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: #if defined(HAVE_CONFIG_H)
Chris@69: #include "config.h"
Chris@69: #endif
Chris@69: 
Chris@69: #include "x86/x86cpu.h"
Chris@69: #include "celt_lpc.h"
Chris@69: #include "pitch.h"
Chris@69: #include "pitch_sse.h"
Chris@69: #include "vq.h"
Chris@69: 
Chris@69: #if defined(OPUS_HAVE_RTCD)
Chris@69: 
Chris@69: # if defined(FIXED_POINT)
Chris@69: 
Chris@69: #if defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)
Chris@69: 
Chris@69: void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:          const opus_val16 *x,
Chris@69:          const opus_val16 *num,
Chris@69:          opus_val16       *y,
Chris@69:          int              N,
Chris@69:          int              ord,
Chris@69:          int              arch
Chris@69: ) = {
Chris@69:   celt_fir_c,                /* non-sse */
Chris@69:   celt_fir_c,
Chris@69:   celt_fir_c,
Chris@69:   MAY_HAVE_SSE4_1(celt_fir), /* sse4.1  */
Chris@69:   MAY_HAVE_SSE4_1(celt_fir)  /* avx  */
Chris@69: };
Chris@69: 
Chris@69: void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:          const opus_val16 *x,
Chris@69:          const opus_val16 *y,
Chris@69:          opus_val32       sum[4],
Chris@69:          int              len
Chris@69: ) = {
Chris@69:   xcorr_kernel_c,                /* non-sse */
Chris@69:   xcorr_kernel_c,
Chris@69:   xcorr_kernel_c,
Chris@69:   MAY_HAVE_SSE4_1(xcorr_kernel), /* sse4.1  */
Chris@69:   MAY_HAVE_SSE4_1(xcorr_kernel)  /* avx  */
Chris@69: };
Chris@69: 
Chris@69: #endif
Chris@69: 
Chris@69: #if (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) ||  \
Chris@69:  (!defined(OPUS_X86_MAY_HAVE_SSE_4_1) && defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2))
Chris@69: 
Chris@69: opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:          const opus_val16 *x,
Chris@69:          const opus_val16 *y,
Chris@69:          int              N
Chris@69: ) = {
Chris@69:   celt_inner_prod_c,                /* non-sse */
Chris@69:   celt_inner_prod_c,
Chris@69:   MAY_HAVE_SSE2(celt_inner_prod),
Chris@69:   MAY_HAVE_SSE4_1(celt_inner_prod), /* sse4.1  */
Chris@69:   MAY_HAVE_SSE4_1(celt_inner_prod)  /* avx  */
Chris@69: };
Chris@69: 
Chris@69: #endif
Chris@69: 
Chris@69: # else
Chris@69: 
Chris@69: #if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)
Chris@69: 
Chris@69: void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:          const opus_val16 *x,
Chris@69:          const opus_val16 *y,
Chris@69:          opus_val32       sum[4],
Chris@69:          int              len
Chris@69: ) = {
Chris@69:   xcorr_kernel_c,                /* non-sse */
Chris@69:   MAY_HAVE_SSE(xcorr_kernel),
Chris@69:   MAY_HAVE_SSE(xcorr_kernel),
Chris@69:   MAY_HAVE_SSE(xcorr_kernel),
Chris@69:   MAY_HAVE_SSE(xcorr_kernel)
Chris@69: };
Chris@69: 
Chris@69: opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:          const opus_val16 *x,
Chris@69:          const opus_val16 *y,
Chris@69:          int              N
Chris@69: ) = {
Chris@69:   celt_inner_prod_c,                /* non-sse */
Chris@69:   MAY_HAVE_SSE(celt_inner_prod),
Chris@69:   MAY_HAVE_SSE(celt_inner_prod),
Chris@69:   MAY_HAVE_SSE(celt_inner_prod),
Chris@69:   MAY_HAVE_SSE(celt_inner_prod)
Chris@69: };
Chris@69: 
Chris@69: void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:                     const opus_val16 *x,
Chris@69:                     const opus_val16 *y01,
Chris@69:                     const opus_val16 *y02,
Chris@69:                     int               N,
Chris@69:                     opus_val32       *xy1,
Chris@69:                     opus_val32       *xy2
Chris@69: ) = {
Chris@69:   dual_inner_prod_c,                /* non-sse */
Chris@69:   MAY_HAVE_SSE(dual_inner_prod),
Chris@69:   MAY_HAVE_SSE(dual_inner_prod),
Chris@69:   MAY_HAVE_SSE(dual_inner_prod),
Chris@69:   MAY_HAVE_SSE(dual_inner_prod)
Chris@69: };
Chris@69: 
Chris@69: void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:               opus_val32 *y,
Chris@69:               opus_val32 *x,
Chris@69:               int         T,
Chris@69:               int         N,
Chris@69:               opus_val16  g10,
Chris@69:               opus_val16  g11,
Chris@69:               opus_val16  g12
Chris@69: ) = {
Chris@69:   comb_filter_const_c,                /* non-sse */
Chris@69:   MAY_HAVE_SSE(comb_filter_const),
Chris@69:   MAY_HAVE_SSE(comb_filter_const),
Chris@69:   MAY_HAVE_SSE(comb_filter_const),
Chris@69:   MAY_HAVE_SSE(comb_filter_const)
Chris@69: };
Chris@69: 
Chris@69: 
Chris@69: #endif
Chris@69: 
Chris@69: #if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)
Chris@69: opus_val16 (*const OP_PVQ_SEARCH_IMPL[OPUS_ARCHMASK + 1])(
Chris@69:       celt_norm *_X, int *iy, int K, int N, int arch
Chris@69: ) = {
Chris@69:   op_pvq_search_c,                /* non-sse */
Chris@69:   op_pvq_search_c,
Chris@69:   MAY_HAVE_SSE2(op_pvq_search),
Chris@69:   MAY_HAVE_SSE2(op_pvq_search),
Chris@69:   MAY_HAVE_SSE2(op_pvq_search)
Chris@69: };
Chris@69: #endif
Chris@69: 
Chris@69: #endif
Chris@69: #endif