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
If a user tries to use double precision but the device does not support it, the run fails rather than either warning or trying to demote everything
#29
Closed
tnowotny opened this issue
Nov 10, 2014
· 5 comments
We need to have a look at this - maybe it's not that urgent as most devices now do support double precision. See corresponding issue 18 in brian2genn as well.
The text was updated successfully, but these errors were encountered:
@tnowotny, If sm < 1.3, maybe we can simply make a typedef to redefine double as float? Or we can leave it as it is as old sm versions will not be supported by new CUDA versions.
I guess throwing a warning and then trying a redefinition might be one way forward - but in a sense it might be more honest and avoiding any confusion on what has actually run to warn and abort. The users can decide to change things to float or find themselves a newer GPU?
Hi shalesh210,
an informative error message and exit may be a good start. Longer term one could think of a warning followed by an automatic demotion of the model precision to single precision floating point; however there would also be downsides to this - users might miss the warning and then take the SP float results as being DP.
We need to have a look at this - maybe it's not that urgent as most devices now do support double precision. See corresponding issue 18 in brian2genn as well.
The text was updated successfully, but these errors were encountered: