[Mesa-dev] [PATCH 6/8] nv50/ir: optimize ADD3(d, 0x0, b, c) to ADD(d, b, c)

[Samuel Pitoiset]

On 07/01/2016 06:19 AM, Ilia Mirkin wrote:
> On Thu, Jun 30, 2016 at 6:54 PM, Samuel Pitoiset
> <samuel.pitoiset at gmail.com> wrote:
>>
>>
>> On 07/01/2016 12:44 AM, Ilia Mirkin wrote:
>>>
>>> If moveSources doesn't move modifiers, we have a serious problem.
>>> However it looks like it does:
>>>
>>> void
>>> Instruction::setSrc(int s, const ValueRef& ref)
>>> {
>>>    setSrc(s, ref.get());
>>>    srcs[s].mod = ref.mod;
>>> }
>>>
>>> which is what moveSources calls.
>>
>>
>> I was not sure about moveSources() because we have two variants and the
>> other one doesn't move the modifiers.
>>>
>>>
>>>
>>> On Thu, Jun 30, 2016 at 6:26 PM, Samuel Pitoiset
>>> <samuel.pitoiset at gmail.com> wrote:
>>>>
>>>> And ADD3(d, a, 0x0, c) to ADD(d, a, c) as well.
>>>>
>>>> Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
>>>> ---
>>>>  src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp | 12
>>>> +++++++++++-
>>>>  1 file changed, 11 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
>>>> b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
>>>> index 1cf1fa3..517f779 100644
>>>> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
>>>> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
>>>> @@ -1032,7 +1032,17 @@ ConstantFolding::opnd(Instruction *i,
>>>> ImmediateValue &imm0, int s)
>>>>              i->src(0).mod = Modifier(0);
>>>>        }
>>>>        break;
>>>> -
>>>> +   case OP_ADD3:
>>>> +      if (i->usesFlags())
>>>> +         break;
>>>
>>>
>>> Why? ADD can produce/consume a flag just fine.
>>
>>
>> Well, this is loosely based on OP_ADD which does exactly the same check.
>
> Right, because you can't get rid of ADD 0, $c. (There is no MOV $c
> variant... or we don't support it ... we probably should, I think
> that's what CSET/CSETP are.) However you can definitely flip the ADD3
> into an ADD2 even if the carry flag is being added in.

Yeah, makes sense.

>
>>
>>>
>>>> +      if (imm0.isInteger(0)) {
>>>> +         i->op = OP_ADD;
>>>> +         for (int k = s; k < 2; k++) {
>>>> +            i->setSrc(k, i->getSrc(k + 1));
>>>> +            i->src(k).mod = i->src(k + 1).mod;
>>>> +         }
>>>
>>>
>>> aka
>>>
>>> i->moveSources(s + 1, -1) ?
>>
>>
>> Yes.
>>
>>
>>>
>>>> +      }
>>>> +      break;
>>>>     case OP_DIV:
>>>>        if (s != 1 || (i->dType != TYPE_S32 && i->dType != TYPE_U32))
>>>>           break;
>>>> --
>>>> 2.8.3
>>>>
>>>> _______________________________________________
>>>> mesa-dev mailing list
>>>> mesa-dev at lists.freedesktop.org
>>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev

-- 
-Samuel