Reusing Buffers

Some Intel IPP functions require internal memory for various optimization strategies. However, memory allocation inside of the function may have a negative impact on performance in some situations, for example, cache misses. To avoid or minimize memory allocation and keep your data in a warm cache, some functions, for example, Fourier transform functions (FFT), can use or reuse memory given as a parameter to the function.

If you call such a function, many times, you can reuse of an external buffer and get better performance.

The following example reuses memory buffers to compute FFT as two FFTs in two separate threads:

            
ippsFFTInitAlloc_C_32fc(&ctxN2, order-1, IPP_FFT_DIV_INV_BY_N, ippAlgHintAccurate);      
ippsFFTGetBufSize_C_32fc( ctxN2, &sz );

buffer = sz > 0 ? ippsMalloc_8u( sz ) : 0;

/// prepare source data for two FFTs                       
int phase = 0;                        
ippsSampleDown_32fc( x, fftlen, xleft, &fftlen2, 2, &phase );                        
phase = 1;                        
ippsSampleDown_32fc( x, fftlen, xrght, &fftlen2, 2, &phase );
                               
ippsFFTFwd_CToC_32fc( xleft, Xleft, ctxN2, buffer );                        
ippsFFTFwd_CToC_32fc( xrght, Xrght, ctxN2, buffer );

The external buffer is not necessary. If the pointer to the buffer is 0, the function allocates memory inside.

Copyright © 2008 - 2011, Intel Corporation. All rights reserved.