Re: [rtl] Problems with RDTSC examples posted here recently

[Chunky Kibbles <gjb105@cs.york.ac.uk>]

On Wed, Oct 31, 2001 at 05:37:18PM -0700, John Regehr wrote:
> > method1  : 23573589435592    - 23573353927542 =   235508050
> > method2  : 23573353927556    - 23573353927556 =           0
> > method2A : 23573353927607    - 23573353927607 =           0
> >  -- also m2_1 - m1_1 =         14
> >  --  and m2A_1- m2_1 =         51
> >  finally m2A_1- m1_1 =         65
> 
> Take a look at the code that gcc generates when compiling your program
> with optimizations.  The "rdtsc" instruction is executed three times,
> the program sleeps, and then rdtsc is executed only once more - hence
> the strange results.  method2 and method2A are broken.  I don't
> understand what the "mov edx, ecx" instruction in these functions is
> supposed to accomplish.  Assuming that it is useful, you'll still need
> to inform gcc that you're clobbering ecx.
> 
> method1 or something like it is what I've always used.

Don't forget to make sure it's __volatile__, elsewise this is the very
epitome of what would produce dodgy results...

Gary (-;