Test Program for Powerstation and Microsoft Fortran

Recently I sent this e-mail to nmusers mailing list: > This is a message to people who use Microsoft Powerstation Fortran. > Jaap Mandema brought to our attention a problem in a generated PRED > routine when it is compiled with option /Ox (full optimization). > The results are incorrect. > > I sent a little test program to Microsoft. The software developement > group is working on it, but it is unlikely that they will have any > bug fix or patch soon. The workaround is: > > Always use /Ox /Op options together. > This should be done for all compiles: NONMEM, PREDPP, FSUBS. > (NMTRAN does not need to be optimized.) > > I don't know if this bug affects the older Microsoft Fortran 5.x > compilers. > > Would anyone like to try? If so, let me know and I'll send the little > test program. > > -- Alison Boeckmann Here is a follow-up. Thanks to Bill Bachman, David Bourne, and Jaap Mandema, who gave this a try. The test program works correctly when compiled with Microsoft Fortran 5.1, at all levels of optimization, including /Ox by itself. It is only the users of Powerstation Fortran who should be careful not to use /Ox by itself. David Bourne also tried the program on a Macintosh which has no floating point chip. He tried "MS Fortran Compiler v5.1 IBM - with Alternate Math Package Library - Running under SoftPC on a Macintosh IIci". The results were correct with /Ox (and also with /Ox /Op, and with no optimization.) David tried "Mac Fortran II v3.1.2 - Absoft - On a Macintosh IIci". The results of the program were incorrect with all compiler options that he tried, including no optimization at all. It appears that Mac Fortran II, with David's configuration, is flawed and should not be used for NONMEM. Alison