-malign-double switch patch?

-malign-double switch patch?

[Davis, Bryan]

Hi,
I am having the same problem as a few others with the gcc -malign-double switch and the fstream seg fault thing.

Is there a patch or work-around out there?

This is causing us big problems.

Thanks
Bryan

Re: -malign-double switch patch?

[Richard Henderson]

On Thu, Aug 08, 2002 at 05:08:31PM -0400, Davis, Bryan wrote:
> I am having the same problem as a few others with the gcc -malign-double
> switch and the fstream seg fault thing.

Apparently you didn't read the documentation, in which it is said
that the switch affects the abi, and therefore should be used only
with extreme caution under controled conditions.

> This is causing us big problems.

Don't use it.


r~

Re: -malign-double switch patch?

[Nix]

On Fri, 9 Aug 2002, Richard Henderson muttered drunkenly:
> On Thu, Aug 08, 2002 at 05:08:31PM -0400, Davis, Bryan wrote:
>> I am having the same problem as a few others with the gcc -malign-double
>> switch and the fstream seg fault thing.
> 
> Apparently you didn't read the documentation, in which it is said
> that the switch affects the abi, and therefore should be used only
> with extreme caution under controled conditions.

GCC-3.1.x's manual says

,----
| `-malign-double'
| `-mno-align-double'
|      Control whether GCC aligns `double', `long double', and `long
|      long' variables on a two word boundary or a one word boundary.
|      Aligning `double' variables on a two word boundary will produce
|      code that runs somewhat faster on a `Pentium' at the expense of
|      more memory.
`----

i.e., it doesn't mention the ABI breakage :( perhaps this is only in the
trunk...

-- 
`There's something satisfying about killing JWZ over and over again.'
                                        -- 1i, personal communication

Re: -malign-double switch patch?

[Jamie Lokier]

Nix wrote:
> > Apparently you didn't read the documentation, in which it is said
> > that the switch affects the abi, and therefore should be used only
> > with extreme caution under controled conditions.
> 
> GCC-3.1.x's manual says
> 
> ,----
> | `-malign-double'
> | `-mno-align-double'
> |      Control whether GCC aligns `double', `long double', and `long
> |      long' variables on a two word boundary or a one word boundary.
> |      Aligning `double' variables on a two word boundary will produce
> |      code that runs somewhat faster on a `Pentium' at the expense of
> |      more memory.
> `----
> 
> i.e., it doesn't mention the ABI breakage :( perhaps this is only in the
> trunk...

The next paragraph says:

     *Warning:* if you use the `-malign-double' switch, structures
     containing the above types will be aligned differently than the
     published application binary interface specifications for the 386.

Now, if you're not using structures, perhaps it isn't obvious that you
can still have ABI breakage...  (Argument alignment on the stack
changes, doesn't it?)

-- Jamie

Re: -malign-double switch patch?

[Ross Smith]

On Sunday, 11 August 2002 13:11, Jamie Lokier wrote:
> Nix wrote:
> > > Apparently you didn't read the documentation, in which it is said
> > > that the switch affects the abi, and therefore should be used
> > > only with extreme caution under controled conditions.
> >
> > GCC-3.1.x's manual says
> >
> > ,----
> >
> > | `-malign-double'
> > | `-mno-align-double'
> > |      Control whether GCC aligns `double', `long double', and
> > | `long long' variables on a two word boundary or a one word
> > | boundary. Aligning `double' variables on a two word boundary will
> > | produce code that runs somewhat faster on a `Pentium' at the
> > | expense of more memory.
> >
> > `----
> >
> > i.e., it doesn't mention the ABI breakage :( perhaps this is only
> > in the trunk...
>
> The next paragraph says:
>
>      *Warning:* if you use the `-malign-double' switch, structures
>      containing the above types will be aligned differently than the
>      published application binary interface specifications for the
> 386.

No it doesn't. The next paragraph is the description of the 
-m128bit-long-double option, which does have a warning similar to that.
But there's nothing to indicate that the same warning applies to 
-malign-double. (This is in the 3.1 docs; as Nix said, the current 
development version may have something different.)

-- 
Ross Smith ..................................... Auckland, New Zealand
r-smith@ihug.co.nz ...................................................

        "A specter is haunting Wall Street; it is the specter
        of honest accounting."           -- Arthur D. Hlavaty

Re: -malign-double switch patch?

[Richard Henderson]

On Sun, Aug 11, 2002 at 02:11:11AM +0100, Jamie Lokier wrote:
> Now, if you're not using structures, perhaps it isn't obvious that you
> can still have ABI breakage...  (Argument alignment on the stack
> changes, doesn't it?)

Stack alignment doesn't change, iirc.  But are you absolutely positively
sure you aren't using structures somewhere?  I wouldn't bet on it.  I bet
there's a 64-bit integer type (e.g. off_t) somewhere in the fstream
structure that's being layed out differently.


r~

Re: -malign-double switch patch?

[Richard Henderson]

On Sun, Aug 11, 2002 at 04:29:52PM +1200, Ross Smith wrote:
> No it doesn't. The next paragraph is the description of the 
> -m128bit-long-double option, which does have a warning similar to that.
> But there's nothing to indicate that the same warning applies to 
> -malign-double.

*Both* options bear that warning.  The warning appears twice.

> (This is in the 3.1 docs; as Nix said, the current 
> development version may have something different.)

I'm looking at the 3.1.1 docs.


r~

Re: -malign-double switch patch?

[Ross Smith]

On Sunday, 11 August 2002 17:25, Richard Henderson wrote:
> On Sun, Aug 11, 2002 at 04:29:52PM +1200, Ross Smith wrote:
> > No it doesn't. The next paragraph is the description of the
> > -m128bit-long-double option, which does have a warning similar to
> > that. But there's nothing to indicate that the same warning applies
> > to -malign-double.
>
> *Both* options bear that warning.  The warning appears twice.
>
> > (This is in the 3.1 docs; as Nix said, the current
> > development version may have something different.)
>
> I'm looking at the 3.1.1 docs.

I just checked the archived docs on the GCC website. The warning was 
added between 3.1 and 3.1.1.

-- 
Ross Smith ..................................... Auckland, New Zealand
r-smith@ihug.co.nz ...................................................

        "A specter is haunting Wall Street; it is the specter
        of honest accounting."           -- Arthur D. Hlavaty

Re: -malign-double switch patch?

[Jamie Lokier]

Ross Smith wrote:
> > I'm looking at the 3.1.1 docs.
> 
> I just checked the archived docs on the GCC website. The warning was 
> added between 3.1 and 3.1.1.

I'm looking at Red Hat's gcc-3.1-4, which does have the warning.
That same manual begins:

    This manual documents how to use the GNU compilers, as well as their
    features and incompatibilities, and how to report bugs.  It corresponds
    to GCC version 3.1.

So I can be forgiven for thinking the warning applies to 3.1 :-)

I'm surprised by one thing: doesn't the stack layout for function calls
change with `-malign-double'?

-- Jamie