[Mesa-dev] [PATCH 16/20] nir: Reduce destination size of ballot intrinsic when possible

[Connor Abbott]

Aside from using nir_src_copy as mentioned by Ken,

Reviewed-by: Connor Abbott <cwabbott0 at gmail.com>

On Mon, Jul 10, 2017 at 10:22 AM, Matt Turner <mattst88 at gmail.com> wrote:
> Some hardware, like i965, doesn't support group sizes greater than 32.
> In that case, we can reduce the destination size of the ballot
> intrinsic, which will simplify our code generation.
> ---
> v2: Just change the intrinsic size, and don't add a new intrinsic (Connor)
>
>  src/compiler/nir/nir.h                |  2 ++
>  src/compiler/nir/nir_opt_intrinsics.c | 18 ++++++++++++++++++
>  src/intel/compiler/brw_compiler.c     |  1 +
>  3 files changed, 21 insertions(+)
>
> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
> index 1e2d7d3cf6..5518807b0b 100644
> --- a/src/compiler/nir/nir.h
> +++ b/src/compiler/nir/nir.h
> @@ -1842,6 +1842,8 @@ typedef struct nir_shader_compiler_options {
>      */
>     bool use_interpolated_input_intrinsics;
>
> +   unsigned max_subgroup_size;
> +
>     unsigned max_unroll_iterations;
>  } nir_shader_compiler_options;
>
> diff --git a/src/compiler/nir/nir_opt_intrinsics.c b/src/compiler/nir/nir_opt_intrinsics.c
> index 0358680aae..d30c1cf6bb 100644
> --- a/src/compiler/nir/nir_opt_intrinsics.c
> +++ b/src/compiler/nir/nir_opt_intrinsics.c
> @@ -62,6 +62,24 @@ opt_intrinsics_impl(nir_function_impl *impl)
>              replacement = nir_imm_int(&b, NIR_TRUE);
>              break;
>           }
> +         case nir_intrinsic_ballot: {
> +            assert(b.shader->options->max_subgroup_size != 0);
> +            if (b.shader->options->max_subgroup_size > 32 ||
> +                intrin->dest.ssa.bit_size <= 32)
> +               continue;
> +
> +            nir_intrinsic_instr *ballot =
> +               nir_intrinsic_instr_create(b.shader, nir_intrinsic_ballot);
> +            nir_ssa_dest_init(&ballot->instr, &ballot->dest, 1, 32, NULL);
> +            ballot->src[0] = intrin->src[0];
> +
> +            nir_builder_instr_insert(&b, &ballot->instr);
> +
> +            replacement = nir_pack_64_2x32_split(&b,
> +                                                 &ballot->dest.ssa,
> +                                                 nir_imm_int(&b, 0));
> +            break;
> +         }
>           default:
>              break;
>           }
> diff --git a/src/intel/compiler/brw_compiler.c b/src/intel/compiler/brw_compiler.c
> index 397c8cccf9..b910fcbc3d 100644
> --- a/src/intel/compiler/brw_compiler.c
> +++ b/src/intel/compiler/brw_compiler.c
> @@ -57,6 +57,7 @@ static const struct nir_shader_compiler_options scalar_nir_options = {
>     .lower_unpack_snorm_4x8 = true,
>     .lower_unpack_unorm_2x16 = true,
>     .lower_unpack_unorm_4x8 = true,
> +   .max_subgroup_size = 64, /* FIXME */
>     .max_unroll_iterations = 32,
>  };
>
> --
> 2.13.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev