- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
DOE (Department of Energy) National Laboratories are working with scientists who are requesting identical numerical accuracy for all optimizing options. Is the IEEE 754 arithmetic operator implementation in IFX more robust than in IFORT?
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
You will want to compile with -fp-model precise for best results. But there are other differences you may not be able to account for - I discussed this in a Supercomputing presentation: Improving Numerical Reproducibility in C/C++/Fortran Changing compilers will almost certainly lead to small differences due to order of operations.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The only way to check is to write a check program and look at the statistical results. But Steve is correct there will be differences, small but different. That is the fun of statistical science, hang on all science is statistical in the end.

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page