From 450d9a6113ce95ad2fe346972f3f7afa63369dd9 Mon Sep 17 00:00:00 2001 From: Markus Multrus Date: Sun, 1 Jun 2025 19:59:46 +0200 Subject: [PATCH] separate changes on MDCT core dec from branch cldfb_opt --- lib_com/ivas_prot_fx.h | 20 +++++-- lib_com/options.h | 2 + lib_dec/ivas_mct_core_dec_fx.c | 71 +++++++++++++++++++++++- lib_dec/ivas_mct_dec_mct_fx_fx.c | 33 +++++++++-- lib_dec/ivas_mdct_core_dec_fx.c | 4 ++ lib_dec/ivas_stereo_mdct_core_dec_fx.c | 6 ++ lib_dec/ivas_stereo_mdct_stereo_dec_fx.c | 20 ++++++- 7 files changed, 141 insertions(+), 15 deletions(-) diff --git a/lib_com/ivas_prot_fx.h b/lib_com/ivas_prot_fx.h index d9ca855be..a31cdf338 100644 --- a/lib_com/ivas_prot_fx.h +++ b/lib_com/ivas_prot_fx.h @@ -1092,10 +1092,14 @@ void ivas_mct_dec_mct_fx( ); void apply_MCT_dec_fx( - MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */ - Decoder_State **sts, /* i/o: decoder state structure */ - Word32 *x[MCT_MAX_CHANNELS][NB_DIV], /* i/o: decoded and dequan. spect. input to MCT */ - Word16 q_x[MCT_MAX_CHANNELS] ); + MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */ + Decoder_State **sts, /* i/o: decoder state structure */ + Word32 *x[MCT_MAX_CHANNELS][NB_DIV] /* i/o: decoded and dequan. spect. input to MCT */ +#ifndef OPT_SBA_DEC_PATH + , + Word16 q_x[MCT_MAX_CHANNELS] +#endif /* OPT_SBA_DEC_PATH */ +); void mctStereoIGF_dec_fx( MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */ @@ -1129,9 +1133,13 @@ void stereo_decoder_tcx_fx( const Word16 mct_on, /* i : flag mct block (1) or stereo (0) */ const Word16 last_core_l, /* i : last core for left channel */ const Word16 last_core_r, /* i : last core for right channel */ - const Word16 tmp_plc_upmix, /* i : indicates temp upmix for PLC decision */ + const Word16 tmp_plc_upmix /* i : indicates temp upmix for PLC decision */ +#ifndef OPT_SBA_DEC_PATH + , Word16 *q_x_ch2, - Word16 *q_x_ch1 ); + Word16 *q_x_ch1 +#endif /* OPT_SBA_DEC_PATH */ +); void v_multc_acc_32_16( const Word32 x[], /* i : Input vector */ diff --git a/lib_com/options.h b/lib_com/options.h index a5d5a4846..f685c5541 100644 --- a/lib_com/options.h +++ b/lib_com/options.h @@ -106,4 +106,6 @@ #define FIX_1713_EXP /* VA: proposed correction to exp in ic-BWE*/ +#define OPT_SBA_DEC_PATH + #endif diff --git a/lib_dec/ivas_mct_core_dec_fx.c b/lib_dec/ivas_mct_core_dec_fx.c index b42eb534b..08359826a 100644 --- a/lib_dec/ivas_mct_core_dec_fx.c +++ b/lib_dec/ivas_mct_core_dec_fx.c @@ -198,7 +198,10 @@ void ivas_mct_core_dec( Word16 tcx_offset; Word16 tcx_offsetFB; Word16 left_rect; - Word16 L_spec, scale, tmp1, tmp2; + Word16 L_spec; +#ifndef OPT_SBA_DEC_PATH + Word16 scale, tmp1, tmp2; +#endif /* OPT_SBA_DEC_PATH */ #ifdef DEBUG_MCT float