commit e6cffc91b0a454507277d44a853c013ee9854fb2
author hayati ayguen <h_ayguen@web.de> Fri Feb 28 19:57:30 2020 +0100
committer hayati ayguen <h_ayguen@web.de> Sat Feb 29 08:49:58 2020 +0100
tree 8bb3c7b12ab9fb0434013d64566a9fd13ba9a23c
parent 3dab35c0c95c7279b29b58d04f6e144fd880dca2

minor additions and bugfix convolution benchmark

* added pffft_zconvolve_no_accu()
* added pffft_min_fft_size()
* pffft_next_power_of_two(), pffft_is_power_of_two()
* added option PFFASTCONV_CORRELATION
* fixed calculation of procSmpPerSec[] in test_pffastconv.c
* minor test additions

Signed-off-by: hayati ayguen <h_ayguen@web.de>

pffastconv.c

diff --git a/pffastconv.c b/pffastconv.c
index daff7ee..36cd7c9 100644
--- a/pffastconv.c
+++ b/pffastconv.c
@@ -49,20 +49,6 @@
 #include "pfsimd_macros.h"
 
 
-static
-unsigned nextPowerOfTwo(unsigned v) {
-  /* https://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2 */
-  /* compute the next highest power of 2 of 32-bit v */
-  v--;
-  v |= v >> 1;
-  v |= v >> 2;
-  v |= v >> 4;
-  v |= v >> 8;
-  v |= v >> 16;
-  v++;
-  return v;
-}
-
 void *pffastconv_malloc(size_t nb_bytes)
 {
   return Valigned_malloc(nb_bytes);
@@ -99,7 +85,7 @@
   PFFASTCONV_Setup * s = NULL;
   const int cplxFactor = ( (flags & PFFASTCONV_CPLX_INP_OUT) && (flags & PFFASTCONV_CPLX_SINGLE_FFT) ) ? 2 : 1;
   const int minFftLen = 2*pffft_simd_size()*pffft_simd_size();
-  int i, Nfft = 2 * nextPowerOfTwo(filterLen -1);
+  int i, Nfft = 2 * pffft_next_power_of_two(filterLen -1);
 #if FASTCONV_DBG_OUT
   const int iOldBlkLen = *blockLen;
 #endif
@@ -114,7 +100,7 @@
 
   if ( *blockLen > Nfft ) {
     Nfft = *blockLen;
-    Nfft = nextPowerOfTwo(Nfft);
+    Nfft = pffft_next_power_of_two(Nfft);
   }
   *blockLen = Nfft;  /* this is in (complex) samples */
 
@@ -136,8 +122,13 @@
   s->scale = (float)( 1.0 / Nfft );
 
   memset( s->Xt, 0, (unsigned)Nfft * sizeof(float) );
-  for ( i = 0; i < filterLen; ++i )
-    s->Xt[ ( Nfft - cplxFactor * i ) & (Nfft -1) ] = filterCoeffs[ filterLen - 1 - i ];
+  if ( flags & PFFASTCONV_CORRELATION ) {
+    for ( i = 0; i < filterLen; ++i )
+      s->Xt[ ( Nfft - cplxFactor * i ) & (Nfft -1) ] = filterCoeffs[ i ];
+  } else {
+    for ( i = 0; i < filterLen; ++i )
+      s->Xt[ ( Nfft - cplxFactor * i ) & (Nfft -1) ] = filterCoeffs[ filterLen - 1 - i ];
+  }
 
   pffft_transform(s->st, s->Xt, s->Hf, /* tmp = */ s->Mf, PFFFT_FORWARD);
 
@@ -219,8 +210,7 @@
         pffft_transform(s->st, s->Xt, s->Xf, /* tmp = */ s->Mf, PFFFT_FORWARD);
       }
 
-      memset( s->Mf, 0, (unsigned)Nfft * sizeof(float) );  /* why does pffft_zconvolve_accumulate() accumulate?? */
-      pffft_zconvolve_accumulate(s->st, s->Xf, s->Hf, /* tmp = */ s->Mf, s->scale);
+      pffft_zconvolve_no_accu(s->st, s->Xf, s->Hf, /* tmp = */ s->Mf, s->scale);
 
       if ( flags & PFFASTCONV_DIRECT_OUT )
       {
@@ -270,8 +260,7 @@
           pffft_transform(s->st, s->Xt, s->Xf, /* tmp = */ s->Mf, PFFFT_FORWARD);
         }
     
-        memset( s->Mf, 0, (unsigned)Nfft * sizeof(float) );  /* why does pffft_zconvolve_accumulate() accumulate?? */
-        pffft_zconvolve_accumulate(s->st, s->Xf, s->Hf, /* tmp = */ s->Mf, s->scale);
+        pffft_zconvolve_no_accu(s->st, s->Xf, s->Hf, /* tmp = */ s->Mf, s->scale);
 
         if ( flags & PFFASTCONV_CPLX_INP_OUT )
         {