[Mesa-dev] [PATCH v3 3/6] nv50/ra: always prefer def == src2 for mad/sad

[Ilia Mirkin]

On Sun, Oct 30, 2016 at 5:20 PM, Karol Herbst <karolherbst at gmail.com> wrote:
> total instructions in shared programs : 3491273 -> 3487792 (-0.10%)
> total gprs used in shared programs    : 453803 -> 453804 (0.00%)
> total local used in shared programs   : 21621 -> 21621 (0.00%)
> total bytes used in shared programs   : 32000072 -> 31968072 (-0.10%)
>
>                 local        gpr       inst      bytes
>     helped           0          26        1937        1937
>       hurt           0          23          81          81
>
> v2: reorder to show the benefit of this patch
>
> Signed-off-by: Karol Herbst <karolherbst at gmail.com>
> ---
>  src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
> index d36c853..d3e04c0 100644
> --- a/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
> +++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp
> @@ -1468,8 +1468,7 @@ GCRA::allocateRegisters(ArrayList& insns)
>           nodes[i].init(regs, lval);
>           RIG.insert(&nodes[i]);
>
> -         if (lval->inFile(FILE_GPR) && lval->getInsn() != NULL &&
> -             prog->getTarget()->getChipset() < 0xc0) {
> +         if (lval->inFile(FILE_GPR) && lval->getInsn() != NULL) {

Drop the flagsDef thing below as well? Or make it conditional on the
chipset? (Should work fine for GF100+ afaik.)

By the way, you could probably get some of the hurt back on nvc0 by
only doing this for GF100+ when src(1) is an immediate. (There's a
benefit either way for nv50.)

>              Instruction *insn = lval->getInsn();
>              if (insn->op == OP_MAD || insn->op == OP_SAD)
>                 // Short encoding only possible if they're all GPRs, no need to
> --
> 2.10.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev