Christian Trott
Christian Trott
I think this is a decent standard benchmark. If this doesn't work well when what would. One improvement we could do is report the fraction of 1D stream we get...
So looking at these tests there is essentially double iteration going on. There is an iteration in the test itself and then google test will do the state based iteration...
Assuming the following points are true right now: * we don't support mixing host/device nodes in the same graph * we don't support multiple GPU * we kinda ignore the...
> I still think it would be good to compare just the two impls of the index calculation and get the memory access fully out of the scope. I am...
Can you try the draft PR Jakob linked?
Alternatively we may wanna just prohibit using GCC 10 as host compiler for NVCC.
@JBludau @masterleinad @nmm0 any more comments?
I think this passed testing now. The failures are seen on other PRs too (The Jenkins CUDA build segfaulting during compilation, and the HPX build segfaulting during the tests).
I will look at it in frankfurt.