Re: [PATCH] aarch64: Fix the documentation on :pg_hi21:

["Martin Storsjö" <martin@martin.st>]

On Thu, 29 May 2014, Yufeng Zhang wrote:

> On 05/29/14 13:19, Yufeng Zhang wrote:
>> Hi Martin,
>> 
>> On 05/29/14 13:00, Martin Storsjo wrote:
>>> The syntax described, including a leading '#', was not supported
>>> in practice. (For :lo12: it is optional.)
>> 
>> I believe this is an issue in the parser; '#' should be optional in
>> either case.
>> 
>> It looks like gas/config/tc-aarch64.c:parse_adrp () is missing the
>> skip_past_char (&p, '#') at its beginning; see parse_half () for an 
>> example.
>
> After spending some time this afternoon reading ARMv8 Architecture Reference 
> Manual (ARMARM) and having a discussion with a colleague, I think I'll have 
> to take back what I just said.  Sorry!
>
> The A64 assembly language allows the optional # character to introduce 
> constant immediate operands, e.g. in LDR instructions with immediate offset:
>
>  LDR <Wt>, [<Xn|SP>], #<simm>
>
> and in the bitfield move instructions:
>
>  BFM <Wd>, <Wn>, #<immr>, #<imms>
>
> The ADRP instruction, however, has the following syntax:
>
>  ADRP <Xd>, <label>
>
> whose second operand is a program label whose 4KB page address is to be 
> calculated, and for such a operand, it doesn't make much sense to allow the # 
> character as an optional prefix, e.g. ADRP X0, #foo would look weird and 
> confusing.
>
> So I think your original patch is the right one to go with, and my only 
> comment is to remove both # characters from the line of
>
> "
>  @samp{#:pg_hi21:} and @samp{#:lo12:} respectively.
> "
>
> Apologize again for the previous wrong direction.
>
> Please get an approval from Marcus.

Ok, thanks - this explanation makes sense.

// Martin