[Mesa-dev] [PATCH] i965/fs: Handle MRF destinations in lower_integer_multiplication().

[Mark Janes]

I verified that this fixed the following tests on snb:

    dEQP-GLES3.functional.shaders.precision.int.highp_mul_fragment
    dEQP-GLES3.functional.shaders.precision.int.mediump_mul_fragment
    dEQP-GLES3.functional.shaders.precision.int.lowp_mul_fragment

However, my testing showed the following regression:
    dEQP-GLES2.functional.shaders.random.texture.vertex.45

Matt, can you check my findings?

-Mark

Matt Turner <mattst88 at gmail.com> writes:

> The lowered code reads from the destination, which isn't possible from
> message registers.
>
> Fixes the dEQP functional.shaders.precision.int.highp_mul_fragment test
> on SNB.
> ---
>  src/mesa/drivers/dri/i965/brw_fs.cpp | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
> index 269914d..e28eb5d 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> @@ -3192,7 +3192,8 @@ fs_visitor::lower_integer_multiplication()
>               * schedule multi-component multiplications much better.
>               */
>  
> -            if (inst->conditional_mod && inst->dst.is_null()) {
> +            fs_reg orig_dst = inst->dst;
> +            if (orig_dst.is_null() || orig_dst.file == MRF) {
>                 inst->dst = fs_reg(GRF, alloc.allocate(dispatch_width / 8),
>                                    inst->dst.type);
>              }
> @@ -3258,10 +3259,9 @@ fs_visitor::lower_integer_multiplication()
>  
>              ibld.ADD(dst, low, high);
>  
> -            if (inst->conditional_mod) {
> -               fs_reg null(retype(ibld.null_reg_f(), inst->dst.type));
> +            if (inst->conditional_mod || orig_dst.file == MRF) {
>                 set_condmod(inst->conditional_mod,
> -                           ibld.MOV(null, inst->dst));
> +                           ibld.MOV(orig_dst, inst->dst));
>              }
>           }
>  
> -- 
> 2.4.6