diff --git a/lib_com/ivas_prot_fx.h b/lib_com/ivas_prot_fx.h index 5ce037b9a2db1bf16bd7491fda0bc13cd25385e2..1682c255b7c47efbb622e03b2ca17a82571d3e46 100644 --- a/lib_com/ivas_prot_fx.h +++ b/lib_com/ivas_prot_fx.h @@ -3341,15 +3341,16 @@ void ivas_omasa_enc_fx( ); void mctStereoIGF_enc_fx( - MCT_ENC_HANDLE hMCT, /* i/o: MCT encoder structure */ - Encoder_State **sts, /* i/o: encoder state structure */ - Word32 *orig_spectrum_fx[MCT_MAX_CHANNELS][2], /* i : MDCT spectrum for ITF */ - Word16 q_origSpec, /* i : Q for MDCT spectrum */ - Word32 powerSpec_fx[MCT_MAX_CHANNELS][L_FRAME48k], /* i/o: MDCT^2 + MDST^2 spectrum,or estimate*/ - Word32 *powerSpecMsInv_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : same as above but for inverse spect.*/ - Word16 q_powerSpec[MCT_MAX_CHANNELS], /* i : Q for powSpec_fx and powSpecMsInv_fx*/ - Word32 *inv_spectrum_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : inverse spectrum */ - const Word16 sp_aud_decision0[MCT_MAX_CHANNELS] /* i : speech audio decision */ + MCT_ENC_HANDLE hMCT, /* i/o: MCT encoder structure */ + Encoder_State **sts, /* i/o: encoder state structure */ + Word32 *orig_spectrum_fx[MCT_MAX_CHANNELS][2], /* i : MDCT spectrum for ITF */ + Word16 q_origSpec, /* i : Q for MDCT spectrum */ + Word32 powerSpec_fx[MCT_MAX_CHANNELS][L_FRAME48k], /* i/o: MDCT^2 + MDST^2 spectrum,or estimate */ + Word16 q_powerSpec[MCT_MAX_CHANNELS], /* i : Q for powSpec_fx */ + Word32 *powerSpecMsInv_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : same as powerSpec_fx but for inverse spect.*/ + Word16 q_powerSpecMsInv[MCT_MAX_CHANNELS], /* i : Q for powSpecMsInv_fx */ + Word32 *inv_spectrum_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : inverse spectrum */ + const Word16 sp_aud_decision0[MCT_MAX_CHANNELS] /* i : speech audio decision */ ); void ivas_mct_core_enc_fx( diff --git a/lib_enc/igf_enc.c b/lib_enc/igf_enc.c index d20169e73002d69c8cb78e00026c3e61cae2eed9..0ce0875086554ce8a9276679ab6f413ed6554edf 100644 --- a/lib_enc/igf_enc.c +++ b/lib_enc/igf_enc.c @@ -943,9 +943,9 @@ static void IGF_CalculateEnvelope_ivas_fx( move16(); sfbEnergyR = add_sat( EPSILON_FX, BASOP_Util_Divide3216_Scale( sum2_32_fx( pMDCTSpectrum_fx + swb_offset[sfb], width, &tmp_e ) /*exp: tmp_e*/, width, &sfbEnergyR_e ) ); // sfbEnergyR_e sfbEnergyR_e = add( sfbEnergyR_e, add( tmp_e, -15 ) ); - gain = sfbEnergyR; // gain_e + gain = L_shl( sfbEnergyR, 16 ); // gain_e move32(); - gain_e = add( sfbEnergyR_e, 16 ); + gain_e = sfbEnergyR_e; IF( element_mode > EVS_MONO ) { diff --git a/lib_enc/ivas_lfe_enc.c b/lib_enc/ivas_lfe_enc.c index e72270e6cb513fef5a0680588777c2ea79366a22..e219e8b37a1b1974f56e1432cc900c152fc57fd7 100644 --- a/lib_enc/ivas_lfe_enc.c +++ b/lib_enc/ivas_lfe_enc.c @@ -199,12 +199,13 @@ static void ivas_lfe_enc_quant_fx( q_lfe_abs_sum = sub( q_lfe_abs_sum, q_tmp ); } - tmp = BASOP_Util_Divide3232_Scale( max_value, W_extract_l( lfe_abs_sum ), &q_tmp ); - tmp = L_shl( tmp, sub( q_lfe_abs_sum, sub( 15, q_tmp ) ) ); /* Q0 (max_value / lfe_abs_sum) */ + tmp = L_deposit_h( BASOP_Util_Divide3232_Scale( max_value, W_extract_l( lfe_abs_sum ), &q_tmp ) ); + // tmp = L_shl( tmp, sub( q_lfe_abs_sum, sub( 15, q_tmp ) ) ); /* Q0 (max_value / lfe_abs_sum) */ + q_tmp = sub( Q16, sub( q_lfe_abs_sum, sub( 15, q_tmp ) ) ); /* log2_f(max_value / lfe_abs_sum) -> Q25 */ tmp = BASOP_Util_Log2( tmp ); - tmp = L_add( tmp, 1040187392 ) /* (31<<25) -> 1040187392 */; /* Q25 */ + tmp = L_add( tmp, L_shl( sub( Q31, q_tmp ), Q25 ) ) /* (31<<25) -> 1040187392 */; /* Q25 */ /* IVAS_LFE_SHIFTS_PER_DOUBLE * log2_f(max_value / lfe_abs_sum) */ tmp = Mpy_32_16_1( tmp, IVAS_LFE_SHIFTS_PER_DOUBLE ); /* 25-15 -> Q10 */ diff --git a/lib_enc/ivas_mct_core_enc.c b/lib_enc/ivas_mct_core_enc.c index f5f179778178b46b5b42d2037eecd9d79b3c7b87..52fd554d8397fb0e916eb1627f4f09d662593c49 100644 --- a/lib_enc/ivas_mct_core_enc.c +++ b/lib_enc/ivas_mct_core_enc.c @@ -247,7 +247,7 @@ void ivas_mct_core_enc_fx( Word32 inv_spectrum_long_fx[MCT_MAX_CHANNELS][L_FRAME48k]; /* quantized MDCT spectrum, inv ms mask mdst spectrum, scratch for MS spectra in the MS decision */ Word16 total_side_bits; Word16 chBitRatios[MCT_MAX_CHANNELS]; - Word16 q_powSpec[MCT_MAX_CHANNELS], q_spec, q_origSpec, tmp_s; + Word16 q_powSpec[MCT_MAX_CHANNELS], q_powerSpecMsInv[MCT_MAX_CHANNELS], q_spec, q_origSpec, tmp_s; Word16 tmp_q_powSpec[L_FRAME48k], tmp_q_powSpecInv[L_FRAME48k], *tmp_q_psi[2]; Word64 W_tmp; Encoder_State *sts[MCT_MAX_CHANNELS]; @@ -437,7 +437,7 @@ void ivas_mct_core_enc_fx( FOR( i = 0; i < L_subframeTCX; i++ ) { W_tmp = W_mac_32_32( W_mult_32_32( mdst_spectrum_fx[ch][n][i], mdst_spectrum_fx[ch][n][i] ), sts[ch]->hTcxEnc->spectrum_fx[n][i], sts[ch]->hTcxEnc->spectrum_fx[n][i] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpec_fx[ch][( i + ( n * L_subframeTCX ) )] = W_extract_h( W_tmp ); tmp_q_powSpec[( i + ( n * L_subframeTCX ) )] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -445,7 +445,7 @@ void ivas_mct_core_enc_fx( move16(); W_tmp = W_mac_32_32( W_mult_32_32( inv_mdst_spectrum_fx[ch][n][i], inv_mdst_spectrum_fx[ch][n][i] ), inv_spectrum_fx[ch][n][i], inv_spectrum_fx[ch][n][i] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpecMsInv_fx[ch][n][i] = W_extract_h( W_tmp ); tmp_q_psi[n][i] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -459,7 +459,7 @@ void ivas_mct_core_enc_fx( { /* power spectrum: MDCT^2 + MDST^2 */ W_tmp = W_mult_32_32( inv_spectrum_fx[ch][n][0], inv_spectrum_fx[ch][n][0] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpecMsInv_fx[ch][n][0] = W_extract_h( W_tmp ); tmp_q_psi[n][0] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -471,7 +471,7 @@ void ivas_mct_core_enc_fx( mdst_fx = L_sub( inv_spectrum_fx[ch][n][i + 1], inv_spectrum_fx[ch][n][i - 1] ); /* An MDST estimate */ W_tmp = W_mac_32_32( W_mult_32_32( mdst_fx, mdst_fx ), inv_spectrum_fx[ch][n][i], inv_spectrum_fx[ch][n][i] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpecMsInv_fx[ch][n][i] = W_extract_h( W_tmp ); tmp_q_psi[n][i] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -480,7 +480,7 @@ void ivas_mct_core_enc_fx( } W_tmp = W_mult_32_32( inv_spectrum_fx[ch][n][L_subframeTCX - 1], inv_spectrum_fx[ch][n][L_subframeTCX - 1] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpecMsInv_fx[ch][n][L_subframeTCX - 1] = W_extract_h( W_tmp ); tmp_q_psi[n][L_subframeTCX - 1] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -490,7 +490,7 @@ void ivas_mct_core_enc_fx( /* power spectrum: MDCT^2 + MDST^2 */ W_tmp = W_mult_32_32( sts[ch]->hTcxEnc->spectrum_fx[n][0], sts[ch]->hTcxEnc->spectrum_fx[n][0] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpec_fx[ch][n * L_subframeTCX] = W_extract_h( W_tmp ); tmp_q_powSpec[n * L_subframeTCX] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -513,7 +513,7 @@ void ivas_mct_core_enc_fx( } W_tmp = W_mult_32_32( sts[ch]->hTcxEnc->spectrum_fx[n][L_subframeTCX - 1], sts[ch]->hTcxEnc->spectrum_fx[n][L_subframeTCX - 1] ); - tmp_s = sub( W_norm( W_tmp ), 1 ); + tmp_s = W_norm( W_tmp ); W_tmp = W_shl( W_tmp, tmp_s ); powerSpec_fx[ch][( ( L_subframeTCX - 1 ) + ( n * L_subframeTCX ) )] = W_extract_h( W_tmp ); tmp_q_powSpec[( ( L_subframeTCX - 1 ) + ( n * L_subframeTCX ) )] = sub( add( imult1616( q_spec, 2 ), tmp_s ), 31 ); // Q = 2 * q_spec + 1 + tmp_s - 32 @@ -524,31 +524,31 @@ void ivas_mct_core_enc_fx( /* Aligning the Q-factors */ { + q_powSpec[ch] = Q31; + move16(); + q_powerSpecMsInv[ch] = Q31; + move16(); /* NOTE: This logic has been added because using a constant headroom while computing `powSpec` and `powSpecMsInv` leads to significant precision loss, which results in poor quality. */ FOR( i = 0; i < L_FRAME48k; i++ ) { - if ( powerSpec_fx[ch][i] == 0 ) + IF( powerSpec_fx[ch][i] != 0 ) { - tmp_q_powSpec[i] = 63; + q_powSpec[ch] = s_min( q_powSpec[ch], add( tmp_q_powSpec[i], norm_l( powerSpec_fx[ch][i] ) ) ); move16(); } - if ( powerSpecMsInv_fx[ch][0][i] == 0 ) + IF( powerSpecMsInv_fx[ch][0][i] != 0 ) { - tmp_q_powSpecInv[i] = 63; + q_powerSpecMsInv[ch] = s_min( q_powerSpecMsInv[ch], add( tmp_q_powSpecInv[i], norm_l( powerSpecMsInv_fx[ch][0][i] ) ) ); move16(); } } - minimum_s( tmp_q_powSpec, L_FRAME48k, &q_powSpec[ch] ); - minimum_s( tmp_q_powSpecInv, L_FRAME48k, &tmp_s ); - q_powSpec[ch] = s_min( q_powSpec[ch], tmp_s ); - move16(); FOR( n = 