This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFA: patch for coalescing stack slots
Dale Johannesen wrote:
This is going to inhibit scheduling in some cases, isn't it, by
reusing the
same memory location? Your numbers are persuasive that it's a win
overall
on x86, but I'm wondering if this is not necessarily true everywhere.
Do you have SPEC results for ppc?
No, I don't. Unfortunately, only slow ppc (450Mhz) machine with few
memory is avaliable for me.
I could run SPEC95 on this machine this night. I could also run
SPEC2000 for Itanic2.
I was awared about scheduling troubles. But my speculation is that
there will be not frequent coalescing for such machines because they
have enough registers. But it is hard to say what will be more
important (more dependencies for insn scheduler or data locality). I
will appriciate if you run SPEC2000 on a faster ppc machine (we should
be more oriented to modern machines) because bigger frequency could make
a difference for the two factors.
In any case, I'll report the results tomorrow or day after tomorrow. If
my theory is wrong, we could make an option to switch the optimization
on and make it default for at least x86.
Vlad