Merge branch 'basop-2336-duplications-of-dot_product' into 'main' (1077d9cc) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -99,6 +99,7 @@
		#define FIX_2196_UNREACHABLE_CODE_IN_OSBA_STEREO_OUTPUT /* FhG: remove unreachable code in ivas_osba_stereo_add_channels */
		#define FIX_OUTPUT_FRAME /* VA: harmonize "output_frame" parameter usage to correspond to FLP */
		#define HARMONIZE_ISSUE_2435_WRITETNSDATA /* FhG basop 2435: Harmonize WriteTnsData_fx(), EncodeTnsData_fx() */
		#define HARM_2336_DOTP /* VA: basop 2336; Harmonisation of some dot_product function + some BE optimisation */

		/* #################### End BE switches ################################## */

		@@ -111,7 +112,6 @@
		/* ##################### End NON-BE switches ########################### */

		/* ################## End MAINTENANCE switches ######################### */

		/* clang-format on */

		#endif

lib_com/prot_fx.h

+6 −6

Original line number	Diff line number	Diff line
		@@ -5687,7 +5687,7 @@ void v_add_fx_me(
		const Word16 N, /* i : Vector length */
		const Word16 hdrm /* i : headroom for when subtraction result > 1 or < -1 */
		);

		#ifndef HARM_2336_DOTP
		void v_add_w64(
		const Word64 x1[], /* i : Input vector 1 */
		const Word64 x2[], /* i : Input vector 2 */
		@@ -5695,7 +5695,7 @@ void v_add_w64(
		const Word16 N, /* i : Vector length */
		const Word16 hdrm /* i : headroom for when subtraction result > 1 or < -1 */
		);

		#endif
		void v_sub_fx(
		const Word32 x1[], /* i : Input vector 1 */
		const Word32 x2[], /* i : Input vector 2 */
		@@ -5725,7 +5725,7 @@ Word32 dotp_fx32_o(
		const Word16 n, /* i : vector length */
		const Word16 log_len, /* i : max factor added to result q after dot product (equal to log2(n)) */
		Word16 *res_q );

		#ifndef HARM_2336_DOTP
		Word32 dotp_fx32_fac(
		const Word32 x[], /* i : vector x[] */
		const Word32 y[], /* i : vector y[] */
		@@ -5733,7 +5733,7 @@ Word32 dotp_fx32_fac(
		const Word16 log_len, /* i : max factor added to result q after dot product (equal to log2(n)) */
		Word16 res_q /stores resultant Q*/
		);

		#endif
		Word32 dotp_fx_ivas_fx(
		const Word32 x[], /* i : vector x[] */
		Word16 x_e,
		@@ -5747,7 +5747,7 @@ Word32 dotp_fx_guarded(
		const Word32 y[], /* i : vector y[] */
		const Word16 n /* i : vector length */
		);

		#ifndef HARM_2336_DOTP
		Word32 dotp_me_fx(
		const Word32 x[], /* i : vector x[] */
		const Word32 y[], /* i : vector y[] */
		@@ -5755,7 +5755,7 @@ Word32 dotp_me_fx(
		Word16 exp_x,
		Word16 exp_y,
		Word16 *exp_suma );

		#endif
		void lsf_end_dec_fx(
		Decoder_State st, / i/o: decoder state structure */
		Word16 mode2_flag, /* Q0 */

lib_com/tools_fx.c

+28 −5

Original line number	Diff line number	Diff line
		@@ -630,7 +630,7 @@ Word32 dotp_fx32_o(
		return W_extract_l( suma );
		}


		#ifndef HARM_2336_DOTP
		Word32 dotp_fx32_fac(
		const Word32 x[], /* i : vector x[] */
		const Word32 y[], /* i : vector y[] */
		@@ -686,6 +686,7 @@ void v_add_w64(

		return;
		}
		#endif


		/-------------------------------------------------------------------
		@@ -703,10 +704,18 @@ void v_sub_fx(
		)
		{
		Word16 i;

