[Mesa-dev] [PATCH 4/6] i965: fs: Add fixup for textureSize on Gen7

[Kenneth Graunke]

On 11/22/2012 04:35 AM, Chris Forbes wrote:
> Signed-off-by: Chris Forbes <chrisf at ijw.co.nz>
> ---
>   src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 16 +++++++++++++++-
>   1 file changed, 15 insertions(+), 1 deletion(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> index 0050252..1fff000 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp
> @@ -1072,10 +1072,24 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
>         }
>         break;
>      }
> -   case ir_txs:
> +   case ir_txs: {
>         emit(MOV(fs_reg(MRF, base_mrf + mlen, BRW_REGISTER_TYPE_UD), lod));
>         mlen += reg_width;

This won't work :)  You haven't reached the emit(SHADER_OPCODE_TXS, dst) 
yet, so...you're emitting the fixup for the destination before filling it.

You might want to move this code to the caller (visit(ir_texture *)), 
which would take care of it in one place for both Gen 6 and Gen 7.

> +
> +      /* fixup #layers for cube arrays */
> +      glsl_type const * type = ir->sampler->variable_referenced()->type;
> +      if (type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE &&
> +            type->sampler_array) {
> +         /* emit_math can't cope with immediates. */
> +         fs_reg * imm = new (mem_ctx) fs_reg(this, glsl_type::ivec4_type);
> +         emit(MOV(*imm, fs_reg(6)));

I can never keep the math workarounds straight.  If MATH can't handle 
immediates, I'd rather see emit_math() emit the MOV rather than 
everybody that wants to emit math having to deal with it.

> +         fs_reg depth = dst;
> +         depth.reg_offset = 2;
> +         emit_math(SHADER_OPCODE_INT_QUOTIENT, depth, depth, *imm);
> +      }
>         break;
> +   }
>      case ir_txf:
>         /* It appears that the ld instruction used for txf does its
>          * address bounds check before adding in the offset.  To work
>