Floating-point underflow

Tino_Zerbini — Fri, 19 Aug 2011 00:40:56 GMT

When I ran debug mode I get a lot of "First-chance exception 0xsomehexnums in ProgramName.exe: 0xanotherhexnum: Floating-point underflow.

I set compiler switch /fpe:0

Question is:

1. What is this exception? (is it a source code problem that need to be fixed)
2. Can I ignore this or does it does it have an effect on the final answer?

Floating-point underflow

Steven_L_Intel1 — Fri, 19 Aug 2011 01:47:03 GMT

When you set /fpe:0, an exception handler is established which, among other things, looks for underflow exceptions and changes the result of the operation to zero. It may or may not be a problem, it depends on what your program is doing. Your program is doing calculations that underflow the normalized range. If you had not set /fpe:0, you would get a "denormalized" number with reduced precision. These can cause odd results if you then feed them back into later calculations.

I suggest that you identify the source of the underflows and decide whether they should be happening or not.

topic Floating-point underflow in Intel® Fortran Compiler

Floating-point underflow

Floating-point underflow