Re: [Qemu-devel] [PATCH risu] ppc64: Fix patterns for rotate doubleword instructions

[G 3]

On May 30, 2017, at 8:01 AM, Peter Maydell wrote:

> On 25 May 2017 at 21:05,  <address@hidden> wrote:
> > What do you think about sharing the same ppc64.risu file to ppc
> > and ppc64/ppc64le. Then, instead of:
> >
> > <NAME> PPC64LE 01111...
> >
> > we'd have:
> >
> > <NAME> PPC 01111...
> >
> > So, specific instructions would be:
> >
> > <NAME> PPC64LE 01111...
> > <NAME> PPC64 01111...
> > <NAME> PPC32 01111...
> >
> > It will allow users to select the arch by using patterns like: ./ 
> > risugen
> > --pattern "PPC64*", or --pattern "PPC32*".
>
> I think we should aim to share the files, but that we should
> have a better mechanism than just a convention on the pattern
> name (which is easy to make mistakes with and also makes it
> harder to use the --pattern argument for the purpose it was
> intended for, since you have to combine your desired pattern
> with the pattern to select the right bit of the instruction set).
> This isn't limited to PPC either -- for ARM we'd like to be
> able to mark instructions as ARMv7/ARMv8/present in some
> optional instruction set extension, etc.

Do you want to add an architecture option?

--pattern 'F.*' --arch PPC32

In this example only the instructions that begin with the letter F  
and have PPC32 as an encodingname are used.

> We could perhaps do that with an optional field in the
> lines in the risu file to mark what part of the instruction
> set they're in, and a command line argument to select the
> instruction set variant(s) desired.

I thought we already had this feature.

From the readme file:

insnname encodingname bitfield ... [ [ !blockname ] { blocktext } ]

Isn't the encodingname field the field we could use to select the  
variant we want?


> Another possible approach would be to have the PPC32 insns
> in their own file, and some sort of #include directive to
> pull them into the PPC64 file.
>
> This is all "adding extra features to risu", though, so I
> don't insist it be done.

The #include directive does sound interesting but if you really want  
one PowerPC risu file then Jose's way would be the way to go. Risu as  
it is can handle this.