p64.cpp

网络MPEG4IP流媒体开发源代码

/*
 * Copyright (c) 1993-1995 Regents of the University of California.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the Network Research
 *	Group at Lawrence Berkeley National Laboratory.
 * 4. Neither the name of the University nor of the Laboratory may be used
 *    to endorse or promote products derived from this software without
 *    specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */
/*
 * This code is derived from the P64 software implementation by the
 * Stanford PVRG group:
 * 
 * Copyright (C) 1990, 1991, 1993 Andy C. Hung, all rights reserved.
 * PUBLIC DOMAIN LICENSE: Stanford University Portable Video Research
 * Group. If you use this software, you agree to the following: This
 * program package is purely experimental, and is licensed "as is".
 * Permission is granted to use, modify, and distribute this program
 * without charge for any purpose, provided this license/ disclaimer
 * notice appears in the copies.  No warranty or maintenance is given,
 * either expressed or implied.  In no event shall the author(s) be
 * liable to you or a third party for any special, incidental,
 * consequential, or other damages, arising out of the use or inability
 * to use the program for any purpose (or the loss of data), even if we
 * have been advised of such possibilities.  Any public reference or
 * advertisement of this source code should refer to it as the Portable
 * Video Research Group (PVRG) code, and not by any author(s) (or
 * Stanford University) name.
 */


#include "systems.h"
#include "p64.h"
#include "p64-huff.h"
#include "dct.h"
#include "bsd-endian.h"

void P64Decoder::err(const char* msg ...) const
{
#ifdef DEVELOPMENT_VERSION
	va_list ap;
	va_start(ap, msg);
	vfprintf(stderr, msg, ap);
	fprintf(stderr, " @g%d m%d %d/%d of %d/%d: %04x %04x %04x %04x|%04x\n",
		gob_, mba_,
		(int)((u_char*)bs_ - (u_char*)ps_), nbb_,
		(int)((u_char*)es_ - (u_char*)ps_), pebit_,
	       bs_[-4], bs_[-3], bs_[-2], bs_[-1], bs_[0]);
#endif
}

P64Decoder::P64Decoder()
{
	mt_ = 0;
	mvdh_ = 0;
	mvdv_ = 0;
	mba_ = 0;
	ngob_ = 0;
	ndblk_ = 0;
	gobquant_ = 0;
	gob_ = 0;
	maxgob_ = 0;
	bad_psc_ = 0;
	bad_bits_ = 0;
	bad_GOBno_ = 0;
	bad_fmt_ = 0;
	fs_ = 0;
	front_ = 0;
	back_ = 0;
	marks_ = 0;
	mark_ = 0;
	fmt_ = IT_CIF;/*XXX*/
	inithuff();
	initquant();
}

P64Decoder::~P64Decoder()
{
	delete fs_;
}

void P64Decoder::init()
{
	if (fmt_ == IT_CIF) {
		ngob_ = 12;
		width_ = 352;
		height_ = 288;
	} else {
		ngob_ = 3;
		width_ = 176;
		height_ = 144;
	}
	size_ = width_ * height_;
	memset(mb_state_, MBST_OLD, sizeof(mb_state_));

	for (u_int gob = 0; gob < 12; ++gob) {
		u_short* p = &base_[gob << 6];
		for (int mba = 0; mba < MBPERGOB; ++mba) {
			u_int x = 2 * (mba % 11);
			u_int y;
			if (fmt_ == IT_CIF) {
				y = 2 * (3 * (gob >> 1) + mba / 11);
				if (gob & 1)
					x += 22;
			} else
				y = 2 * (3 * gob + mba / 11);
			
			p[mba] = (x << 8) | y;
		}
	}
	minx_ = width_;
	miny_ = height_;
	maxx_ = 0;
	maxy_ = 0;

	allocate();
}

#if BYTE_ORDER == LITTLE_ENDIAN
#define HUFFRQ(bs, bb) \
 { \
	register int t = *bs++; \
	bb <<= 16; \
	bb |= (t & 0xff) << 8; \
	bb |= t >> 8; \
}
#else
#define HUFFRQ(bs, bb) \
 { \
	bb <<= 16; \
	bb |= *bs++; \
}
#endif

