[Mesa-dev] [PATCH 2/2] nir: Don't try to scalarize unpack ops.

[Matt Turner]

On Tue, Aug 11, 2015 at 11:25 AM, Eric Anholt <eric at anholt.net> wrote:
> Avoids regressions in vc4 when trying to do our blending in NIR.
>
> v2: Add the other unpack ops I meant to when writing the original commit
>     message.
> ---
>  src/glsl/nir/nir_lower_alu_to_scalar.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
>
> diff --git a/src/glsl/nir/nir_lower_alu_to_scalar.c b/src/glsl/nir/nir_lower_alu_to_scalar.c
> index 5d15fb2..efbe9e7 100644
> --- a/src/glsl/nir/nir_lower_alu_to_scalar.c
> +++ b/src/glsl/nir/nir_lower_alu_to_scalar.c
> @@ -100,6 +100,21 @@ lower_alu_instr_scalar(nir_alu_instr *instr, void *mem_ctx)
>         */
>        return;
>
> +   case nir_op_unpack_unorm_4x8:
> +   case nir_op_unpack_snorm_4x8:
> +   case nir_op_unpack_unorm_2x16:
> +   case nir_op_unpack_snorm_2x16:
> +      /* There is no scalar version of these ops, unless we were to break it
> +       * down to bitshifts and math (which is definitely not intended).
> +       */
> +      return;
> +
> +   case nir_op_unpack_half_2x16:
> +      /* We could split this into unpack_half_2x16_split_[xy], but should
> +       * we?
> +       */
> +      return;
> +

Reviewed-by: Matt Turner <mattst88 at gmail.com>

Seems like we should we add the pack opcodes as well?