issues
search
eth-cscs
/
conflux
Distributed Communication-Optimal LU-factorization Algorithm
BSD 3-Clause "New" or "Revised" License
12
stars
3
forks
source link
Additional Optimizations
#13
Closed
kabicm
closed
3 years ago
kabicm
commented
3 years ago
This PR brings the following improvements:
[bugfix] Negative message size in MPI fixed
[bugfix] Missing MPI_Irecv in tournament pivoting fixed
[bugfix] Fixed n_reqs in tournament pivoting when Px not a power of 2
[smallfix] The matrix reinitialized in each step, to get more consistent benchmarking results.
[optimization] Removed a redundant A00 broadcast after pivoting
[optimization] Removed the broadcast of
curPivOrder
, which is now sent with
curPivots
[optimization] Further optimizations
This PR brings the following improvements:
curPivOrder
, which is now sent withcurPivots