#define P64MASK(s) ((1 << (s)) - 1)

#define HUFF_DECODE(bs, ht, nbb, bb, result) { \
	register int s__, v__; \
 \
	if (nbb < 16) { \
		HUFFRQ(bs, bb); \
		nbb += 16; \
	} \
	s__ = ht.maxlen; \
	v__ = (bb >> (nbb - s__)) & P64MASK(s__); \
	s__ = (ht.prefix)[v__]; \
	nbb -= (s__ & 0x1f); \
	result = s__ >> 5; \
 }

#define GET_BITS(bs, n, nbb, bb, result) \
{ \
	nbb -= n; \
	if (nbb < 0)  { \
		HUFFRQ(bs, bb); \
		nbb += 16; \
	} \
	(result) = ((bb >> nbb) & P64MASK(n)); \
}

#define SKIP_BITS(bs, n, nbb, bb) \
{ \
	nbb -= n; \
	if (nbb < 0)  { \
		HUFFRQ(bs, bb); \
		nbb += 16; \
	} \
}

/*
 * Set up the huffman tables.
 */
void P64Decoder::inithuff()
{
	ht_mtype_.prefix = htd_mtype;
	ht_mtype_.maxlen = htd_mtype_width;
	ht_mba_.prefix = htd_mba;
	ht_mba_.maxlen = htd_mba_width;
	ht_mvd_.prefix = htd_dvm;
	ht_mvd_.maxlen = htd_dvm_width;
	ht_cbp_.prefix = htd_cbp;
	ht_cbp_.maxlen = htd_cbp_width;
	ht_tcoeff_.prefix = htd_tcoeff;
	ht_tcoeff_.maxlen = htd_tcoeff_width;
}

int P64Decoder::quantize(int v, int q)
{
	if (v > 0) 
		return (((v << 1) + 1) * q) - (~q & 1);
	else 
		return (((v << 1) - 1) * q) + (~q & 1);
}

/*
 * Build quantization lookup table.
 * One for each possible MQUANT paramenter.
 */
void P64Decoder::initquant()
{
	for (int mq = 0; mq < 32; ++mq) {
		short* qt = &quant_[mq << 8];
		for (int v = 0; v < 256; ++v) {
			int s = (v << 24) >> 24;
			qt[v] = quantize(s, mq);
		}
	}
}

/*
 * Decode the next block of transform coefficients
 * from the input stream.
 * Return number of non-zero ac coefficients.
 */
#ifdef INT_64
int P64Decoder::parse_block(short* blk, INT_64* mask)
#else
int P64Decoder::parse_block(short* blk, u_int* mask)
#endif
{
#ifdef INT_64
	INT_64 m0 = 0;
#else
	u_int m1 = 0, m0 = 0;
#endif
	/*
	 * Cache bit buffer in registers.
	 */
	register int nbb = nbb_;
	register int bb = bb_;
	register short* qt = qt_;

	int k;
	if ((mt_ & MT_CBP) == 0) {
		int v;
		GET_BITS(bs_, 8, nbb, bb, v);
		if (v == 255)
			v = 128;
		if (mt_ & MT_INTRA)
			v <<= 3;
		else
			v = qt[v];
		blk[0] = v;
		k = 1;
		m0 |= 1;
	} else if ((bb >> (nbb - 1)) & 1) {
		/*
		 * In CBP blocks, the first block present must be
		 * non-empty (otherwise it's mask bit wouldn't
		 * be set), so the first code cannot be an EOB.
		 * CCITT optimizes this case by using a huffman
		 * table equivalent to ht_tcoeff_ but without EOB,
		 * in which 1 is coded as "1" instead of "11".
		 * We grab two bits, the first bit is the code
		 * and the second is the sign.
		 */
		int v;
		GET_BITS(bs_, 2, nbb, bb, v);
		/*XXX quantize?*/
		blk[0] = qt[(v & 1) ? 0xff : 1];
		k = 1;
		m0 |= 1;
	} else {
		k = 0;
#ifndef INT_64
		blk[0] = 0;/*XXX need this because the way we set bits below*/
#endif
	}
	int nc = 0;
	for (;;) {
		int r, v;
		HUFF_DECODE(bs_, ht_tcoeff_, nbb, bb, r);
		if (r <= 0) {
			/* SYM_EOB, SYM_ILLEGAL, or SYM_ESCAPE */
			if (r == SYM_ESCAPE) {
				GET_BITS(bs_, 14, nbb, bb, r);
				v = r & 0xff;
				r >>= 8;
			} else {
				if (r == SYM_ILLEGAL) {
					bb_ = bb;
					nbb_ = nbb;
					err("illegal symbol in block");
				}
				/* EOB */
				break;
			}
		} else {
			v = (r << 22) >> 27;
			r = r & 0x1f;
		}
		k += r;
		if (k >= 64) {
			bb_ = bb;
			nbb_ = nbb;
			err("bad run length %d (r %d, v %d)", k, r, v);
			break;
		}
		r = COLZAG[k++];
		blk[r] = qt[v & 0xff];
		++nc;
#ifdef INT_64
		m0 |= (INT_64)1 << r;
#else
		if (r < 32)
			m0 |= 1 << r;
		else
			m1 |= 1 << (r - 32);
#endif
	}
	/*
	 * Done reading input.  Update bit buffer.
	 */
	bb_ = bb;
	nbb_ = nbb;

