mbtransquant.c

这是一个压缩解压包,用C语言进行编程的,里面有详细的源代码.

		limit *= 3;

	/* Quantize the block */
	cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);

	/* DeQuantize the block */
	MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);

	/* Perform inverse DCT*/
	MBiDCT(data, cbp);

 	/* Transfer back the data -- Add the data */
	MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);

	return(cbp);
}

uint8_t
MBTransQuantInterBVOP(const MBParam * pParam,
					  FRAMEINFO * frame,
					  MACROBLOCK * pMB,
					  const uint32_t x_pos,
					  const uint32_t y_pos,
					  int16_t data[6 * 64],
					  int16_t qcoeff[6 * 64])
{
	uint8_t cbp;
	uint32_t limit;

	/* There is no MBTrans8to16 for Inter block, that's done in motion compensation
	 * already */

	/* Perform DCT (and field decision) */
	MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);

	/* Set the limit threshold */
	limit = BVOP_TOOSMALL_LIMIT;

	if (frame->vop_flags & XVID_VOP_CARTOON)
		limit *= 2;

	/* Quantize the block */
	cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);

	/*
	 * History comment:
	 * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
	 *
	 * BUT some plugins require the rebuilt original frame to be passed so we
	 * have to take care of that here
	 */
	if((pParam->plugin_flags & XVID_REQORIGINAL)) {

		/* DeQuantize the block */
		MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);

		/* Perform inverse DCT*/
		MBiDCT(data, cbp);

		/* Transfer back the data -- Add the data */
		MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
	}

	return(cbp);
}

/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */
uint32_t
MBFieldTest_c(int16_t data[6 * 64])
{
	const uint8_t blocks[] =
		{ 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 };
	const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 };

	int frame = 0, field = 0;
	int i, j;

	for (i = 0; i < 7; ++i) {
		for (j = 0; j < 8; ++j) {
			frame +=
				abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]);
			frame +=
				abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]);
			frame +=
				abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]);
			frame +=
				abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]);

			field +=
				abs(data[blocks[i + 1] + lines[i + 1] + j] -
					data[blocks[i] + lines[i] + j]);
			field +=
				abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] -
					data[blocks[i] + lines[i] + 8 + j]);
			field +=
				abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] -
					data[blocks[i] + 64 + lines[i] + j]);
			field +=
				abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] -
					data[blocks[i] + 64 + lines[i] + 8 + j]);
		}
	}

	return (frame >= (field + 350));
}


/* deinterlace Y blocks vertically */

#define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp))
#define LINE(X,Y)	&data[X*64 + Y*8]

void
MBFrameToField(int16_t data[6 * 64])
{
	int16_t tmp[8];

	/* left blocks */

	/* 1=2, 2=4, 4=8, 8=1 */
	MOVLINE(tmp, LINE(0, 1));
	MOVLINE(LINE(0, 1), LINE(0, 2));
	MOVLINE(LINE(0, 2), LINE(0, 4));
	MOVLINE(LINE(0, 4), LINE(2, 0));
	MOVLINE(LINE(2, 0), tmp);

	/* 3=6, 6=12, 12=9, 9=3 */
	MOVLINE(tmp, LINE(0, 3));
	MOVLINE(LINE(0, 3), LINE(0, 6));
	MOVLINE(LINE(0, 6), LINE(2, 4));
	MOVLINE(LINE(2, 4), LINE(2, 1));
	MOVLINE(LINE(2, 1), tmp);

	/* 5=10, 10=5 */
	MOVLINE(tmp, LINE(0, 5));
	MOVLINE(LINE(0, 5), LINE(2, 2));
	MOVLINE(LINE(2, 2), tmp);

	/* 7=14, 14=13, 13=11, 11=7 */
	MOVLINE(tmp, LINE(0, 7));
	MOVLINE(LINE(0, 7), LINE(2, 6));
	MOVLINE(LINE(2, 6), LINE(2, 5));
	MOVLINE(LINE(2, 5), LINE(2, 3));
	MOVLINE(LINE(2, 3), tmp);

	/* right blocks */

	/* 1=2, 2=4, 4=8, 8=1 */
	MOVLINE(tmp, LINE(1, 1));
	MOVLINE(LINE(1, 1), LINE(1, 2));
	MOVLINE(LINE(1, 2), LINE(1, 4));
	MOVLINE(LINE(1, 4), LINE(3, 0));
	MOVLINE(LINE(3, 0), tmp);

