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

Samuel Pitoiset samuel.pitoiset at gmail.com
Thu Jun 30 22:54:08 UTC 2016



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.

>
>> +      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


More information about the mesa-dev mailing list