This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: i386 attribute memory bug
On Wed, Oct 28, 1998 at 12:06:17AM -0800, Richard Henderson wrote:
> On Wed, Oct 28, 1998 at 12:26:20AM -0700, Jeffrey A Law wrote:
> > How about fixing return_internal instead to specify "other" instead of
> > changing the default for everything.
>
> And how many other instructions are similarly mis-labeled?
> I guarantee that number is non-zero.
>
> The only correct thing to do is to go through each of the
> instructions that ought to be "integer" and so mark them.
> But that is outside the scope of this patch.
This is approximately what my patch does.
Since last time I've changed attribute scheme little bit so I have an
type attribute with some new types (like fpload, fpstore, alu etc.)
opcode attribute specifying whteher resulting opcode has no, one or two
operands and memory attribute specifing expected load/store or both.
I also have prefix attribute guessing whether opcode has 16->32 prefix.
To make then really exact is long piece of work, but I am making some progress
on it.
I've taken care to avoid accesing of non-existend operands
(my attributes isn't exact at the moment, but they works quite well
in many cases).
This solution is IMO more scalable than my first one, since I can specify
pentium pairing attributes using one large condition instead and fix just
few most weird cases in patterns. I plan to use them to specify behaviour of
other CPUs as well.
So please don't take much care to fixing this problem. The "other" solution
should be OK as temporary solution, if my patch will be applied (hope so)
Honza
>
>
> r~
--
------------------------------------------------------------------------------
Have you browsed my www pages? Look at:
http://www.paru.cas.cz/~hubicka
Koules-the game for Svgalib,X11 and OS/2, Xonix-the game for X11
czech documentation for linux index, original 2D computer art and
funny 100 years old photos and articles are there!