Skip to content

Optimization of Memory usage by L1TGC LUT for Run-2 (TGCTriggerDbAlg)

Junpei Maeda requested to merge junpei/athena:TileLUTMemoryOptimization into master

As discussed in ATR-22282, it is pointed out that TGC L1 simulation packages have huge memory usage.

For the first step, to reduce malloc size by TGCTriggerDbAlg, the variable structures of Look-Up Tables for the L1TGC system is redesigned. TGCTriggerDbAlg reads databases of Coincidence-Window of TGC-BW (Run-1/2), EIFI, and Tile.

Since the fullCW option is generally not assigned to CW, especially BW, the LUT usually is symmetric for each side and octant. The array of fixed size of size*octant is changed to the variable vector/map.

The qtest of q221 of ttbar and private single-muon sample is passed. The malloc size has been reduced as expected. The processing time at LVL1Trigger (TrigT1TGC) is a bit increased. This will be optimized in the different commits/MR.

Edited by Junpei Maeda

Merge request reports