[Mesa-dev] [PATCH 3/3] i965/fs: Read all components of a SSBO field with one send

[Iago Toral]

On Sun, 2015-10-18 at 21:31 -0700, Kristian Høgsberg Kristensen wrote:
> Instead of looping through single-component reads, read all components
> in one go.
> 
> Signed-off-by: Kristian Høgsberg Kristensen <krh at bitplanet.net>
> ---
>  src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 25 +++++++------------------
>  1 file changed, 7 insertions(+), 18 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> index 792663f..f09578c 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> @@ -1509,24 +1509,13 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
>        }
>  
>        /* Read the vector */
> -      for (int i = 0; i < instr->num_components; i++) {
> -         fs_reg read_result = emit_untyped_read(bld, surf_index, offset_reg,
> -                                                1 /* dims */, 1 /* size */,
> -                                                BRW_PREDICATE_NONE);
> -         read_result.type = dest.type;
> -         bld.MOV(dest, read_result);
> -         dest = offset(dest, bld, 1);
> -
> -         /* Vector components are stored contiguous in memory */
> -         if (i < instr->num_components) {
> -            if (!has_indirect) {
> -               const_offset_bytes += 4;
> -               bld.MOV(offset_reg, fs_reg(const_offset_bytes));
> -            } else {
> -               bld.ADD(offset_reg, offset_reg, brw_imm_ud(4));
> -            }
> -         }
> -      }
> +      fs_reg read_result = emit_untyped_read(bld, surf_index, offset_reg,
> +                                             1 /* dims */,
> +                                             instr->num_components,
> +                                             BRW_PREDICATE_NONE);
> +      read_result.type = dest.type;
> +      for (int i = 0; i < instr->num_components; i++)
> +         bld.MOV(offset(dest, bld, i), offset(read_result, bld, i));
>  
>        break;
>     }

Thanks Kristian, this was silly, I wonder why I decided to do it like
this to begin with, specially since I do full vector reads for vec4...

Reviewed-by: Iago Toral Quiroga <itoral at igalia.com>

Iago