Subject: Re: [boost] proposal - Statistically robust function timing for performance decisions.
From: Edward Grace (ej.grace_at_[hidden])
Date: 2009-07-18 05:14:25
On 18 Jul 2009, at 07:40, joel wrote:
> Add me to the bandwagon ;)
>
> How do you handle the fact that sometimes, compiler may otpimize out
> benchamrk code because it has no side-effects ?
Easy. I don't!
When I started this I was concerned that might happen and was quite
surprised to see that the compiler (g++ 4.0.1) didn't optimise out
the function.
More generally, I don't think a function such as:
void f();
can be considered free of side-effects unless you can guarantee that
it does not access global variables (or otherwise sneakily modify
state). From the prototype of a compiled function I don't see how
there's any way to know this. With fully templated code it may be
far more subtle since the compiler can potentially 'see' everything.
That said, at the moment I want to concentrate on making the timings
reliable on the assumption that problems such as you describe can be
circumvented later on by someone who knows more about compilers than me!
-ed