[sakras]

A while ago at my company we switched from GCC to Clang, and noticed a couple of massive regressions (on the order of 50%?) in performance having to do with floating point.

After profiling for a bit, I discovered that suddenly a lot of time was spent in isinf on Clang and no time in GCC… Clang was emitting a function call where GCC wasn’t. I happened to randomly change isinf to std::isinf (it’s a random habit of mine to put std:: in front of these C functions). Suddenly the regression disappeared! I guess on Clang only std::isinf was a compiler intrinsic while GCC recognized both? Anyway, that’s my small-change optimization story.

[10000truths]

C defines isinf as a macro, whereas C++’s std::isinf is a function. Perhaps the discrepancy has to do with differences in how they’re evaluated?

[jeffrallen]

> random habit of mine to put std:: in front of these C functions

And did you learn your lesson about making random changes that "shouldn't matter" without proving they don't matter? :)

I find that once I spend the time to make these changes correctly, they are not worth the time to make correctly.