0; n < nSubframes; n++ ) { FOR( i = 0; i < L_subframeTCX; i++ ) { - powerSpecMsInv_fx[ch][n][i] = L_shr( powerSpecMsInv_fx[ch][n][i], sub( tmp_q_psi[n][i], q_powSpec[ch] ) ); - powerSpec_fx[ch][( i + ( n * L_subframeTCX ) )] = L_shr( powerSpec_fx[ch][( i + ( n * L_subframeTCX ) )], sub( tmp_q_powSpec[i], q_powSpec[ch] ) ); + powerSpecMsInv_fx[ch][n][i] = L_shr( powerSpecMsInv_fx[ch][n][i], sub( tmp_q_psi[n][i], q_powerSpecMsInv[ch] ) ); + powerSpec_fx[ch][( i + ( n * L_subframeTCX ) )] = L_shr( powerSpec_fx[ch][( i + ( n * L_subframeTCX ) )], sub( tmp_q_powSpec[i + ( n * L_subframeTCX )], q_powSpec[ch] ) ); move32(); move32(); } @@ -606,7 +606,7 @@ void ivas_mct_core_enc_fx( { IF( hMCT->currBlockDataCnt > 0 ) { - mctStereoIGF_enc_fx( hMCT, sts, orig_spectrum_fx, q_origSpec, powerSpec_fx, powerSpecMsInv_fx, q_powSpec, inv_spectrum_fx, sp_aud_decision0 ); + mctStereoIGF_enc_fx( hMCT, sts, orig_spectrum_fx, q_origSpec, powerSpec_fx, q_powSpec, powerSpecMsInv_fx, q_powerSpecMsInv, inv_spectrum_fx, sp_aud_decision0 ); } ELSE { diff --git a/lib_enc/ivas_mct_enc_mct.c b/lib_enc/ivas_mct_enc_mct.c index 4b84ea9cb607b05c7e6cd50ab7a198d9f8795b11..654c00db47d71332731b8986f75ed08fc5c70fa0 100644 --- a/lib_enc/ivas_mct_enc_mct.c +++ b/lib_enc/ivas_mct_enc_mct.c @@ -981,15 +981,16 @@ void write_mct_bitstream_fx( * IGF analysis of channels after MCT processing *--------------------------------------------------------------------*/ void mctStereoIGF_enc_fx( - MCT_ENC_HANDLE hMCT, /* i/o: MCT encoder structure */ - Encoder_State **sts, /* i/o: encoder state structure */ - Word32 *orig_spectrum_fx[MCT_MAX_CHANNELS][2], /* i : MDCT spectrum for ITF */ - Word16 q_origSpec, /* i : Q for MDCT spectrum */ - Word32 powerSpec_fx[MCT_MAX_CHANNELS][L_FRAME48k], /* i/o: MDCT^2 + MDST^2 spectrum,or estimate*/ - Word32 *powerSpecMsInv_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : same as above but for inverse spect.*/ - Word16 q_powerSpec[MCT_MAX_CHANNELS], /* i : Q for powSpec_fx and powSpecMsInv_fx*/ - Word32 *inv_spectrum_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : inverse spectrum */ - const Word16 sp_aud_decision0[MCT_MAX_CHANNELS] /* i : speech audio decision */ + MCT_ENC_HANDLE hMCT, /* i/o: MCT encoder structure */ + Encoder_State **sts, /* i/o: encoder state structure */ + Word32 *orig_spectrum_fx[MCT_MAX_CHANNELS][2], /* i : MDCT spectrum for ITF */ + Word16 q_origSpec, /* i : Q for MDCT spectrum */ + Word32 powerSpec_fx[MCT_MAX_CHANNELS][L_FRAME48k], /* i/o: MDCT^2 + MDST^2 spectrum,or estimate */ + Word16 q_powerSpec[MCT_MAX_CHANNELS], /* i : Q for powSpec_fx */ + Word32 *powerSpecMsInv_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : same as powerSpec_fx but for inverse spect.