annotate libmpcodecs/vf_ivtc.c @ 13720:821f464b4d90

adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64

#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 "../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
};

static inline void *my_memcpy_pic(void * dst, void * src, int bytesPerLine, int height, int dstStride, int srcStride)
{
	int i;
	void *retval=dst;

	for(i=0; i<height; i++)
	{
		memcpy(dst, src, bytesPerLine);
		src+= srcStride;
		dst+= dstStride;
	}

	return retval;
}

#ifdef HAVE_MMX
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
		
		".balign 16 \n\t"
		"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
		
		".balign 16 \n\t"
		"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"
		
		".balign 16 \n\t"
		"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)
{
	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_s* vf, mp_image_t *dmpi)
{
	struct vf_priv_s *p = vf->priv;
	int dropflag;

	switch (p->drop && !p->dropnext) {
	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) {
		//mp_msg(MSGT_VFILTER, MSGL_V, "drop! [%d/%d=%g]\n",
		//	p->outframes, p->inframes, (float)p->outframes/p->inframes);
		mp_msg(MSGT_VFILTER, MSGL_V, "!");
		p->lastdrop = 0;
		return 0;
	}

	p->outframes++;
	return vf_next_put_image(vf, dmpi);
}

static int put_image(struct vf_instance_s* vf, mp_image_t *mpi)
{
	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 = 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;
		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);
		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);
		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);
		mp_msg(MSGT_VFILTER, MSGL_V, "OK\n");
		p->dmpi = NULL;
		break;
	}
	return ret;
}

static int query_format(struct vf_instance_s* vf, unsigned int fmt)
{
	switch (fmt) {
	case IMGFMT_YV12:
	case IMGFMT_IYUV:
	case IMGFMT_I420:
		return vf_next_query_format(vf, fmt);
	}
	return 0;
}

static void uninit(struct vf_instance_s* vf)
{
	free(vf->priv);
}

static int 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;
#ifdef HAVE_MMX
	if(gCpuCaps.hasMMX) block_diffs = block_diffs_MMX;
#endif
	return 1;
}

vf_info_t vf_info_ivtc = {
    "inverse telecine, take 2",
    "ivtc",
    "Rich Felker",
    "",
    open,
    NULL
};