Re: [patch] gcc/testsuite/gcc.c-torture/compile/pr27528.c

[Mike Stump <mrs at apple dot com>]

On Jan 6, 2007, at 3:12 PM, Andrew Pinski wrote:
> On Sat, 2007-01-06 at 13:56 -0800, Hui-May Chang wrote:
> > gcc/testsuite/ChangeLog:
> >          * gcc.c-torture/compile/pr27528.c: Add -fno-pic option.
>
> I don't think this is the correct fix.  The "i" constraint should  
> accept
> this, even if we are in PIC mode.

Could you elaborate some?  Is it because it is really a legitimate  
pic operand?  If so, could you discuss just why it is.  How does one  
encode a LABEL_DECL in x86 land's addressing modes when the  
instruction doesn't take a pc-relative address but you want read only  
code for pic code?

> What the "i" constraint should do for these asms

For just these asm, or for all uses of the "i" constraint?

> is just give them the asm label and nothing more.

And how do you expect that to be encoded in an instruction that  
doesn't have pc-relative addressing for pic mode when the desire is  
to be able to not have dynamic reloc fixups for the code so that it  
can be read only and shared?

> This is the reason behind this bug report and testcase originally  
> in fact.

And what PR was that exactly?  PR27528?  I didn't see any mention of - 
fPIC anywhere in that PR.