Commit bfdca295 authored by Dominik Weckbecker's avatar Dominik Weckbecker 💬
Browse files

optimize calculation of q_cy_auto_dir_smooth and cy_cross_dir_smooth

parent 6c0f3d8e

lib_rend/ivas_dirac_output_synthesis_dec.c

+43 −11
Original line number Diff line number Diff line
@@ -4050,6 +4050,33 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx( 
        /*Only direct prototype*/

        FOR( ; l < num_freq_bands; l++ )

        {

#ifdef FIX_867_DBG

			{

				static FILE *f = 0;

				if (f == 0)

					f = fopen("cy_auto_dir_smooth_prev", "w");



				Word32 m = *p_cy_auto_dir_smooth;

				Word16 e = h_dirac_output_synthesis_state->q_cy_auto_dir_smooth[k*num_freq_bands + l];

				float f1_m = (float)m;

				float f1_e = powf(2, e);

				float f1 = f1_m / f1_e;

				fprintf(f, "%d %d %d %f\n", l, m, e, f1);

			}

			{

				static FILE *f = 0;

				if (f == 0)

					f = fopen("p_power_smooth", "w");



				Word32 m = *p_power_smooth;

				Word16 e = h_dirac_output_synthesis_state->proto_power_smooth_q;

				float f1_m = (float)m;

				float f1_e = powf(2, e);

				float f1 = f1_m / f1_e;

				fprintf(f, "%d %d %d %f\n", l, m, e, f1);

			}

#endif



            /*Direct*/

            g1 = alpha[l]; // Q31

            move32();
@@ -4070,6 +4097,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx( 
                *( p_cy_cross_dir_smooth_prev ) = 1;

                move32();

            }



            ( p_cy_cross_dir_smooth++ );

            power_smooth_temp = L_shl( *p_power_smooth, norm_l( *p_power_smooth ) );

            L_tmp = Mpy_32_32( power_smooth_temp, L_tmp ); // proto_power_smooth_q + norm_l( *p_power_smooth ) ) + q_cy_auto_dir_smooth_prev_local - 31
@@ -6820,7 +6848,19 @@ static void computeTargetPSDs_direct_subframe_fx( 
    FOR( ch_idx = 0; ch_idx < num_channels; ++ch_idx )

    {

        cur_idx = imult1616( ch_idx, num_freq_bands );

#ifdef FIX_867

		FOR(i = 0; i < num_freq_bands; i++)

		{

			W_tmp[i]  = W_mult_32_32(direct_power[i], direct_responses_square[cur_idx + i]); // q_reference_power + Q31 + 1

			q_tmp = W_norm(W_tmp[i]);

			cy_auto_dir_smooth[cur_idx + i] = W_extract_h( W_shl(W_tmp[i], q_tmp) );

			q_cy_auto_dir_smooth[ch_idx + i] = add(q_reference_power[i], q_tmp);

			move16();



			cy_cross_dir_smooth[cur_idx + i] = Mpy_32_32(direct_power[i], direct_responses[cur_idx+i]);

			q_cy_cross_dir_smooth[cur_idx + i] = q_reference_power[i];

		}

#else

        W_max = 0;

        move64();

        FOR( i = 0; i < num_freq_bands; i++ )
@@ -6837,25 +6877,17 @@ static void computeTargetPSDs_direct_subframe_fx( 
        {

            cy_auto_dir_smooth[cur_idx + i] = W_extract_h( W_shl( W_tmp[i], q_tmp ) ); /*q_reference_power+q_tmp*/

            move32();

#ifdef FIX_867

            q_cy_auto_dir_smooth[cur_idx + i] = add( q_reference_power[i], q_tmp );

            move16();



            q_cy_cross_dir_smooth[cur_idx + i] = q_reference_power[i];

            move16();

#endif

        }

#ifndef FIX_867

        q_cy_auto_dir_smooth[ch_idx] = add( *q_reference_power, q_tmp );

        move16();

#endif

        v_mult_fixed( direct_power, &direct_responses[cur_idx], &cy_cross_dir_smooth[cur_idx], num_freq_bands ); // (q_reference_power, Q31) -> q_reference_power

#endif

    }



#ifndef FIX_867

    *q_cy_cross_dir_smooth = *q_reference_power;

#endif

	move16();

#endif



    return;

}