On 07/13/2010 05:49 PM, Vladimir Makarov wrote:
I checked Bernd's patch on SPEC2000 on x86 (32-bit modes) last week.
Code of few benchmarks changed (only vpr, gcc, crafty, parser, eon,
vortex, mesa, facerec, sixtrack, apsi). The codes size for SPECINT2000
was improved by 0.03% with patch and code size for SPECFP2000 was worsen
by 0.006%. There is no visible improvement/degradation on changed
benchmarks. Compilation time increase was quite insignificant (less
than 0.2%). So as I wrote Ken Zadeck two years ago who tried to solve
the problem for the old RA, the complexity of the problem solution does
not justify the solution benefits.
I feel one of the major benefits is enhanced predictability - it becomes
easier to tune other code, e.g. to add splitters to help the
lower-subreg pass. If IRA is able to handle multiword conflicts more
gracefully, it becomes easier to demonstrate that such changes are
beneficial since the results are more well-behaved. It would also
eliminate most of the need for the clobbers patch I sent earlier today.