/*
 * Copyright 1992 by Jutta Degener and Carsten Bormann, Technische
 * Universitaet Berlin.  See the accompanying file "COPYRIGHT" for
 * details.  THERE IS ABSOLUTELY NO WARRANTY FOR THIS SOFTWARE.
 */

/* $Header$ */

#include <stdio.h>
#include <assert.h>

#include "private.h"

#include "gsm.h"
#include "proto.h"
#ifdef K6OPT
#include "k6opt.h"

#define Short_term_analysis_filtering Short_term_analysis_filteringx

#endif
/*
 *  SHORT TERM ANALYSIS FILTERING SECTION
 */

/* 4.2.8 */

static void Decoding_of_the_coded_Log_Area_Ratios P2((LARc,LARpp),
	word 	* LARc,		/* coded log area ratio	[0..7] 	IN	*/
	word	* LARpp)	/* out: decoded ..			*/
{
	register word	temp1 /* , temp2 */;

	/*  This procedure requires for efficient implementation
	 *  two tables.
 	 *
	 *  INVA[1..8] = integer( (32768 * 8) / real_A[1..8])
	 *  MIC[1..8]  = minimum value of the LARc[1..8]
	 */

	/*  Compute the LARpp[1..8]
	 */

	/* 	for (i = 1; i <= 8; i++, B++, MIC++, INVA++, LARc++, LARpp++) {
	 *
	 *		temp1  = GSM_ADD( *LARc, *MIC ) << 10;
	 *		temp2  = *B << 1;
	 *		temp1  = GSM_SUB( temp1, temp2 );
	 *
	 *		assert(*INVA != MIN_WORD);
	 *
	 *		temp1  = GSM_MULT_R( *INVA, temp1 );
	 *		*LARpp = GSM_ADD( temp1, temp1 );
	 *	}
	 */

#undef	STEP
#define	STEP( B, MIC, INVA )	\
		temp1    = GSM_ADD( *LARc++, MIC ) << 10;	\
		temp1    = GSM_SUB( temp1, B << 1 );		\
		temp1    = (word)GSM_MULT_R( INVA, temp1 );		\
		*LARpp++ = GSM_ADD( temp1, temp1 );

	STEP(      0,  -32,  13107 );
	STEP(      0,  -32,  13107 );
	STEP(   2048,  -16,  13107 );
	STEP(  -2560,  -16,  13107 );

	STEP(     94,   -8,  19223 );
	STEP(  -1792,   -8,  17476 );
	STEP(   -341,   -4,  31454 );
	STEP(  -1144,   -4,  29708 );

	/* NOTE: the addition of *MIC is used to restore
	 * 	 the sign of *LARc.
	 */
}

/* 4.2.9 */
/* Computation of the quantized reflection coefficients 
 */

/* 4.2.9.1  Interpolation of the LARpp[1..8] to get the LARp[1..8]
 */

/*
 *  Within each frame of 160 analyzed speech samples the short term
 *  analysis and synthesis filters operate with four different sets of
 *  coefficients, derived from the previous set of decoded LARs(LARpp(j-1))
 *  and the actual set of decoded LARs (LARpp(j))
 *
 * (Initial value: LARpp(j-1)[1..8] = 0.)
 */

static void Coefficients_0_12 P3((LARpp_j_1, LARpp_j, LARp),
	register word * LARpp_j_1,
	register word * LARpp_j,
	register word * LARp)
{
	register int 	i;

	for (i = 1; i <= 8; i++, LARp++, LARpp_j_1++, LARpp_j++) {
		*LARp = GSM_ADD( SASR( *LARpp_j_1, 2 ), SASR( *LARpp_j, 2 ));
		*LARp = GSM_ADD( *LARp,  SASR( *LARpp_j_1, 1));
	}
}

static void Coefficients_13_26 P3((LARpp_j_1, LARpp_j, LARp),
	register word * LARpp_j_1,
	register word * LARpp_j,
	register word * LARp)
{
	register int i;
	for (i = 1; i <= 8; i++, LARpp_j_1++, LARpp_j++, LARp++) {
		*LARp = GSM_ADD( SASR( *LARpp_j_1, 1), SASR( *LARpp_j, 1 ));
	}
}

static void Coefficients_27_39 P3((LARpp_j_1, LARpp_j, LARp),
	register word * LARpp_j_1,
	register word * LARpp_j,
	register word * LARp)
{
	register int i;

	for (i = 1; i <= 8; i++, LARpp_j_1++, LARpp_j++, LARp++) {
		*LARp = GSM_ADD( SASR( *LARpp_j_1, 2 ), SASR( *LARpp_j, 2 ));
		*LARp = GSM_ADD( *LARp, SASR( *LARpp_j, 1 ));
	}
}


static void Coefficients_40_159 P2((LARpp_j, LARp),
	register word * LARpp_j,
	register word * LARp)
{
	register int i;

	for (i = 1; i <= 8; i++, LARp++, LARpp_j++)
		*LARp = *LARpp_j;
}

/* 4.2.9.2 */

static void LARp_to_rp P1((LARp),
	register word * LARp)	/* [0..7] IN/OUT  */
/*
 *  The input of this procedure is the interpolated LARp[0..7] array.
 *  The reflection coefficients, rp[i], are used in the analysis
 *  filter and in the synthesis filter.
 */
{
	register int 		i;
	register word		temp;

	for (i = 1; i <= 8; i++, LARp++) {

		/* temp = GSM_ABS( *LARp );
	         *
		 * if (temp < 11059) temp <<= 1;
		 * else if (temp < 20070) temp += 11059;
		 * else temp = GSM_ADD( temp >> 2, 26112 );
		 *
		 * *LARp = *LARp < 0 ? -temp : temp;
		 */

		if (*LARp < 0) {
			temp = *LARp == MIN_WORD ? MAX_WORD : -(*LARp);
			*LARp = - ((temp < 11059) ? temp << 1
				: ((temp < 20070) ? temp + 11059
				:  GSM_ADD( temp >> 2, 26112 )));
		} else {
			temp  = *LARp;
			*LARp =    (temp < 11059) ? temp << 1
				: ((temp < 20070) ? temp + 11059
				:  GSM_ADD( temp >> 2, 26112 ));
		}
	}
}


/* 4.2.10 */
#ifndef Short_term_analysis_filtering

/* SJB Remark:
 * I tried 2 MMX versions of this function, neither is significantly
 * faster than the C version which follows.  MMX might be useful if
 * one were processing 2 input streams in parallel.
 */
static void Short_term_analysis_filtering P4((u0,rp0,k_n,s),
	register word * u0,
	register word	* rp0,	/* [0..7]	IN	*/
	register int 	k_n, 	/*   k_end - k_start	*/
	register word	* s	/* [0..n-1]	IN/OUT	*/
)
/*
 *  This procedure computes the short term residual signal d[..] to be fed
 *  to the RPE-LTP loop from the s[..] signal and from the local rp[..]
 *  array (quantized reflection coefficients).  As the call of this
 *  procedure can be done in many ways (see the interpolation of the LAR
 *  coefficient), it is assumed that the computation begins with index
 *  k_start (for arrays d[..] and s[..]) and stops with index k_end
 *  (k_start and k_end are defined in 4.2.9.1).  This procedure also
 *  needs to keep the array u0[0..7] in memory for each call.
 */
{
	register word		* u_top = u0 + 8;
	register word		* s_top = s + k_n;

	while (s < s_top) {
		register word		*u, *rp ;
		register longword		di, u_out;
		di = u_out = *s;
		for (rp=rp0, u=u0; u<u_top;) {
			register longword	ui, rpi;
			ui    = *u;
			*u++  = (word)u_out;
			rpi   = *rp++;
			u_out = ui + (((rpi*di)+0x4000)>>15);
			di    = di + (((rpi*ui)+0x4000)>>15);
			/* make the common case fastest: */
			if ((u_out == (word)u_out) && (di == (word)di)) continue;
			/* otherwise do slower fixup (saturation) */
			if (u_out>MAX_WORD) u_out=MAX_WORD;
			else if (u_out<MIN_WORD) u_out=MIN_WORD;
			if (di>MAX_WORD) di=MAX_WORD;
			else if (di<MIN_WORD) di=MIN_WORD;
		}
		*s++ = (word)di;
	}
}
#endif

#if defined(USE_FLOAT_MUL) && defined(FAST)

static void Fast_Short_term_analysis_filtering P4((u,rp,k_n,s),
	register word * u;
	register word	* rp,	/* [0..7]	IN	*/
	register int 	k_n, 	/*   k_end - k_start	*/
	register word	* s	/* [0..n-1]	IN/OUT	*/
)
{
	register int		i;

	float 	  uf[8],
		 rpf[8];

	register float scalef = 3.0517578125e-5;
	register float		sav, di, temp;

	for (i = 0; i < 8; ++i) {
		uf[i]  = u[i];
		rpf[i] = rp[i] * scalef;
	}
	for (; k_n--; s++) {
		sav = di = *s;
		for (i = 0; i < 8; ++i) {
			register float rpfi = rpf[i];
			register float ufi  = uf[i];

			uf[i] = sav;
			temp  = rpfi * di + ufi;
			di   += rpfi * ufi;
			sav   = temp;
		}
		*s = di;
	}
	for (i = 0; i < 8; ++i) u[i] = uf[i];
}
#endif /* ! (defined (USE_FLOAT_MUL) && defined (FAST)) */

/*
 * SJB Remark: modified Short_term_synthesis_filtering() below
 *  for significant (abt 35%) speedup of decompression.
 *    (gcc-2.95, k6 cpu)
 *  Please don't change this without benchmarking decompression
 *  to see that you haven't harmed speed.
 *  This function burns most of CPU time for untoasting.
 *  Unfortunately, didn't see any good way to benefit from mmx.
 */
static void Short_term_synthesis_filtering P5((S,rrp,k,wt,sr),
	struct gsm_state * S,
	register word	* rrp,	/* [0..7]	IN	*/
	register int	k,	/* k_end - k_start	*/
	register word	* wt,	/* [0..k-1]	IN	*/
	register word	* sr	/* [0..k-1]	OUT	*/
)
{
	register word		* v = S->v;
	register int		i;
	register longword		sri;

	while (k--) {
		sri = *wt++;
		for (i = 8; i--;) {
			register longword		tmp1, tmp2;

			/* sri = GSM_SUB( sri, gsm_mult_r( rrp[i], v[i] ) );
			 */
			tmp1 = rrp[i];
			tmp2 = v[i];

			tmp2 = (( tmp1 * tmp2 + 16384) >> 15) ;
			/* saturation done below */
			sri  -= tmp2;
			if (sri != (word)sri) {
				sri = (sri<0)? MIN_WORD:MAX_WORD;
			}
			/* v[i+1] = GSM_ADD( v[i], gsm_mult_r( rrp[i], sri ) );
			 */

			tmp1 = (( tmp1 * sri + 16384) >> 15) ;
			/* saturation done below */
			tmp1 += v[i];
			if (tmp1 != (word)tmp1) {
				tmp1 = (tmp1<0)? MIN_WORD:MAX_WORD;
			}
			v[i+1] = (word)tmp1;
		}
		*sr++ = v[0] = (word)sri;
	}
}


