bryancatanzaro / inplace

CUDA and OpenMP implementations of C2R/R2C inplace transposition
GNU General Public License v2.0
44 stars 7 forks source link

Smaller block size test #1

Open neon5d opened 9 years ago

neon5d commented 9 years ago

I ran visual_test for smaller block size ( m, n < 32 ). The result is not correct. If block size is larger than 32, it looks fine. I am wondering whether this is an algorithm limitation or an implementation issue for smaller block size.