		#ifdef HARM_2336_DOTP
		Word32 wtmp;
		assert( hdrm < 31 && hdrm > 0 );
		wtmp = L_shl_sat( 1, sub( 31, hdrm ) );
		#endif
		FOR( i = 0; i < N; i++ )
		{
		#ifndef HARM_2336_DOTP
		y[i] = L_sub( L_shr( x1[i], hdrm ), L_shr( x2[i], hdrm ) );
		#else
		y[i] = Msub_32_32( Mpy_32_32( x1[i], wtmp ), x2[i], wtmp );
		#endif
		move32();
		}

		@@ -5143,7 +5152,7 @@ UWord32 mvl2s_r(

		return noClipping;
		}

		#ifndef HARM_2336_DOTP
		Word32 dotp_me_fx(
		const Word32 x[], /* i : vector x[] */
		const Word32 y[], /* i : vector y[] */
		@@ -5166,7 +5175,7 @@ Word32 dotp_me_fx(

		return suma;
		}

		#endif
		Word32 dotp_fx_guarded(
		const Word32 x[], /* i : vector x[] */
		const Word32 y[], /* i : vector y[] */
		@@ -5195,6 +5204,7 @@ Word32 dotp_fx_ivas_fx(
		const Word16 n, /* i : vector length */
		Word16 *out_e )
		{
		#ifndef HARM_2336_DOTP
		Word16 i, exp;
		Word32 suma = 0;

		@@ -5208,7 +5218,20 @@ Word32 dotp_fx_ivas_fx(

		*out_e = exp;
		move16();

		#else
		Word16 i;
		Word32 suma;
		Word32 mul;
		Word16 mul_exp = add( x_e, y_e );
		suma = Mpy_32_32( x[0], y[0] );
		*out_e = mul_exp;
		move16();
		FOR( i = 1; i < n; i++ )
		{
		mul = Mpy_32_32( x[i], y[i] );
		suma = BASOP_Util_Add_Mant32Exp( suma, *out_e, mul, mul_exp, out_e ); // exp_x+exp_A
		}
		#endif
		return suma;
		}

lib_enc/ivas_mct_enc_mct_fx.c

+9 −0

Original line number	Diff line number	Diff line
		@@ -235,8 +235,17 @@ static void getCorrelationMatrix_fx(

		FOR( n = 0; n < nSubframes; n++ )
		{
		#ifndef HARM_2336_DOTP
		res_q = sub( 31, sts[0]->hTcxEnc->spectrum_e[0] );
		L_tmp = dotp_fx32_fac( sts[ch1]->hTcxEnc->spectrum_fx[n], sts[ch2]->hTcxEnc->spectrum_fx[n], L_subframe, 1, &res_q );
		#else
		Word16 tmp_Q;
		res_q = sub( 31, sts[0]->hTcxEnc->spectrum_e[0] );
		tmp_Q = res_q;
		move16();
		L_tmp = dotp_fx32_o( sts[ch1]->hTcxEnc->spectrum_fx[n], sts[ch2]->hTcxEnc->spectrum_fx[n], L_subframe, 1, &res_q );
		res_q = add( res_q, tmp_Q );
		#endif
		xCorrMatrix[ch1][ch2] = BASOP_Util_Add_Mant32Exp( xCorrMatrix[ch1][ch2], xCorrMatrix_exp[ch1][ch2], L_tmp, sub( 31, res_q ), &xCorrMatrix_exp[ch1][ch2] );
		move32();
		}

lib_enc/speech_music_classif_fx.c

+4 −0

Original line number	Diff line number	Diff line
		@@ -1740,7 +1740,11 @@ Word16 ivas_smc_gmm_fx(
		move16();
		len = mel_fb_len[i];
		move16();
		#ifndef HARM_2336_DOTP
		temp32 = dotp_me_fx( &PS_fx[j], pt_mel_fb_fx, len, 31 - Qfact_PS, Q1, &dotp_exp );
		#else
		temp32 = dotp_fx_ivas_fx( &PS_fx[j], 31 - Qfact_PS, pt_mel_fb_fx, Q1, len, &dotp_exp );
		#endif
		IF( LT_16( dotp_exp, -17 ) ) /-18 is exponent of 10737:to avoid overflow when left shifting 10737/
		{
		temp32 = L_shr( temp32, sub( -17, dotp_exp ) );