	/* 3=6, 6=12, 12=9, 9=3 */
	MOVLINE(tmp, LINE(1, 3));
	MOVLINE(LINE(1, 3), LINE(1, 6));
	MOVLINE(LINE(1, 6), LINE(3, 4));
	MOVLINE(LINE(3, 4), LINE(3, 1));
	MOVLINE(LINE(3, 1), tmp);

	/* 5=10, 10=5 */
	MOVLINE(tmp, LINE(1, 5));
	MOVLINE(LINE(1, 5), LINE(3, 2));
	MOVLINE(LINE(3, 2), tmp);

	/* 7=14, 14=13, 13=11, 11=7 */
	MOVLINE(tmp, LINE(1, 7));
	MOVLINE(LINE(1, 7), LINE(3, 6));
	MOVLINE(LINE(3, 6), LINE(3, 5));
	MOVLINE(LINE(3, 5), LINE(3, 3));
	MOVLINE(LINE(3, 3), tmp);
}

/*****************************************************************************
 *               Trellis based R-D optimal quantization
 *
 *   Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
 *
 ****************************************************************************/

/*----------------------------------------------------------------------------
 *
 *        Trellis-Based quantization
 *
 * So far I understand this paper:
 *
 *  "Trellis-Based R-D Optimal Quantization in H.263+"
 *    J.Wen, M.Luttrell, J.Villasenor
 *    IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
 *
 * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
 * Source Shortest Path algo. But due to the underlying graph structure
 * ("Trellis"), it can be turned into a dynamic programming algo,
 * partially saving the explicit graph's nodes representation. And
 * without using a heap, since the open frontier of the DAG is always
 * known, and of fixed size.
 *--------------------------------------------------------------------------*/



/* Codes lengths for relevant levels. */

/* let's factorize: */
static const uint8_t Code_Len0[64] = {
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len1[64] = {
	20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len2[64] = {
	19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len3[64] = {
	18,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len4[64] = {
	17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len5[64] = {
	16,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len6[64] = {
	15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len7[64] = {
	13,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len8[64] = {
	11,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len9[64] = {
	12,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len10[64] = {
	12,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len11[64] = {
	12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len12[64] = {
	11,17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len13[64] = {
	11,15,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len14[64] = {
	10,12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len15[64] = {
	10,13,17,19,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len16[64] = {
	9,12,13,18,18,19,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
static const uint8_t Code_Len17[64] = {
	8,11,13,14,14,14,15,19,19,19,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len18[64] = {
	7, 9,11,11,13,13,13,15,15,15,16,22,22,22,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len19[64] = {
	5, 7, 9,10,10,11,11,11,11,11,13,14,16,17,17,18,18,18,18,18,18,18,18,20,20,21,21,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
static const uint8_t Code_Len20[64] = {
	3, 4, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9,10,10,10,10,10,10,10,10,12,12,13,13,12,13,14,15,15,
	15,16,16,16,16,17,17,17,18,18,19,19,19,19,19,19,19,19,21,21,22,22,30,30,30,30,30,30,30,30,30,30 };

/* a few more table for LAST table: */
static const uint8_t Code_Len21[64] = {
	13,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
static const uint8_t Code_Len22[64] = {
	12,15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
	30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
static const uint8_t Code_Len23[64] = {
	10,12,15,15,15,16,16,16,16,17,17,17,17,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,20,20,20,
	20,21,21,21,21,21,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
static const uint8_t Code_Len24[64] = {
	5, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,10,10,10,10,10,10,10,10,11,11,11,11,12,12,12,
	12,13,13,13,13,13,13,13,13,14,16,16,16,16,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,19,19};


static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */
	Code_Len20,Code_Len19,Code_Len18,Code_Len17,
	Code_Len16,Code_Len15,Code_Len14,Code_Len13,
	Code_Len12,Code_Len11,Code_Len10,Code_Len9,
	Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5,
	Code_Len4, Code_Len3, Code_Len3 ,Code_Len2,
	Code_Len2, Code_Len1, Code_Len1, Code_Len1,
};

static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */
	Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
};

