Re: [Qemu-devel] [PATCH v2] target-mips: Fix incorrect shift for SHILO and SHILOV

[Aurelien Jarno]

On Wed, Dec 05, 2012 at 04:38:22PM +0000, Peter Maydell wrote:
> On 5 December 2012 15:51, Richard Henderson <address@hidden> wrote:
> > On 2012-12-05 09:49, Peter Maydell wrote:
> >> On 5 December 2012 15:36, Richard Henderson <address@hidden> wrote:
> >>> On 2012-12-04 17:29, Petar Jovanovic wrote:
> >>>>      rs5_0 = (int8_t)(rs5_0 << 2) >> 2;
> >>>
> >>> This is more portably written as
> >>>
> >>>   rs5_0 = (rs5_0 ^ 0x20) - 0x20;
> >>
> >> ...but way more obscurely. If we want to play that
> >> kind of game can we have a sign-extension function in
> >> a header somewhere?
> >
> > I dunno about more obscurely.  It took me a minute to figure out
> > what was wanted in the original.
> >
> > As for a helper function... sure.
> 
> I don't think we should block this patch on that general
> cleanup, though. All the sign extensions in target-mips/translate.c
> are done in the double-shift way, so this is consistent with
> the existing code.
> 

While it might be a good idea to make QEMU even more portable, it should
be noticed that currently QEMU assumes that a left shift followed by the
same arithmetic right shift. This is the case in at least the alpha,
arm, m68k, mips, ppc, s390 and unicore32 targets, as well as for
implementing sign extension and division in TCG.


-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
address@hidden                 http://www.aurel32.net