[Mesa-dev] [PATCH 8/9] tgsi: use separate structure for indirect address v2

[Christian König]

Am 18.03.2013 15:12, schrieb Roland Scheidegger:
> Am 18.03.2013 14:36, schrieb Christian König:
>> From: Christian König <christian.koenig at amd.com>
>>
>> To further improve the optimization of source and destination
>> indirect addressing we need the ability to store a reference
>> to the declaration of the addressed operands.
>>
>> Since most of the fields in tgsi_src_register doesn't apply for
>> an indirect addressing operand replace it with a separate
>> tgsi_ind_register structure and so make room for extra information.
>>
>> v2: rename Declaration to ArrayID, put the ArrayID into () instead of []
>>
>> Signed-off-by: Christian König <christian.koenig at amd.com>
>> ---
>> diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
>> index cd140de..be99ae0 100644
>> --- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
>> +++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
>> @@ -62,6 +62,7 @@ struct ureg_src
>>      int      IndirectIndex    : 16; /* SINT */
>>      int      DimensionIndex   : 16; /* SINT */
>>      int      DimIndIndex      : 16; /* SINT */
>> +   unsigned ArrayID          : 10; /* UINT */
>>   };
>>   
>>   /* Very similar to a tgsi_dst_register, removing unsupported fields
>> @@ -84,6 +85,7 @@ struct ureg_dst
>>      int      Index           : 16; /* SINT */
>>      int      IndirectIndex   : 16; /* SINT */
>>      int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
>> +   unsigned ArrayID         : 10; /* UINT */
>>   };
>>   
> Not sure I like the ordering of these fields...
> In any case it would be difficult (but possible) to make ureg_dst fit
> into a 64bit struct.
> FWIW (but that's not your fault) IndirectSwizzle really ought to be a
> unsigned too, looks to me like this misdeclaration really could cause
> bugs somewhere.

Noticed the strange declaration of IndirectSwizzle also, and I have it 
on my todo list (as very low priority) to figure out if this causes any 
problems or not, but as far as I have seen we currently only use 
indirect addressing with swizzle X, and then it probably won't hurt.

Regarding putting everything into a 64bit struct, well I'm pretty sure 
that I don't like that idea. Bitfields might make sense for TGSI tokens, 
but for ureg_src/ureg_dst saving some bytes of memory probably doesn't 
judge the extra overhead related to them.

Christian.

>
> Roland
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev