View page source

warp.config.enable_mathdx_fft#

warp.config.enable_mathdx_fft: bool = True#

Use libmathdx (cuFFTDx) for tile_fft and tile_ifft on GPU when available.

When False, these ops use a fallback that supports power-of-two FFT sizes only, trading runtime performance for faster kernel compile times.

This setting can be overridden at the module level by setting the "enable_mathdx_fft" module option.