You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fix pycuda initialisation during accuracy tests (pyvkff-test).
Version 2024.1 (2024-02-06)
Based on VkFFT 1.3.4
Add support for direct sine transforms (DST)
R2C, DST and DCT now support arbitrary sizes (up to ~2^32,
same as C2C)
Odd lengths for the fast axis is now supported for all R2C
transforms. Inplace transforms require using
the r2c_odd=True parameter
Custom transform axes (strided) are now allowed also for R2C,
as long as the fast axis is transformed.
added functions to access the size of the temporary buffer
created by VkFFT (if any), the type of algorithm used along
each axis (radix, Rader, Bluestein), and the number of
uploads for each transformed axis.
DCT and DST now support F-ordered arrays
Longer default test including multi-upload using radix,
Rader and Bluestein algorithms.
The full test suite (including c2c, r2c, dct, dst, radix
and non-radix transforms, single and double precision)
now includes about 1.5 million unit tests
The pyvkff-benchmark script can also test R2C, DCT and DST
transforms, and will give more details about the algorithm
used for performance tuning.