Closed kubagalecki closed 10 months ago
Attention: 70 lines
in your changes are missing coverage. Please review.
Comparison is base (
365770c
) 42.81% compared to head (0890776
) 44.52%.
:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
@llaniewski I think this is nearly everything we discussed. One outstanding item is renaming GetThreads.X
. My suggestion would be LaunchUtils.X
.
By the way, I ran the benchmark Travis gave me on Gadi, there was a 0.737% slowdown, which I'd write off as noise. This is without __grid_constant__
. Full log: taylor-bubble-benchmark.log
GPU data pointers stored in
LatticeData
and similar classes have thus far been stored in constant device memory and manually updated when needed. The idea is to now pass them directly to kernels viaconst
arguments, which should automatically get placed in constant memory. This significantly simplifies the code and allows for greater flexibility going forward.