ghost · GitHub

Merged

Merged

Conversation

ghost

(int)istrides[0], (int)istrides[rank],
(ctype_t *)out.get(), out_embed, (int)ostrides[0],
(int)ostrides[rank], FFTW_ESTIMATE);
(int)ostrides[rank], FFTW_ESTIMATE | FFTW_PRESERVE_INPUT);

@umar456

@umar456

The fftC2R function was modifying the input array when the fftw library
was used. This was happening because the fftw library does not have
and algorithm that can compute the fft without modifying the input array.
This problem does not seem to be happening on the MKL based fft.

@umar456

9prady9

to_test_params);

// Does not work well with CUDA 10.1
// TEST_P(FFTC2R2D, Complex32ToRealInputsPreserved) {

@9prady9

Labels

None yet

Read the original on github.com ↗