I found to be rather interesting, somewhat amusing, and a little disturbing.
Indeed. One of the comments suggests that the mapping of real10 to real8 is done "(to cheat on benchmarks)", but I am under the impression that the (Intel) floating point processor does all it computations in 80-bits, even if the results are truncated when they are stored to ram.
If that's the case, and I am pretty sure that I read that on a reliable source which I will attempt to relocate, then there would be no performance advantage in not storing the 80-bits after the computation?
In reply to Re^5: XS, C doubles and -Duselongdouble
by BrowserUk
in thread XS, C doubles and -Duselongdouble
by syphilis
| For: | Use: | ||
| & | & | ||
| < | < | ||
| > | > | ||
| [ | [ | ||
| ] | ] |