Is it possible to get the source code to java.exe runtime vm and recompile it with the “stop on NaN” floating point option? The default is NaN & Inf and try to continue which never ends well but hides where the initial error occurred.
What currently happens is our Intel Fortran for Windows DLLs inherit the setting from the process exe--we spend a lot of time debugging during testing rather than having it tell us where it hurts (divide by zero, etc).
The Intel Fortran discussion forum is like “Dear Abby” with frustrated maintenance programmers with “Ask Dr Fortran” the moderator black belt. This is one of the FAQs.
Developer’s Guide seems to leave out the where is the source code repo and how to compile from source specifics? I assume there is a git or svn site somewhere?
Rebuilding everything locally would be one way to do it—we’d be happy to let the build machine make a debug version of the JDK for testing JNI. Please add this suggestion to the list.
Darrell
San Diego, California, USA
p.s. to make python run faster they recommend creating a library in Fortran and using f2py! To make gfortran faster, Intel has ported and optimized all the math libraries and claims to be 80% faster. Maybe the same can be done with the java codebase? Optimize the compiler to make the compiler and runtime work faster…