Floating point precision

aliirmak · Post by **aliirmak** » Wed Jan 23, 2008 6:00 am

Hi all,

What is the float point precision of opensees? Is it the same as c++?

Post by **fmk** » Fri Jan 25, 2008 3:18 pm

OpenSees uses double variables for virtually everything internally. the output
precision varies, there are commands to change the defaults.

aliirmak · Post by **aliirmak** » Mon Jan 28, 2008 10:04 am

If we could decrease the precision , would it be in favor of shorter(it should be noticeable) analysis times?

Post by **fmk** » Tue Jan 29, 2008 10:52 am

depends on your computers memory and model size .. float versus double arithmetic is about the same, depends on cpu's fpu but some just treat float as a double .. only benifit i see if in reduced memory size and if that results in less paging.

aliirmak · Post by **aliirmak** » Tue Jan 29, 2008 11:17 am

so that means, reducing precision may be advantageous if we were analyzing having a structure one billion elements.

Anyway, do you think to publish opensees in future compiled for specific processors such as intel or amd, mips risc etc...

At least I know that, instead of gcc, intel c++ compiler gives better run times(faster analysis) due to optimized compiler algorithm for intel processors.