	*mask = m0;
#ifndef INT_64
	mask[1] = m1;
#endif
	return (nc);
}

/*
 * Mix in a motion-compensated, filtered block.  Note that
 * the input block may be misaligned so we cannot try fancy,
 * word-at-a-time accesses without being careful.  The output
 * block is, of course, aligned.
 * 
 * The 2-D loop filter is separable into 1-D FIR (0.25 0.5 0.25)
 * horizontal and vertical passes.  At the block edge, the filter
 * taps are (0 1 0).  Full arithmetic precision must be maintained,
 * until the output stage, where values are rounded (up).
 *
 * The code below tries to be efficient by caching the input
 * rows in registers, and running the filter on 3x3 chunks.
 * Multiple columns can be computed in parallel by using 
 * two 16-bit adds in a 32-bit register, or four 16-bit adds
 * in a 64-bit register.
 */
void P64Decoder::filter(u_char* in, u_char* out, u_int stride)
{
	/* Corner pixel has filter coef 1 */
	u_int s = in[0];
	u_int o = 0;
	SPLICE(o, s, 24);

	u_int r00 = s << 24 | in[1] << 16 | in[2] << 8 | in[3];
	u_int r01 = in[4] << 24 | in[5] << 16 | in[6] << 8 | in[7];
	in += stride;

	/*
	 * First row.
	 */
	s += (r00 >> 15) & 0x1fe;
	s += (r00 >> 8) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 16);

	s = (r00 >> 16) & 0xff;
	s += (r00 >> 7) & 0x1fe;
	s += r00 & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 8);
	
	s = (r00 >> 8) & 0xff;
	s += (r00 & 0xff) << 1;
	s += r01 >> 24;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 0);
	*(u_int*)out = o;
	
	s = r00 & 0xff;
	s += (r01 >> 23) & 0x1fe;
	s += (r01 >> 16) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	o = 0;
	SPLICE(o, s, 24);
	
	s = r01 >> 24;
	s += (r01 >> 15) & 0x1fe;
	s += (r01 >> 8) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 16);
	
	s = (r01 >> 16) & 0xff;
	s += (r01 >> 7) & 0x1fe;
	s += r01 & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 8);

	/* corner has filter coef 1 */
	s = r01 & 0xff;
	SPLICE(o, s, 0);
	*(u_int*)(out + 4) = o;
	out += stride;

	/* load next rows into cache */
	u_int r10 = in[0] << 24 | in[1] << 16 | in[2] << 8 | in[3];
	u_int r11 = in[4] << 24 | in[5] << 16 | in[6] << 8 | in[7];
	in += stride;

	u_int r20, r21;
	u_int mask = 0xff00ff;
	for (int k = 6; --k >= 0; ) {
		/* load next row */
		r20 = in[0] << 24 | in[1] << 16 | in[2] << 8 | in[3];
		r21 = in[4] << 24 | in[5] << 16 | in[6] << 8 | in[7];
		in += stride;

