You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
CUDA Throughput Improvement, main branch (2024.05.06.) (#576)
* Made I/O functions use traccc::io::get_absolute_path(...) during reading.
This is to make it possible to pick up files from anywhere, as long
as the user provides an absolute file name.
* Returning a reduced amount of info about the fitted tracks.
This is to avoid the (currently) very expensive copy of the jagged
vector of track states back to the host.
* Fixed the geometry file locations for the tests.
So that they would not be treated as absolute path names
by the updated I/O code.
* Removed the comparison of chi2 values from the fit results.
This is a temporary measure, until the chi2 values "start behaving".
But for now it's not completely ridiculous that the CPU and GPU
code would produce different values for that parameter.
0 commit comments