HARMONIZE_DCT (2ac14933) · Commits · SA4 / Audio / IVAS BASOP

lib_com/basop_util.c

+42 −0

Original line number	Diff line number	Diff line
		@@ -758,9 +758,51 @@ Word16 getScaleFactor32( /* o: measured headroom in range [

		i = s_and( s_min( i_max, i_min ), 0x1F );

		return i;
		}

		#ifdef OPTIMIZE_FFT_STACK
		/* o: measured headroom in range [0..31], 0 if all x[i] == 0 */
		Word16 getScaleFactor32_cmplx(
		cmplx x, / i: array containing 32-bit data */
		const Word16 len_x /* i: length of the array to scan */
		)
		{
		Word16 i, i_min, i_max;
		Word32 x_min, x_max;

		x_max = 0;
		move32();
		x_min = 0;
		move32();
		FOR( i = 0; i < len_x; i++ )
		{
		if ( x[i].re >= 0 )
		x_max = L_max( x_max, x[i].re );
		if ( x[i].re < 0 )
		x_min = L_min( x_min, x[i].re );
		if ( x[i].im >= 0 )
		x_max = L_max( x_max, x[i].im );
		if ( x[i].im < 0 )
		x_min = L_min( x_min, x[i].im );
		}

		i_max = 0x20;
		move16();
		i_min = 0x20;
		move16();

		if ( x_max != 0 )
		i_max = norm_l( x_max );

		if ( x_min != 0 )
		i_min = norm_l( x_min );

		i = s_and( s_min( i_max, i_min ), 0x1F );

		return i;
		}
		#endif

		Word16 getScaleFactor32_copy( /* o: measured headroom in range [0..31], 0 if all x[i] == 0 */
		const Word32 x, / i: array containing 32-bit data */

lib_com/basop_util.h

+8 −0

Original line number	Diff line number	Diff line
		@@ -235,6 +235,14 @@ Word16 getScaleFactor32(
		const Word32 x, / i : array containing 32-bit data */
		const Word16 len_x ); /* i : length of the array to scan */

		#ifdef OPTIMIZE_FFT_STACK
		/* o: measured headroom in range [0..31], 0 if all x[i] == 0 */
		Word16 getScaleFactor32_cmplx(
		cmplx x, / i: array containing 32-bit data */
		const Word16 len_x /* i: length of the array to scan */
		);
		#endif

		Word16 getScaleFactor32_copy( /* o: measured headroom in range [0..31], 0 if all x[i] == 0 */
		const Word32 x, / i: array containing 32-bit data */
		const Word32 len_x ); /* i: length of the array to scan */

lib_com/edct_fx.c

+160 −22

Original line number	Diff line number	Diff line
		@@ -207,14 +207,23 @@ void edct_fx(
		const Word32 x, / i : input signal Qq */
		Word32 y, / o : output transform Qq */
		Word16 length, /* i : length Q0*/
		#ifdef HARMONIZE_DCT
		Word16 q, / i : Q value of input signal */
		const Word16 element_mode /* i : element mode */
		#else
		Word16 q / i : Q value of input signal */
		#endif
		)
		{
		Word16 i;
		Word32 re;
		Word32 im;
		const Word16 edct_table; /Q16 */
		#ifdef OPTIMIZE_FFT_STACK
		cmplx spec[L_FRAME_PLUS / 2];
		#else
		Word32 complex_buf[2 * ( L_FRAME48k / 2 + 240 )];
		#endif
		Word32 L_tmp;
		Word16 tmp;
		Word16 len1;
		@@ -225,36 +234,93 @@ void edct_fx(
		FOR( i = 0; i < len1; i++ )
		{
		L_tmp = Mult_32_16( x[2 * i], edct_table[i] ); /Q(q+1) /
		#ifdef OPTIMIZE_FFT_STACK
		spec[i].re = Madd_32_16( L_tmp, x[( length - ( 1 + ( i * 2 ) ) )], edct_table[( len1 - ( 1 + i ) )] ); /Q(q+1) /
		#else
		complex_buf[2 * i] = Madd_32_16( L_tmp, x[( length - ( 1 + ( i * 2 ) ) )], edct_table[( len1 - ( 1 + i ) )] ); /Q(q+1) /
		#endif
		move32();

