Brian Paul wrote:
> Brian Paul wrote:
>> Ian Romanick wrote:
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> I've noticed some odd bugs when accessing data packets sent via batch
>>> buffers from the PPE to SPE. Every now and then one of the pointers
>>> sent would be NULL or similar, low value. I tracked this down to
>>> unaligned accesses. If 64-bit values aren't aligned, reads return
>>> garbage.
>>>
>>> I managed to work around this by making the opcode field 64-bits and
>>> ensuring that each batch is 64-bit aligned. This seems to also
>>> eliminate the need for the dummy fields in cell_command_render.
>>>
>>> Patch attached.
>>
>> The patch breaks some things for me. Let me sort through that.
>>
>> I'm seeing some alignment-related problems in some new code too so I
>> think you've got the right idea.
>>
>> I'll probably send you a new patch in a bit...
>
> OK, things are running again. I'll just commit and push what I've got,
> including your patches.
Done.
But I omitted your changes to configs/linux-cell, Ian:
'ccache' doesn't exist
-DSPU_MAIN_PARAM_LONG_LONG doesn't work for SDK 2.1
/opt/powerpc/usr/include doesn't exist
How about we make a new linux-cell-sdk3 (or rename linux-cell to
linux-cell-sdk2.1) to avoid conflicts?
I'll probably upgrade to SDK 3.0 eventually, but not for a little while.
-Brian
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Mesa3d-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
