XSum C++

This is the C++ implementation of xsum which is fast exact summation using small and large superaccumulators.

This code include test cases in TestRunner.cpp.

This code is based on this commit from Radford M. Neal's xsum repository.

How to Run

• Make sure that cmake is installed
• To initialize build directory, run cmake -DCMAKE_BUILD_TYPE:STRING=Release -DCMAKE_EXPORT_COMPILE_COMMANDS:BOOL=TRUE --no-warn-unused-cli -S./ -B./build
  • No need to run after the first time
• In the root dir, run cmake --build ./build --config Release --target all -j 12 --
• Then run ./build/xsum

Note

• xsum wont return -0 (negative zero), however, this handle -0. See test cases in TestRunner.cpp.
  • https://gitlab.com/radfordneal/xsum/-/blob/master/api-doc?ref_type=heads#L203-L206

References

• https://arxiv.org/abs/1505.05571
• https://gitlab.com/radfordneal/xsum
• https://github.com/tc39/proposal-math-sum