Chris@69: /* Copyright (c) 2014, Cisco Systems, INC
Chris@69:    Written by XiangMingZhu WeiZhou MinPeng YanWang
Chris@69: 
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: #ifndef SIGPROC_FIX_SSE_H
Chris@69: #define SIGPROC_FIX_SSE_H
Chris@69: 
Chris@69: #ifdef HAVE_CONFIG_H
Chris@69: #include "config.h"
Chris@69: #endif
Chris@69: 
Chris@69: #if defined(OPUS_X86_MAY_HAVE_SSE4_1)
Chris@69: void silk_burg_modified_sse4_1(
Chris@69:     opus_int32                  *res_nrg,           /* O    Residual energy                                             */
Chris@69:     opus_int                    *res_nrg_Q,         /* O    Residual energy Q value                                     */
Chris@69:     opus_int32                  A_Q16[],            /* O    Prediction coefficients (length order)                      */
Chris@69:     const opus_int16            x[],                /* I    Input signal, length: nb_subfr * ( D + subfr_length )       */
Chris@69:     const opus_int32            minInvGain_Q30,     /* I    Inverse of max prediction gain                              */
Chris@69:     const opus_int              subfr_length,       /* I    Input signal subframe length (incl. D preceding samples)    */
Chris@69:     const opus_int              nb_subfr,           /* I    Number of subframes stacked in x                            */
Chris@69:     const opus_int              D,                  /* I    Order                                                       */
Chris@69:     int                         arch                /* I    Run-time architecture                                       */
Chris@69: );
Chris@69: 
Chris@69: #if defined(OPUS_X86_PRESUME_SSE4_1)
Chris@69: #define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \
Chris@69:     ((void)(arch), silk_burg_modified_sse4_1(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch))
Chris@69: 
Chris@69: #else
Chris@69: 
Chris@69: extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:     opus_int32                  *res_nrg,           /* O    Residual energy                                             */
Chris@69:     opus_int                    *res_nrg_Q,         /* O    Residual energy Q value                                     */
Chris@69:     opus_int32                  A_Q16[],            /* O    Prediction coefficients (length order)                      */
Chris@69:     const opus_int16            x[],                /* I    Input signal, length: nb_subfr * ( D + subfr_length )       */
Chris@69:     const opus_int32            minInvGain_Q30,     /* I    Inverse of max prediction gain                              */
Chris@69:     const opus_int              subfr_length,       /* I    Input signal subframe length (incl. D preceding samples)    */
Chris@69:     const opus_int              nb_subfr,           /* I    Number of subframes stacked in x                            */
Chris@69:     const opus_int              D,                  /* I    Order                                                       */
Chris@69:     int                         arch                /* I    Run-time architecture                                       */);
Chris@69: 
Chris@69: #  define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \
Chris@69:     ((*SILK_BURG_MODIFIED_IMPL[(arch) & OPUS_ARCHMASK])(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch))
Chris@69: 
Chris@69: #endif
Chris@69: 
Chris@69: opus_int64 silk_inner_prod16_aligned_64_sse4_1(
Chris@69:     const opus_int16 *inVec1,
Chris@69:     const opus_int16 *inVec2,
Chris@69:     const opus_int   len
Chris@69: );
Chris@69: 
Chris@69: 
Chris@69: #if defined(OPUS_X86_PRESUME_SSE4_1)
Chris@69: 
Chris@69: #define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \
Chris@69:     ((void)(arch),silk_inner_prod16_aligned_64_sse4_1(inVec1, inVec2, len))
Chris@69: 
Chris@69: #else
Chris@69: 
Chris@69: extern opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:                     const opus_int16 *inVec1,
Chris@69:                     const opus_int16 *inVec2,
Chris@69:                     const opus_int   len);
Chris@69: 
Chris@69: #  define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \
Chris@69:     ((*SILK_INNER_PROD16_ALIGNED_64_IMPL[(arch) & OPUS_ARCHMASK])(inVec1, inVec2, len))
Chris@69: 
Chris@69: #endif
Chris@69: #endif
Chris@69: #endif