I have found a performance problem on abs() that I don't understand:

http://codepad.org/guZyeFOK

In that little program core.stdc.stdlib.abs leads to a much faster (11 or 12 times faster with DMD on Windows32, compiling with -O -release -inline -noboundscheck) program compared to using std.math.abs, despite (as shown in the asm) core.stdc.stdlib.abs is not inlined.

Do you understand what's going on?

Bye,
bearophile