mirror/kissfft
1
0
Fork 0
mirror of https://github.com/mborgerding/kissfft.git synced 2025-07-18 21:14:24 -04:00
Code Issues Packages Projects Releases Wiki Activity

polishing SIMD changes

Browse Source
This commit is contained in:
Mark Borgerding
2005-06-25 04:38:19 +00:00
parent 95fd244f93
commit 4be23bffa4
10 changed files with 75 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tools/kfc.c
View File

@ -42,7 +42,7 @@ static kiss_fft_cfg find_cached_fft(int nfft,int inverse)
if (cur== NULL) {
/* no cached node found, need to create a new one*/
kiss_fft_alloc(nfft,inverse,0,&len);
cur = (kfc_cfg)malloc(sizeof(struct cached_fft) + len );
cur = (kfc_cfg)KISS_FFT_MALLOC((sizeof(struct cached_fft) + len ));
if (cur == NULL)
return NULL;
cur->cfg = (kiss_fft_cfg)(cur+1);

22
tools/kiss_fftr.c
View File

@ -40,11 +40,7 @@ kiss_fftr_cfg kiss_fftr_alloc(int nfft,int inverse_fft,void * mem,size_t * lenme
memneeded = sizeof(struct kiss_fftr_state) + subsize + sizeof(kiss_fft_cpx) * ( nfft * 2);
if (lenmem == NULL) {
#ifdef USE_SIMD
st = (kiss_fftr_cfg) memalign (sizeof(kiss_fft_cpx),memneeded);
#else
st = (kiss_fftr_cfg) malloc (memneeded);
#endif
st = (kiss_fftr_cfg) KISS_FFT_MALLOC (memneeded);
} else {
if (*lenmem >= memneeded)
st = (kiss_fftr_cfg) mem;
@ -109,18 +105,10 @@ void kiss_fftr(kiss_fftr_cfg st,const kiss_fft_scalar *timedata,kiss_fft_cpx *fr
C_MUL( tw , f2k , st->super_twiddles[k]);
C_ADD( freqdata[k] , f1k ,tw);
#ifdef USE_SIMD
freqdata[k].r = (f1k.r + tw.r) * _mm_set1_ps(.5);
freqdata[k].i = (f1k.i + tw.i) * _mm_set1_ps(.5);
freqdata[N-k].r = (f1k.r - tw.r) * _mm_set1_ps(.5);
freqdata[N-k].i = - (f1k.i - tw.i) * _mm_set1_ps(.5);
#else
freqdata[k].r = (f1k.r + tw.r) / 2;
freqdata[k].i = (f1k.i + tw.i) / 2;
freqdata[N-k].r = (f1k.r - tw.r)/2;
freqdata[N-k].i = - (f1k.i - tw.i)/2;
#endif
freqdata[k].r = HALF_OF(f1k.r + tw.r);
freqdata[k].i = HALF_OF(f1k.i + tw.i);
freqdata[N-k].r = HALF_OF(f1k.r - tw.r);
freqdata[N-k].i = HALF_OF(tw.i - f1k.i);
}
CHECK_OVERFLOW_OP(tdc.r ,-, tdc.i);