		/* columns 0,2 */
		u_int v = (r00 >> 8) & mask;
		v += ((r10 >> 8) & mask) << 1;
		v += (r20 >> 8) & mask;

		/* first pixel */
		s = v >> 16;
		/* round */
		s += 2;
		s >>= 2;
		o = 0;
		SPLICE(o, s, 24);
		
		/* columns 1,3 */
		u_int w = r00 & mask;
		w += (r10 & mask) << 1;
		w += r20 & mask;

		/* row */
		s = v >> 16;
		s += v & 0xffff;
		s += w >> (16-1);
		/* round */
		s += 8;
		s >>= 4;
		SPLICE(o, s, 16);

		s = w >> 16;
		s += w & 0xffff;
		s += (v & 0xffff) << 1;
		/* round */
		s += 8;
		s >>= 4;
		SPLICE(o, s, 8);

		/* start next row */
		s = v & 0xffff;
		s += (w & 0xffff) << 1;
		/* but first do columns 4,6 */
		v = (r01 >> 8) & mask;
		v += ((r11 >> 8) & mask) << 1;
		v += (r21 >> 8) & mask;
		/* finish row */
		s += v >> 16;
		/* round */
		s += 8;
		s >>= 4;
		SPLICE(o, s, 0);
		*(u_int*)out = o;

		/* start next row */
		s = w & 0xffff;
		s += (v >> 16) << 1;
		/* but first do columns 5,7 */
		w = r01 & mask;
		w += (r11 & mask) << 1;
		w += r21 & mask;
		/* finish row */
		s += w >> 16;
		/* round */
		s += 8;
		s >>= 4;
		o = 0;
		SPLICE(o, s, 24);

		s = v >> 16;
		s += v & 0xffff;
		s += w >> (16-1);
		/* round */
		s += 8;
		s >>= 4;
		SPLICE(o, s, 16);

		s = w >> 16;
		s += w & 0xffff;
		s += (v & 0xffff) << 1;
		/* round */
		s += 8;
		s >>= 4;
		SPLICE(o, s, 8);

		s = w & 0xffff;
		/* round */
		s += 2;
		s >>= 2;
		SPLICE(o, s, 0);
		*(u_int*)(out + 4) = o;

		out += stride;

		/* roll lines up cache */
		r00 = r10;
		r01 = r11;
		r10 = r20;
		r11 = r21;
	}
	/*
	 * last row
	 */
	s = r20 >> 24;
	o = 0;
	SPLICE(o, s, 24);

	s += (r20 >> 15) & 0x1fe;
	s += (r20 >> 8) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 16);

	s = (r20 >> 16) & 0xff;
	s += (r20 >> 7) & 0x1fe;
	s += r20 & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 8);
	
	s = (r20 >> 8) & 0xff;
	s += (r20 & 0xff) << 1;
	s += r21 >> 24;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 0);
	*(u_int*)out = o;
	
	s = r20 & 0xff;
	s += (r21 >> 23) & 0x1fe;
	s += (r21 >> 16) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	o = 0;
	SPLICE(o, s, 24);
	
	s = r21 >> 24;
	s += (r21 >> 15) & 0x1fe;
	s += (r21 >> 8) & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 16);
	
	s = (r21 >> 16) & 0xff;
	s += (r21 >> 7) & 0x1fe;
	s += r21 & 0xff;
	/* round */
	s += 2;
	s >>= 2;
	SPLICE(o, s, 8);

	/* corner has filter coef 1 */
	s = r21 & 0xff;
	SPLICE(o, s, 0);
	*(u_int*)(out + 4) = o;
}

void P64Decoder::mvblka(u_char* in, u_char* out, u_int stride)
{
#ifdef INT_64
	*(INT_64*)out = *(INT_64*)in;
	out += stride; in += stride;
	*(INT_64*)out = *(INT_64*)in;
	out += stride; in += stride;
	*(INT_64*)out = *(INT_64*)in;
	out += stride; in += stride;
	*(INT_64*)out = *(INT_64*)in;
	out += stride; in += stride;
	*(INT_64*)out = *(INT_64*)in;
	out += stride; in += stride;
	*(INT_64*)out = *(INT_64*)in;