		L_tmp = Mult_32_16( x[( length - ( 1 + ( i * 2 ) ) )], edct_table[i] ); /Q(q+1) /

		#ifdef OPTIMIZE_FFT_STACK
		spec[i].im = Msub_32_16( L_tmp, x[( i * 2 )], edct_table[( len1 - ( 1 + i ) )] ); /Q(q+1) /
		#else
		complex_buf[( ( i * 2 ) + 1 )] = Msub_32_16( L_tmp, x[( i * 2 )], edct_table[( len1 - ( 1 + i ) )] ); /Q(q+1) /
		#endif
		move32();
		}

		#ifdef HARMONIZE_DCT
		IF( element_mode == EVS_MONO )
		{
		#endif
		q = sub( 15, q );
		move16();
		#ifdef OPTIMIZE_FFT_STACK
		BASOP_cfft( spec, len1, q, y );
		#else
		BASOP_cfft( (cmplx *) complex_buf, len1, q, y );
		#endif

		tmp = div_s( 1, length ); /Q15 /
		tmp = round_fx( L_shl( L_mult( tmp, 19302 /0.75f EVS_PI in Q13/ ), 2 ) ); /Q15 */
		#ifdef HARMONIZE_DCT
		}
		ELSE
		{
		q = sub( 31, q );
		move16();
		tmp = sub( getScaleFactor32_cmplx( spec, len1 ), find_guarded_bits_fx( len1 ) );
		scale_sig32_cmplx( spec, len1, tmp );

		fft_cmplx_fx( spec, len1 );
		q = sub( q, tmp );
		move16();

		tmp = div_s( 4, length ); /Q17 /
		tmp = round_fx( L_mult( tmp, 19302 /0.75f EVS_PI in Q13/ ) ); /Q15 */
		}
		#endif

		FOR( i = 0; i < len1; i++ )
		{
		#ifdef OPTIMIZE_FFT_STACK
		re = Msub_32_16( spec[i].re, spec[i].im, tmp ); /Q(q+1) /
		im = Madd_32_16( spec[i].im, spec[i].re, tmp ); /Q(q+1) /
		#else
		re = Msub_32_16( complex_buf[2 * i], complex_buf[( ( i * 2 ) + 1 )], tmp ); /Q(q+1) /
		im = Madd_32_16( complex_buf[( ( i * 2 ) + 1 )], complex_buf[2 * i], tmp ); /Q(q+1) /
		#endif
		y[2 * i] = L_add( Mult_32_16( re, edct_table[i] ), Mult_32_16( im, edct_table[( len1 - ( 1 + i ) )] ) ); /Q(q+2)/
		move32();
		y[( length - ( 1 + ( i * 2 ) ) )] = L_sub( Mult_32_16( re, edct_table[( len1 - ( 1 + i ) )] ), Mult_32_16( im, edct_table[i] ) ); /Q(q+2)/
		move32();
		} /Q(q-2) /

		#ifdef HARMONIZE_DCTaa
		IF( element_mode == EVS_MONO )
		{
		q = sub( 15 + 2, q );
		}
		ELSE
		{
		q = sub( 31 + 2, q );
		}
		#else
		q = sub( 15 + 2, q );
		#ifdef HARMONIZE_DCT
		IF( element_mode != EVS_MONO )
		{
		q = add( q, Q16 );
		}
		#endif
		#endif
		move16();

		return;
		}

		#ifndef HARMONIZE_DCT
		void edct_ivas_fx(
		const Word32 x, / i : input signal Qq */
		Word32 y, / o : output transform Qq */
		@@ -314,8 +380,11 @@ void edct_ivas_fx(

		q = sub( 31 + 2, q );
		move16();

