resample_sse.h File Reference

Resampler functions (SSE version) More...

#include <xmmintrin.h>

Include dependency graph for resample_sse.h:

Go to the source code of this file.

Macros
#define	OVERRIDE_INNER_PRODUCT_SINGLE
#define	OVERRIDE_INTERPOLATE_PRODUCT_SINGLE

Functions
static float	inner_product_single (const float *a, const float *b, unsigned int len)
static float	interpolate_product_single (const float *a, const float *b, unsigned int len, const spx_uint32_t oversample, float *frac)

Detailed Description

Resampler functions (SSE version)

Definition in file resample_sse.h.

Macro Definition Documentation

OVERRIDE_INNER_PRODUCT_SINGLE

#define OVERRIDE_INNER_PRODUCT_SINGLE

Definition at line 39 of file resample_sse.h.

OVERRIDE_INTERPOLATE_PRODUCT_SINGLE

#define OVERRIDE_INTERPOLATE_PRODUCT_SINGLE

Definition at line 56 of file resample_sse.h.

Function Documentation

inner_product_single()

static float inner_product_single ( const float *  a, const float *  b, unsigned int  len  )

inlinestatic

Definition at line 40 of file resample_sse.h.

References len().

Referenced by resampler_basic_direct_single().

{
   int i;
   float ret;
   __m128 sum = _mm_setzero_ps();
   for (i=0;i<len;i+=8)
   {
      sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i), _mm_loadu_ps(b+i)));
      sum = _mm_add_ps(sum, _mm_mul_ps(_mm_loadu_ps(a+i+4), _mm_loadu_ps(b+i+4)));
   }
   sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
   sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
   _mm_store_ss(&ret, sum);
   return ret;
}

interpolate_product_single()

static float interpolate_product_single ( const float *  a, const float *  b, unsigned int  len, const spx_uint32_t  oversample, float *  frac  )

inlinestatic

Definition at line 57 of file resample_sse.h.

References a, b, len(), and QualityMapping::oversample.

Referenced by resampler_basic_interpolate_single().

                                                                                                                                             {
  int i;
  float ret;
  __m128 sum = _mm_setzero_ps();
  __m128 f = _mm_loadu_ps(frac);
  for(i=0;i<len;i+=2)
  {
    sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i), _mm_loadu_ps(b+i*oversample)));
    sum = _mm_add_ps(sum, _mm_mul_ps(_mm_load1_ps(a+i+1), _mm_loadu_ps(b+(i+1)*oversample)));
  }
   sum = _mm_mul_ps(f, sum);
   sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
   sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
   _mm_store_ss(&ret, sum);
   return ret;
}