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
Compiling the OpenCL program on every startup is very slow for large programs. Dumping a binary at first start and reloading it later is almost instant, but currently a manual process. It would be better if there was a C API function for telling a Futhark program to manage such caching in some specific file. The program would then automatically take care of cache invalidation and the like as needed. A similar feature can be implemented for the CUDA backend with cuModuleLoadData().
The text was updated successfully, but these errors were encountered:
Compiling the OpenCL program on every startup is very slow for large programs. Dumping a binary at first start and reloading it later is almost instant, but currently a manual process. It would be better if there was a C API function for telling a Futhark program to manage such caching in some specific file. The program would then automatically take care of cache invalidation and the like as needed. A similar feature can be implemented for the CUDA backend with
cuModuleLoadData()
.The text was updated successfully, but these errors were encountered: