Tags

use the ABA free lifo from P99

after a cleanup campaign

a version that works without error on my laptop with the matrix
multiplication code, even with a lot of processes and threads.

new version with a first interfaced CUDA support

a working version

there is still some bug that has everything block when there are
several processes and a lot of strain by running a lot of processes

Finally we found that the malfunctionning of the gpupricer is related
to the initialization of the GPU that blocks the whole machine for one
second. If this is launched a bit delayed on different nodes of an
ORWL run, the effect is catastrophic: the waiting time is taken as a
sum of the waiting times on each node.

But this clearly is a problen of the gpupricer and not of ORWL.

synch with P99

and old version that worked at that time

Tested on the supelec cluster

after a long struggle this one seems to work, at last

a stable version that we used in September