[PATCH] nv50/ir: we can't replace 0x0 with zero reg for SHLADD

[PATCH] nv50/ir: we can't replace 0x0 with zero reg for SHLADD

[Karol Herbst]

fixes a crash in Alien Isolation

Signed-off-by: Karol Herbst <karolherbst@gmail.com>
---
 src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
index 732e1a93b4..4815d6df07 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -740,7 +740,7 @@ NVC0LegalizePostRA::visit(BasicBlock *bb)
                next = hi;
          }
 
-         if (i->op != OP_MOV && i->op != OP_PFETCH)
+         if (i->op != OP_MOV && i->op != OP_PFETCH && i->op != OP_SHLADD)
             replaceZero(i);
       }
    }
-- 
2.12.2

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/nouveau

Re: [PATCH] nv50/ir: we can't replace 0x0 with zero reg for SHLADD

[Ilia Mirkin]

On Sat, Apr 29, 2017 at 10:41 AM, Karol Herbst <karolherbst@gmail.com> wrote:
> fixes a crash in Alien Isolation

What crash? How did the zero get there? Does this only happen if you
do your optimization loop thing?

In either case, we still want the replaceZero() logic. However that
logic should be aware that the middle argument of a SHLADD is not to
be touched. Otherwise we could end up with an un-emittable
instruction.

>
> Signed-off-by: Karol Herbst <karolherbst@gmail.com>
> ---
>  src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> index 732e1a93b4..4815d6df07 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
> @@ -740,7 +740,7 @@ NVC0LegalizePostRA::visit(BasicBlock *bb)
>                 next = hi;
>           }
>
> -         if (i->op != OP_MOV && i->op != OP_PFETCH)
> +         if (i->op != OP_MOV && i->op != OP_PFETCH && i->op != OP_SHLADD)
>              replaceZero(i);
>        }
>     }
> --
> 2.12.2
>
_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/nouveau

Re: [PATCH] nv50/ir: we can't replace 0x0 with zero reg for SHLADD

[Karol Herbst]

2017-04-29 17:18 GMT+02:00 Ilia Mirkin <imirkin@alum.mit.edu>:
> On Sat, Apr 29, 2017 at 10:41 AM, Karol Herbst <karolherbst@gmail.com> wrote:
>> fixes a crash in Alien Isolation
>
> What crash?

assertion, because shladd requires an immediate, there can't be a reg at src1

"shladd u32 $r0 $r0 $r63 $r36" is invalid for the emiter so we have to
use 0x0 here

> How did the zero get there?

by replaceZero

> Does this only happen if you do your optimization loop thing?

no, it happens on master and the stable branches without any modifications

>
> In either case, we still want the replaceZero() logic. However that
> logic should be aware that the middle argument of a SHLADD is not to
> be touched. Otherwise we could end up with an un-emittable
> instruction.

ohh right, you mean for the other args... true.

>
>>
>> Signed-off-by: Karol Herbst <karolherbst@gmail.com>
>> ---
>>  src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>> index 732e1a93b4..4815d6df07 100644
>> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
>> @@ -740,7 +740,7 @@ NVC0LegalizePostRA::visit(BasicBlock *bb)
>>                 next = hi;
>>           }
>>
>> -         if (i->op != OP_MOV && i->op != OP_PFETCH)
>> +         if (i->op != OP_MOV && i->op != OP_PFETCH && i->op != OP_SHLADD)
>>              replaceZero(i);
>>        }
>>     }
>> --
>> 2.12.2
>>
_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/nouveau