fltconsistency vs -fp-model

[qrio_ithotmail_com]

I'm trying to get my debug code to match the optimized code (O2), and I'm able to do so using the fltconsistency and fmath-errno flag. However, I'm unable to get them to match using any of the fp-model flags. The user's manual recommends using the fp-model flags and the performance drop from using the fltconsistency flag is unacceptable. What are the differences between the two flags and is there a set of flags that can mimic what fltconsistency does?

Thanks!

[TimP]

I had to read this about 3 times to begin to guess what you may be looking at. -fltconsistency probably promotes many operations to double precision, which is time-consuming for SSE code. If you are depending on promotion to double precision for desired accuracy, you should find out where and write it explicitly into your source code. Then -fp-model source should be sufficient.