Towards a 64-bit version of scientific libraries

While most of the current scientific computers are still 32-bit machines, research and computation requirements, specially in the field of High Energy Physics, already require greater precision. The authors frame the problem in the context of FORTRAN source compatibility versus object-code performance, and evaluate the automatic promotion/padding features of the FORTRAN compilers on the different computer platforms available at CERN. They make a comparative study taking into account different factors, including implementation of extended INTEGER and floating-point arithmetic. They propose one of the studied vendors' implementations as the right intermediate choice while waiting for true 64-bit hardware platforms