topic Re: Problems usind DFT in Intel® Integrated Performance Primitives

Problems usind DFT

Golvin — Wed, 07 Jul 2021 10:48:28 GMT

I want to compare the performance of Fourier Transforms between IPPS library and cuFFT library. Now Im comparing the Real to Complex case. For IPPS, I implemented the FFT (inputs size equal to 2^n) and then I tryed with the DFT (inputs of any size).

The FFT works fine but when I run the DFT I get an output of 0s. I read all the documentation about the DFT and did a very similiar code from https://software.intel.com/content/www/us/en/develop/articles/how-to-use-intel-ipp-s-1d-fourier-transform-functions.html but it stills dont work.

Im using IPPS version 2020.0.1 and Debian 10 as OS.

In the code I have a matrix as input and I will do as many DFT as row in the matrix. For each row I perform a DFT.

// NX == amount of columns in matrix // BATCH == amounto of rows in matrix int NX = 4; int BATCH = 1; int SEED = 1; int flag = IPP_FFT_NODIV_BY_ANY; int sizeSpec= 0; int sizeInit = 0; int sizeBuffer = 0; Ipp8u *pMemInit = 0; Ipp8u *pMemBuffer = 0; // For this case, BATCH is 1 so only the NX is important float **datos = (float**) malloc(sizeof(float*) * NX * BATCH); // Generates a NX x BATCH matrix with random numbers of the 4 possibles //(-3.533, -1.000, 1.000, 3.533) using the SEED datos = genMatrix(NX, BATCH, SEED); Ipp32fc **dst = (Ipp32fc**) malloc(sizeof(Ipp32fc*) * (NX / 2 + 1) * BATCH); // Check is NX is power of 2 if (!((NX) & (NX - 1))) { // FFT // ... } else { IppsDFTSpec_R_32f *pSpec = 0; ippsDFTGetSize_R_32f(NX, flag, ippAlgHintNone, &sizeSpec, &sizeInit, &sizeBuffer); /// allocate memory for required buffers if (sizeInit > 0) { pMemInit = (Ipp8u*) ippMalloc_8u(sizeInit); } if (sizeBuffer > 0) { pMemBuffer = (Ipp8u*) ippMalloc_8u(sizeBuffer); } ippsDFTInit_R_32f(NX, flag, ippAlgHintNone, pSpec, pMemInit); if (sizeInit > 0) { ippFree(pMemInit); } // Do the DFT for (int i = 0; i < BATCH; i++) { dst[i] = (Ipp32fc*) malloc(sizeof(Ipp32fc) * (NX / 2 + 1)); /// perform forward DFT ippsDFTFwd_RToCCS_32f((Ipp32f*) datos[i], (Ipp32f*) dst[i], pSpec, pMemBuffer); } }

In the FFT option (power of 2) I have a similar code than in DFT. I run this code for both cases and in FFT it works but in DFT I only get 0s.

P.D: excuse me for my english

Re:Problems usind DFT

VidyalathaB_Intel — Thu, 08 Jul 2021 13:20:35 GMT

Hi,

Thanks for reaching out to us.

>>Im using IPPS version 2020.0.1

Could you please try with latest ipp version which comes with oneAPI toolkit 2021.3.0.

and also share a complete reproducer if possible (along with the compiler that you are using) so that we can work on it from our end.

Thanks & Regards,

Vidya.

Re: Problems usind DFT

VidyalathaB_Intel — Thu, 15 Jul 2021 04:17:32 GMT

Hi,

Reminder:

Is your issue resolved ? Please let us know if the issue still persists by providing above mentioned details.

Regards,

Vidya.

Re: Problems usind DFT

Golvin — Fri, 16 Jul 2021 07:47:25 GMT

Hi,

Yes, I solved the problem.

I forgot to allocate memory for the pSpec variable.

Re: Problems usind DFT

VidyalathaB_Intel — Fri, 16 Jul 2021 13:25:17 GMT

Hi,

>>Yes, I solved the problem.

Thanks for the confirmation!

Glad to know that you have figured it out.

we are closing this thread as your issue is resolved.

If you require any additional assistance from Intel, please start a new thread.

Have a good day!

Regards,

Vidya.