[ros-dev] [ros-diffs] [jgardou] 58660: [MESA32] * Disable SSE optimizations, as they only cause mayhem.

Sylvain Petreolle spetreolle at yahoo.fr
Fri Apr 5 13:10:36 UTC 2013


While we have to fix the kernel check asap, jerome's commit is not a hack since it uses existing MESA defines.
 
Kind regards,
Sylvain Petreolle


>________________________________
> De : Aleksey Bragin <aleksey at reactos.org>
>À : ReactOS Development List <ros-dev at reactos.org> 
>Envoyé le : Vendredi 5 avril 2013 10h14
>Objet : Re: [ros-dev] [ros-diffs] [jgardou] 58660: [MESA32] * Disable SSE optimizations, as they only cause mayhem.
> 
>I think he's referring to the actual commit, which I find weird too.
>
>Regards,
>Aleksey Bragin
>
>On 05.04.2013 2:42, Kamil Hornicek wrote:
>> "LOL", I'm not "trying" to "fix" "anything" - disabling the whole SSE support is on par with disabling just the check. You fix the kernel and you better do it soon.
>> 
>> Dne 4.4.2013 21:58, Timo Kreuzer napsal(a):
>>> 
>>> LOL, we have a bug in the kernel and you try to "fix" this in MESA?
>>> 
>>> 
>>> Am 04.04.2013 14:33, schrieb Kamil Hornicek:
>>>> If I recall correctly there's a check whether the OS can handle
>>>> masked/unmasked sse exceptions. It causes trouble even in Windows if
>>>> the app using Mesa has it's own exception handlers installed IIRC. SSE
>>>> works just fine. Just disable the (useless) _mesa_check_os_sse_suppor
>>>> stuff (ReactOS supports this, no need to do the check) or find a way
>>>> to stop the exception from propagating.
>>>> 
>>>> Regards,
>>>> Kamil
>>>> 
>>>> Dne 4.4.2013 12:34, Jérôme Gardou napsal(a):
>>>>> It causes some kernel mode exception. The code deliberately throws an
>>>>> SSE exception to see if the OS supports them. The trap handler considers
>>>>> this as a k-mode exception and bug checks.
>>>>> 
>>>>> See http://jira.reactos.org/browse/CORE-6776
>>>>> 
>>>>> Timo Kreuzer a écrit :
>>>>>> 
>>>>>> What exactly does it cause? And shouldn't we rather fix that,
>>>>>> instead of
>>>>>> disabling optimizations? mesa is already slow enough.
>>>>>> 
>>>>>> Am 03.04.2013 14:02, schrieb jgardou at svn.reactos.org:
>>>>>>> Author: jgardou
>>>>>>> Date: Wed Apr  3 12:02:58 2013
>>>>>>> New Revision: 58660
>>>>>>> 
>>>>>>> URL: http://svn.reactos.org/svn/reactos?rev=58660&view=rev
>>>>>>> Log:
>>>>>>> [MESA32]
>>>>>>>   * Disable SSE optimizations, as they only cause mayhem.
>>>>>>> 
>>>>>>> Modified:
>>>>>>> trunk/reactos/dll/opengl/mesa/src/mesa/CMakeLists.txt
>>>>>>> 
>>>>>>> Modified: trunk/reactos/dll/opengl/mesa/src/mesa/CMakeLists.txt
>>>>>>> URL:
>>>>>>> http://svn.reactos.org/svn/reactos/trunk/reactos/dll/opengl/mesa/src/mesa/CMakeLists.txt?rev=58660&r1=58659&r2=58660&view=diff 
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> ============================================================================== 
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> --- trunk/reactos/dll/opengl/mesa/src/mesa/CMakeLists.txt [iso-8859-1]
>>>>>>> (original)
>>>>>>> +++ trunk/reactos/dll/opengl/mesa/src/mesa/CMakeLists.txt [iso-8859-1]
>>>>>>> Wed Apr  3 12:02:58 2013
>>>>>>> @@ -33,17 +33,18 @@
>>>>>>>           x86/3dnow_xform3.S
>>>>>>>           x86/3dnow_xform4.S
>>>>>>>           x86/3dnow_normal.S
>>>>>>> -        x86/sse_xform1.S
>>>>>>> -        x86/sse_xform2.S
>>>>>>> -        x86/sse_xform3.S
>>>>>>> -        x86/sse_xform4.S
>>>>>>> -        x86/sse_normal.S
>>>>>>> +        # x86/sse_xform1.S
>>>>>>> +        # x86/sse_xform2.S
>>>>>>> +        # x86/sse_xform3.S
>>>>>>> +        # x86/sse_xform4.S
>>>>>>> +        # x86/sse_normal.S
>>>>>>>           x86/read_rgba_span_x86.S)
>>>>>>>       add_definitions(
>>>>>>>           -DUSE_X86_ASM
>>>>>>>           -DUSE_MMX_ASM
>>>>>>>           -DUSE_3DNOW_ASM
>>>>>>> -        -DUSE_SSE_ASM)
>>>>>>> +        # -DUSE_SSE_ASM
>>>>>>> +    )
>>>>>>>   endif()
>>>>>>>   list(APPEND SOURCE
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>> 
>
>
>_______________________________________________
>Ros-dev mailing list
>Ros-dev at reactos.org
>http://www.reactos.org/mailman/listinfo/ros-dev
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.reactos.org/pipermail/ros-dev/attachments/20130405/d0d939e3/attachment.html>


More information about the Ros-dev mailing list