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: #ifdef HAVE_CONFIG_H
Chris@69: #include "config.h"
Chris@69: #endif
Chris@69: 
Chris@69: #include <xmmintrin.h>
Chris@69: #include <emmintrin.h>
Chris@69: #include <smmintrin.h>
Chris@69: #include "celt_lpc.h"
Chris@69: #include "stack_alloc.h"
Chris@69: #include "mathops.h"
Chris@69: #include "pitch.h"
Chris@69: #include "x86cpu.h"
Chris@69: 
Chris@69: #if defined(FIXED_POINT)
Chris@69: 
Chris@69: void celt_fir_sse4_1(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:     int i,j;
Chris@69:     VARDECL(opus_val16, rnum);
Chris@69: 
Chris@69:     __m128i vecNoA;
Chris@69:     opus_int32 noA ;
Chris@69:     SAVE_STACK;
Chris@69: 
Chris@69:    ALLOC(rnum, ord, opus_val16);
Chris@69:    for(i=0;i<ord;i++)
Chris@69:       rnum[i] = num[ord-i-1];
Chris@69:    noA = EXTEND32(1) << SIG_SHIFT >> 1;
Chris@69:    vecNoA = _mm_set_epi32(noA, noA, noA, noA);
Chris@69: 
Chris@69:    for (i=0;i<N-3;i+=4)
Chris@69:    {
Chris@69:       opus_val32 sums[4] = {0};
Chris@69:       __m128i vecSum, vecX;
Chris@69: 
Chris@69:       xcorr_kernel(rnum, x+i-ord, sums, ord, arch);
Chris@69: 
Chris@69:       vecSum = _mm_loadu_si128((__m128i *)sums);
Chris@69:       vecSum = _mm_add_epi32(vecSum, vecNoA);
Chris@69:       vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT);
Chris@69:       vecX = OP_CVTEPI16_EPI32_M64(x + i);
Chris@69:       vecSum = _mm_add_epi32(vecSum, vecX);
Chris@69:       vecSum = _mm_packs_epi32(vecSum, vecSum);
Chris@69:       _mm_storel_epi64((__m128i *)(y + i), vecSum);
Chris@69:    }
Chris@69:    for (;i<N;i++)
Chris@69:    {
Chris@69:       opus_val32 sum = 0;
Chris@69:       for (j=0;j<ord;j++)
Chris@69:          sum = MAC16_16(sum, rnum[j], x[i+j-ord]);
Chris@69:       y[i] = SATURATE16(ADD32(EXTEND32(x[i]), PSHR32(sum, SIG_SHIFT)));
Chris@69:    }
Chris@69: 
Chris@69:    RESTORE_STACK;
Chris@69: }
Chris@69: 
Chris@69: #endif