[Mesa-dev] [PATCH] ac/nir: mark some texture intrinsics as convergent

[Marek Olšák]

On Thu, May 30, 2019, 3:54 PM Rhys Perry <pendingchaos02 at gmail.com> wrote:

> Otherwise LLVM can sink them and their texture coordinate calculations
> into divergent branches.
>
> v2: simplify the conditions on which the intrinsic is marked as convergent
>
> Cc: <mesa-stable at lists.freedesktop.org>
> Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
> Reviewed-By: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
> ---
>  src/amd/common/ac_nir_to_llvm.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
>
> diff --git a/src/amd/common/ac_nir_to_llvm.c
> b/src/amd/common/ac_nir_to_llvm.c
> index 265e3b636c4..b1a191ac24c 100644
> --- a/src/amd/common/ac_nir_to_llvm.c
> +++ b/src/amd/common/ac_nir_to_llvm.c
> @@ -1394,6 +1394,18 @@ static LLVMValueRef build_tex_intrinsic(struct
> ac_nir_context *ctx,
>         }
>
>         args->attributes = AC_FUNC_ATTR_READNONE;
> +       /* Prevent texture instructions with implicit derivatives from
> being
> +        * sinked into branches. */
> +       switch (instr->op) {
> +       case nir_texop_tex:
> +       case nir_texop_txb:
> +       case nir_texop_lod:
> +               args->attributes |= AC_FUNC_ATTR_CONVERGENT;
> +               break;
> +       default:
> +               break;
> +       }
>

I think this should only apply to the fragment shader.

Marek

+
>         return ac_build_image_opcode(&ctx->ac, args);
>  }
>
> --
> 2.21.0
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20190530/dfd61155/attachment-0001.html>