1/* Copyright (C) 2004 Jean-Marc Valin */
2/**
3   @file vq_sse.h
4   @brief SSE-optimized vq routine
5*/
6/*
7   Redistribution and use in source and binary forms, with or without
8   modification, are permitted provided that the following conditions
9   are met:
10
11   - Redistributions of source code must retain the above copyright
12   notice, this list of conditions and the following disclaimer.
13
14   - Redistributions in binary form must reproduce the above copyright
15   notice, this list of conditions and the following disclaimer in the
16   documentation and/or other materials provided with the distribution.
17
18   - Neither the name of the Xiph.org Foundation nor the names of its
19   contributors may be used to endorse or promote products derived from
20   this software without specific prior written permission.
21
22   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
26   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
27   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
29   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
30   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
31   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
32   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33*/
34
35#define OVERRIDE_VQ_NBEST
36void vq_nbest(spx_word16_t *_in, const __m128 *codebook, int len, int entries, __m128 *E, int N, int *nbest, spx_word32_t *best_dist, char *stack)
37{
38   int i,j,k,used;
39   VARDECL(float *dist);
40   VARDECL(__m128 *in);
41   __m128 half;
42   used = 0;
43   ALLOC(dist, entries, float);
44   half = _mm_set_ps1(.5f);
45   ALLOC(in, len, __m128);
46   for (i=0;i<len;i++)
47      in[i] = _mm_set_ps1(_in[i]);
48   for (i=0;i<entries>>2;i++)
49   {
50      __m128 d = _mm_mul_ps(E[i], half);
51      for (j=0;j<len;j++)
52         d = _mm_sub_ps(d, _mm_mul_ps(in[j], *codebook++));
53      _mm_storeu_ps(dist+4*i, d);
54   }
55   for (i=0;i<entries;i++)
56   {
57      if (i<N || dist[i]<best_dist[N-1])
58      {
59         for (k=N-1; (k >= 1) && (k > used || dist[i] < best_dist[k-1]); k--)
60         {
61            best_dist[k]=best_dist[k-1];
62            nbest[k] = nbest[k-1];
63         }
64         best_dist[k]=dist[i];
65         nbest[k]=i;
66         used++;
67      }
68   }
69}
70
71
72
73
74#define OVERRIDE_VQ_NBEST_SIGN
75void vq_nbest_sign(spx_word16_t *_in, const __m128 *codebook, int len, int entries, __m128 *E, int N, int *nbest, spx_word32_t *best_dist, char *stack)
76{
77   int i,j,k,used;
78   VARDECL(float *dist);
79   VARDECL(__m128 *in);
80   __m128 half;
81   used = 0;
82   ALLOC(dist, entries, float);
83   half = _mm_set_ps1(.5f);
84   ALLOC(in, len, __m128);
85   for (i=0;i<len;i++)
86      in[i] = _mm_set_ps1(_in[i]);
87   for (i=0;i<entries>>2;i++)
88   {
89      __m128 d = _mm_setzero_ps();
90      for (j=0;j<len;j++)
91         d = _mm_add_ps(d, _mm_mul_ps(in[j], *codebook++));
92      _mm_storeu_ps(dist+4*i, d);
93   }
94   for (i=0;i<entries;i++)
95   {
96      int sign;
97      if (dist[i]>0)
98      {
99         sign=0;
100         dist[i]=-dist[i];
101      } else
102      {
103         sign=1;
104      }
105      dist[i] += .5f*((float*)E)[i];
106      if (i<N || dist[i]<best_dist[N-1])
107      {
108         for (k=N-1; (k >= 1) && (k > used || dist[i] < best_dist[k-1]); k--)
109         {
110            best_dist[k]=best_dist[k-1];
111            nbest[k] = nbest[k-1];
112         }
113         best_dist[k]=dist[i];
114         nbest[k]=i;
115         used++;
116         if (sign)
117            nbest[k]+=entries;
118      }
119   }
120}
121