This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [ARM] signed zero


On Tue, 24 May 2011, Matthew Gretton-Dann wrote:

> > > Looking at the source code and tests I think I agree with most of the
> > > times you elide +0.  However, you seem to elide the offset when doing a
> > > literal load (for instance ldr r0, [pc, #0] is being disassembled as ldr
> > > r0, [pc]).  In this case I think the offset should be shown, and this is
> > > also what the ARMARM does (see section A8.6.59 for example).
> > 
> > I think those bits were Maciej's changes, for consistency with displaying the 
> > same addressing mode with other base registers.  I'm not sure which version of 
> > the ARM ARM you're looking at -- the one I have (ARM DDI 0100I) doesn't have a 
> > section A8.
> 
> I'm looking at the v7-AR ARMARM (ARM DDI 0406B) - this is available on
> the ARM documentation website (http://infocenter.arm.com/).

 I have a vague recollection of sorting these bits out while merging 
conflicting versions of upstream and our code base.  It's been a while, so 
I'd have do dig out the details, but my understanding is you find the 
presence of the insignificant zero immediate offset in disassembly 
important in the specific case of the PC or r15 being used as the base 
register (as opposed to where one of the other registers is used as the 
base).  Would you please elaborate why this is the case?

 In particular given the following source:

$ cat ldr.s
	ldr	r0, [pc, #0]
	ldr	r0, [pc]
	ldr	r0, [r1, #0]
	ldr	r0, [r1]

assembles as this:

$ arm-none-eabi-as -o ldr.o ldr.s
$ arm-none-eabi-objdump -d ldr.o

ldr.o:     file format elf32-littlearm


Disassembly of section .text:

00000000 <.text>:
   0:	e59f0000 	ldr	r0, [pc]	; 8 <.text+0x8>
   4:	e59f0000 	ldr	r0, [pc]	; c <.text+0xc>
   8:	e5910000 	ldr	r0, [r1]
   c:	e5910000 	ldr	r0, [r1]

i.e. there is absolutely no difference whatever between the two 
instruction forms.

 Furthermore the omission of the insignificant zero offset is explicitly 
permitted by the architecture manual with the following note found on the 
page defining this particular addressing mode being used:

"5.2.2 Load and Store Word or Unsigned Byte -- Immediate offset
[...]
Offset of zero	The syntax [<Rn>] is treated as an abbreviation for
		[<Rn>, #0], unless the instruction is one that only allows 
		post-indexed addressing modes (LDRBT, LDRT, STRBT or
		STRT)."

and given the disassembly is typically meant to be examined by a human 
being, omitting redundant information is usually good for readability.

  Maciej


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]