os, int ns)
{
int x, y;
struct metrics l;
struct metrics *peak=&fi->p, *rel=&fi->r, *mean=&fi->m;
memset(peak, 0, sizeof(struct metrics));
memset(rel, 0, sizeof(struct metrics));
memset(mean, 0, sizeof(struct metrics));
for (y = 0; y < h-7; y += 8) {
for (x = 8; x < w-8-7; x += 8) {
block_diffs(&l, old+x+y*os, new+x+y*ns, os, ns);
mean->d += l.d;
mean->e += l.e;
mean->o += l.o;
mean->s += l.s;
mean->p += l.p;
mean->t += l.t;
MAXUP(peak->d, l.d);
MAXUP(peak->e, l.e);
MAXUP(peak->o, l.o);
MAXUP(peak->s, l.s);
MAXUP(peak->p, l.p);
MAXUP(peak->t, l.t);
MAXUP(rel->e, l.e-l.o);
MAXUP(rel->o, l.o-l.e);
MAXUP(rel->s, l.s-l.t);
MAXUP(rel->p, l.p-l.t);
MAXUP(rel->t, l.t-l.p);
MAXUP(rel->d, l.t-l.s); /* hack */
}
}
x = (w/8-2)*(h/8);
mean->d /= x;
mean->e /= x;
mean->o /= x;
mean->s /= x;
mean->p /= x;
mean->t /= x;
}
static void diff_fields(struct frameinfo *fi, mp_image_t *old, mp_image_t *new)
{
diff_planes(fi, old->planes[0], new->planes[0],
new->w, new->h, old->stride[0], new->stride[0]);
}
static void stats(struct frameinfo *f)
{
ff_mp_msg(MSGT_VFILTER, MSGL_V, " pd=%d re=%d ro=%d rp=%d rt=%d rs=%d rd=%d pp=%d pt=%d ps=%d\r",
f->p.d, f->r.e, f->r.o, f->r.p, f->r.t, f->r.s, f->r.d, f->p.p, f->p.t, f->p.s);
}
static int foo(struct vf_priv_s *p, mp_image_t *new, mp_image_t *cur)
{
struct frameinfo *f = p->fi;
f[0] = f[1];
diff_fields(&f[1], cur, new);
stats(&f[1]);
// Immediately drop this frame if it's already been used.
if (p->dropnext) {
p->dropnext = 0;
return F_DROP;
}
// Sometimes a pulldown frame comes all by itself, so both
// its top and bottom field are duplicates from the adjacent
// two frames. We can just drop such a frame, but we
// immediately show the next frame instead to keep the frame
// drops evenly spaced during normal 3:2 pulldown sequences.
if ((3*f[1].r.o < f[1].r.e) && (f[1].r.s < f[1].r.d)) {
p->dropnext = 1;
return F_NEXT;
}
// If none of these conditions hold, we will consider the frame
// progressive and just show it as-is.
if (!( (3*f[0].r.e < f[0].r.o) ||
((2*f[0].r.d < f[0].r.s) && (f[0].r.s > 1200)) ||
((2*f[1].r.t < f[1].r.p) && (f[1].r.p > 1200)) ))
return F_SHOW;
// Otherwise, we have to decide whether to merge or drop.
// If the noise metric only increases minimally, we're off
// to a good start...
if (((2*f[1].r.t < 3*f[1].r.p) && (f[1].r.t < 3600)) ||
(f[1].r.t < 900) || (f[1].r.d < 900)) {
// ...and if noise decreases or the duplicate even field
// is detected, we go ahead with the merge.
if ((3*f[0].r.e < f[0].r.o) || (2*f[1].r.t < f[1].r.p)) {
p->dropnext = 1;
return F_MERGE;
}
}
return F_DROP;
}
static void copy_image(mp_image_t *dmpi, mp_image_t *mpi, int field)
{
switch (field) {
case 0:
my_memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h/2,
dmpi->stride[0]*2, mpi->stride[0]*2);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(dmpi->planes[1], mpi->planes[1],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[1]*2, mpi->stride[1]*2);
my_memcpy_pic(dmpi->planes[2], mpi->planes[2],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[2]*2, mpi->stride[2]*2);
}
break;
case 1:
my_memcpy_pic(dmpi->planes[0]+dmpi->stride[0],
mpi->planes[0]+mpi->stride[0], mpi->w, mpi->h/2,
dmpi->stride[0]*2, mpi->stride[0]*2);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
my_memcpy_pic(dmpi->planes[1]+dmpi->stride[1],
mpi->planes[1]+mpi->stride[1],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[1]*2, mpi->stride[1]*2);
my_memcpy_pic(dmpi->planes[2]+dmpi->stride[2],
mpi->planes[2]+mpi->stride[2],
mpi->chroma_width, mpi->chroma_height/2,
dmpi->stride[2]*2, mpi->stride[2]*2);
}
break;
case 2:
memcpy_pic(dmpi->planes[0], mpi->planes[0], mpi->w, mpi->h,
dmpi->stride[0], mpi->stride[0]);
if (mpi->flags & MP_IMGFLAG_PLANAR) {
memcpy_pic(dmpi->planes[1], mpi->planes[1],
mpi->chroma_width, mpi->chroma_height,
dmpi->stride[1], mpi->stride[1]);
memcpy_pic(dmpi->planes[2], mpi->planes[2],
mpi->chroma_width, mpi->chroma_height,
dmpi->stride[2], mpi->stride[2]);
