Re: [PATCH] PowerPC VLE changes

[Alexander Fedotov <alfedotov@gmail.com>]

Well, I think while such aliases are not in PowerISA and may be
changed in any moment - better to move such ones to macros table.
Performance should not harm dramatically if at all.

Anyway, don't miss decorated storage instruction for e200z4 from my patch.


On Wed, Mar 29, 2017 at 2:33 AM, Alan Modra <amodra@gmail.com> wrote:
> On Mon, Mar 27, 2017 at 03:32:23PM +0100, Andrew Jenner wrote:
>> Hi Alexander,
>>
>> On 24/03/2017 15:32, Александр Федотов wrote:
>> >I have question in continuation of
>> >https://sourceware.org/ml/binutils/2016-07/msg00342.html
>> >
>> >Why do you need to have "e_cmpwi" etc in vle_opcodes table ? It seems
>> >they are just alias names because of the same opcodes and objdump
>> >never show them.
>> >Why don't to use powerpc_macros instead ?
>
> It is the other way around.  If there is something in the macro table
> that could go in the main opcode table then it should be moved there.
>
> Mnemonic lookup in the main opcode table happens first.  Macro lookup
> is later.  So putting something in the macro table unnecessarily, just
> slows does the assembler a litte.  The macro table should really only
> be used where some manipulation of arguments is needed.
>
> Hmm, looking at the macros, that means e_rotlwi and e_clrlwi probably
> ought to be moved since the macros just provide a default argument to
> the underlying machine insn rather than some arithmetic manipulation.
> Fixing this isn't at all important though, so don't see these comments
> as a request for someone to tidy the VLE support.  Note that moving
> those insns to the main table would also allow the disassembler to
> produce e_rotlwi or e_clrlwi rather than e_rlwinm, depending on
> whether you place them before e_rlwinm or after.  I recognize that
> some people (more hardware oriented) would prefer to see rlwinm rather
> than rotlwi (likely prefered by those more software oriented).  I've
> just been working on a patch that will add -Mraw to the powerpc
> disassembler to maybe satisfy both camps.
>
> --
> Alan Modra
> Australia Development Lab, IBM



-- 
Best regards,
AF