[Mesa-dev] [PATCH 2/2] gk110/ir: use separate signed expressions for dst/src with IMUL32I

[Ilia Mirkin]

On Wed, Apr 20, 2016 at 3:14 PM, Samuel Pitoiset
<samuel.pitoiset at gmail.com> wrote:
>
>
> On 04/20/2016 07:47 PM, Ilia Mirkin wrote:
>>
>> Presumably you'd want to touch up the non-limm side of this as well?
>> Although TBH, I can't really think of a time when it'd matter. I think
>> you're pretty much guaranteed to have stype == dtype for mul's. Maybe
>> I'm wrong?
>
>
> Well, the non-limm part is wrong because 0xc1c won't be used anymore with
> the previous patch, as emitForm_21() uses opc1 when src(1) is an immediate.
> The different flags was only useful for 0xc1c but now they are useless,
> because 0x21c is actually:
>
> /*0000*/   @P0 FMUL32I.FTZ R0.CC, R0, -0;      /* 0x21c0000000000002 */

huh? It or's some bits depending on arg types, so it'll be 61c or whatever.

>
> I was thinking to make a new patch to remove them, but this one doesn't need
> to be backported.
>
> I think we are pretty guaranteed that stype == dtype, but does this is
> correct all the time? Not sure. If you look at NVC0::emitUMUL() you will see
> that the signed expressions for dst/src are separated.
>
>
>>
>>    -ilia
>>
>> On Wed, Apr 20, 2016 at 1:06 PM, Samuel Pitoiset
>> <samuel.pitoiset at gmail.com> wrote:
>>>
>>> Forcing the destination type to be signed when the source is signed
>>> is not totally correct.
>>>
>>> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
>>> Cc: "11.1 11.2" <mesa-stable at lists.freedesktop.org>
>>> ---
>>>   src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp | 4 +++-
>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
>>> b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
>>> index f69567a..90b90bc 100644
>>> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
>>> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
>>> @@ -617,8 +617,10 @@ CodeEmitterGK110::emitIMUL(const Instruction *i)
>>>
>>>         if (i->subOp == NV50_IR_SUBOP_MUL_HIGH)
>>>            code[1] |= 1 << 24;
>>> +      if (i->dType == TYPE_S32)
>>> +         code[1] |= 1 << 25;
>>>         if (i->sType == TYPE_S32)
>>> -         code[1] |= 3 << 25;
>>> +         code[1] |= 1 << 26;
>>>      } else {
>>>         emitForm_21(i, 0x21c, 0xc1c);
>>>
>>> --
>>> 2.8.0
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev