FLAC/src/deduplication/lpc_compute_autocorrelation_intrin_sse2.c

   1 /* This code is imported several times in lpc_intrin_sse2.c with different
   2  * values for MAX_LAG. Comments are for MAX_LAG == 14 */
   3         int i;
   4         __m128d sum0, sum1, sum2, sum3;
   5         __m128d d0, d1, d2, d3;
   6 #if MAX_LAG > 8
   7         __m128d d4;
   8         __m128d sum4;
   9 #endif
  10 #if MAX_LAG > 10
  11         __m128d d5, d6;
  12         __m128d sum5, sum6;
  13 #endif
  14
  15         (void) lag;
  16         FLAC__ASSERT(lag <= MAX_LAG);
  17
  18         /* Initialize all sum vectors with zero */
  19         sum0 = _mm_setzero_pd();
  20         sum1 = _mm_setzero_pd();
  21         sum2 = _mm_setzero_pd();
  22         sum3 = _mm_setzero_pd();
  23         d0 = _mm_setzero_pd();
  24         d1 = _mm_setzero_pd();
  25         d2 = _mm_setzero_pd();
  26         d3 = _mm_setzero_pd();
  27 #if MAX_LAG > 8
  28         sum4 = _mm_setzero_pd();
  29         d4 = _mm_setzero_pd();
  30 #endif
  31 #if MAX_LAG > 10
  32         sum5 = _mm_setzero_pd();
  33         sum6 = _mm_setzero_pd();
  34         d5 = _mm_setzero_pd();
  35         d6 = _mm_setzero_pd();
  36 #endif
  37
  38         /* Loop backwards through samples from data_len to limit */
  39         for(i = data_len-1; i >= 0; i--) {
  40                 __m128d d = _mm_set1_pd(data[i]);
  41
  42                 /* The next lines of code work like a queue. For more
  43                  * information see the lag8 version of this function */
  44 #if MAX_LAG > 10
  45                 d6 = _mm_shuffle_pd(d5, d6, _MM_SHUFFLE(0,0,0,1));
  46                 d5 = _mm_shuffle_pd(d4, d5, _MM_SHUFFLE(0,0,0,1));
  47 #endif
  48 #if MAX_LAG > 8
  49                 d4 = _mm_shuffle_pd(d3, d4, _MM_SHUFFLE(0,0,0,1));
  50 #endif
  51                 d3 = _mm_shuffle_pd(d2, d3, _MM_SHUFFLE(0,0,0,1));
  52                 d2 = _mm_shuffle_pd(d1, d2, _MM_SHUFFLE(0,0,0,1));
  53                 d1 = _mm_shuffle_pd(d0, d1, _MM_SHUFFLE(0,0,0,1));
  54                 d0 = _mm_shuffle_pd(d,  d0, _MM_SHUFFLE(0,0,0,1));
  55
  56                 /* sumn += d*dn */
  57                 sum0 = _mm_add_pd(sum0, _mm_mul_pd(d, d0));
  58                 sum1 = _mm_add_pd(sum1, _mm_mul_pd(d, d1));
  59                 sum2 = _mm_add_pd(sum2, _mm_mul_pd(d, d2));
  60                 sum3 = _mm_add_pd(sum3, _mm_mul_pd(d, d3));
  61 #if MAX_LAG > 8
  62                 sum4 = _mm_add_pd(sum4, _mm_mul_pd(d, d4));
  63 #endif
  64 #if MAX_LAG > 10
  65                 sum5 = _mm_add_pd(sum5, _mm_mul_pd(d, d5));
  66                 sum6 = _mm_add_pd(sum6, _mm_mul_pd(d, d6));
  67 #endif
  68         }
  69
  70         /* Store sum0..sum6 in autoc[0..14] */
  71         _mm_storeu_pd(autoc,   sum0);
  72         _mm_storeu_pd(autoc+2, sum1);
  73         _mm_storeu_pd(autoc+4, sum2);
  74         _mm_storeu_pd(autoc+6 ,sum3);
  75 #if MAX_LAG > 8
  76         _mm_storeu_pd(autoc+8, sum4);
  77 #endif
  78 #if MAX_LAG > 10
  79         _mm_storeu_pd(autoc+10,sum5);
  80         _mm_storeu_pd(autoc+12,sum6);
  81 #endif