Tags: davidson16807/tectonics.cpp
Tags
splitting profiling/ into folders for single & multi thread optimization gprof works well for diagnosing single thread situations but openmp requires switching to std::chrono. The two track runtime in different ways so their results cannot be directly compaired. The two situations require different compiler flags so switching between them is painful. Keeping separate folders lets a developer easily switch between optimization for both cases
breaking footprint of out-of-order traversal into several smaller ones The idea is that multiple out-of-order traversals over smaller footprints will be easier to cache than a single out-of-order traversal over a large footprint. In this case this only seems to decrease performance, regardless of grid resolution. This seems to validate my original design.
reestablishing baseline for new phase of profiling experiments existing notation was becoming a burden and microoptimizations became so numerous that test cases no longer measured just one thing
miscellaneous improvements and cleanup in dymaxion::
significant performance gain by switching math::modulus with % operator unfortunate, admittedly unexpected, but I'll still gladly take the performance gain
significant performance gains from caching vector for polarity check
performance gains by caching coordinate bases for each triangle in Grid
dropping call to standardize when able improves runtime on all -O levels
adding and applying replacement script to profiling output
PreviousNext