The implementation using 32 bit precision, whilst working ok for the tests, seems to be unstable for certain configurations. This isn't a big deal since most people will still use floating point 64. What's more worrying is that using fp64 seemed to throw some issues, even though this should be identical to the dev branch.
timcallow
commented
1 year ago
The implementation using 32 bit precision, whilst working ok for the tests, seems to be unstable for certain configurations. This isn't a big deal since most people will still use floating point 64. What's more worrying is that using fp64 seemed to throw some issues, even though this should be identical to the dev branch.