How to proceed on the Thumb-2 assembler patches?

How to proceed on the Thumb-2 assembler patches?

[Zack Weinberg]

I'm not sure how best to proceed to get the Thumb-2 assembler patches
checked into mainline binutils.  (To remind, they are already present
on binutils-csl-arm-2005q1-branch - so if anyone wants to play with
the modified assembler without applying all seventy-odd patches, you
can get it from there.)

It seems to me that it should all be checked in at once, because the
intermediate stages are not that interesting and some of them contain
false starts and/or flat-out bugs.  Assuming that that is an
acceptable procedure, I then have two questions for the team:

1) Are there any objections at the design or principle level to the
   changes I have made?

2) Are there any outstanding issues which need to be fixed before the
   patch series goes into mainline?  I would prefer to keep this set
   to the minimum - basically only issues which affect use of the
   assembler for plain old ARM or Thumb code generation, or somehow
   affect other targets than arm-*-*.

zw

Re: How to proceed on the Thumb-2 assembler patches?

[Daniel Jacobowitz]

On Mon, May 02, 2005 at 11:20:47PM -0700, Zack Weinberg wrote:
> 2) Are there any outstanding issues which need to be fixed before the
>    patch series goes into mainline?  I would prefer to keep this set
>    to the minimum - basically only issues which affect use of the
>    assembler for plain old ARM or Thumb code generation, or somehow
>    affect other targets than arm-*-*.

I saw the patches to allow conditional suffixes on Thumb code go by,
but not the validation patch to reject them if we're in classic Thumb
mode that you said would be forthcoming.  I'd rather not merge until
that's done... a test for it would be nice, too.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: How to proceed on the Thumb-2 assembler patches?

[Daniel Jacobowitz]

On Mon, May 02, 2005 at 11:20:47PM -0700, Zack Weinberg wrote:
> 2) Are there any outstanding issues which need to be fixed before the
>    patch series goes into mainline?  I would prefer to keep this set
>    to the minimum - basically only issues which affect use of the
>    assembler for plain old ARM or Thumb code generation, or somehow
>    affect other targets than arm-*-*.

I saw the patches to allow conditional suffixes on Thumb code go by,
but not the validation patch to reject them if we're in classic Thumb
mode that you said would be forthcoming.  I'd rather not merge until
that's done... a test for it would be nice, too.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: How to proceed on the Thumb-2 assembler patches?

[Zack Weinberg]

Daniel Jacobowitz <drow@false.org> writes:

> On Mon, May 02, 2005 at 11:20:47PM -0700, Zack Weinberg wrote:
>> 2) Are there any outstanding issues which need to be fixed before the
>>    patch series goes into mainline?
>
> I saw the patches to allow conditional suffixes on Thumb code go by,
> but not the validation patch to reject them if we're in classic Thumb
> mode that you said would be forthcoming.  I'd rather not merge until
> that's done... a test for it would be nice, too.

Do you want the whole validator now, or just rejection of conditional
suffixes in classic Thumb mode?  The former is quite a bit of code.

zw

Re: How to proceed on the Thumb-2 assembler patches?

[Daniel Jacobowitz]

On Tue, May 03, 2005 at 08:32:18AM -0700, Zack Weinberg wrote:
> Daniel Jacobowitz <drow@false.org> writes:
> 
> > On Mon, May 02, 2005 at 11:20:47PM -0700, Zack Weinberg wrote:
> >> 2) Are there any outstanding issues which need to be fixed before the
> >>    patch series goes into mainline?
> >
> > I saw the patches to allow conditional suffixes on Thumb code go by,
> > but not the validation patch to reject them if we're in classic Thumb
> > mode that you said would be forthcoming.  I'd rather not merge until
> > that's done... a test for it would be nice, too.
> 
> Do you want the whole validator now, or just rejection of conditional
> suffixes in classic Thumb mode?  The former is quite a bit of code.

I have no idea.  It seems to me that we aren't in a desparate hurry to
merge this code to mainline; why not wait until it is finished, since
it is already so large?

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: How to proceed on the Thumb-2 assembler patches?

[Zack Weinberg]

Daniel Jacobowitz <drow@false.org> writes:

>> Do you want the whole validator now, or just rejection of conditional
>> suffixes in classic Thumb mode?  The former is quite a bit of code.
>
> I have no idea.  It seems to me that we aren't in a desparate hurry to
> merge this code to mainline; why not wait until it is finished, since
> it is already so large?

Um, no, actually, I am in a hurry to merge this to mainline; the
larger the patch set gets, the harder it gets to work on outside the
tree.  If, for instance, someone decided to rip out ARM_OLD_ABI
(which, on the merits, should happen) I'd be in a world of hurt.

zw

Re: How to proceed on the Thumb-2 assembler patches?

[Daniel Jacobowitz]

On Tue, May 03, 2005 at 09:10:39AM -0700, Zack Weinberg wrote:
> Daniel Jacobowitz <drow@false.org> writes:
> 
> >> Do you want the whole validator now, or just rejection of conditional
> >> suffixes in classic Thumb mode?  The former is quite a bit of code.
> >
> > I have no idea.  It seems to me that we aren't in a desparate hurry to
> > merge this code to mainline; why not wait until it is finished, since
> > it is already so large?
> 
> Um, no, actually, I am in a hurry to merge this to mainline; the
> larger the patch set gets, the harder it gets to work on outside the
> tree.  If, for instance, someone decided to rip out ARM_OLD_ABI
> (which, on the merits, should happen) I'd be in a world of hurt.

OK, that makes sense.  I would want at least the rejection of invalid
classic Thumb.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: How to proceed on the Thumb-2 assembler patches?

[Zack Weinberg]

Daniel Jacobowitz <drow@false.org> writes:

>> Um, no, actually, I am in a hurry to merge this to mainline; the
>> larger the patch set gets, the harder it gets to work on outside the
>> tree.  If, for instance, someone decided to rip out ARM_OLD_ABI
>> (which, on the merits, should happen) I'd be in a world of hurt.
>
> OK, that makes sense.  I would want at least the rejection of invalid
> classic Thumb.

Alright, here it is.

zw

        * gas/config/tc-arm.c (md_assemble): In classic syntax, reject
        conditional suffixes on Thumb instructions other than branch.
        * gas/testsuite/gas/arm/t16-bad.s: Add test for rejection of
        conditional suffixes on non-branch instructions.
        * gas/testsuite/gas/arm/t16-bad.l: Update to match.

===================================================================
Index: gas/config/tc-arm.c
--- gas/config/tc-arm.c	(revision 88)
+++ gas/config/tc-arm.c	(working copy)
@@ -7574,6 +7574,12 @@
 	  as_bad (_("selected processor does not support `%s'"), str);
 	  return;
 	}
+      if (inst.cond != COND_ALWAYS && !unified_syntax
+	  && opcode->tencode != do_t_branch)
+	{
+	  as_bad (_("Thumb does not support conditional execution"));
+	  return;
+	}
 
       mapping_state (MAP_THUMB);
       inst.instruction = opcode->tvalue;
===================================================================
Index: gas/testsuite/gas/arm/t16-bad.s
--- gas/testsuite/gas/arm/t16-bad.s	(revision 87)
+++ gas/testsuite/gas/arm/t16-bad.s	(working copy)
@@ -133,3 +133,6 @@
 	bkpt	#257
 	cpsie	ai,#5
 	cpsid	ai,#5
+
+	@ Conditional suffixes
+	addeq	r0,r1,r2
===================================================================
Index: gas/testsuite/gas/arm/t16-bad.l
--- gas/testsuite/gas/arm/t16-bad.l	(revision 87)
+++ gas/testsuite/gas/arm/t16-bad.l	(working copy)
@@ -183,3 +183,4 @@
 [^:]*:133: Error: immediate value out of range -- `bkpt #257'
 [^:]*:134: Error: Thumb does not support the 2-argument form of this instruction -- `cpsie ai,#5'
 [^:]*:135: Error: Thumb does not support the 2-argument form of this instruction -- `cpsid ai,#5'
+[^:]*:138: Error: Thumb does not support conditional execution