cannam@154: /* Copyright (c) 2014, Cisco Systems, INC cannam@154: Written by XiangMingZhu WeiZhou MinPeng YanWang cannam@154: cannam@154: Redistribution and use in source and binary forms, with or without cannam@154: modification, are permitted provided that the following conditions cannam@154: are met: cannam@154: cannam@154: - Redistributions of source code must retain the above copyright cannam@154: notice, this list of conditions and the following disclaimer. cannam@154: cannam@154: - Redistributions in binary form must reproduce the above copyright cannam@154: notice, this list of conditions and the following disclaimer in the cannam@154: documentation and/or other materials provided with the distribution. cannam@154: cannam@154: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS cannam@154: ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT cannam@154: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR cannam@154: A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER cannam@154: OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, cannam@154: EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, cannam@154: PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR cannam@154: PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF cannam@154: LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING cannam@154: NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS cannam@154: SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. cannam@154: */ cannam@154: cannam@154: #if defined(HAVE_CONFIG_H) cannam@154: #include "config.h" cannam@154: #endif cannam@154: cannam@154: #include "celt/x86/x86cpu.h" cannam@154: #include "structs.h" cannam@154: #include "SigProc_FIX.h" cannam@154: #include "pitch.h" cannam@154: #include "main.h" cannam@154: cannam@154: #if !defined(OPUS_X86_PRESUME_SSE4_1) cannam@154: cannam@154: #if defined(FIXED_POINT) cannam@154: cannam@154: #include "fixed/main_FIX.h" cannam@154: cannam@154: opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: const opus_int16 *inVec1, cannam@154: const opus_int16 *inVec2, cannam@154: const opus_int len cannam@154: ) = { cannam@154: silk_inner_prod16_aligned_64_c, /* non-sse */ cannam@154: silk_inner_prod16_aligned_64_c, cannam@154: silk_inner_prod16_aligned_64_c, cannam@154: MAY_HAVE_SSE4_1( silk_inner_prod16_aligned_64 ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_inner_prod16_aligned_64 ) /* avx */ cannam@154: }; cannam@154: cannam@154: #endif cannam@154: cannam@154: opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: silk_encoder_state *psEncC, cannam@154: const opus_int16 pIn[] cannam@154: ) = { cannam@154: silk_VAD_GetSA_Q8_c, /* non-sse */ cannam@154: silk_VAD_GetSA_Q8_c, cannam@154: silk_VAD_GetSA_Q8_c, cannam@154: MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ) /* avx */ cannam@154: }; cannam@154: cannam@154: #if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ cannam@154: void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ) = { cannam@154: silk_NSQ_c, /* non-sse */ cannam@154: silk_NSQ_c, cannam@154: silk_NSQ_c, cannam@154: MAY_HAVE_SSE4_1( silk_NSQ ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_NSQ ) /* avx */ cannam@154: }; cannam@154: #endif cannam@154: cannam@154: #if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ cannam@154: void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: opus_int8 *ind, /* O index of best codebook vector */ cannam@154: opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ cannam@154: opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ cannam@154: const opus_int16 *in_Q14, /* I input vector to be quantized */ cannam@154: const opus_int32 *W_Q18, /* I weighting matrix */ cannam@154: const opus_int8 *cb_Q7, /* I codebook */ cannam@154: const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ cannam@154: const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ cannam@154: const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ cannam@154: const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ cannam@154: opus_int L /* I number of vectors in codebook */ cannam@154: ) = { cannam@154: silk_VQ_WMat_EC_c, /* non-sse */ cannam@154: silk_VQ_WMat_EC_c, cannam@154: silk_VQ_WMat_EC_c, cannam@154: MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ) /* avx */ cannam@154: }; cannam@154: #endif cannam@154: cannam@154: #if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ cannam@154: void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ) = { cannam@154: silk_NSQ_del_dec_c, /* non-sse */ cannam@154: silk_NSQ_del_dec_c, cannam@154: silk_NSQ_del_dec_c, cannam@154: MAY_HAVE_SSE4_1( silk_NSQ_del_dec ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_NSQ_del_dec ) /* avx */ cannam@154: }; cannam@154: #endif cannam@154: cannam@154: #if defined(FIXED_POINT) cannam@154: cannam@154: void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )( cannam@154: opus_int32 *res_nrg, /* O Residual energy */ cannam@154: opus_int *res_nrg_Q, /* O Residual energy Q value */ cannam@154: opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ cannam@154: const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ cannam@154: const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ cannam@154: const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ cannam@154: const opus_int nb_subfr, /* I Number of subframes stacked in x */ cannam@154: const opus_int D, /* I Order */ cannam@154: int arch /* I Run-time architecture */ cannam@154: ) = { cannam@154: silk_burg_modified_c, /* non-sse */ cannam@154: silk_burg_modified_c, cannam@154: silk_burg_modified_c, cannam@154: MAY_HAVE_SSE4_1( silk_burg_modified ), /* sse4.1 */ cannam@154: MAY_HAVE_SSE4_1( silk_burg_modified ) /* avx */ cannam@154: }; cannam@154: cannam@154: #endif cannam@154: #endif