[Mesa-dev] [PATCH 3/5] glsl: Implement the required built-in functions when OES_shader_image_atomic is enabled.

[Ilia Mirkin]

On Mon, Feb 22, 2016 at 9:40 PM, Francisco Jerez <currojerez at riseup.net> wrote:
> This is basically just the same atomic functions exposed by
> ARB_shader_image_load_store, with one exception:
>
>     "highp float imageAtomicExchange(
>          coherent IMAGE_PARAMS,
>          float data);"
>
> There's no float atomic exchange overload in the original
> ARB_shader_image_load_store or GL 4.2, so this seems like new
> functionality that requires specific back-end support and a separate
> availability condition in the built-in function generator.
> ---
>  src/compiler/glsl/builtin_functions.cpp | 23 +++++++++++++++++++----
>  1 file changed, 19 insertions(+), 4 deletions(-)
>
> diff --git a/src/compiler/glsl/builtin_functions.cpp b/src/compiler/glsl/builtin_functions.cpp
> index f488434..db74672 100644
> --- a/src/compiler/glsl/builtin_functions.cpp
> +++ b/src/compiler/glsl/builtin_functions.cpp
> @@ -448,8 +448,16 @@ shader_image_load_store(const _mesa_glsl_parse_state *state)
>  static bool
>  shader_image_atomic(const _mesa_glsl_parse_state *state)
>  {
> -   return (state->is_version(420, 0) ||
> -           state->ARB_shader_image_load_store_enable);
> +   return (state->is_version(420, 320) ||
> +           state->ARB_shader_image_load_store_enable ||
> +           state->OES_shader_image_atomic_enable);
> +}
> +
> +static bool
> +shader_image_atomic_exchange_float(const _mesa_glsl_parse_state *state)
> +{
> +   return (state->is_version(450, 320) ||
> +           state->OES_shader_image_atomic_enable);
>  }
>
>  static bool
> @@ -586,6 +594,7 @@ private:
>        IMAGE_FUNCTION_WRITE_ONLY = (1 << 5),
>        IMAGE_FUNCTION_AVAIL_ATOMIC = (1 << 6),
>        IMAGE_FUNCTION_MS_ONLY = (1 << 7),
> +      IMAGE_FUNCTION_AVAIL_ATOMIC_EXCHANGE = (1 << 8)
>     };
>
>     /**
> @@ -2981,7 +2990,9 @@ builtin_builder::add_image_functions(bool glsl)
>     add_image_function((glsl ? "imageAtomicExchange" :
>                         "__intrinsic_image_atomic_exchange"),
>                        "__intrinsic_image_atomic_exchange",
> -                      &builtin_builder::_image_prototype, 1, atom_flags);
> +                      &builtin_builder::_image_prototype, 1,
> +                      (flags | IMAGE_FUNCTION_AVAIL_ATOMIC_EXCHANGE |
> +                       IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE));
>
>     add_image_function((glsl ? "imageAtomicCompSwap" :
>                         "__intrinsic_image_atomic_comp_swap"),
> @@ -5250,7 +5261,11 @@ builtin_builder::_image_prototype(const glsl_type *image_type,
>        glsl_type::ivec(image_type->coordinate_components()), "coord");
>
>     const builtin_available_predicate avail =
> -      (flags & IMAGE_FUNCTION_AVAIL_ATOMIC ? shader_image_atomic :
> +      (((flags & IMAGE_FUNCTION_AVAIL_ATOMIC_EXCHANGE) &&
> +        image_type->sampled_type == GLSL_TYPE_FLOAT) ?
> +       shader_image_atomic_exchange_float :
> +       flags & (IMAGE_FUNCTION_AVAIL_ATOMIC_EXCHANGE |
> +                IMAGE_FUNCTION_AVAIL_ATOMIC) ? shader_image_atomic :
>         shader_image_load_store);

Wow, this is a monster expression. I know how much you like "pure"
const values, but perhaps this might be more readable as a bunch of
if's? I think you can actually do

const foo; // no init
if () foo = a
else foo = b

type of stuff. Your call though.

For some reason in my version I had decided to add a separate
add_image_float_functions, but in hindsight that may have been
misguided, as this seems to be sufficient.

>     ir_function_signature *sig = new_sig(ret_type, avail, 2, image, coord);
>
> --
> 2.7.0
>