		return;
		}
		#endif

		/-------------------------------------------------------------------------
		* FUNCTION : edst_fx()
		*
		@@ -340,7 +409,11 @@ void edst_fx(
		Word32 re;
		Word32 im;
		const Word16 edct_table; /Q16 */
		#ifdef OPTIMIZE_FFT_STACK
		cmplx complex_buf[L_FRAME_PLUS / 2];
		#else
		Word32 complex_buf[2 * ( L_FRAME48k / 2 + 240 )];
		#endif
		Word32 L_tmp;
		Word16 tmp;
		Word16 len1;
		@@ -351,24 +424,41 @@ void edst_fx(
		FOR( i = 0; i < len1; i++ )
		{
		L_tmp = Mult_32_16( x[( length - ( 1 + ( i * 2 ) ) )], edct_table[i] ); /Qq+1/
		#ifdef OPTIMIZE_FFT_STACK
		complex_buf[i].re = Madd_32_16( L_tmp, x[2 * i], edct_table[( len1 - ( 1 + i ) )] ); /Qq+1/
		#else
		complex_buf[2 * i] = Madd_32_16( L_tmp, x[2 * i], edct_table[( len1 - ( 1 + i ) )] ); /Qq+1/
		#endif
		move32();

		L_tmp = Mult_32_16( x[2 * i], edct_table[i] ); /Qq+1/
		#ifdef OPTIMIZE_FFT_STACK
		complex_buf[i].im = Msub_32_16( L_tmp, x[( length - ( 1 + ( i * 2 ) ) )], edct_table[( len1 - ( 1 + i ) )] ); /Qq+1/
		#else
		complex_buf[( ( i * 2 ) + 1 )] = Msub_32_16( L_tmp, x[( length - ( 1 + ( i * 2 ) ) )], edct_table[( len1 - ( 1 + i ) )] ); /Qq+1/
		#endif
		move32();
		}

		q = sub( 15, q );
		move16();
		#ifdef OPTIMIZE_FFT_STACK
		BASOP_cfft( complex_buf, len1, q, y );
		#else
		BASOP_cfft( (cmplx *) complex_buf, len1, q, y );
		#endif

		tmp = div_s( 1, length ); /Q15 /
		tmp = round_fx( L_shl( L_mult( tmp, 19302 /0.75f EVS_PI in Q13/ ), 2 ) ); /Q15 */
		FOR( i = 0; i < len1; i++ )
		{
		#ifdef OPTIMIZE_FFT_STACK
		re = Msub_32_16( complex_buf[i].re, complex_buf[i].im, tmp ); /Qq+1/
		im = Madd_32_16( complex_buf[i].im, complex_buf[i].re, tmp ); /Qq+1/
		#else
		re = Msub_32_16( complex_buf[2 * i], complex_buf[( ( i * 2 ) + 1 )], tmp ); /Qq+1/
		im = Madd_32_16( complex_buf[( ( i * 2 ) + 1 )], complex_buf[2 * i], tmp ); /Qq+1/
		#endif
		y[2 * i] = L_add( Mult_32_16( re, edct_table[i] ), Mult_32_16( im, edct_table[( len1 - ( 1 + i ) )] ) ); /Qq+2/
		move32();
		y[( length - ( 1 + ( i * 2 ) ) )] = L_sub( Mult_32_16( im, edct_table[i] ), Mult_32_16( re, edct_table[( len1 - ( 1 + i ) )] ) ); /Qq+2/
		@@ -380,6 +470,8 @@ void edst_fx(

		return;
		}


		/========================================================================/
		/* FUNCTION : edct_fx() */
		/------------------------------------------------------------------------/
		@@ -404,26 +496,35 @@ void edct_16fx(
		const Word16 x, / i : input signal Qx */
		Word16 y, / o : output transform Qx */
		Word16 length, /* i : length */
		Word16 bh, /* bit-headroom */
		Word16 bh /* bit-headroom */
		#ifndef HARMONIZE_DCT
		,
		const Word16 element_mode

		#endif
		)
		{
		Word16 i;
		Word16 re[L_FRAME48k / 2];
		Word16 im[L_FRAME48k / 2];
		const Word16 *edct_table = NULL;
		#ifndef OPTIMIZE_FFT_STACK
		Word16 re2[L_FRAME48k / 2];
		Word16 im2[L_FRAME48k / 2];

		#endif
		Word32 L_tmp, Lacc, Lmax;
		#ifdef OPTIMIZE_FFT_STACK
		Word16 tmp, tmp_re, fact;
		#else
		Word16 tmp, fact;
		#endif
		Word16 Q_edct;
		Word16 Len2, i2;
		const Word16 px, pt;
		Word16 *py;
		#ifndef HARMONIZE_DCT
		(void) element_mode;
		/COMPLETE: some eDCT sub function are missing /
		#endif

		IF( EQ_16( length, L_FRAME32k ) )
		{
		@@ -479,26 +580,46 @@ void edct_16fx(

		L_tmp = L_mult( x[i2], edct_table[i] ); /Q(Qx+16) /
		Lacc = L_mac_sat( L_tmp, px, pt ); /Q(Qx+16) /
		#ifdef OPTIMIZE_FFT_STACK
		re[i] = round_fx_sat( L_shl_sat( Lacc, Q_edct ) ); /* Q(Qx+Q_edct) */
		#else
		re2[i] = round_fx_sat( L_shl_sat( Lacc, Q_edct ) ); /* Q(Qx+Q_edct) */
		#endif
		move16();
		L_tmp = L_mult( px, edct_table[i] ); /Q(Qx+16) */
		Lacc = L_msu_sat( L_tmp, x[i2], pt ); /Q(Qx+16) */
		#ifdef OPTIMIZE_FFT_STACK
		im[i] = round_fx_sat( L_shl_sat( Lacc, Q_edct ) ); /* Q(Qx+Q_edct) */
		#else
		im2[i] = round_fx_sat( L_shl_sat( Lacc, Q_edct ) ); /* Q(Qx+Q_edct) */
		#endif
		move16();
		px -= 2;
		pt--;
		}
		IF( EQ_16( length, L_FRAME32k ) )
		{
		#ifdef OPTIMIZE_FFT_STACK
		DoRTFT320_16fx( re, im );
		#else
		DoRTFT320_16fx( re2, im2 );
		#endif
		}
		ELSE IF( EQ_16( length, L_FRAME ) )
		{
		#ifdef OPTIMIZE_FFT_STACK
		DoRTFT128_16fx( re, im );
		#else
		DoRTFT128_16fx( re2, im2 );
		#endif
		}
		ELSE IF( EQ_16( length, L_FRAME16k ) )
		{
		#ifdef OPTIMIZE_FFT_STACK
		DoRTFT160_16fx( re, im );
		#else
		DoRTFT160_16fx( re2, im2 );
		#endif
		}
		ELSE
		{
		@@ -508,6 +629,18 @@ void edct_16fx(
		fact = round_fx( L_shl( L_tmp, 2 ) ); /Q15 /
		FOR( i = 0; i < shr( length, 1 ); i++ )
		{
		#ifdef OPTIMIZE_FFT_STACK
		tmp = mult_r( im[i], fact ); /Q(Qx+Q_edct) /
		tmp_re = sub_sat( re[i], tmp ); /Q(Qx+Q_edct) /
		move16();

		tmp = mult_r( re[i], fact ); /Q(Qx+Q_edct) /
		im[i] = add_sat( im[i], tmp ); /Q(Qx+Q_edct) /
		move16();

		re[i] = tmp_re;
		move16();
		#else
		tmp = mult_r( im2[i], fact ); /Q(Qx+Q_edct) /
		re[i] = sub_sat( re2[i], tmp ); /Q(Qx+Q_edct) /
		move16();
		@@ -515,6 +648,7 @@ void edct_16fx(
		tmp = mult_r( re2[i], fact ); /Q(Qx+Q_edct) /
		im[i] = add_sat( im2[i], tmp ); /Q(Qx+Q_edct) /
		move16();
		#endif
		}

		/* Post-rotate and obtain the output data */
		@@ -567,7 +701,11 @@ void iedct_short_fx(
		seg_len_div4 = shr( segment_length, 2 ); /Q0/
		seg_len_3mul_div4 = add( seg_len_div2, seg_len_div4 );

		#ifdef HARMONIZE_DCT
		edct_fx( in, alias, seg_len_div2, Q, EVS_MONO );
		#else
		edct_fx( in, alias, seg_len_div2, Q );
		#endif
		FOR( i = 0; i < seg_len_div2; i++ )
		{
		IF( alias[i] != 0 )

lib_com/fft_fx.c

+12 −1

Original line number	Diff line number	Diff line
		@@ -109,11 +109,13 @@ static void fft5_8( Word16 n1, Word32 zRe, Word32 zIm, const Word16 *Idx );
		static void fft4_5( Word32 x, Word32 y, const Word16 *Idx );
		static void fft5_4( Word16 n1, Word32 zRe, Word32 zIm, const Word16 *Idx );

		#ifndef HARMONIZE_DCT
		void DoRTFTn_fx_ivas(
		Word32 x, / i/o: real part of input and output data */
		Word32 y, / i/o: imaginary part of input and output data */
		const Word16 n /* i : size of the FFT n=(2^k) up to 1024 */
		);
		#endif
		/-----------------------------------------------------------------
		* fft15_shift2()
		* 15-point FFT with 2-point circular shift
		@@ -2630,6 +2632,7 @@ static void dctsub(
		return;
		}

		#ifndef HARMONIZE_DCT
		/-----------------------------------------------------------------
		* edct2_fx_ivas()
		*
		@@ -2790,7 +2793,7 @@ void DoRTFTn_fx_ivas(

		return;
		}

		#endif
		#ifndef HQ_ALIGN_DUPLICATED_CODE
		void fft3_fx_ivas(
		const Word32 X[], // Qx
		@@ -4274,7 +4277,11 @@ void DoFFT_fx(
		DoRTFT320_fx( re2, im2 );
		BREAK;
		case 256:
		#ifdef HARMONIZE_DCT
		DoRTFTn_fx( re2, im2, NULL, 256 );
		#else
		DoRTFTn_fx_ivas( re2, im2, 256 );
		#endif
		BREAK;
		case 240:
		DoRTFT240( re2, im2 );
		@@ -4298,7 +4305,11 @@ void DoFFT_fx(
		DoRTFT80_fx( re2, im2 );
		BREAK;
		case 64:
		#ifdef HARMONIZE_DCT
		DoRTFTn_fx( re2, im2, NULL, 64 );
		#else
		DoRTFTn_fx_ivas( re2, im2, 64 );
		#endif
		BREAK;
		case 40:
		DoRTFT40_fx( re2, im2 );

lib_com/gs_inact_switching_fx.c

+12 −0

Original line number	Diff line number	Diff line
		@@ -89,7 +89,11 @@ void Inac_switch_ematch_fx(
		ELSE IF( EQ_16( coder_type, VOICED ) \|\| EQ_16( coder_type, GENERIC ) \|\| EQ_16( coder_type, TRANSITION ) \|\| ( last_core != ACELP_CORE ) \|\| NE_16( last_codec_mode, MODE1 ) \|\| ( ( element_mode > EVS_MONO ) && EQ_16( coder_type, UNVOICED ) ) )
		{
		/* Find spectrum and energy per band for GC and VC frames */
		#ifdef HARMONIZE_DCT
		edct_16fx( exc2, dct_exc_tmp, L_frame, 5 );
		#else
		edct_16fx( exc2, dct_exc_tmp, L_frame, 5, element_mode );
		#endif

		Ener_per_band_comp_fx( dct_exc_tmp, Ener_per_bd, Q_exc, MBANDS_GN, 1, L_frame );

		@@ -103,7 +107,11 @@ void Inac_switch_ematch_fx(
		ELSE IF( ( coder_type == INACTIVE ) && inactive_coder_type_flag )
		{
		/* Find spectrum and energy per band for inactive frames */
		#ifdef HARMONIZE_DCT
		edct_16fx( exc2, dct_exc_tmp, L_frame, 5 );
		#else
		edct_16fx( exc2, dct_exc_tmp, L_frame, 5, element_mode );
		#endif

		Ener_per_band_comp_fx( dct_exc_tmp, Ener_per_bd, Q_exc, MBANDS_GN, 1, L_frame );

		@@ -188,7 +196,11 @@ void Inac_switch_ematch_fx(
		Scale_sig( dct_exc_tmp, 240, 1 ); // Q_exc
		Scale_sig( exc2, 240, 1 ); // Q_exc
		}
		#ifdef HARMONIZE_DCT
		edct_16fx( dct_exc_tmp, exc2, L_frame, 5 );
		#else
		edct_16fx( dct_exc_tmp, exc2, L_frame, 5, element_mode );
		#endif
		}

		return;