Alternative to Bresenham's raytracing #167

Activity

1. glpuga

   added

   enhancementNew feature or requestNew feature or request

   on May 1, 2023
2. hidmic commented on May 3, 2023

   

   hidmic

   on May 3, 2023

   Last edited by hidmic

   Collaborator

   More actions

   Evaluate the performance of Bresenham's algorithm for the Bean Sensor Model to determine if it's worth optimizing

   Note benchmarks are included in #160.

   Integer arithmetic-based Bresenham algorithm benchmark

   Run on (16 X 3200 MHz CPU s)
   CPU Caches:
     L1 Data 32 KiB (x8)
     L1 Instruction 32 KiB (x8)
     L2 Unified 512 KiB (x8)
     L3 Unified 16384 KiB (x1)
   Load Average: 0.52, 1.47, 2.08
   ***WARNING*** CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead.
   -----------------------------------------------------------------------
   Benchmark                             Time             CPU   Iterations
   -----------------------------------------------------------------------
   BM_Bresenham2i/Standard/128        92.1 ns         92.1 ns      7475814
   BM_Bresenham2i/Standard/256         181 ns          181 ns      3802972
   BM_Bresenham2i/Standard/512         360 ns          360 ns      1933997
   BM_Bresenham2i/Standard/1024        704 ns          704 ns       974018
   BM_Bresenham2i/Standard/2048       1411 ns         1410 ns       495547
   BM_Bresenham2i/Standard/4096       2800 ns         2799 ns       250246
   BM_Bresenham2i/Standard_BigO       0.69 N          0.68 N    
   BM_Bresenham2i/Standard_RMS           1 %             1 %    
   BM_Bresenham2i/Modified/128         390 ns          390 ns      1794217
   BM_Bresenham2i/Modified/256         764 ns          764 ns       900910
   BM_Bresenham2i/Modified/512        1527 ns         1526 ns       457221
   BM_Bresenham2i/Modified/1024       3054 ns         3054 ns       228170
   BM_Bresenham2i/Modified/2048       6136 ns         6135 ns       113027
   BM_Bresenham2i/Modified/4096      12276 ns        12275 ns        56818
   BM_Bresenham2i/Modified_BigO       3.00 N          3.00 N    
   BM_Bresenham2i/Modified_RMS           0 %             0 %

   Ray casting benchmark

   Run on (16 X 3200 MHz CPU s)
   CPU Caches:
     L1 Data 32 KiB (x8)
     L1 Instruction 32 KiB (x8)
     L2 Unified 512 KiB (x8)
     L3 Unified 16384 KiB (x1)
   Load Average: 0.89, 1.00, 1.61
   ***WARNING*** CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead.
   ---------------------------------------------------------------
   Benchmark                     Time             CPU   Iterations
   ---------------------------------------------------------------
   BM_RayCasting2d/128         175 ns          175 ns      3972583
   BM_RayCasting2d/256         298 ns          298 ns      2362109
   BM_RayCasting2d/512         589 ns          589 ns      1169115
   BM_RayCasting2d/1024       1168 ns         1168 ns       589929
   BM_RayCasting2d_BigO       1.15 N          1.15 N    
   BM_RayCasting2d_RMS           3 %             3 %

   These numbers don't look terrible, but when ray tracing 100 times for 2000 different states in a ~400 x ~400 grid, it adds up quickly to ~50 ms. I did some profiling and while we could shave off some of it by eliminating Eigen types, it'll probably take a different algorithm (approximate, multi-resolution, whatever) to get it in control for denser grids.
3. glpuga commented on May 4, 2023

   

   glpuga

   on May 4, 2023

   CollaboratorAuthor

   More actions

   It's great that we have the benchmarks already.

   I think when deciding if it's worth to pursue further optimization we need to evaluate the execution time in the context of the total execution time per iteration.

   A 5x improvement in Bresenham will be barely be noticeable if bresenham only accounts for 15% of the total iteration time, but it may be crucial if it represents 80% of the iteration time.
4. hidmic commented on May 4, 2023

   

   hidmic

   on May 4, 2023

   Collaborator

   More actions

   FWIW I don't think we can improve the computation itself much further. We need to reduce the number of computations.
5. nahueespinosa commented on May 25, 2023

   

   nahueespinosa

   on May 25, 2023

   Last edited by nahueespinosa

   Member

   More actions

   @glpuga FYI, here are the flamegraphs from likelihood field and beam models for comparison.
   Downloading the files and re-opening them in Chrome should enable the zoom feature.

   Main version: e72e148

   
   

   🚀React with 🚀1Reacted by Gerardo Puga
6. nahueespinosa

   added a commit that references this issue on May 31, 2023

   Extend raycast microbenchmarks (#201)

   ...

   9436030
7. hidmic

   changed the title [-]Research alternative to Bresenham's raytracing algorithm[/-] [+]Alternative to Bresenham's raytracing[/+] on Sep 11, 2025
8. Ekumen-OS

   locked and limited conversation to collaborators on Sep 11, 2025
9. hidmic

   converted this issue into a discussion #506 on Sep 11, 2025