Re: [Qemu-devel] [PATCH 1/9] TCG "sync" op

[Ulrich Hecht <uli@suse.de>]

On Saturday 17 October 2009, Edgar E. Iglesias wrote:
> I looked at the s390 patches and was also unsure about this sync op.
> I'm not convinced it's bad but my first feeling was as Aurelien points
> out that the translator shoud take care of it.

Indeed. I would have expected it to, in fact. But it doesn't. The sync op 
is the simplest and quickest way to get what I want that I could come up 
with. I'd be perfectly happy if TCG could handle aliases on its own, but 
doing a lot of ALU operations on every register access is not an option.

> Another thing I noticed was the large amount of helpers. Without
> looking at the details my feeling was that you could probably do more
> at translation time.

My experience is that helper functions have an undeservedly bad image. A 
pure const helper call is very easy to optimize away for TCG. Random bit 
shifting, comparing and branching isn't.

> Nice work!

Thank you.

CU
Uli

-- 
SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg)