Command Line Options Cleanup, main branch (2024.03.05.)

[krasznaa]

This all started out from a report on Mattermost that the throughput measuring executables are not working at the moment. But instead of merely fixing a typo in the "options code", I decided that it was time to do a thorough cleanup of it.

I did a number of things:

• Got rid of all templating from the "options structs". I do not believe that we would have any need for using either float or double precision when specifying runtime parameters on the command line. So these options are always just specified as float-s, and then those configuration parameters may be used in double precision calculations later on.
• Changed the name of some of the structs. Weirdly enough the names of the source files were in all cases how I wanted to call the structs themselves. So this was a pretty unfortunate inconsistency in the code so far. 😦
• Switched all of the structs to a formalism where they would make boost::program_options write the configured parameters into C++ variables "directly". Instead of declaring the option first, and then extracting the option ourselves later on.
  • I still kept all of the read(...) functions, since in a few cases we still need to do some non-trivial operations while interpreting the user-provided arguments. Though much of the code from these read(...) functions could be removed.
  • This is mainly to reduce code repetition. Which was the reason for the typo in the first place. 🤔
• Introduced "output operators" for all of the options structs. Similar to what we had already for a few of them.
  • Then updated all of the example applications to print their configured parameters using these helper operators. So that they would present a slightly more uniform look-and-feel. Like:

[bash][atspot01]:traccc > ./out/build/sycl/bin/traccc_seq_example_sycl --detector-file=tml_detector/trackml-detector.csv --digitization-config-file=tml_detector/default-geometric-config-generic.json --input-directory=tml_full/ttbar_mu20/ --events=2

Running the full tracking chain using SYCL

>>> Common options <<<
  Input data format            : csv
  Input directory              : tml_full/ttbar_mu20/
  Events                       : 2
  Skip                         : 0
  Target cells per partition   : 1024
  Check performance            : 0
  Perform ambiguity resolution : 1
>>> Detector options <<<
  Detector file : tml_detector/trackml-detector.csv
  Material file : 
  Grid file     : 
>>> Full tracking chain options <<<
  Digitization configuration file: tml_detector/default-geometric-config-generic.json

Running Seeding on device: NVIDIA GeForce RTX 2060
==> Statistics ... 
- read    95678 cells from 12529 modules
- created (cpu)  0 measurements
- created (cpu)  0 spacepoints
- created (sycl) 27862 spacepoints     
- created  (cpu) 0 seeds
- created (sycl) 3924 seeds
==>Elapsed times...
           File reading  (cpu)  228 ms
         Clusterization (sycl)  4 ms
                Seeding (sycl)  4 ms
           Track params (sycl)  0 ms
                     Wall time  238 ms
[bash][atspot01]:traccc >

This whole cleanup, I believe, was in the end very useful. Since I now have a new idea for why track finding in #509 would not be working. 🤔

[beomki-yeo]

In general I agree with all points so will hit approval once it passes the CI

[krasznaa]

Note that I also removed all of the "required options" as well. 🤔 Providing default values for absolutely all of the options. Because I've always found it pretty painful that none of our executables can just be run as-is. 😦 Well, now they can. At least most of them... (Since for some of the options it's not easy to find universal defaults... 😦)

So, something like this:

[bash][atspot01]:traccc > ./out/build/sycl/bin/traccc_seq_example

Running the full tracking chain on the host

>>> Common options <<<
  Input data format            : csv
  Input directory              : tml_full/ttbar_mu20/
  Events                       : 1
  Skipped events               : 0
  Target cells per partition   : 1024
  Check performance            : 0
  Perform ambiguity resolution : 1
>>> Detector options <<<
  Detector file : tml_detector/trackml-detector.csv
  Material file : 
  Grid file     : 
>>> Full tracking chain options <<<
  Digitization configuration file: tml_detector/default-geometric-config-generic.json

==> Statistics ... 
- read    54988 cells from 6997 modules
- created 15919 measurements. 
- created 15919 space points. 
- created 2340 seeds
[bash][atspot01]:traccc > ./out/build/sycl/bin/traccc_seq_example_cuda 

Running the full tracking chain using CUDA

>>> Common options <<<
  Input data format            : csv
  Input directory              : tml_full/ttbar_mu20/
  Events                       : 1
  Skipped events               : 0
  Target cells per partition   : 1024
  Check performance            : 0
  Perform ambiguity resolution : 1
>>> Detector options <<<
  Detector file : tml_detector/trackml-detector.csv
  Material file : 
  Grid file     : 
>>> Full tracking chain options <<<
  Digitization configuration file: tml_detector/default-geometric-config-generic.json

==> Statistics ... 
- read    54988 cells from 6997 modules
- created (cpu)  0 measurements     
- created (cpu)  0 spacepoints     
- created (cuda) 15919 spacepoints     
- created  (cpu) 0 seeds
- created (cuda) 2340 seeds
==>Elapsed times...
           File reading  (cpu)  136 ms
         Clusterization (cuda)  59 ms
                Seeding (cuda)  4 ms
           Track params (cuda)  2 ms
                     Wall time  203 ms
[bash][atspot01]:traccc >

[krasznaa]

It's ready now, after having updated the Alpaka and Kokkos executables as well. 😉