This projects tests how well different paralization strategies and libraries (TBB, OpenMP) scale scale and impacts throughput.
The capture node is a rewrite of a previous hybrid implementation (HDK/For each sop ) found on OdForce (see the references)
References:
- https://www.cse.wustl.edu/~taoju/research/meanvalue.pdf
- https://forums.odforce.net/topic/8877-gmvc-general-mean-value-coordinates/?page=2
mkdir -p __build && cd "$_"
cmake .. -DCMAKE_INSTALL_PREFIX=~/houdini18.0/
make install
source ~/hfs18.0.416/houdini_setup
run -foreground -n $HOME/PRJ/pgmvc/example.hipnc
env_gcc650 # building with more 7.5.0 gcc leaks while debugging in gdb
cmake .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS="-g -O0"
gdb bin/hmaster-bin
run -foreground -n $HOME/PRJ/pgmvc/example.hipnc
b SOP_PGMVC_captAtt::cookMySop
%clang
-I$HOME/hfs18.0.416/toolkit/include
| 10000 points - i5-2520M CPU @ 2.50GHz | outer loop | inner loop |
|---|---|---|
| 19.8 | serial | serial |
| 12.0 | tbb | serial |
| 8.7 | tbb | serial |
| 10.5 | openmp | tbb |
| 8.3 | tbb | tbb |