Re: expanding 1 insn to 2

[Aurelien Buhrig <aurelien dot buhrig dot gcc at gmail dot com>]

Should I consider this silence as a "it's not possible" answer, as a
"cgen is not maintained anymore", or as "everybody is on Holiday" ?

Aurelien

2011/8/26 Aurelien Buhrig <aurelien.buhrig.gcc@gmail.com>:
> Hi,
>
> thanks for your reply.
>
>
>> It's possible (and documented in the cgen manual to do this inside cgen -
>> see section 5.4.14 Writing define-macro-insn. and section 3.17
>> Macro-instructions
>
> I've already used dnmi to define 1 pseudo insn which represents
> another insn (such as push --> move -(sp)).
> My problem here is to define one pseudo instruction such as
> ld.a #$abs32 --> r0r1
> to be expanded into 2 real instructions such as
> move.w #$abs16_hi --> r0
> move.w #$abs16_lo --> r1
>
> I can't see how to emit 2 instructions using define-macro-insn, and
> how to tell cgen how to fill abs16_hi and abs16_lo operands from abs32
> (all 3 operands refering to 3 different bfd howtos).
>
>
>> You could also do this in the assembler. ?I assume use used cgen to build a
>> gas assembler from the binutils distribution.
> yes
>
>> You can either define a .macro that will do what you want assuming that your
>> addressing is simple enough, or you can tweak the md_assemble() function in
>> your assember to parse the operands and break them down appropriately if you
>> need.
>
> md_assemble only apply on string source and I don't know how to handle
> splitting symbol references (to be correctly relocated).
> Idem with .macro.
>
> A workaround can be to define 2 instructions such as ld_hi/ld_lo,
> implemented as move with different howtos for relocs. but I would like
> to define only one instruction.using cgen if possible...
>
> Aurelien
>