The Art Of The Spiral ...

Goal for mapping library

• Simple to use
• Fast (-er than current one, at least)
• Self-contained (i.e. no need to read in ASCII or OCDB files)
• [ Swappable implementation ]

Detours

• Boost test, TDD
• Modern CMake
• Google Benchmark
• Boost Geometry Spatial Indices
• Hourglass interfaces
• Two Implementations
• C++ Templates
• Refactorings

Reasons Behind Detours (1)

• Boost test, TDD
  • code must be testeable, no question here
• Modern CMake
  • to understand how to deal properly with dependencies in vanilla CMake
• Google Benchmark
  • to learn how to (micro)benchmark code, as eventually we'll deal with performance...

Reasons Behind Detours (2)

• Boost Geometry Spatial Indices
  • to try not reinvent the algorithmic wheel
• Hourglass interfaces
  • to offer clean library, easy to evolve, accessible to other languages
• Two Implementations
  • to get a fighting chance to have a good interface

Reasons Behind Detours (3)

• C++ Templates
  • for performance
  • need(ed) a refresh as used to be a no-go in AliRoot
• Refactorings
  • safe with tests in place
  • is that even a detour btw ?

Simplifications Or Complexifications

• Current (AliRoot) implementations differ for St12 and St345

  • different implementers, different times
  • but also different detectors
  • how much difference is justified ?

• Initial tests with Boost Geometry Indices show that maybe it's a good fit only for St12 ?

The Norm And The Deviants

• "Regular" motif = (up to 64) adjacent pads of same size

• 210 motif types in total

  • but only 5 leading to non-regular motifs
  • localized in 2 (over 21) segmentation types

• Trying to avoid the minority of deviants to drive the implementation

Current Status (finally...)

alo branch "hourglass" for latest wip.

• Good : have a vs-aliroot test to check correctness of the implementations

• Good : have a vs-aliroot bench to check speed

• (Almost) good : have a C-like interface (almost) defined

• Bad : currently two incomplete implementations

  • impl1 is missing full channel access and is a template madness
  • impl2 is WIP (but promising ;-))

• Bad : I'm late

Planning ...

JIRA "Agile" planning

Reconstruction planning (1)

Reconstruction planning (2)

The Notion of Done

• Well underway is not the same as done...
• Done would be : in O2 repo, tests and documentation ready

Integration

• should put what we have in AliceO2 repo
• the sooner the better
• MID first ?

Next Hackathon

• @Subatech ?
• Would week 5 (Jan 30 - Feb 1) work ?

or

Backup

Hourglass pattern

impl1 : find pad by (x,y) / [segtype] / [bending]

❯ ./benchSegmentation --benchmark_filter="(0|16)/(0|1)"
Run on (8 X 2800 MHz CPU s)
2018-01-11 11:32:18
---------------------------------------------------------------------------------------
Benchmark                                Time           CPU Iterations UserCounters...
---------------------------------------------------------------------------------------
BenchO2/hasPadByPosition/0/1          2000 us       1999 us        316 n=500.126k/s nin=428.108k/s
BenchO2/hasPadByPosition/0/0          2139 us       2139 us        320 n=467.447k/s nin=387.981k/s
BenchO2/hasPadByPosition/16/1          346 us        346 us       1881 n=2.88809M/s nin=2.80433M/s
BenchO2/hasPadByPosition/16/0          246 us        246 us       2791 n=4.05993M/s nin=3.95437M/s
BenchO2/hasPadByPosition/20/1          177 us        177 us       3934 n=5.65121M/s nin=5.65121M/s
BenchO2/hasPadByPosition/20/0          126 us        126 us       5234 n=7.9437M/s nin=7.9437M/s
BenchAliRoot/PadByPosition/0/1         353 us        353 us       1932 n=2.83682M/s nin=2.33754M/s
BenchAliRoot/PadByPosition/0/0        1611 us       1611 us        413 n=620.81k/s nin=516.514k/s
BenchAliRoot/PadByPosition/16/1        202 us        202 us       3287 n=4.94486M/s nin=4.7619M/s
BenchAliRoot/PadByPosition/16/0        172 us        172 us       3707 n=5.80993M/s nin=5.58334M/s
BenchAliRoot/PadByPosition/20/1        176 us        176 us       3471 n=5.68363M/s nin=5.68363M/s
BenchAliRoot/PadByPosition/20/0        156 us        156 us       4431 n=6.39301M/s nin=6.39301M/s

Test project /Users/laurent/alice/sw/BUILD/alo-latest-clion/alo

    Start 1: testJsonMapCreator
1/8 Test #1: testJsonMapCreator ...............   Passed    9.11 sec
    Start 2: testContourCreatorVsAliRoot
2/8 Test #2: testContourCreatorVsAliRoot ......   Passed    4.81 sec
    Start 3: testSegmentationVsAliroot
3/8 Test #3: testSegmentationVsAliroot ........   Passed   23.14 sec
    Start 4: testSomeGlobalNumbers
4/8 Test #4: testSomeGlobalNumbers ............   Passed    2.04 sec
    Start 5: testContour
5/8 Test #5: testContour ......................   Passed    0.06 sec
    Start 6: testimpl1
6/8 Test #6: testimpl1 ........................   Passed    0.46 sec
    Start 7: testSegmentation1
7/8 Test #7: testSegmentation1 ................   Passed    1.77 sec
    Start 8: testSegmentation2
8/8 Test #8: testSegmentation2 ................***Failed    0.07 sec

88% tests passed, 1 tests failed out of 8

Total Test time (real) =  41.48 sec

The following tests FAILED:
          8 - testSegmentation2 (Failed)