[LinearAlgebra] Speedup accumulation on BTDMatrix

[alxbilger]

Validated with the following benchmark. The benchmark BM_BTDMatrix_addBlock uses the new implemented accumulation function and must be compared to BM_BTDMatrix_add.

-------------------------------------------------------------------------------
Benchmark                                     Time             CPU   Iterations
-------------------------------------------------------------------------------
BM_BTDMatrix_add<6, SReal>/64              6212 ns         3069 ns       224000
BM_BTDMatrix_add<6, SReal>/512            49857 ns        32087 ns        22400
BM_BTDMatrix_add<6, SReal>/1024          103536 ns        54408 ns        11200
BM_BTDMatrix_addBlock<6, SReal>/64          602 ns          391 ns      1600000
BM_BTDMatrix_addBlock<6, SReal>/512        5607 ns         3432 ns       186667
BM_BTDMatrix_addBlock<6, SReal>/1024      12176 ns         6417 ns       112000

alxbilger/SofaBenchmark#36

---

By submitting this pull request, I acknowledge that
I have read, understand, and agree SOFA Developer Certificate of Origin (DCO).

---

Reviewers will merge this pull-request only if

• it builds with SUCCESS for all platforms on the CI.
• it does not generate new warnings.
• it does not generate new unit test failures.
• it does not generate new scene test failures.
• it does not break API compatibility.
• it is more than 1 week old (or has fast-merge label).

[alxbilger]

[ci-build][with-all-tests]

[hugtalbot]

Wow, @fredroy and @epernod might enjoy it!

[fredroy]

Wow, @fredroy and @epernod might enjoy it!

I tried on the 3instrument_collis (BeamAdapter) and there is no noticeable speed-up (or speed-down for that matter) ; but maybe this is not a good use case

[alxbilger]

I tried on the 3instrument_collis (BeamAdapter) and there is no noticeable speed-up (or speed-down for that matter) ; but maybe this is not a good use case

I am not sure that this new function is called. I think #4253 must be merged first.

Anyway, I don't expect a huge speedup in a scene. This kind of speed up is significant with a high number of DoFs, and it's not the case when using beams. I expect that you save a few microseconds by time step, which is really not significant.

[fredroy]

I tried on the 3instrument_collis (BeamAdapter) and there is no noticeable speed-up (or speed-down for that matter) ; but maybe this is not a good use case

I am not sure that this new function is called. I think #4253 must be merged first.

Indeed ! 🚀

master;
[BatchGUI] 5000 iterations done in 66.0261 s ( 75.7276 FPS)

master+ this branch + #4253 :
[BatchGUI] 5000 iterations done in 58.863 s ( 84.943 FPS)

[alxbilger]

@fredroy thanks for the benchmark. I am pleased with the results.

[fredroy]

I tried on the 3instrument_collis (BeamAdapter) and there is no noticeable speed-up (or speed-down for that matter) ; but maybe this is not a good use case

I am not sure that this new function is called. I think #4253 must be merged first.

Indeed ! 🚀

master; [BatchGUI] 5000 iterations done in 66.0261 s ( 75.7276 FPS)

master+ this branch + #4253 : [BatchGUI] 5000 iterations done in 58.863 s ( 84.943 FPS)

This was on WIndows11+MSVC2022 ; on ubuntu/gcc no speedups 😓