guillaumetousignant / NDG_cuda

MIT License
1 stars 0 forks source link

Non power of 2 number of elements output nan #5

Closed guillaumetousignant closed 3 years ago

guillaumetousignant commented 3 years ago

Seems more complicated than that: 64 & 62 elements work, 128 elements work, but not 66.

guillaumetousignant commented 3 years ago

Seems to be linked to #4 , 94 works well, 96 gives bad results and 98 doesn't work.

guillaumetousignant commented 3 years ago

96 elements on debug gives good results, but doesn't work on release??

guillaumetousignant commented 3 years ago

When the program is recompiled the results are different.

guillaumetousignant commented 3 years ago

The results are different each time the program is run oh no.

guillaumetousignant commented 3 years ago

96 on release is fixed with a lower timestep.

guillaumetousignant commented 3 years ago

Not fixed in debug for 126 elements, and results are still non-deterministic.

guillaumetousignant commented 3 years ago

Fixed along with #4 by correctly swapping pointers when using the move constructor.