- Dec 21, 2017
-
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
- Dec 20, 2017
-
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Benedikt Hegner authored
Added a new event loop manager for HLT like usage, that is very simple and very fast scheduling (mr !537)
-
Benedikt Hegner authored
-
Benedikt Hegner authored
Closes #1
-
-
Benedikt Hegner authored
-
Benedikt Hegner authored
-
Benedikt Hegner authored
-
-
Benedikt Hegner authored
-
Benedikt Hegner authored
-
Benedikt Hegner authored
-
-
Benedikt Hegner authored
-
Benedikt Hegner authored
Fixed use of zlib See merge request !553
-
Sebastien Ponce authored
-
- Dec 18, 2017
-
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
Marco Clemencic authored
-
-
Marco Clemencic authored
-
Marco Clemencic authored
See GAUDI-1312
-
- Dec 15, 2017
-
-
Charles Leggett authored
Fix a bug in data/condition object state detection in temporal and topological tracer's (3T) graph assembling See merge request !543
-
- Dec 14, 2017
-
-
Illya Shapoval authored
-
- Dec 13, 2017
-
-
Benedikt Hegner authored
See merge request !533
-
- Dec 12, 2017
-
-
Sebastien Ponce authored
-
Sebastien Ponce authored
-
- Dec 08, 2017
-
-
Sebastien Ponce authored
This includes in particular : - the use of none_of for ordering the algorithms - dropping of duplicated warning messages - usage of unique_ptr for the EventContext - code simplifications
-
Sebastien Ponce authored
-
- Dec 06, 2017
-
-
Sebastien Ponce authored
This new event loop manager has an integrated scheduler that is very simple : it sequentializes the list of algorithms to run according to Data Dependencies and schedules full events in tbb, so that one task runs serially all algorithms one after the other. The scheduling of events is pure fifo : schedule as many events as slots in the white board initially and triggers the next one when oen finishes. In comparison with the Avalanche scheduler, many features are not present : no resources, no dynamic rearrangement of the sequence, no IO bound algos in particular. On the other hand, the scheduling is much faster and allows to sustain high rates of events.
-
Sebastien Ponce authored
This was doing a linear search on the slots, taking a lock on each of them to check their availability. Not only the algorithm was O(n) while it could be O(1) but the locking was not scaling when many threads were at work. Putting a fifo queue with its own small lock solves the issue.
-