/* TL_SHIFT controls the precision of the RD optimizations in trellis
 * valid range is [10..16]. The bigger, the more trellis is vulnerable
 * to overflows in cost formulas.
 *  - 10 allows ac values up to 2^11 == 2048
 *  - 16 allows ac values up to 2^8 == 256
 */
#define TL_SHIFT 11
#define TL(q) ((0xfe00>>(16-TL_SHIFT))/(q*q))

static const int Trellis_Lambda_Tabs[31] = {
	TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7),
	TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15),
	TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23),
	TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31)
};
#undef TL

static int __inline
Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
{
	while(i>=0)
		if (C[Zigzag[i]])
			return i;
		else i--;
	return -1;
}

/* this routine has been strippen of all debug code */
static int
dct_quantize_trellis_c(int16_t *const Out,
					   const int16_t *const In,
					   int Q,
					   const uint16_t * const Zigzag,
					   const uint16_t * const QuantMatrix,
					   int Non_Zero,
					   int Sum)
{

	/* Note: We should search last non-zero coeffs on *real* DCT input coeffs
	 * (In[]), not quantized one (Out[]). However, it only improves the result
	 * *very* slightly (~0.01dB), whereas speed drops to crawling level :)
	 * Well, actually, taking 1 more coeff past Non_Zero into account sometimes
	 * helps. */
	typedef struct { int16_t Run, Level; } NODE;

	NODE Nodes[65], Last = { 0, 0};
	uint32_t Run_Costs0[64+1];
	uint32_t * const Run_Costs = Run_Costs0 + 1;

	/* it's 1/lambda, actually */
	const int Lambda = Trellis_Lambda_Tabs[Q-1];

	int Run_Start = -1;
	uint32_t Min_Cost = 2<<TL_SHIFT;

	int Last_Node = -1;
	uint32_t Last_Cost = 0;

	int i, j;

	/* source (w/ CBP penalty) */
	Run_Costs[-1] = 2<<TL_SHIFT;

	Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
	if (Non_Zero<0)
		return 0; /* Sum is zero if there are only zero coeffs */

	for(i=0; i<=Non_Zero; i++) {
		const int q = ((Q*QuantMatrix[Zigzag[i]])>>4);
		const int Mult = 2*q;
		const int Bias = (q-1) | 1;
		const int Lev0 = Mult + Bias;

		const int AC = In[Zigzag[i]];
		const int Level1 = Out[Zigzag[i]];
		const unsigned int Dist0 = Lambda* AC*AC;
		uint32_t Best_Cost = 0xf0000000;
		Last_Cost += Dist0;

		/* very specialized loop for -1,0,+1 */
		if ((uint32_t)(Level1+1)<3) {
			int dQ;
			int Run;
			uint32_t Cost0;

			if (AC<0) {
				Nodes[i].Level = -1;
				dQ = Lev0 + AC;
			} else {
				Nodes[i].Level = 1;
				dQ = Lev0 - AC;
			}
			Cost0 = Lambda*dQ*dQ;

			Nodes[i].Run = 1;
			Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0;
			for(Run=i-Run_Start; Run>0; --Run) {
				const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
				const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT);
				const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT);

				/* TODO: what about tie-breaks? Should we favor short runs or
				 * long runs? Although the error is the same, it would not be
				 * spread the same way along high and low frequencies... */

				/* Gruel: I'd say, favour short runs => hifreq errors (HVS) */

				if (Cost<Best_Cost) {
					Best_Cost	 = Cost;
					Nodes[i].Run = Run;
				}

				if (lCost<Last_Cost) {
					Last_Cost  = lCost;
					Last.Run   = Run;
					Last_Node  = i;
				}
			}
			if (Last_Node==i)
				Last.Level = Nodes[i].Level;
		} else if (51U>(uint32_t)(Level1+25)) {
			/* "big" levels (not less than ESC3, though) */
			const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
			int Level2;
			int dQ1, dQ2;
			int Run;
			uint32_t Dist1,Dist2;
			int dDist21;

			if (Level1>1) {
				dQ1 = Level1*Mult-AC + Bias;
				dQ2 = dQ1 - Mult;
				Level2 = Level1-1;
				Tbl_L1		= (Level1<=24) ? B16_17_Code_Len[Level1-1]	   : Code_Len0;
				Tbl_L2		= (Level2<=24) ? B16_17_Code_Len[Level2-1]	   : Code_Len0;
				Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
				Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
			} else { /* Level1<-1 */
				dQ1 = Level1*Mult-AC - Bias;