diff options
author | Felicia Lim <flim@google.com> | 2017-07-05 17:36:56 -0700 |
---|---|---|
committer | Felicia Lim <flim@google.com> | 2017-07-07 09:42:55 -0700 |
commit | 0c2090c324e4f2ba2a8621c8b083559bab74c7c5 (patch) | |
tree | fea1ab0038bc4102569d1ab4ee57a0f973895570 /silk/x86 | |
parent | a7703b70699299f078a189e19b6915120cded732 (diff) | |
download | libopus-0c2090c324e4f2ba2a8621c8b083559bab74c7c5.tar.gz |
Opus 1.2.1android-o-iot-preview-5o-iot-preview-5
Change-Id: I551f1de5c5e121ac1275334e67c7e0f96ab18114
Test: - verified builds for arm*/mips*/x86*
- checked functionality using an emulator and stagefright
Diffstat (limited to 'silk/x86')
-rw-r--r-- | silk/x86/NSQ_del_dec_sse.c | 18 | ||||
-rw-r--r-- | silk/x86/NSQ_sse.c | 3 | ||||
-rw-r--r-- | silk/x86/main_sse.h | 45 | ||||
-rw-r--r-- | silk/x86/x86_silk_map.c | 26 |
4 files changed, 27 insertions, 65 deletions
diff --git a/silk/x86/NSQ_del_dec_sse.c b/silk/x86/NSQ_del_dec_sse.c index 21d4a8bc..c5212bee 100644 --- a/silk/x86/NSQ_del_dec_sse.c +++ b/silk/x86/NSQ_del_dec_sse.c @@ -107,12 +107,12 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( opus_int predictLPCOrder, /* I Prediction filter order */ opus_int warping_Q16, /* I */ opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ - opus_int *smpl_buf_idx, /* I Index to newest samples in buffers */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ opus_int decisionDelay /* I */ ); void silk_NSQ_del_dec_sse4_1( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -234,7 +234,8 @@ void silk_NSQ_del_dec_sse4_1( psDD = &psDelDec[ Winner_ind ]; last_smple_idx = smpl_buf_idx + decisionDelay; for( i = 0; i < decisionDelay; i++ ) { - last_smple_idx = ( last_smple_idx - 1 ) & DECISION_DELAY_MASK; + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gains_Q16[ 1 ] ), 14 ) ); @@ -285,7 +286,8 @@ void silk_NSQ_del_dec_sse4_1( last_smple_idx = smpl_buf_idx + decisionDelay; Gain_Q10 = silk_RSHIFT32( Gains_Q16[ psEncC->nb_subfr - 1 ], 6 ); for( i = 0; i < decisionDelay; i++ ) { - last_smple_idx = ( last_smple_idx - 1 ) & DECISION_DELAY_MASK; + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gain_Q10 ), 8 ) ); @@ -299,7 +301,6 @@ void silk_NSQ_del_dec_sse4_1( NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; /* Save quantized speech signal */ - /* DEBUG_STORE_DATA( enc.pcm, &NSQ->xq[psEncC->ltp_mem_length], psEncC->frame_length * sizeof( opus_int16 ) ) */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); RESTORE_STACK; @@ -333,7 +334,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( opus_int predictLPCOrder, /* I Prediction filter order */ opus_int warping_Q16, /* I */ opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ - opus_int *smpl_buf_idx, /* I Index to newest samples in buffers */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ opus_int decisionDelay /* I */ ) { @@ -638,8 +639,9 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( psSS[ 1 ].xq_Q14 = xq_Q14; } } - *smpl_buf_idx = ( *smpl_buf_idx - 1 ) & DECISION_DELAY_MASK; /* Index to newest samples */ - last_smple_idx = ( *smpl_buf_idx + decisionDelay ) & DECISION_DELAY_MASK; /* Index to decisionDelay old samples */ + *smpl_buf_idx = ( *smpl_buf_idx - 1 ) % DECISION_DELAY; + if( *smpl_buf_idx < 0 ) *smpl_buf_idx += DECISION_DELAY; + last_smple_idx = ( *smpl_buf_idx + decisionDelay ) % DECISION_DELAY; /* Find winner */ RDmin_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; diff --git a/silk/x86/NSQ_sse.c b/silk/x86/NSQ_sse.c index bb3c5f19..6a9e6e96 100644 --- a/silk/x86/NSQ_sse.c +++ b/silk/x86/NSQ_sse.c @@ -71,7 +71,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( ); void silk_NSQ_sse4_1( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -233,7 +233,6 @@ void silk_NSQ_sse4_1( NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; /* Save quantized speech and noise shaping signals */ - /* DEBUG_STORE_DATA( enc.pcm, &NSQ->xq[ psEncC->ltp_mem_length ], psEncC->frame_length * sizeof( opus_int16 ) ) */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); RESTORE_STACK; diff --git a/silk/x86/main_sse.h b/silk/x86/main_sse.h index d8d61310..2f15d448 100644 --- a/silk/x86/main_sse.h +++ b/silk/x86/main_sse.h @@ -34,6 +34,7 @@ # if defined(OPUS_X86_MAY_HAVE_SSE4_1) +#if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ # define OVERRIDE_silk_VQ_WMat_EC void silk_VQ_WMat_EC_sse4_1( @@ -79,11 +80,13 @@ extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( mu_Q9, max_gain_Q7, L)) #endif +#endif +#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ # define OVERRIDE_silk_NSQ void silk_NSQ_sse4_1( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -110,7 +113,7 @@ void silk_NSQ_sse4_1( #else extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -137,7 +140,7 @@ extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( # define OVERRIDE_silk_NSQ_del_dec void silk_NSQ_del_dec_sse4_1( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -164,7 +167,7 @@ void silk_NSQ_del_dec_sse4_1( #else extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -187,6 +190,7 @@ extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) #endif +#endif void silk_noise_shape_quantizer( silk_nsq_state *NSQ, /* I/O NSQ state */ @@ -238,39 +242,6 @@ extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( silk_encoder_state *psEnC, const opus_int16 pIn[]); -# define OVERRIDE_silk_warped_LPC_analysis_filter_FIX - -#endif - -void silk_warped_LPC_analysis_filter_FIX_sse4_1( - opus_int32 state[], /* I/O State [order + 1] */ - opus_int32 res_Q2[], /* O Residual signal [length] */ - const opus_int16 coef_Q13[], /* I Coefficients [order] */ - const opus_int16 input[], /* I Input signal [length] */ - const opus_int16 lambda_Q16, /* I Warping factor */ - const opus_int length, /* I Length of input signal */ - const opus_int order /* I Filter order (even) */ -); - -#if defined(OPUS_X86_PRESUME_SSE4_1) -#define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \ - ((void)(arch),silk_warped_LPC_analysis_filter_FIX_c(state, res_Q2, coef_Q13, input, lambda_Q16, length, order)) - -#else - -extern void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[OPUS_ARCHMASK + 1])( - opus_int32 state[], /* I/O State [order + 1] */ - opus_int32 res_Q2[], /* O Residual signal [length] */ - const opus_int16 coef_Q13[], /* I Coefficients [order] */ - const opus_int16 input[], /* I Input signal [length] */ - const opus_int16 lambda_Q16, /* I Warping factor */ - const opus_int length, /* I Length of input signal */ - const opus_int order /* I Filter order (even) */ -); - -# define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \ - ((*SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[(arch) & OPUS_ARCHMASK])(state, res_Q2, coef_Q13, input, lambda_Q16, length, order)) - #endif # endif diff --git a/silk/x86/x86_silk_map.c b/silk/x86/x86_silk_map.c index 818841f2..32dcc3ca 100644 --- a/silk/x86/x86_silk_map.c +++ b/silk/x86/x86_silk_map.c @@ -66,8 +66,9 @@ opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ) /* avx */ }; +#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -89,7 +90,9 @@ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_NSQ ), /* sse4.1 */ MAY_HAVE_SSE4_1( silk_NSQ ) /* avx */ }; +#endif +#if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( opus_int8 *ind, /* O index of best codebook vector */ opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ @@ -109,9 +112,11 @@ void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ), /* sse4.1 */ MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ) /* avx */ }; +#endif +#if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )( - const silk_encoder_state *psEncC, /* I/O Encoder State */ + const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ SideInfoIndices *psIndices, /* I/O Quantization Indices */ const opus_int32 x_Q3[], /* I Prefiltered input signal */ @@ -133,25 +138,10 @@ void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )( MAY_HAVE_SSE4_1( silk_NSQ_del_dec ), /* sse4.1 */ MAY_HAVE_SSE4_1( silk_NSQ_del_dec ) /* avx */ }; +#endif #if defined(FIXED_POINT) -void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[ OPUS_ARCHMASK + 1 ] )( - opus_int32 state[], /* I/O State [order + 1] */ - opus_int32 res_Q2[], /* O Residual signal [length] */ - const opus_int16 coef_Q13[], /* I Coefficients [order] */ - const opus_int16 input[], /* I Input signal [length] */ - const opus_int16 lambda_Q16, /* I Warping factor */ - const opus_int length, /* I Length of input signal */ - const opus_int order /* I Filter order (even) */ -) = { - silk_warped_LPC_analysis_filter_FIX_c, /* non-sse */ - silk_warped_LPC_analysis_filter_FIX_c, - silk_warped_LPC_analysis_filter_FIX_c, - MAY_HAVE_SSE4_1( silk_warped_LPC_analysis_filter_FIX ), /* sse4.1 */ - MAY_HAVE_SSE4_1( silk_warped_LPC_analysis_filter_FIX ) /* avx */ -}; - void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )( opus_int32 *res_nrg, /* O Residual energy */ opus_int *res_nrg_Q, /* O Residual energy Q value */ |