[Mesa-dev] [PATCH 12/12] i965/gen7+: Use NIR for lowering of pack/unpack opcodes.

[Iago Toral]

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

On Mon, 2016-01-25 at 15:18 -0800, Matt Turner wrote:
> ---
>  src/mesa/drivers/dri/i965/brw_compiler.c           | 15 +++++++++++++
>  .../dri/i965/brw_fs_channel_expressions.cpp        |  8 +++++++
>  src/mesa/drivers/dri/i965/brw_link.cpp             | 25 ++++++----------------
>  3 files changed, 29 insertions(+), 19 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/brw_compiler.c b/src/mesa/drivers/dri/i965/brw_compiler.c
> index 21fff1d..f9e22d1 100644
> --- a/src/mesa/drivers/dri/i965/brw_compiler.c
> +++ b/src/mesa/drivers/dri/i965/brw_compiler.c
> @@ -87,7 +87,15 @@ shader_perf_log_mesa(void *data, const char *fmt, ...)
>  static const struct nir_shader_compiler_options scalar_nir_options = {
>     COMMON_OPTIONS,
>     .lower_pack_half_2x16 = true,
> +   .lower_pack_snorm_2x16 = true,
> +   .lower_pack_snorm_4x8 = true,
> +   .lower_pack_unorm_2x16 = true,
> +   .lower_pack_unorm_4x8 = true,
>     .lower_unpack_half_2x16 = true,
> +   .lower_unpack_snorm_2x16 = true,
> +   .lower_unpack_snorm_4x8 = true,
> +   .lower_unpack_unorm_2x16 = true,
> +   .lower_unpack_unorm_4x8 = true,
>  };
>  
>  static const struct nir_shader_compiler_options vector_nir_options = {
> @@ -98,6 +106,13 @@ static const struct nir_shader_compiler_options vector_nir_options = {
>      * instructions because it can optimize better for us.
>      */
>     .fdot_replicates = true,
> +
> +   .lower_pack_snorm_2x16 = true,
> +   .lower_pack_unorm_2x16 = true,
> +   .lower_unpack_snorm_2x16 = true,
> +   .lower_unpack_unorm_2x16 = true,
> +   .lower_extract_byte = true,
> +   .lower_extract_word = true,
>  };
>  
>  struct brw_compiler *
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_channel_expressions.cpp b/src/mesa/drivers/dri/i965/brw_fs_channel_expressions.cpp
> index b16dd2f..cbad47e 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_channel_expressions.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_channel_expressions.cpp
> @@ -73,6 +73,10 @@ channel_expressions_predicate(ir_instruction *ir)
>  
>     switch (expr->operation) {
>        case ir_unop_pack_half_2x16:
> +      case ir_unop_pack_snorm_2x16:
> +      case ir_unop_pack_snorm_4x8:
> +      case ir_unop_pack_unorm_2x16:
> +      case ir_unop_pack_unorm_4x8:
>           return false;
>  
>        /* these opcodes need to act on the whole vector,
> @@ -166,6 +170,10 @@ ir_channel_expressions_visitor::visit_leave(ir_assignment *ir)
>  
>     switch (expr->operation) {
>        case ir_unop_pack_half_2x16:
> +      case ir_unop_pack_snorm_2x16:
> +      case ir_unop_pack_snorm_4x8:
> +      case ir_unop_pack_unorm_2x16:
> +      case ir_unop_pack_unorm_4x8:
>        case ir_unop_interpolate_at_centroid:
>        case ir_binop_interpolate_at_offset:
>        case ir_binop_interpolate_at_sample:
> diff --git a/src/mesa/drivers/dri/i965/brw_link.cpp b/src/mesa/drivers/dri/i965/brw_link.cpp
> index 8f2d760..ab9d792 100644
> --- a/src/mesa/drivers/dri/i965/brw_link.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_link.cpp
> @@ -73,26 +73,13 @@ brw_lower_packing_builtins(struct brw_context *brw,
>                             gl_shader_stage shader_type,
>                             exec_list *ir)
>  {
> -   const struct brw_compiler *compiler = brw->intelScreen->compiler;
> -
> -   int ops = LOWER_PACK_SNORM_2x16
> -           | LOWER_UNPACK_SNORM_2x16
> -           | LOWER_PACK_UNORM_2x16
> -           | LOWER_UNPACK_UNORM_2x16;
> -
> -   if (compiler->scalar_stage[shader_type]) {
> -      ops |= LOWER_UNPACK_UNORM_4x8
> -           | LOWER_UNPACK_SNORM_4x8
> -           | LOWER_PACK_UNORM_4x8
> -           | LOWER_PACK_SNORM_4x8;
> -   }
> -
> -   if (brw->gen < 7) {
> -      ops |= LOWER_PACK_HALF_2x16
> -          |  LOWER_UNPACK_HALF_2x16;
> -   }
> +   /* Gens < 7 don't have instructions to convert to or from half-precision,
> +    * and Gens < 6 don't expose that functionality.
> +    */
> +   if (brw->gen != 6)
> +      return;
>  
> -   lower_packing_builtins(ir, ops);
> +   lower_packing_builtins(ir, LOWER_PACK_HALF_2x16 | LOWER_UNPACK_HALF_2x16);
>  }
>  
>  static void