}
break;
}
}
static int do_put_image(struct vf_instance *vf, mp_image_t *dmpi)
{
struct vf_priv_s *p = vf->priv;
int dropflag=0;
if (!p->dropnext) switch (p->drop) {
case 0:
dropflag = 0;
break;
case 1:
dropflag = (++p->lastdrop >= 5);
break;
case 2:
dropflag = (++p->lastdrop >= 5) && (4*p->inframes <= 5*p->outframes);
break;
}
if (dropflag) {
//ff_mp_msg(MSGT_VFILTER, MSGL_V, "drop! [%d/%d=%g]\n",
// p->outframes, p->inframes, (float)p->outframes/p->inframes);
ff_mp_msg(MSGT_VFILTER, MSGL_V, "!");
p->lastdrop = 0;
return 0;
}
p->outframes++;
return ff_vf_next_put_image(vf, dmpi, MP_NOPTS_VALUE);
}
static int put_image(struct vf_instance *vf, mp_image_t *mpi, double pts)
{
int ret=0;
struct vf_priv_s *p = vf->priv;
p->inframes++;
if (p->first) { /* hack */
p->first = 0;
return 1;
}
if (!p->dmpi) p->dmpi = ff_vf_get_image(vf->next, mpi->imgfmt,
MP_IMGTYPE_STATIC, MP_IMGFLAG_ACCEPT_STRIDE |
MP_IMGFLAG_PRESERVE | MP_IMGFLAG_READABLE,
mpi->width, mpi->height);
/* FIXME -- not correct, off by one frame! */
p->dmpi->qscale = mpi->qscale;
p->dmpi->qstride = mpi->qstride;
p->dmpi->qscale_type = mpi->qscale_type;
switch (foo(p, mpi, p->dmpi)) {
case F_DROP:
copy_image(p->dmpi, mpi, 2);
ret = 0;
p->lastdrop = 0;
ff_mp_msg(MSGT_VFILTER, MSGL_V, "DROP\n");
break;
case F_MERGE:
copy_image(p->dmpi, mpi, 0);
ret = do_put_image(vf, p->dmpi);
copy_image(p->dmpi, mpi, 1);
ff_mp_msg(MSGT_VFILTER, MSGL_V, "MERGE\n");
p->dmpi = NULL;
break;
case F_NEXT:
copy_image(p->dmpi, mpi, 2);
ret = do_put_image(vf, p->dmpi);
ff_mp_msg(MSGT_VFILTER, MSGL_V, "NEXT\n");
p->dmpi = NULL;
break;
case F_SHOW:
ret = do_put_image(vf, p->dmpi);
copy_image(p->dmpi, mpi, 2);
ff_mp_msg(MSGT_VFILTER, MSGL_V, "OK\n");
p->dmpi = NULL;
break;
}
return ret;
}
static int query_format(struct vf_instance *vf, unsigned int fmt)
{
switch (fmt) {
case IMGFMT_YV12:
case IMGFMT_IYUV:
case IMGFMT_I420:
return ff_vf_next_query_format(vf, fmt);
}
return 0;
}
static void uninit(struct vf_instance *vf)
{
free(vf->priv);
}
static int vf_open(vf_instance_t *vf, char *args)
{
struct vf_priv_s *p;
vf->put_image = put_image;
vf->query_format = query_format;
vf->uninit = uninit;
vf->default_reqs = VFCAP_ACCEPT_STRIDE;
vf->priv = p = calloc(1, sizeof(struct vf_priv_s));
p->drop = 0;
p->first = 1;
if (args) sscanf(args, "%d", &p->drop);
block_diffs = block_diffs_C;
#if HAVE_MMX && HAVE_EBX_AVAILABLE
if(ff_gCpuCaps.hasMMX) block_diffs = block_diffs_MMX;
#endif
return 1;
}
const vf_info_t ff_vf_info_ivtc = {
"inverse telecine, take 2",
"ivtc",
"Rich Felker",
"",
vf_open,
NULL
};
......@@ -30,7 +30,7 @@
#define LIBAVFILTER_VERSION_MAJOR 3
#define LIBAVFILTER_VERSION_MINOR 56
#define LIBAVFILTER_VERSION_MICRO 101
#define LIBAVFILTER_VERSION_MICRO 102
#define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
LIBAVFILTER_VERSION_MINOR, \
......
......@@ -123,17 +123,13 @@ static const struct {
{0, AV_PIX_FMT_NONE}
};
extern const vf_info_t ff_vf_info_detc;
extern const vf_info_t ff_vf_info_dint;
extern const vf_info_t ff_vf_info_divtc;
extern const vf_info_t ff_vf_info_down3dright;
extern const vf_info_t ff_vf_info_eq2;
extern const vf_info_t ff_vf_info_eq;
extern const vf_info_t ff_vf_info_fil;
//extern const vf_info_t ff_vf_info_filmdint;
extern const vf_info_t ff_vf_info_fspp;
extern const vf_info_t ff_vf_info_ilpack;
extern const vf_info_t ff_vf_info_ivtc;
extern const vf_info_t ff_vf_info_mcdeint;
extern const vf_info_t ff_vf_info_ow;
extern const vf_info_t ff_vf_info_perspective;
......@@ -149,17 +145,13 @@ extern const vf_info_t ff_vf_info_uspp;
static const vf_info_t* const filters[]={
&ff_vf_info_detc,
&ff_vf_info_dint,
&ff_vf_info_divtc,
&ff_vf_info_down3dright,
&ff_vf_info_eq2,
&ff_vf_info_eq,
&ff_vf_info_fil,
// &ff_vf_info_filmdint, cmmx.h vd.h ‘opt_screen_size_x’
&ff_vf_info_fspp,
&ff_vf_info_ilpack,
&ff_vf_info_ivtc,
&ff_vf_info_mcdeint,
&ff_vf_info_ow,
&ff_vf_info_perspective,
......
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