Re: further behavioral issues with ARM assembler

[Richard Earnshaw <rearnsha at arm dot com>]

On 08/04/13 16:05, Jan Beulich wrote:
> All,
>
> alongside the fixes I sent during the day, I've also got questions on
> whether to other aspects are intentional or bugs:
>
> The first is the fact that ".arch all" doesn't work. The code specifically
> suppresses this, but it doesn't become clear why. Obviously it would
> be nice to have a way to return to this command line specifiable
> mode via some directive.

It's tricky because of the potential interaction with the attributes 
marking on object files.  The whole code in this area could do with an 
overhaul; I'm not sure that tinkering at the edges is going to solve the 
problems.  Part of the problem is that we've got overlapping 
incompatible extensions in the co-processor instruction space.

> Further there are several instructions where the specification says
> PC-relative addresses are okay, yet the assembler rejects them. A
> likely incomplete list is

Re-ordering your list slightly.

> 	ldrd	r0, r1, [r15, r2]
> 	ldrh	r0, [r15, r0]
> 	strd	r0, r1, [r15, r2]
> 	strh	r0, [r15, r0]

On first inspection, these appear to be errors in GAS.


> 	smlad	r0, r0, r0, r15

SMLAD with Ra=15 isn't permitted, if you read the pseudo code carefully, 
you'll see that if Ra = 15, then the instruction encodes SMUAD.

> 	smmla	r0, r0, r0, r15

Similarly this encodes SMMUL.

> 	vld1.8 	{d0}, [r15]

Section A7.7.1 "Advanced SIMD addressing mode" bans Rn=15 for all simd 
addresses.

R.
> Is this discrepancy intentional?
>
> Thanks, Jan