Validation and benchmarking of latest examples with single-precision geometry

[agheata]

AdePT can be built against VecGeom compiled with -DSINGLE_PRECISION=ON. When this happens, vecgeom::Precision typedef becomes float and all geometry interfaces work in single precision automatically. The examples were adapted to use automatically this feature when available. To validate and test performance in this mode, we should:

• verify that the latest examples (starting with example12) compile and run in this mode (may not be the case for full CMS geometry)
• validate the output against standard double precision mode, understand differences and provide fixes if possible
• benchmark TestEM3 and more complex geometry examples

[hahnjo]

Also please validate with TestEm3, I currently see percent-level disagreement with double precision and Geant4.

[hahnjo]

The correctness of TestEm3 should be addressed by #163. However, more complex geometries such as cms2018 still require a (temporary) push on every step, which is potentially bad.

[agheata]

more complex geometries such as cms2018 still require a (temporary) push on every step, which is potentially bad

We expected this since not all solids are for the moment single-precision proof, in particular the complex ones (polycone, polyhedra). We will need to continue the case-by-case shape_testXXX single-precision debugging to make the single-precision mode more robust.

[hahnjo]

The correctness of TestEm3 should be addressed by #163. However, more complex geometries such as cms2018 still require a (temporary) push on every step, which is potentially bad.

This turns out to be the reason for the wrong results when activating magnetic field in TestEm3: During field propagation, especially of low energy particles, the chordLen will get smaller than kPush. In that case, even though the pushing is temporary, the navigator has to assume that there is no boundary, just because it doesn't make sense to query a negative distance. This essentially leads to particles not seeing any boundary anymore, and depositing their energy in the wrong volume / material.