Nick Sabalausky wrote:
Great release! :)
Although I'm curious, where is the 3x speed improvement from? Just general
misc optimizations, or something specific?
It's just the exp(), expm1(), and exp2() functions. Previously they used
some extremely slow asm instructions. I wrote completely new
implementations that are fast. It was done primarily because the C std
math library was inaccurate on Mac OSX. Most people won't care. But in
the past, someone (Robert Frazer, I think) had complained that DMD's
exp() was the bottleneck in his code, and was less than half as fast as
MSVC/Intel, so the speed of these functions is actually critical sometimes.