we can mull over one example to see if there's anything inspriational

think outside the box here
even to the point of applying a preprocessing step
e.g., comment-like hints in the code for a preprocessor

why a preprocessor?
all of the esoteric goop can be restricted to one spot
including e.g., the extra speedup that uses the magic X instruction on 
processor Y for compiler Z
and the original code is intact except for the hints

I don't know if that's workable
but that's the kind of approach that is readable and maintainable
*and* when the compiler optimizers finally catch up can be eliminated from the 
source with sed

On Mon, 21 May 2012 09:57:37 +0200 Irek Szczesniak wrote:
> On Tue, May 1, 2012 at 7:22 AM, Glenn Fowler <[email protected]> wrote:
> >
> > I appreciate the effort you put into coding the prefetch example
> > but there's no way we would sprinkle #ifdef'd code like that across ast
> > it would be a maintenance headache
> > and what will the next speedup dujour do to the code
> > this feels really close to coding in asm
> > (although we have done manual unrolling and asm in a few spots)

> Glenn, can we make a bargain where you accept the patch when Roland
> removes the #ifdef and does some other restructuring? We've tested a
> newer patch from Roland and found that every platform we benchmarked
> (IA64, x86-64/AMD, x86-64/Sandybridge, x86-32, ARM, MIPS64, SPARC64,
> Oracle Ultrasparc T2/T3/T3+) show significant performance gains and I
> start to think it is worth taking, even if the design highlights the
> 'portable assembler' heritage of the C language :)

> Irek

_______________________________________________
ast-developers mailing list
[email protected]
https://mailman.research.att.com/mailman/listinfo/ast-developers

Reply via email to