*/ + Word16 q_powerSpecMsInv[MCT_MAX_CHANNELS], /* i : Q for powSpecMsInv_fx */ + Word32 *inv_spectrum_fx[MCT_MAX_CHANNELS][NB_DIV], /* i : inverse spectrum */ + const Word16 sp_aud_decision0[MCT_MAX_CHANNELS] /* i : speech audio decision */ ) { Word32 *p_powerSpecMsInv_fx[CPE_CHANNELS][NB_DIV]; @@ -999,6 +1000,7 @@ void mctStereoIGF_enc_fx( Word16 b, nSubframes, L_subframeTCX; Word16 p_ch[2], n, ch, ch1, ch2, s = 31; + Word16 q_pSI_ch[2]; Word16 q_pS_ch[2]; Encoder_State *p_st[NB_DIV]; Encoder_State *st; @@ -1062,7 +1064,9 @@ void mctStereoIGF_enc_fx( p_inv_spectrum_fx[0][n] = inv_spectrum_fx[ch1][n]; p_inv_spectrum_fx[1][n] = inv_spectrum_fx[ch2][n]; q_pS_ch[0] = q_powerSpec[ch1]; + q_pSI_ch[0] = q_powerSpecMsInv[ch1]; q_pS_ch[1] = q_powerSpec[ch2]; + q_pSI_ch[1] = q_powerSpecMsInv[ch2]; move16(); move16(); @@ -1075,6 +1079,7 @@ void mctStereoIGF_enc_fx( { s = s_min( s, sub( 31, p_st[ch]->hTcxEnc->spectrum_e[n] ) ); s = s_min( s, q_pS_ch[ch] ); + s = s_min( s, q_pSI_ch[ch] ); } FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { @@ -1085,7 +1090,7 @@ void mctStereoIGF_enc_fx( move16(); move16(); - scale_sig32( p_powerSpecMsInv_fx[ch][0], L_FRAME48k, sub( s, q_pS_ch[ch] ) ); + scale_sig32( p_powerSpecMsInv_fx[ch][0], L_FRAME48k, sub( s, q_pSI_ch[ch] ) ); scale_sig32( &p_powerSpec_fx[ch][0], sts[ch]->hTcxEnc->L_frameTCX, sub( s, q_pS_ch[ch] ) ); q_powerSpec[ch1] = s; q_powerSpec[ch2] = s; diff --git a/lib_enc/tcx_utils_enc_fx.c b/lib_enc/tcx_utils_enc_fx.c index 1e35c78c4c590651080c7e4bb8877217efafdb25..27c8e6f5f2356a3e13762e5622e94e74c9a2a65c 100644 --- a/lib_enc/tcx_utils_enc_fx.c +++ b/lib_enc/tcx_utils_enc_fx.c @@ -3842,7 +3842,7 @@ void ProcessIGF_ivas_fx( ITF_Detect_ivas_fx( hIGFEnc->spec_be_igf, hIGFEnc->infoStartLine, hIGFEnc->infoStopLine, 8 /*maxOrder*/, A, &q_A, &predictionGain, &curr_order, sub( 31, hIGFEnc->spec_be_igf_e ) ); test(); - IF( LT_32( L_deposit_l( hIGFEnc->tns_predictionGain ), 9646899 /* 1.15 in Q23 */ ) && LT_32( L_deposit_l( predictionGain ), 9646899 /* 1.15 in Q23 */ ) ) + IF( LT_16( hIGFEnc->tns_predictionGain, ONE_POINT_ONE_FIVE_Q7 ) && LT_16( predictionGain, ONE_POINT_ONE_FIVE_Q7 ) ) { hIGFEnc->flatteningTrigger = 1; move16(); @@ -3949,7 +3949,7 @@ void ProcessStereoIGF_fx( ITF_Detect_ivas_fx( hIGFEnc[ch]->spec_be_igf, hIGFEnc[ch]->infoStartLine, hIGFEnc[ch]->infoStopLine, 8 /*maxOrder*/, A, &Q_A, &predictionGain, &curr_order, sub( 31, hIGFEnc[ch]->spec_be_igf_e ) ); test(); - hIGFEnc[ch]->flatteningTrigger = LT_32( hIGFEnc[ch]->tns_predictionGain, ONE_POINT_ONE_FIVE_Q23 ) && LT_32( predictionGain, ONE_POINT_ONE_FIVE_Q7 ); + hIGFEnc[ch]->flatteningTrigger = LT_16( hIGFEnc[ch]->tns_predictionGain, ONE_POINT_ONE_FIVE_Q7 ) && LT_16( predictionGain, ONE_POINT_ONE_FIVE_Q7 ); move16(); hIGFEnc[ch]->infoTotalBitsPerFrameWritten = 0;