nrg[MCT_MAX_CHANNELS]; #endif @@ -218,11 +221,18 @@ void ivas_mct_core_dec( move16(); FOR( cpe_id = 0; cpe_id < nCPE; cpe_id++ ) { +#ifndef OPT_SBA_DEC_PATH FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { sts[i] = hCPE[cpe_id]->hCoreCoder[ch]; i = add( i, 1 ); } +#else /* OPT_SBA_DEC_PATH */ + FOR( ch = 0; ch < CPE_CHANNELS; ( ch++, i++ ) ) + { + sts[i] = hCPE[cpe_id]->hCoreCoder[ch]; + } +#endif /* OPT_SBA_DEC_PATH */ } i = 0; @@ -268,7 +278,61 @@ void ivas_mct_core_dec( { CONTINUE; } +#ifdef OPT_SBA_DEC_PATH + IF( EQ_16( st->core, TCX_10_CORE ) ) + { + Word16 L_frame_2 = shr( st->L_frame, 1 ); + Word16 L_frameTCX_2 = shr( st->hTcxDec->L_frameTCX, 1 ); + nSubframes = NB_DIV; // assuming NB_DIV is 2. Below logic needs to be modified if NB_DIV changes. + move16(); + FOR( k = 0; k < nSubframes; k++ ) + { + L_spec = shr( st->hTcxCfg->tcx_coded_lines, 1 ); + move32(); + init_tcx_info_fx( st, L_frame_2, L_frameTCX_2, k, bfi, &tcx_offset, &tcx_offsetFB, &L_frame, &L_frameTCX, &left_rect, &L_spec ); + + Word16 x_e, x_len; + /* mono or dual mono IGF decoding */ + x_e = sub( 31, q_x[ch] ); + decoder_tcx_IGF_mono_fx( st, x_fx[ch][k], &x_e, &x_len, L_frame, left_rect, bfi, k ); + q_x[ch] = sub( 31 - 12, x_e ); // Shift to bring x to Q12 + move16(); + FOR( i = 0; i < x_len; i++ ) + { + x_fx[ch][k][i] = L_shr( x_fx[ch][k][i], q_x[ch] ); + move32(); + } + q_x[ch] = Q12; + move16(); + } + } + ELSE + { + nSubframes = 1; + move16(); + { + L_spec = st->hTcxCfg->tcx_coded_lines; + move32(); + + init_tcx_info_fx( st, st->L_frame, st->hTcxDec->L_frameTCX, 0, bfi, &tcx_offset, &tcx_offsetFB, &L_frame, &L_frameTCX, &left_rect, &L_spec ); + + Word16 x_e, x_len; + /* mono or dual mono IGF decoding */ + x_e = sub( 31, q_x[ch] ); + decoder_tcx_IGF_mono_fx( st, x_fx[ch][0], &x_e, &x_len, L_frame, left_rect, bfi, 0 ); + q_x[ch] = sub( 31 - 12, x_e ); // Shift to bring x to Q12 + move16(); + FOR( i = 0; i < x_len; i++ ) + { + x_fx[ch][0][i] = L_shr( x_fx[ch][0][i], q_x[ch] ); + move32(); + } + q_x[ch] = Q12; + move16(); + } + } +#else /* OPT_SBA_DEC_PATH */ IF( EQ_16( st->core, TCX_10_CORE ) ) { nSubframes = NB_DIV; @@ -304,6 +368,7 @@ void ivas_mct_core_dec( q_x[ch] = Q12; move16(); } +#endif /* OPT_SBA_DEC_PATH */ } } ELSE @@ -318,6 +383,9 @@ void ivas_mct_core_dec( * MCT processing *--------------------------------------------------------------------------------*/ +#ifdef OPT_SBA_DEC_PATH + apply_MCT_dec_fx( hMCT, sts, x_fx ); +#else /* OPT_SBA_DEC_PATH */ apply_MCT_dec_fx( hMCT, sts, x_fx, q_x ); FOR( ch = 0; ch < nChannels; ch++ ) { @@ -331,6 +399,7 @@ void ivas_mct_core_dec( q_x[ch] = Q12; move16(); } +#endif /* OPT_SBA_DEC_PATH */ } pop_wmops(); diff --git a/lib_dec/ivas_mct_dec_mct_fx_fx.c b/lib_dec/ivas_mct_dec_mct_fx_fx.c index 0f20c1ec0..0d998b355 100644 --- a/lib_dec/ivas_mct_dec_mct_fx_fx.c +++ b/lib_dec/ivas_mct_dec_mct_fx_fx.c @@ -165,8 +165,12 @@ void ivas_mct_dec_mct_fx( static void applyGlobalILD_fx( Decoder_State **sts, MCT_DEC_HANDLE hMCT, - Word32 *x[MCT_MAX_CHANNELS][NB_DIV], - Word16 q_x[MCT_MAX_CHANNELS] ) + Word32 *x[MCT_MAX_CHANNELS][NB_DIV] +#ifndef OPT_SBA_DEC_PATH + , + Word16 q_x[MCT_MAX_CHANNELS] +#endif /* OPT_SBA_DEC_PATH */ +) { Word16 ch, k; Word16 nSubframes, L_subframeTCX; @@ -207,9 +211,14 @@ static void applyGlobalILD_fx( FOR( k = 0; k < nSubframes; k++ ) { v_multc_fixed( x[ch][k], qratio, x[ch][k], L_subframeTCX ); // Qx - 5 + q_qratio +#ifdef OPT_SBA_DEC_PATH + scale_sig32( x[ch][k], L_subframeTCX, sub( 5, q_qratio ) ); +#endif /* OPT_SBA_DEC_PATH */ } +#ifndef OPT_SBA_DEC_PATH q_x[ch] = add( sub( q_x[ch], 5 ), q_qratio ); move16(); +#endif /* OPT_SBA_DEC_PATH */ } ELSE { @@ -228,10 +237,14 @@ static void applyGlobalILD_fx( *----------------------------------------------------------*/ void apply_MCT_dec_fx( - MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */ - Decoder_State **sts, /* i/o: decoder state structure */ - Word32 *x[MCT_MAX_CHANNELS][NB_DIV], /* i/o: decoded and dequan. spect. input to MCT */ - Word16 q_x[MCT_MAX_CHANNELS] ) + MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */ + Decoder_State **sts, /* i/o: decoder state structure */ + Word32 *x[MCT_MAX_CHANNELS][NB_DIV] /* i/o: decoded and dequan. spect. input to MCT */ +#ifndef OPT_SBA_DEC_PATH + , + Word16 q_x[MCT_MAX_CHANNELS] +#endif /* OPT_SBA_DEC_PATH */ +) { Word16 pair; MCT_DEC_BLOCK_DATA_HANDLE hBlock; @@ -240,10 +253,18 @@ void apply_MCT_dec_fx( { hBlock = hMCT->hBlockData[pair]; +#ifndef OPT_SBA_DEC_PATH stereo_decoder_tcx_fx( hBlock->hStereoMdct, hBlock->mask, &x[hBlock->ch2][0], &x[hBlock->ch1][0], &x[hBlock->ch2][0], hBlock->hStereoMdct->mdct_stereo_mode, sts[hBlock->ch1]->core, sts[hBlock->ch2]->core, sts[0]->igf, sts[0]->hTcxDec->L_frameTCX, sts[1]->hTcxDec->L_frameTCX, 1, TCX_20_CORE, TCX_20_CORE, 0, &q_x[hBlock->ch2], &q_x[hBlock->ch1] ); +#else /* OPT_SBA_DEC_PATH */ + stereo_decoder_tcx_fx( hBlock->hStereoMdct, hBlock->mask, &x[hBlock->ch2][0], &x[hBlock->ch1][0], &x[hBlock->ch2][0], hBlock->hStereoMdct->mdct_stereo_mode, sts[hBlock->ch1]->core, sts[hBlock->ch2]->core, sts[0]->igf, sts[0]->hTcxDec->L_frameTCX, sts[1]->hTcxDec->L_frameTCX, 1, TCX_20_CORE, TCX_20_CORE, 0 ); +#endif /* OPT_SBA_DEC_PATH */ } +#ifndef OPT_SBA_DEC_PATH applyGlobalILD_fx( sts, hMCT, x, q_x ); +#else /* OPT_SBA_DEC_PATH */ + applyGlobalILD_fx( sts, hMCT, x ); +#endif /* OPT_SBA_DEC_PATH */ return; } diff --git a/lib_dec/ivas_mdct_core_dec_fx.c b/lib_dec/ivas_mdct_core_dec_fx.c index fabebede4..4e90792fb 100644 --- a/lib_dec/ivas_mdct_core_dec_fx.c +++ b/lib_dec/ivas_mdct_core_dec_fx.c @@ -737,7 +737,11 @@ void ivas_mdct_core_invQ_fx( q_r = sub( 30, sts[1]->hTonalMDCTConc->lastBlockData.spectralData_exp ); /* both input in same Q */ +#ifndef OPT_SBA_DEC_PATH stereo_decoder_tcx_fx( hCPE->hStereoMdct, ms_mask, x_0[1], &spectralData_tmp[0], &spectralData_tmp[1], &hCPE->hStereoMdct->mdct_stereo_mode[0], sts[0]->core, sts[1]->core, sts[0]->igf, L_frameTCX[0], L_frameTCX[1], 0, sts[0]->last_core, sts[1]->last_core, 1, &q_r, &q_l ); +#else /* OPT_SBA_DEC_PATH */ + stereo_decoder_tcx_fx( hCPE->hStereoMdct, ms_mask, x_0[1], &spectralData_tmp[0], &spectralData_tmp[1], &hCPE->hStereoMdct->mdct_stereo_mode[0], sts[0]->core, sts[1]->core, sts[0]->igf, L_frameTCX[0], L_frameTCX[1], 0, sts[0]->last_core, sts[1]->last_core, 1 ); +#endif /* OPT_SBA_DEC_PATH */ Copy_Scale_sig_32_16( spectralData_tmp[0], sts[0]->hTonalMDCTConc->lastBlockData.spectralData, L_frameTCX[0], -15 ); // q_l - 15 Copy_Scale_sig_32_16( spectralData_tmp[1], sts[1]->hTonalMDCTConc->lastBlockData.spectralData, L_frameTCX[1], -15 ); // q_r - 15 diff --git a/lib_dec/ivas_stereo_mdct_core_dec_fx.c b/lib_dec/ivas_stereo_mdct_core_dec_fx.c index 89cb92a4d..85ae00c90 100644 --- a/lib_dec/ivas_stereo_mdct_core_dec_fx.c +++ b/lib_dec/ivas_stereo_mdct_core_dec_fx.c @@ -465,9 +465,11 @@ void stereo_mdct_core_dec_fx( test(); IF( ( !bfi || ( sts[0]->core != ACELP_CORE || sts[1]->core != ACELP_CORE ) ) && !hCPE->hStereoMdct->isSBAStereoMode ) { +#ifndef OPT_SBA_DEC_PATH Word16 q_x_1 = 11, q_x_0 = 11; move16(); move16(); +#endif /* OPT_SBA_DEC_PATH */ FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { @@ -477,7 +479,11 @@ void stereo_mdct_core_dec_fx( move16(); } } +#ifdef OPT_SBA_DEC_PATH + stereo_decoder_tcx_fx( hCPE->hStereoMdct, ms_mask, x_0_fx[1], x_fx[0], x_fx[1], &hCPE->hStereoMdct->mdct_stereo_mode[0], sts[0]->core, sts[1]->core, sts[0]->igf, L_frameTCX[0], L_frameTCX[1], 0, sts[0]->last_core, sts[1]->last_core, 0 ); +#else /* OPT_SBA_DEC_PATH */ stereo_decoder_tcx_fx( hCPE->hStereoMdct, ms_mask, x_0_fx[1], x_fx[0], x_fx[1], &hCPE->hStereoMdct->mdct_stereo_mode[0], sts[0]->core, sts[1]->core, sts[0]->igf, L_frameTCX[0], L_frameTCX[1], 0, sts[0]->last_core, sts[1]->last_core, 0, &q_x_1, &q_x_0 ); +#endif /* OPT_SBA_DEC_PATH */ } ivas_mdct_core_tns_ns_fx( hCPE, fUseTns, tnsData, x_fx, Aq_fx_32, 0, x_e ); diff --git a/lib_dec/ivas_stereo_mdct_stereo_dec_fx.c b/lib_dec/ivas_stereo_mdct_stereo_dec_fx.c index 8842a6792..492060a67 100644 --- a/lib_dec/ivas_stereo_mdct_stereo_dec_fx.c +++ b/lib_dec/ivas_stereo_mdct_stereo_dec_fx.c @@ -313,9 +313,13 @@ void stereo_decoder_tcx_fx( const Word16 mct_on, /* i : flag mct block (1) or stereo (0) Q0*/ const Word16 last_core_l, /* i : last core for left channel Q0*/ const Word16 last_core_r, /* i : last core for right channel Q0*/ - const Word16 tmp_plc_upmix, /* i : indicates temp upmix for PLC decision Q0*/ + const Word16 tmp_plc_upmix /* i : indicates temp upmix for PLC decision Q0*/ +#ifndef OPT_SBA_DEC_PATH + , Word16 *q_x_ch2, - Word16 *q_x_ch1 ) + Word16 *q_x_ch1 +#endif /* OPT_SBA_DEC_PATH */ +) { Word16 i, k, sfb, nSubframes; STEREO_MDCT_BAND_PARAMETERS *sfbConf = NULL; @@ -361,10 +365,12 @@ void stereo_decoder_tcx_fx( } } inverseMS_fx( sfbConf->sfbOffset[sfbConf->nBandsStereoCore], spec_l[k], spec_r[k], SQRT2_OVER_2_FIXED ); +#ifndef OPT_SBA_DEC_PATH *q_x_ch2 = *q_x_ch2; move16(); *q_x_ch1 = *q_x_ch1; move16(); +#endif /* OPT_SBA_DEC_PATH */ } ELSE IF( EQ_16( mdct_stereo_mode[k], SMDCT_BW_MS ) ) { @@ -381,10 +387,12 @@ void stereo_decoder_tcx_fx( } } inverseBwMS_fx( sfbConf->sfbOffset[sfb], sfbConf->sfbOffset[sfb + 1], spec_l[k], spec_r[k], SQRT2_OVER_2_FIXED ); +#ifndef OPT_SBA_DEC_PATH *q_x_ch2 = *q_x_ch2; move16(); *q_x_ch1 = *q_x_ch1; move16(); +#endif /* OPT_SBA_DEC_PATH */ } } } @@ -402,10 +410,12 @@ void stereo_decoder_tcx_fx( } } inverseMS_fx( sub( sfbConf->sfbOffset[sfbConf->sfbCnt], sfbConf->sfbOffset[sfbConf->nBandsStereoCore] ), &spec_l[k][sfbConf->sfbOffset[sfbConf->nBandsStereoCore]], &spec_r[k][sfbConf->sfbOffset[sfbConf->nBandsStereoCore]], SQRT2_OVER_2_FIXED ); +#ifndef OPT_SBA_DEC_PATH *q_x_ch2 = *q_x_ch2; move16(); *q_x_ch1 = *q_x_ch1; move16(); +#endif /* OPT_SBA_DEC_PATH */ } ELSE IF( EQ_16( hStereoMdct->IGFStereoMode[k], SMDCT_BW_MS ) ) { @@ -422,10 +432,12 @@ void stereo_decoder_tcx_fx( } } inverseBwMS_fx( sfbConf->sfbOffset[sfb], sfbConf->sfbOffset[sfb + 1], spec_l[k], spec_r[k], SQRT2_OVER_2_FIXED ); +#ifndef OPT_SBA_DEC_PATH *q_x_ch2 = *q_x_ch2; move16(); *q_x_ch1 = *q_x_ch1; move16(); +#endif /* OPT_SBA_DEC_PATH */ } } } @@ -483,8 +495,10 @@ void stereo_decoder_tcx_fx( nrgRatio = L_shl( nrgRatio, shift ); /* Q26 + shift */ v_multc_fixed( spec_r[k], nrgRatio, spec_r[k], L_frameTCX_r ); /* spec_r will be in Qx + shift - Q5 */ Scale_sig32( spec_r[k], L_frameTCX_r, sub( 5, shift ) ); /* Qx */ +#ifndef OPT_SBA_DEC_PATH *q_x_ch2 = *q_x_ch2; move16(); +#endif /* OPT_SBA_DEC_PATH */ } ELSE IF( ( LT_32( nrgRatio, ONE_IN_Q26 ) ) && ( LT_16( k, tmp2 ) ) ) { @@ -492,8 +506,10 @@ void stereo_decoder_tcx_fx( shift = sub( 5, tmp_e ); v_multc_fixed( spec_l[k], inv_nrgRatio, spec_l[k], L_frameTCX_l ); /* spec_r will be in Qx - tmp_e */ Scale_sig32( spec_l[k], L_frameTCX_l, sub( 5, shift ) ); /* Qx */ +#ifndef OPT_SBA_DEC_PATH *q_x_ch1 = *q_x_ch1; move16(); +#endif /* OPT_SBA_DEC_PATH */ } } } /* for k */ -- GitLab