[Nouveau] [Mesa-dev] [PATCH] nv50/ir: avoid messing up arg1 of PFETCH

[Tobias Klausmann]

On 23.05.2015 08:06, Ilia Mirkin wrote:
> There can be scenarios where the "indirect" arg of a PFETCH becomes
> known, and so the code will attempt to propagate it. Use this
> opportunity to just fold it into the first argument, and prevent the
> load propagation pass from touching PFETCH further.
>
> This fixes gs-input-array-vec4-index-rd.shader_test and
> vs-output-array-vec4-index-wr-before-gs.shader_test on nvc0 at least.
>
> Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
> Cc: "10.5 10.6" <mesa-stable at lists.freedesktop.org>
> ---
>   src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp | 10 ++++++++++
>   1 file changed, 10 insertions(+)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
> index 72dd31e..98e3d1f 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
> @@ -236,6 +236,9 @@ LoadPropagation::visit(BasicBlock *bb)
>         if (i->op == OP_CALL) // calls have args as sources, they must be in regs
>            continue;
>   
> +      if (i->op == OP_PFETCH) // pfetch expects arg1 to be a reg
> +         continue;
> +
>         if (i->srcExists(1))
>            checkSwapSrc01(i);
>   
> @@ -581,6 +584,11 @@ ConstantFolding::expr(Instruction *i,
>      case OP_POPCNT:
>         res.data.u32 = util_bitcount(a->data.u32 & b->data.u32);
>         break;
> +   case OP_PFETCH:
> +      // The two arguments to pfetch are logically added together. Normally
> +      // the second argument will not be constant, but that can happen.
> +      res.data.u32 = a->data.u32 + b->data.u32;
> +      break;
>      default:
>         return;
>      }
> @@ -610,6 +618,8 @@ ConstantFolding::expr(Instruction *i,
>            bld.setPosition(i, false);
>            i->setSrc(1, bld.loadImm(NULL, res.data.u32));
>         }
> +   } else if (i->op == OP_PFETCH) {
> +      // Leave PFETCH alone... we just folded its 2 args into 1.
>      } else {
>         i->op = i->saturate ? OP_SAT : OP_MOV; /* SAT handled by unary() */
>      }
this last part sure works, but it gets ugly, while you are at it, can 
you change it to a switch statement?