Operators#
Operators are used to describe which FFT operation to solve, and to configure the execution.
Note
Here we only document the operators that are new to the preview. For existing operators, please refer to the Operators section of the official cuFFTDx documentation.
Description Operators#
Operator |
Default value |
Description |
---|---|---|
experimental::CodeType< experimental::code_type > |
|
Code type of the FFT compute device function, either |
CodeType Operator#
cufftdx::experimental::CodeType<cufftdx::experimental::code_type>()
Sets the type of code to use for kernel generation. Available options are:
experimental::code_type::ptx
(default) - Use inlined-PTX implementation of the FFT compute device functions.experimental::code_type::ltoir
- If available, use LTOIR implementation of the FFT compute device functions.
The default code type is experimental::code_type::ptx
.
Note
If experimental::code_type::ltoir
code type is specified but not available, cuFFTDx will fall back to using inlined-PTX implementation.
On the other hand, if experimental::code_type::ptx
code type is specified, cuFFTDx will not use LTOIR implementation even if it is available.