Re: [Qemu-ppc] TCG PPC performance regression?

[Mark Cave-Ayland]

On 06/03/12 09:58, Jan Kiszka wrote:

> > Since I can reproduce this problem 100% of the time, I'm now trying to
> > verify whether or not this is an actual bug with just my setup, or a bug
> > in QEMU. I've had a look at the Jan's patch and can't see anything
> > immediately obvious that would cause this, so I'm wondering if this is
> > perhaps somehow compiler related? Any hints or reproduction of results
> > would be greatly appreciated.
>
> Would I could imagine spontaneously is that my patch changed some memory
> allocation ordering and, thus, maybe some alignment. But this is just a
> blind guess.

Okay - so it looks like there are multiple bugs here, but it's primary 
cause is not your commit. The reason commit 
d5ab9713d2d4037fd56b0adddd26c8d4dc11cf09 struggles for any kind of 
performance is because of a separate ram_size bug which you fixed in 
commit 0ac543de94f4f5e7e79d59a56df60937cdd3cd24.

Unfortunately I haven't been able to do a comprehensive bisection 
because just after this point the memory API breaks PPC for quite some 
time, but I get the old (fast) speed at commit 
2dd3022826bb1ced27d12493a8f1f4b6d4bc71b7 (+0ac543 backport) while the 
next commit I can find that works which is notably slower is 
6c731dc2af1e80c410e8605098deb1ac16ab72f0.

I think I might try generating some gprof profiles to see if I can find 
out where the time is going.


ATB,

Mark.