FP16_Base compilation fixes

[slava77]

I tested by adding -DFP16_Base to RecoTracker/LSTCore/standalone/Makefile (in ALPAKA_CUDA) and RecoTracker/LSTCore/standalone/LST/Makefile (in ALPAKACUDA) as well as ROCM.

ref before this PR

with the changes in Makefile:

There is a hint that the code is a bit faster.

The memory reduction is probably more important currently.

[slava77]

also, this PR is relative to a new branch CMSSW_14_1_0_pre3_LST_X_LSTCore_realfiles_batch9

[slava77]

with the changes in Makefile:

comparisons also looked OK when running on GPU L4 cgpu-1 dev=0
(somewhat older ref from sometime in batch7) https://uaf-10.t2.ucsd.edu/~slava77/sdl/efficiencies/testSOA_00013d-PU200/summary/fakerate.html
vs this PR
https://uaf-10.t2.ucsd.edu/~slava77/sdl/efficiencies/testFP16_7c3358-PU200/summary/fakerate.html

[slava77]

/run all

[slava77]

running the tests, although results will not be visible

[github-actions]

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     39.6    390.8    194.9    127.4    135.7    544.1    117.7    243.4    105.8      3.4    1902.6    1318.9+/- 364.7     502.6   explicit[s=4] (target branch)
   avg     39.8    388.8    190.5    128.4    135.7    546.0    118.2    242.3    105.8      3.1    1898.6    1312.8+/- 367.0     501.9   explicit[s=4] (this PR)

[github-actions]

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.

[VourMa]

I will force-push the updated realfiles branch also to batch9, so that they are consistent.

[slava77]

I will force-push the updated realfiles branch also to batch9, so that they are consistent.

do we need updates to the CI? (before trying to rerun the tests)

[VourMa]

do we need updates to the CI? (before trying to rerun the tests)

I didn't think of that. Let's see what happens to #119, where I naively issued the command.

[slava77]

@VourMa
if this looks OK, perhaps it can be merged

[slava77]

/run all

[slava77]

@ariostas
in case you were working on Matti's comments, I added the bool part here (the other part about CopyToDevice seemed less trivial)

[VourMa]

@VourMa
if this looks OK, perhaps it can be merged

Once the tests finish, I will merge.

[github-actions]

The PR was built and ran successfully in standalone mode. Here are some of the comparison plots.

The full set of validation and comparison plots can be found here.

Here is a timing comparison:

   Evt    Hits       MD       LS      T3       T5       pLS       pT5      pT3      TC       Reset    Event     Short             Rate
   avg     42.6    393.7    196.6    135.0    140.8    544.4    119.1    247.5    106.4      3.3    1929.5    1342.4+/- 368.2     511.1   explicit[s=4] (target branch)
   avg     41.4    395.8    197.4    136.0    139.8    540.0    118.1    247.4    105.5      3.4    1924.8    1343.5+/- 377.0     519.8   explicit[s=4] (this PR)

[github-actions]

The PR was built and ran successfully with CMSSW. Here are some plots.

OOTB All Tracks

The full set of validation and comparison plots can be found here.