Jerome Kieffer
Jerome Kieffer
Hi Jon, I spent a bit of time on this topic: we know OpenMP does not scale well and I wonder if the CSC approach + multithreading could provide better...
I tried to parallelize with mutithreading (`ThreadPoolExecutor.map`) and integrate a set of 1000 random images: * 1 thread: 23ms (consistent with timeit) * 2 threads: 12ms * 4 threads: 6ms...
Nop, I am on remote work, thus playing with what is available locally :)
I does scale nicely for now but it will hit the I/O bound of the CPU memory like all other algorithms.
> This looks great, did you already try it on a 40 or 64-core machine on the cluster? Thanks! have a look at #1733
TODO ... Probably a lot of atomic adds
On Tue, 09 Oct 2018 09:27:55 +0000 (UTC) picca wrote: > Hello, we are trying to deal with integration anomaly using the multi geometry and the s140 detector. > In...
When compiling in "No OpenMP" mode there are often warnings about: - cdummy value not initialized - min and max values not initlialized in splitpixel - area_n defined but unused...
TODO ... Probably a lot of atomic adds
Merging two calibrant files is "trivial", at least not more complicated than mixing two powder and using it for calibration. The second option is much more complicated, especially on the...