activate only be speedups - is really be? (c64a0a02) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+8 −1

Original line number	Diff line number	Diff line
		@@ -53,7 +53,7 @@

		#define SUPPORT_JBM_TRACEFILE /* Support for JBM tracefile, which is needed for 3GPP objective/subjective testing, but not relevant for real-world implementations */

		/#define WMOPS/ /* Activate complexity and memory counters */
		//#define WMOPS /* Activate complexity and memory counters */
		#ifdef WMOPS
		/#define WMOPS_PER_FRAME/ /* Output per-frame complexity (writes one float value per frame to the file "wmops_analysis") */
		/#define WMOPS_DETAIL/ /* Output detailed complexity printout for every function. Increases runtime overhead */
		@@ -78,4 +78,11 @@
		#define HARM_PUSH_BIT
		#define HARM_ENC_INIT
		//#define HARM_SCE_INIT

		// new speedups - pipe running
		#define FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat /FhG: reduces WMOPS - bit-exact/
		#define FIX_1439_SPEEDUP_stereo_icBWE_dec_fx /FhG: reduces WMOPS - bit-exact/
		#define FIX_1439_SPEEDUP_ivas_swb_tbe_dec_fx_0 /FhG: reduces WMOPS - bit-exact/
		#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig /FhG: reduces WMOPS - bit-exact/

		#endif

+1 −8

Original line number	Diff line number	Diff line
		@@ -61,17 +61,10 @@


		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE1 // nonbe // \|
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2 // nonbe // \| > 3.1 WOPS, pipe testing https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/pipelines/50562
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2 // nonbe // \| > 3.1 WMOPS, pipe testing https://forge.3gpp.org/rep/sa4/audio/ivas-basop/-/pipelines/50562
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE3 // nonbe // \|
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig // BE // .9 WMOPS -- all pipes green
		// ^^^^^^^^
		// Numbers ALL: 84,189
		// Numbers beOnly: 87,344
		// NUmbers none: 88,236


		//-----------------------------------------------------------------------
		//#define FIX_1439_SPEEDUP_ivas_swb_tbe_dec_fx_0 // .8 WMOPS BE, all pipes green!
		// OPT_STEREO_32KBPS_V1 switch is inactive - why? (3.2 WMOPS)

+0 −13

Original line number	Diff line number	Diff line
		@@ -760,7 +760,6 @@ void Copy_Scale_sig_16_32_DEPREC(
		}
		}

		#define FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat
		void Copy_Scale_sig_16_32_no_sat(
		const Word16 x[], /* i : signal to scale input Qx */
		Word32 y[], /* o : scaled signal output Qx */
		@@ -796,7 +795,6 @@ void Copy_Scale_sig_16_32_no_sat(
		return;
		}
		#ifdef FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat
		//#define FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat_TEST
		L_tmp = L_shl_o( 1, exp0 - 1, &Overflow );

		IF( L_tmp >= 0x7FFF )
		@@ -806,15 +804,9 @@ void Copy_Scale_sig_16_32_no_sat(
		// y[i] = L_mult0(x[i], L_tmp);
		y[i] = W_extract_l( W_mult_32_16( L_tmp, x[i] ) );
		move32(); /* saturation can occur here */
		#ifdef FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat_TEST
		Word32 tmp32 = W_extract_l( W_mult_32_16( L_tmp, x[i] ) );
		if ( tmp32 != y[i] )
		assert( 0 );
		#endif
		}
		return;
		}

		// ELSE
		{
		// Word16 tmp = extract_l( L_tmp );
		@@ -822,11 +814,6 @@ void Copy_Scale_sig_16_32_no_sat(
		{
		y[i] = L_mult( x[i], L_tmp );
		move32(); /* saturation can occur here */
		#ifdef FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat_TEST
		Word32 tmp32 = W_extract_l( W_mult_32_16( L_tmp, x[i] ) );
		if ( tmp32 != y[i] )
		assert( 0 );
		#endif
		}
		}
		#else

+15 −0

Original line number	Diff line number	Diff line
		@@ -904,8 +904,22 @@ void stereo_icBWE_dec_fx(
		winSlope_fx = div_s( 1, winLen_fx ); /* Q15 */
		alpha_fx = winSlope_fx; /* Q15 */
		move16();
		#ifdef FIX_1439_SPEEDUP_stereo_icBWE_dec_fx
		Word16 winSlope_fx_ = sub( 32767 /* 1.0 in Q15*/, winSlope_fx );
		#endif
		FOR( i = 0; i < winLen_fx; i++ )
		{
		#ifdef FIX_1439_SPEEDUP_stereo_icBWE_dec_fx
		L_tmp = L_mult0( alpha_fx, icbweM2Ref_fx ); /* Q29 */
		L_tmp = L_mac0( L_tmp, sub( 32767 /* 1.0 in Q15/, alpha_fx ), hStereoICBWE->icbweM2Ref_prev_fx ); / Q29 */
		tmp = shl( round_fx( L_tmp ), 1 ); /* Q14 */
		synthRef_fx[i] = Mpy_32_16_1( synthRef_fx[i], tmp ); /* Qsyn - 1 */
		move32();
		if ( LE_16( alpha_fx, winSlope_fx_ ) )
		{
		alpha_fx = add( alpha_fx, winSlope_fx ); /* Q15 */
		}
		#else
		L_tmp = L_mult0( alpha_fx, icbweM2Ref_fx ); /* Q29 */
		L_tmp = L_mac0( L_tmp, sub( 32767 /* 1.0 in Q15/, alpha_fx ), hStereoICBWE->icbweM2Ref_prev_fx ); / Q29 */
		tmp = shl( round_fx( L_tmp ), 1 ); /* Q14 */
		@@ -915,6 +929,7 @@ void stereo_icBWE_dec_fx(
		{
		alpha_fx = add( alpha_fx, winSlope_fx ); /* Q15 */
		}
		#endif
		}

		FOR( ; i < NS2SA_FX2( st->output_Fs, FRAME_SIZE_NS ); i++ )