|
Chris@69
|
1 /* Copyright (c) 2014, Cisco Systems, INC
|
|
Chris@69
|
2 Written by XiangMingZhu WeiZhou MinPeng YanWang
|
|
Chris@69
|
3
|
|
Chris@69
|
4 Redistribution and use in source and binary forms, with or without
|
|
Chris@69
|
5 modification, are permitted provided that the following conditions
|
|
Chris@69
|
6 are met:
|
|
Chris@69
|
7
|
|
Chris@69
|
8 - Redistributions of source code must retain the above copyright
|
|
Chris@69
|
9 notice, this list of conditions and the following disclaimer.
|
|
Chris@69
|
10
|
|
Chris@69
|
11 - Redistributions in binary form must reproduce the above copyright
|
|
Chris@69
|
12 notice, this list of conditions and the following disclaimer in the
|
|
Chris@69
|
13 documentation and/or other materials provided with the distribution.
|
|
Chris@69
|
14
|
|
Chris@69
|
15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
Chris@69
|
16 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
Chris@69
|
17 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
Chris@69
|
18 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
|
|
Chris@69
|
19 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
|
Chris@69
|
20 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
|
Chris@69
|
21 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
Chris@69
|
22 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
|
|
Chris@69
|
23 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
|
|
Chris@69
|
24 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
Chris@69
|
25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
Chris@69
|
26 */
|
|
Chris@69
|
27
|
|
Chris@69
|
28 #ifdef HAVE_CONFIG_H
|
|
Chris@69
|
29 #include "config.h"
|
|
Chris@69
|
30 #endif
|
|
Chris@69
|
31
|
|
Chris@69
|
32 #include <xmmintrin.h>
|
|
Chris@69
|
33 #include <emmintrin.h>
|
|
Chris@69
|
34 #include <smmintrin.h>
|
|
Chris@69
|
35 #include "celt_lpc.h"
|
|
Chris@69
|
36 #include "stack_alloc.h"
|
|
Chris@69
|
37 #include "mathops.h"
|
|
Chris@69
|
38 #include "pitch.h"
|
|
Chris@69
|
39 #include "x86cpu.h"
|
|
Chris@69
|
40
|
|
Chris@69
|
41 #if defined(FIXED_POINT)
|
|
Chris@69
|
42
|
|
Chris@69
|
43 void celt_fir_sse4_1(const opus_val16 *x,
|
|
Chris@69
|
44 const opus_val16 *num,
|
|
Chris@69
|
45 opus_val16 *y,
|
|
Chris@69
|
46 int N,
|
|
Chris@69
|
47 int ord,
|
|
Chris@69
|
48 int arch)
|
|
Chris@69
|
49 {
|
|
Chris@69
|
50 int i,j;
|
|
Chris@69
|
51 VARDECL(opus_val16, rnum);
|
|
Chris@69
|
52
|
|
Chris@69
|
53 __m128i vecNoA;
|
|
Chris@69
|
54 opus_int32 noA ;
|
|
Chris@69
|
55 SAVE_STACK;
|
|
Chris@69
|
56
|
|
Chris@69
|
57 ALLOC(rnum, ord, opus_val16);
|
|
Chris@69
|
58 for(i=0;i<ord;i++)
|
|
Chris@69
|
59 rnum[i] = num[ord-i-1];
|
|
Chris@69
|
60 noA = EXTEND32(1) << SIG_SHIFT >> 1;
|
|
Chris@69
|
61 vecNoA = _mm_set_epi32(noA, noA, noA, noA);
|
|
Chris@69
|
62
|
|
Chris@69
|
63 for (i=0;i<N-3;i+=4)
|
|
Chris@69
|
64 {
|
|
Chris@69
|
65 opus_val32 sums[4] = {0};
|
|
Chris@69
|
66 __m128i vecSum, vecX;
|
|
Chris@69
|
67
|
|
Chris@69
|
68 xcorr_kernel(rnum, x+i-ord, sums, ord, arch);
|
|
Chris@69
|
69
|
|
Chris@69
|
70 vecSum = _mm_loadu_si128((__m128i *)sums);
|
|
Chris@69
|
71 vecSum = _mm_add_epi32(vecSum, vecNoA);
|
|
Chris@69
|
72 vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT);
|
|
Chris@69
|
73 vecX = OP_CVTEPI16_EPI32_M64(x + i);
|
|
Chris@69
|
74 vecSum = _mm_add_epi32(vecSum, vecX);
|
|
Chris@69
|
75 vecSum = _mm_packs_epi32(vecSum, vecSum);
|
|
Chris@69
|
76 _mm_storel_epi64((__m128i *)(y + i), vecSum);
|
|
Chris@69
|
77 }
|
|
Chris@69
|
78 for (;i<N;i++)
|
|
Chris@69
|
79 {
|
|
Chris@69
|
80 opus_val32 sum = 0;
|
|
Chris@69
|
81 for (j=0;j<ord;j++)
|
|
Chris@69
|
82 sum = MAC16_16(sum, rnum[j], x[i+j-ord]);
|
|
Chris@69
|
83 y[i] = SATURATE16(ADD32(EXTEND32(x[i]), PSHR32(sum, SIG_SHIFT)));
|
|
Chris@69
|
84 }
|
|
Chris@69
|
85
|
|
Chris@69
|
86 RESTORE_STACK;
|
|
Chris@69
|
87 }
|
|
Chris@69
|
88
|
|
Chris@69
|
89 #endif
|