#if defined(FAST) && defined(USE_FLOAT_MUL)

static void Fast_Short_term_synthesis_filtering P5((S,rrp,k,wt,sr),
	struct gsm_state * S,
	register word	* rrp,	/* [0..7]	IN	*/
	register int	k,	/* k_end - k_start	*/
	register word	* wt,	/* [0..k-1]	IN	*/
	register word	* sr	/* [0..k-1]	OUT	*/
)
{
	register word		* v = S->v;
	register int		i;

	float va[9], rrpa[8];
	register float scalef = 3.0517578125e-5, temp;

	for (i = 0; i < 8; ++i) {
		va[i]   = v[i];
		rrpa[i] = (float)rrp[i] * scalef;
	}
	while (k--) {
		register float sri = *wt++;
		for (i = 8; i--;) {
			sri -= rrpa[i] * va[i];
			if     (sri < -32768.) sri = -32768.;
			else if (sri > 32767.) sri =  32767.;

			temp = va[i] + rrpa[i] * sri;
			if     (temp < -32768.) temp = -32768.;
			else if (temp > 32767.) temp =  32767.;
			va[i+1] = temp;
		}
		*sr++ = va[0] = sri;
	}
	for (i = 0; i < 9; ++i) v[i] = va[i];
}

#endif /* defined(FAST) && defined(USE_FLOAT_MUL) */

void Gsm_Short_Term_Analysis_Filter P3((S,LARc,s),

	struct gsm_state * S,

	word	* LARc,		/* coded log area ratio [0..7]  IN	*/
	word	* s		/* signal [0..159]		IN/OUT	*/
)
{
	word		* LARpp_j	= S->LARpp[ S->j      ];
	word		* LARpp_j_1	= S->LARpp[ S->j ^= 1 ];

	word		LARp[8];

#undef	FILTER
#if 	defined(FAST) && defined(USE_FLOAT_MUL)
# 	define	FILTER 	(* (S->fast			\
			   ? Fast_Short_term_analysis_filtering	\
		    	   : Short_term_analysis_filtering	))

#else
# 	define	FILTER	Short_term_analysis_filtering
#endif

	Decoding_of_the_coded_Log_Area_Ratios( LARc, LARpp_j );

	Coefficients_0_12(  LARpp_j_1, LARpp_j, LARp );
	LARp_to_rp( LARp );
	FILTER( S->u, LARp, 13, s);

	Coefficients_13_26( LARpp_j_1, LARpp_j, LARp);
	LARp_to_rp( LARp );
	FILTER( S->u, LARp, 14, s + 13);

	Coefficients_27_39( LARpp_j_1, LARpp_j, LARp);
	LARp_to_rp( LARp );
	FILTER( S->u, LARp, 13, s + 27);

	Coefficients_40_159( LARpp_j, LARp);
	LARp_to_rp( LARp );
	FILTER( S->u, LARp, 120, s + 40);
	
}

void Gsm_Short_Term_Synthesis_Filter P4((S, LARcr, wt, s),
	struct gsm_state * S,

	word	* LARcr,	/* received log area ratios [0..7] IN  */
	word	* wt,		/* received d [0..159]		   IN  */

	word	* s		/* signal   s [0..159]		  OUT  */
)
{
	word		* LARpp_j	= S->LARpp[ S->j     ];
	word		* LARpp_j_1	= S->LARpp[ S->j ^=1 ];

	word		LARp[8];

#undef	FILTER
#if 	defined(FAST) && defined(USE_FLOAT_MUL)

# 	define	FILTER 	(* (S->fast			\
			   ? Fast_Short_term_synthesis_filtering	\
		    	   : Short_term_synthesis_filtering	))
#else
#	define	FILTER	Short_term_synthesis_filtering
#endif

	Decoding_of_the_coded_Log_Area_Ratios( LARcr, LARpp_j );

	Coefficients_0_12( LARpp_j_1, LARpp_j, LARp );
	LARp_to_rp( LARp );
	FILTER( S, LARp, 13, wt, s );

	Coefficients_13_26( LARpp_j_1, LARpp_j, LARp);
	LARp_to_rp( LARp );
	FILTER( S, LARp, 14, wt + 13, s + 13 );

	Coefficients_27_39( LARpp_j_1, LARpp_j, LARp);
	LARp_to_rp( LARp );
	FILTER( S, LARp, 13, wt + 27, s + 27 );

	Coefficients_40_159( LARpp_j, LARp );
	LARp_to_rp( LARp );
	FILTER(S, LARp, 120, wt + 40, s + 40);
}