Effect of -fno-reorder-blocks?

[Jan Hubicka]

> >>>>> Hans-Peter Nilsson writes:
> 
> Hans-Peter> I see unexpected results with a ss of gcc-3.2 (gcc-20020902).  I
> Hans-Peter> get better code with -fno-reorder-blocks *especially*
> Hans-Peter> considering effects on the cache (a simple one: 8k unified
> Hans-Peter> 1-way, 32-byte lines) for cris-axis-linux-gnu.  I'll add more
> Hans-Peter> substance to this claim once I've done more thorough testing,
> Hans-Peter> with newer sources, though I believe no changes have been done
> Hans-Peter> that would improve this.  I wonder if there are similar effects
> Hans-Peter> for other architectures, with more intricate branch timings and
> Hans-Peter> cache semantics.  I know about the the messages about SH and
> Hans-Peter> -fno-reorder-blocks; that's the reason I tried it.
> 
> 	This affects PowerPC as well. See PR optmization/6990.
> 
> 	When I pointed this out to Honza last July , he mentioned that
> this type of problem is addressed in the STC implementation on the CFG
> branch.  I am not sure about the current status of merging those
> improvements. 
THe branch didn't bring consistent improvements, so we didn't merged it
yet.  Josef is working on it right now and his new code brings small,
but at least quite consistent, improvements on Athlon now.
Athlon is not perfect testing platform (neighter P4 is) for such type of
pass.  Perhaps tunning for some RISC architecture first would help us to
better understand what is going on.

Honza
> 
> David