Christian Trott
Christian Trott
yeah we need to get this fixed.
Hey all: we should just add create_mirror and create_mirror_view for UnorderedMap. Note there is already a function we can use: create_copy_views in the UnorderedMap. So something like this: ``` template...
On Blake 4.3.01 is consistently slower than 4.4.01 with GCC ~11~ 8.5 (I build for SKX) - I do see the slowdown with GCC 11
Not sure what to make of this, or what the action item here is ... | Compiler | KK 4.4 | KK 4.3 | |---|---|---| | GCC 8.5 | 5.7...
Yeah with gcc 9.1 on Kokkos dev its 5.3 vs 6.0s with commenting in/out the lock for MDRange parallel_for
This is used in LAMMPS. In particular they need the capability to initialize the number of random generators to the size of UniqueToken thing. I am not opposed to deprecating...
Agree with Damien, this is too much time for a perf test we run all the time for not a lot of benefit I would significantly cut back on this...
Retest this please!
And I see 4seconds on Serial on my laptop.
I want to see how long these are running before I approve.