Re: [ros-dev] [ros-diffs] [dgorbachev] 52916: [CDFS] Workaround for GCC 4.6 optimizer bug (ICE when compiling with -O3).

[Timo Kreuzer]

Can you explain this one in detail?

Am 27.07.2011 04:37, schrieb [e-mail]:
> Author: dgorbachev
> Date: Wed Jul 27 02:37:40 2011
> New Revision: 52916
>
> URL: http://svn.reactos.org/svn/reactos?rev=52916&view=rev
> Log:
> [CDFS]
> Workaround for GCC 4.6 optimizer bug (ICE when compiling with -O3).
>
> Modified:
>      trunk/reactos/drivers/filesystems/cdfs/fsctl.c
>
> Modified: trunk/reactos/drivers/filesystems/cdfs/fsctl.c
> URL: http://svn.reactos.org/svn/reactos/trunk/reactos/drivers/filesystems/cdfs/fs...
> ==============================================================================
> --- trunk/reactos/drivers/filesystems/cdfs/fsctl.c [iso-8859-1] (original)
> +++ trunk/reactos/drivers/filesystems/cdfs/fsctl.c [iso-8859-1] Wed Jul 27 02:37:40 2011
> @@ -63,10 +63,10 @@
>       for (i = 0; i<  2048; i += 4)
>       {
>           /* DON'T optimize this to ULONG!!! (breaks overflow) */
>    Serial.Part[3] += Buffer[i+0];
>    Serial.Part[2] += Buffer[i+1];
>    Serial.Part[1] += Buffer[i+2];
>     Serial.Part[0] += Buffer[i+3];
>    Serial.Part[1] += Buffer[i+2];
>    Serial.Part[2] += Buffer[i+1];
>    Serial.Part[3] += Buffer[i+0];
> }
> CdInfo->SerialNumber = Serial.Value;

[Dmitry Gorbachev]

Hi Timo,

There are no changes in the logic, it's just to avoid a crash
("internal compiler error: in vect_transform_stmt, at
tree-vect-stmts.c:4887"). It's only in GCC versions 4.6.x.

Incidentally, was it through an oversight, or with a design, that
KiConvertToGuiThread (one which is in trap.s, committed in r52509),
does not save/restore EBX?

[Pierre Schweitzer]

Hi,
> There are no changes in the logic, it's just to avoid a crash
> ("internal compiler error: in vect_transform_stmt, at
> tree-vect-stmts.c:4887"). It's only in GCC versions 4.6.x.

I really don't get the logic behind the commit.

GCC is not supported yet by any of the BEs. Which implies that ReactOS is planned to be built using GCC 4.4.x (3 IIRC).
So, you're changing code to workaround a bug in an unsupported compiler?

This reminds me Alex & Magnus commits fights regarding GCC 4.4.x...

[Dmitry Gorbachev]

Hi,

The logic behind the commit is simply to allow to build ReactOS with
GCC 4.6. The next RosBE release could include it.

[Amine Khaldi]

The next RosBE release will include 4.6.x

Regards,
Amine.

[Timo Kreuzer]

Am 27.07.2011 12:38, schrieb Dmitry Gorbachev:
> Incidentally, was it through an oversight, or with a design, that
> KiConvertToGuiThread (one which is in trap.s, committed in r52509),
> does not save/restore EBX?

Ehhhm, it was to test your attention :D