nvmath.device.FFTOptions¶

class nvmath.device.FFTOptions(size, precision, fft_type, code_type, execution, *, direction=None, ffts_per_block=None, elements_per_thread=None, real_fft_options=None)[source]¶

A class that encapsulates a partial FFT device function. A partial device function can be queried for available or optimal values for the some knobs (such as leading_dimension or block_dim). It does not contain a compiled, ready-to-use, device function until finalized using create().

Parameters:

size (int) – The size of the FFT to calculate.
precision (str) – The computation precision specified as a numpy float dtype, currently supports numpy.float16, numpy.float32 and numpy.float64.
fft_type (str) – A string specifying the type of FFT operation, can be 'c2c', 'c2r' or 'r2c'.
code_type (CodeType) – The target GPU code and compute-capability.
execution (str) – A string specifying the execution method, can be 'Block' or 'Thread'.
direction (str) – A string specifying the direction of FFT, can be 'forward' or 'inverse'. If not provided, will be 'forward' if complex-to-real FFT is specified and 'inverse' if real-to-complex FFT is specified.
ffts_per_block (int) – The number of FFTs calculated per CUDA block, optional. The default is 1. Alternatively, if provided as 'suggested', will be set to a suggested value
elements_per_thread (int) – The number of elements per thread, optional. The default is 1. Alternatively, if provided as 'suggested', will be set to a suggested value.
real_fft_options (dict) –
A dictionary specifying the options for real FFT operation, optional. User may specify the following options in the dictionary:
- 'complex_layout', currently supports 'natural', 'packed', and 'full'.
- 'real_mode', currently supports 'normal' and 'folded.

Note

The class is not meant to used directly with its constructor. Users are instead advised to use fft() create the object.