qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH] target-ppc: add Altivec instructions, patch-bom


From: Aurelien Jarno
Subject: Re: [Qemu-devel] [PATCH] target-ppc: add Altivec instructions, patch-bomb version
Date: Mon, 15 Dec 2008 07:40:59 +0100
User-agent: Mutt/1.5.13 (2006-08-11)

On Sun, Dec 14, 2008 at 06:14:33PM -0800, Nathan Froyd wrote:
> [Thiemo Seufer asked for a patch-bomb so people can conveniently comment
> on things and so the mail archive has a record of Signed-off-by.]

Yes, that's a good idea, it makes them a lot easier for comments.

> This patch series adds support for integer Altivec instructions to QEMU,
> including element-wise loads and stores.  It's a long patch series,
> since each instruction "family" (element-wise loads/stores, modulo
> arithmetic instructions, saturating arithmetic instructions, etc.) is
> separated out into its own patch.  Hopefully since the individual
> patches are so short, they will be more-or-less self-explanatory: the
> explanation for the patches is generally fairly short, on the order of a
> single line.

Very nice work :)

> The patch series is slightly cleaned up from the one I originally
> posted: there were a few problems with my tcg usage, and I didn't
> faithfully use some convenience functions.
> 
> Why only the integer instructions?  I originally wrote support for the
> whole instructions set, but I did it in the days of dyngen.  So a
> straight forward-port was out of the question.  The original patch
> sloppily used native floats everywhere, rather than the float32
> abstraction.  It also used C99 math functions to implement some of the
> more exotic Altivec instructions.  Both of these decisions mean that
> some care has to be taken in porting the floating-point instructions.
> 
> I figured it'd be better to push out the integer instructions now and
> the floating-point instructions later, rather than waiting for some
> unspecified time for full support.

Good idea.

Given the series is very long, don't expect me to comment all the
patches now. I'll comment them one by one and it make takes a few
days...

Don't hesitate to resend only one patch after I send a comment, this way
we can merge the series part by part as soon as we agree.

And a question for the whole series: how have you tested those 
instructions?

-- 
  .''`.  Aurelien Jarno             | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   address@hidden         | address@hidden
   `-    people.debian.org/~aurel32 | www.aurel32.net




reply via email to

[Prev in Thread] Current Thread [Next in Thread]