activate speedup 09 10 11 for testing (7eb74ead) · Commits · SA4 / Audio / IVAS BASOP

lib_rend/ivas_dirac_dec_binaural_functions_fx.c

+62 −21

Original line number	Diff line number	Diff line
		@@ -45,16 +45,25 @@

		#include "wmc_auto.h"

		//#define FIX_1326_SPEEDUP_00 //make sqrt(1) a const - catch bitstreams //no occurence in current bitstream
		#define FIX_1326_SPEEDUP_01 // optimize matrixTransp1Mul_fx -> eig2x2_fx : .4 WMOPS
		#define FIX_1326_SPEEDUP_02 // speedup eig2x2_fx // .3 WMOPS
		#define FIX_1326_SPEEDUP_03 // speedup eig2x2_fx // .1 WMOPS
		#define FIX_1326_SPEEDUP_04 // speedup eig2x2_fx // .2 WMOPS
		// MHZ NUMBERS:
		// NULL: 179.292


		//#define FIX_1326_SPEEDUP_00 //make sqrt(1) a const - catch bitstreams //no occurence --> DONT USE
		//#define FIX_1326_SPEEDUP_01 // optimize matrixT1mul->eig2x2_fx // .4 WMOPS --> USE
		//#define FIX_1326_SPEEDUP_02 // speedup eig2x2_fx // .3 WMOPS --> USE
		//#define FIX_1326_SPEEDUP_03 // speedup eig2x2_fx // .1 WMOPS --> USE
		//#define FIX_1326_SPEEDUP_04 // speedup eig2x2_fx // .2 WMOPS --> USE
		//#define FIX_1326_SPEEDUP_05 // div->sqrt =>isqrt // 3.5 WMOPS //Quite bad diffs --> DONT USE
		//#define FIX_1326_SPEEDUP_06 // div->sqrt =>isqrt // 3.0 WMOPS //Quite bad diffs --> DONT USE
		#define FIX_1326_SPEEDUP_07 // div->sqrt =>isqrt // ? WMOPS //Big DIffs , no replacement of divSqrt , PIPELINE GREEN, --> USE
		#define FIX_1326_SPEEDUP_08 // "-" // 3.0 WMOPS //small diffs, PIPELINE GREEN! -- > USE
		#define FIX_1326_SPEEDUP_09 // Relocate matrixMul
		//#define FIX_1326_SPEEDUP_07 // div->sqrt =>isqrt // 0 WMOPS --> DONT USE
		//#define FIX_1326_SPEEDUP_08 // "-" // 3.0 WMOPS //small diffs --> USE
		#define FIX_1326_SPEEDUP_09 // tiny speedup // .1 WMOPS
		#define FIX_1326_SPEEDUP_10 // tiny speedup // .1 WMOPS
		#define FIX_1326_SPEEDUP_11 // tiny speedup // .1 WMOPS
		//#define FIX_1326_SPEEDUP_12 // tiny speedup // <.1 WMOPS -->DONTUSE
		//#define FIX_1326_SPEEDUP_13 // replace div/sqrt->isqrt // 2.9 WMOPS
		//#define FIX_1326_SPEEDUP_14 //
		Word16 slot_fx[4] = { 32767, 16384, 10922, 8192 };

		/*-------------------------------------------------------------------------
		@@ -2124,6 +2133,14 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx(
		exp = sub( get_min_scalefactor( resultMtxRe_fx[0][0], resultMtxRe_fx[1][1] ), 2 );
		tmp2 = L_add( L_shl( resultMtxRe_fx[0][0], exp ), L_shl( resultMtxRe_fx[1][1], exp ) );
		q_tmp2 = add( q_res, exp );
		#ifdef FIX_1326_SPEEDUP_11
		{
		Word16 shift1 = s_max( 0, sub( q_tmp2, q_CrEne ) );
		Word16 shift2 = s_max( 0, sub( q_CrEne, q_tmp2 ) );
		realizedOutputEne_fx = L_add( L_shr( tmp1, shift2 ), L_shr( tmp2, shift1 ) );
		q_realizedOutputEne = s_min( q_CrEne, q_tmp2 );
		}
		#else
		IF( LT_16( q_CrEne, q_tmp2 ) )
		{
		realizedOutputEne_fx = L_add( tmp1, L_shr( tmp2, sub( q_tmp2, q_CrEne ) ) );
		@@ -2136,7 +2153,7 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx(
		q_realizedOutputEne = q_tmp2;
		move16();
		}

		#endif
		exp = sub( get_min_scalefactor( hDiracDecBin->ChEneOut_fx[0][bin], hDiracDecBin->ChEneOut_fx[1][bin] ), 1 );
		targetOutputEne_fx = L_add( L_shl( hDiracDecBin->ChEneOut_fx[0][bin], exp ), L_shl( hDiracDecBin->ChEneOut_fx[1][bin], exp ) );
		q_targetOutputEne = add( hDiracDecBin->q_ChEneOut, exp );
		@@ -2159,15 +2176,26 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx(
		q_missingOutputEne = q_targetOutputEne;
		move16();
		}

		tmp1 = BASOP_Util_Add_Mant32Exp( tmp2, sub( 31, q_tmp2 ), missingOutputEne_fx, sub( 31, q_missingOutputEne ), &exp1 );

		#ifdef FIX_1326_SPEEDUP_13
		{
		Word16 exp_temp;
		tmp2 = BASOP_Util_Add_Mant32Exp( tmp2, sub( 31, q_tmp2 ), EPSILON_MANT, EPSILON_EXP, &exp_temp );
		tmp2 = ISqrt32( tmp2, &exp_temp );
		gain_fx = Mpy_32_32(tmp2, Sqrt32(tmp1, &exp1));
		q_gain = sub( 31, add( exp_temp, exp1 ) );
		}
		#else
		{
		tmp2 = BASOP_Util_Add_Mant32Exp( tmp2, sub( 31, q_tmp2 ), EPSILON_MANT, EPSILON_EXP, &exp2 );
		tmp2 = ISqrt32( tmp2, &exp2 );
		gain_fx = Mpy_32_32( tmp2, Sqrt32( tmp1, &exp1 ) );
		q_gain = sub( 31, add( exp2, exp1 ) );
		}
		gain_fx = Sqrt32( tmp2, &exp2 );
		q_gain = sub( 31, exp2 );
		#endif


		// 1073741824 = 4 in Q28
		@@ -4415,17 +4443,11 @@ static void formulate2x2MixingMatrix_fx(


		eig2x2_fx( tmpRe_fx[0][0], tmpRe_fx[1][1], q_temp, tmpRe_fx[1][0], tmpIm_fx[1][0], q_temp, Ure_fx, Uim_fx, &q_U, D_fx, &q_D );
		#ifdef FIX_1326_SPEEDUP_09
		matrixMul_fx( Are_fx, Aim_fx, &q_A, Ure_fx, Uim_fx, &q_U, tmpRe_fx, tmpIm_fx, &q_temp );
		#endif
		#else
		matrixTransp1Mul_fx( Are_fx, Aim_fx, q_A, Are_fx, Aim_fx, q_A, tmpRe_fx, tmpIm_fx, &q_temp );

		eig2x2_fx( tmpRe_fx[0][0], tmpRe_fx[1][1], q_temp, tmpRe_fx[1][0], tmpIm_fx[1][0], q_temp, Ure_fx, Uim_fx, &q_U, D_fx, &q_D );

		#ifdef FIX_1326_SPEEDUP_09
		matrixMul_fx( Are_fx, Aim_fx, &q_A, Ure_fx, Uim_fx, &q_U, tmpRe_fx, tmpIm_fx, &q_temp );
		#endif
		#endif
		pop_wmops(); /push_wmops( "oPtoA MT1M" );/

		@@ -4512,7 +4534,19 @@ static void formulate2x2MixingMatrix_fx(
		div_fx[1] = L_shr( div_fx[1], sub( sub( 31, exp1 ), q_div ) ); // q_div
		move32();


		// 1310720000 = 10,000.0f in Q17
		#ifdef FIX_1326_SPEEDUP_09
		{
		Word16 shift1 = s_max( sub( Q17, q_div ), 0 );
		Word16 shift2 = s_max( sub( q_div, Q17 ), 0 );

		div_fx[0] = L_min( L_shr( 1310720000, shift1 ), L_shr( div_fx[0], shift2 ) ); // q_div
		move32();
		div_fx[1] = L_min( L_shr( 1310720000, shift1 ), L_shr( div_fx[1], shift2 ) ); // q_div
		move32();
		}
		#else
		IF( LT_16( q_div, Q17 ) )
		{
		div_fx[0] = L_min( L_shr( 1310720000, sub( Q17, q_div ) ), div_fx[0] ); // q_div
		@@ -4529,10 +4563,9 @@ static void formulate2x2MixingMatrix_fx(
		q_div = Q17;
		move16();
		}
		#endif

		#ifndef FIX_1326_SPEEDUP_09
		matrixMul_fx( Are_fx, Aim_fx, &q_A, Ure_fx, Uim_fx, &q_U, tmpRe_fx, tmpIm_fx, &q_temp );
		#endif

		exp = L_norm_arr( div_fx, BINAURAL_CHANNELS );
		scale_sig32( div_fx, BINAURAL_CHANNELS, exp );
		@@ -4547,7 +4580,11 @@ static void formulate2x2MixingMatrix_fx(
		W_tmp = W_mult0_32_32( tmpRe_fx[chA][chB], div_fx[chB] );
		IF( W_tmp != 0 )
		{
		#ifdef FIX_1326_SPEEDUP_10
		hdrm_re[chA][chB] = W_norm( W_tmp );
		#else
		hdrm_re[chA][chB] = sub( W_norm( W_tmp ), 0 );
		#endif
		move16();
		W_tmp = W_shl( W_tmp, hdrm_re[chA][chB] );
		tmpRe_fx[chA][chB] = W_extract_h( W_tmp );
		@@ -4564,7 +4601,11 @@ static void formulate2x2MixingMatrix_fx(
		W_tmp = W_mult0_32_32( tmpIm_fx[chA][chB], div_fx[chB] );
		IF( W_tmp != 0 )
		{
		#ifdef FIX_1326_SPEEDUP_10
		hdrm_im[chA][chB] = W_norm( W_tmp );
		#else
		hdrm_im[chA][chB] = sub( W_norm( W_tmp ), 0 );
		#endif
		move16();
		W_tmp = W_shl( W_tmp, hdrm_im[chA][chB] );
		tmpIm_fx[chA][chB] = W_extract_h( W_tmp );