Re: [committed] pa10 load/store opcodes and FLAG_STRICT

[Carlos O'Donell <carlos at systemhalted dot org>]

On Mon, Jul 11, 2005 at 11:51:53AM -0400, John David Anglin wrote:
> > I like the FLAG_STRICT mode, I think using '0' for the space register is
> > rubish. I've cleaned up glibc under the strict mode.
> 
> I agree but the HP assembler accepts stuff like this, so people are
> going to complain.  There's a PR from Randolph regarding the pdtlb
> instruction.  The HP assembler allows the use of a '0' index for '%r0'.

I replaced a couple of other uses of '0' for '%r0' in glibc too,
somewhere in the math routines.

> I'm glad the glibc builds under the strict mode.  Do you know if
> the parisc kernel builds with FLAG_STRICT mode?

The kernel should build with FLAG_STRICT, though no promises. Lazy use
of assembly in the kernel leads to weird bugs, so I would like
FLAG_STRICT to be the default.
 
> > Are you adverse to making people cleanup poorly formated assembly? It's
> > a mechanistic change :)
> 
> I got tired editing gmp yesterday.  I need it for gfortran testing ;(

That's what perl is for.

> Possibly, we could have a '-mstrict' option with the default on.  Then,
> we would get strict checking on all instructions, not just loads and stores.
> Users could still build legacy code with '-mno-strict'.

I'm all for that.

c.