[Mesa-dev] [PATCH 2/2] tgsi: handle bitwise opcodes in tgsi_opcode_infer_type

[Marek Olšák]

On Mon, Mar 2, 2015 at 5:12 PM, Roland Scheidegger <sroland at vmware.com> wrote:
> Am 02.03.2015 um 12:52 schrieb Marek Olšák:
>> From: Marek Olšák <marek.olsak at amd.com>
>>
>> ---
>>  src/gallium/auxiliary/tgsi/tgsi_info.c | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
>> index e6e0a60..258173d 100644
>> --- a/src/gallium/auxiliary/tgsi/tgsi_info.c
>> +++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
>> @@ -341,6 +341,8 @@ tgsi_opcode_infer_type( uint opcode )
>>     case TGSI_OPCODE_USNE:
>>     case TGSI_OPCODE_SVIEWINFO:
>>     case TGSI_OPCODE_UMUL_HI:
>> +   case TGSI_OPCODE_UBFE:
>> +   case TGSI_OPCODE_UMSB:
>>        return TGSI_TYPE_UNSIGNED;
>>     case TGSI_OPCODE_ARL:
>>     case TGSI_OPCODE_ARR:
>> @@ -362,6 +364,12 @@ tgsi_opcode_infer_type( uint opcode )
>>     case TGSI_OPCODE_IABS:
>>     case TGSI_OPCODE_ISSG:
>>     case TGSI_OPCODE_IMUL_HI:
>> +   case TGSI_OPCODE_POPC:
>> +   case TGSI_OPCODE_IBFE:
>> +   case TGSI_OPCODE_BFI:
>> +   case TGSI_OPCODE_BREV:
>> +   case TGSI_OPCODE_LSB:
>> +   case TGSI_OPCODE_IMSB:
>>        return TGSI_TYPE_SIGNED;
>>     default:
>>        return TGSI_TYPE_FLOAT;
>>
>
> This doesn't look all that consistent to me, in particular it doesn't
> match what tgsi_exec.c thinks these arguments are wrt signed/unsigned.
> Granted it probably doesn't make much of a difference, but it should be
> fixed imho - though not sure if sticking to what tgsi_exec.c thinks
> these arguments are (sometimes with different result and src type) is
> always better.

So how should it look like then? Note that the main purpose of this
code is for gallivm to choose between f32 and i32 correctly to avoid
type failures. LLVM doesn't care about the signedness and it looks
like gallivm doesn't care either. At least I can't find code in
gallivm that behaves differently for SIGNED vs UNSIGNED.

Marek