search

ICLDisco / dplasma

DPLASMA is a highly optimized, accelerator-aware, implementation of a dense linear algebra package for distributed heterogeneous systems. It is designed to deliver sustained performance for distributed systems where each node featuring multiple sockets of multicore processors, and if available, accelerators, using the PaRSEC runtime as a backend.
Other
11 stars 9 forks source link

stack-based variables are not allowed on async tasks. #46

Closed bosilca closed 2 years ago

bosilca commented 2 years ago

When starting an async task we cannot use pointers to variables on the stack, because when they will eventually be updated it will corrupt the stack of some other task. Instead, move the variable in the task itself, it will then become safe to alter at any moment.

Second thing fixed in this patch is the handling of the INFO parameter for POTRF. This INFO should always be scaled to the correct diagonal tile if an error is detected.

Signed-off-by: George Bosilca bosilca@icl.utk.edu

abouteiller commented 2 years ago

commits show as 'unverified' you may need to resign them