[Mesa-dev] [PATCH v3 19/23] glsl: Add image built-in function generator.

Paul Berry stereotype441 at gmail.com
Tue Dec 17 08:39:33 PST 2013


On 17 December 2013 02:10, Francisco Jerez <currojerez at riseup.net> wrote:

> Because of the combinatorial explosion of different image built-ins
> with different image dimensionalities and base data types, enumerating
> all the 242 possibilities would be annoying and a waste of .text
> space.  Instead use a special path in the built-in builder that loops
> over all the known image types.
>
> v2: Generate built-ins on GLSL version 4.20 too.  Rename
>     '_has_float_data_type' to '_supports_float_data_type'.  Avoid
>     duplicating enumeration of image built-ins in create_intrinsics()
>     and create_builtins().
> v3: Use a more orthodox approach for passing image built-in generator
>     parameters.
> ---
>  src/glsl/builtin_functions.cpp | 267
> +++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 267 insertions(+)
>
> diff --git a/src/glsl/builtin_functions.cpp
> b/src/glsl/builtin_functions.cpp
> index 4251948..90a9336 100644
> --- a/src/glsl/builtin_functions.cpp
> +++ b/src/glsl/builtin_functions.cpp
> @@ -334,6 +334,13 @@ shader_atomic_counters(const _mesa_glsl_parse_state
> *state)
>     return state->ARB_shader_atomic_counters_enable;
>  }
>
> +static bool
> +shader_image_load_store(const _mesa_glsl_parse_state *state)
> +{
> +   return (state->is_version(420, 0) ||
> +           state->ARB_shader_image_load_store_enable);
> +}
> +
>  /** @} */
>
>
>  /******************************************************************************/
> @@ -407,6 +414,33 @@ private:
>     /** Create a new function and add the given signatures. */
>     void add_function(const char *name, ...);
>
> +   enum image_function_flags {
> +      IMAGE_FUNCTION_EMIT_STUB = (1 << 0),
> +      IMAGE_FUNCTION_HAS_RETURN = (1 << 1),
> +      IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE = (1 << 2),
> +      IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE = (1 << 3),
> +      IMAGE_FUNCTION_READ_ONLY = (1 << 4),
> +      IMAGE_FUNCTION_WRITE_ONLY = (1 << 5)
> +   };
> +
> +   /**
> +    * Create a new image built-in function for all known image types.
> +    * \p flags is a bitfield of \c image_function_flags flags.
> +    */
> +   void add_image_function(const char *name,
> +                           const char *intrinsic_name,
> +                           unsigned num_arguments,
> +                           unsigned flags);
> +
> +   /**
> +    * Create new functions for all known image built-ins and types.
> +    * If \p glsl is \c true, use the GLSL built-in names and emit code
> +    * to call into the actual compiler intrinsic.  If \p glsl is
> +    * false, emit a function prototype with no body for each image
> +    * intrinsic name.
> +    */
> +   void add_image_functions(bool glsl);
> +
>     ir_function_signature *new_sig(const glsl_type *return_type,
>                                    builtin_available_predicate avail,
>                                    int num_params, ...);
> @@ -570,6 +604,20 @@ private:
>     ir_function_signature *_atomic_op(const char *intrinsic,
>                                       builtin_available_predicate avail);
>
> +   ir_function_signature *_image_prototype(const glsl_type *image_type,
> +                                           const char *intrinsic_name,
> +                                           unsigned num_arguments,
> +                                           unsigned flags);
> +   ir_function_signature *_image(const glsl_type *image_type,
> +                                 const char *intrinsic_name,
> +                                 unsigned num_arguments,
> +                                 unsigned flags);
> +
> +   ir_function_signature *_memory_barrier_intrinsic(
> +      builtin_available_predicate avail);
> +   ir_function_signature *_memory_barrier(
> +      builtin_available_predicate avail);
> +
>  #undef B0
>  #undef B1
>  #undef B2
> @@ -684,6 +732,12 @@ builtin_builder::create_intrinsics()
>     add_function("__intrinsic_atomic_predecrement",
>                  _atomic_intrinsic(shader_atomic_counters),
>                  NULL);
> +
> +   add_image_functions(false);
> +
> +   add_function("__intrinsic_memory_barrier",
> +                _memory_barrier_intrinsic(shader_image_load_store),
> +                NULL);
>  }
>
>  /**
> @@ -2106,6 +2160,12 @@ builtin_builder::create_builtins()
>                             shader_atomic_counters),
>                  NULL);
>
> +   add_image_functions(true);
> +
> +   add_function("memoryBarrier",
> +                _memory_barrier(shader_image_load_store),
> +                NULL);
> +
>  #undef F
>  #undef FI
>  #undef FIU
> @@ -2139,6 +2199,120 @@ builtin_builder::add_function(const char *name,
> ...)
>     shader->symbols->add_function(f);
>  }
>
> +void
> +builtin_builder::add_image_function(const char *name,
> +                                    const char *intrinsic_name,
> +                                    unsigned num_arguments,
> +                                    unsigned flags)
> +{
> +   static const glsl_type *const types[] = {
> +      glsl_type::image1D_type,
> +      glsl_type::image2D_type,
> +      glsl_type::image3D_type,
> +      glsl_type::image2DRect_type,
> +      glsl_type::imageCube_type,
> +      glsl_type::imageBuffer_type,
> +      glsl_type::image1DArray_type,
> +      glsl_type::image2DArray_type,
> +      glsl_type::imageCubeArray_type,
> +      glsl_type::image2DMS_type,
> +      glsl_type::image2DMSArray_type,
> +      glsl_type::iimage1D_type,
> +      glsl_type::iimage2D_type,
> +      glsl_type::iimage3D_type,
> +      glsl_type::iimage2DRect_type,
> +      glsl_type::iimageCube_type,
> +      glsl_type::iimageBuffer_type,
> +      glsl_type::iimage1DArray_type,
> +      glsl_type::iimage2DArray_type,
> +      glsl_type::iimageCubeArray_type,
> +      glsl_type::iimage2DMS_type,
> +      glsl_type::iimage2DMSArray_type,
> +      glsl_type::uimage1D_type,
> +      glsl_type::uimage2D_type,
> +      glsl_type::uimage3D_type,
> +      glsl_type::uimage2DRect_type,
> +      glsl_type::uimageCube_type,
> +      glsl_type::uimageBuffer_type,
> +      glsl_type::uimage1DArray_type,
> +      glsl_type::uimage2DArray_type,
> +      glsl_type::uimageCubeArray_type,
> +      glsl_type::uimage2DMS_type,
> +      glsl_type::uimage2DMSArray_type
> +   };
> +   ir_function *f = new(mem_ctx) ir_function(name);
> +
> +   for (unsigned i = 0; i < Elements(types); ++i) {
> +      if (types[i]->fields.image.type != GLSL_TYPE_FLOAT ||
> +          (flags & IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE))
> +         f->add_signature(_image(types[i], intrinsic_name,
> +                                 num_arguments, flags));
> +   }
> +
> +   shader->symbols->add_function(f);
> +}
> +
> +void
> +builtin_builder::add_image_functions(bool glsl)
> +{
> +   add_image_function(glsl ? "imageLoad" : "__intrinsic_image_load",
> +                      "__intrinsic_image_load", 0,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN |
> +                       IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE |
> +                       IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE |
> +                       IMAGE_FUNCTION_READ_ONLY));
> +
> +   add_image_function(glsl ? "imageStore" : "__intrinsic_image_store",
> +                      "__intrinsic_image_store", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE |
> +                       IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE |
> +                       IMAGE_FUNCTION_WRITE_ONLY));
> +
> +   add_image_function(glsl ? "imageAtomicAdd" :
> "__intrinsic_image_atomic_add",
> +                      "__intrinsic_image_atomic_add", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function(glsl ? "imageAtomicMin" :
> "__intrinsic_image_atomic_min",
> +                      "__intrinsic_image_atomic_min", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function(glsl ? "imageAtomicMax" :
> "__intrinsic_image_atomic_max",
> +                      "__intrinsic_image_atomic_max", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function(glsl ? "imageAtomicAnd" :
> "__intrinsic_image_atomic_and",
> +                      "__intrinsic_image_atomic_and", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function(glsl ? "imageAtomicOr" :
> "__intrinsic_image_atomic_or",
> +                      "__intrinsic_image_atomic_or", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function(glsl ? "imageAtomicXor" :
> "__intrinsic_image_atomic_xor",
> +                      "__intrinsic_image_atomic_xor", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function((glsl ? "imageAtomicExchange" :
> +                       "__intrinsic_image_atomic_exchange"),
> +                      "__intrinsic_image_atomic_exchange", 1,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
> +
> +   add_image_function((glsl ? "imageAtomicCompSwap" :
> +                       "__intrinsic_image_atomic_comp_swap"),
> +                      "__intrinsic_image_atomic_comp_swap", 2,
> +                      ((glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) |
> +                       IMAGE_FUNCTION_HAS_RETURN));
>

This is definitely an improvement, thanks.  I still think there are some
minor tweaks that would reduce duplication of information, though:

- Precompute the value (glsl ? IMAGE_FUNCTION_EMIT_STUB : 0) rather than
duplicating it at each add_image_function call site.  Alternatively, just
make emit_stub a bool parameter to add_image_function rather than including
it in flags.

- Rather than repeating the pattern add_image_function((glsl ? BUILTIN_NAME
: INTRINSIC_NAME), INTRINSIC_NAME, ...) everywhere, make BUILTIN_NAME and
INTRINSIC_NAME the first two arguments to add_image_function(), and have
add_image_function() pick which name to use based on emit_stub.

- Since IMAGE_FUNCTION_HAS_RETURN is by far the more common case, invert
the sense of the flag (e.g. change it to IMAGE_FUNCTION_RETURNS_VOID).

With those three changes, the code would look like this:

   const unsigned base_flags = glsl ? IMAGE_FUNCTION_EMIT_STUB : 0;

   add_image_function("imageLoad", "__intrinsic_image_load", 0,
                      (base_flags |
                       IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE |
                       IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE |
                       IMAGE_FUNCTION_READ_ONLY));

   add_image_function("imageStore", "__intrinsic_image_store", 1,
                      (base_flags |
                       IMAGE_FUNCTION_RETURNS_VOID |
                       IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE |
                       IMAGE_FUNCTION_SUPPORTS_FLOAT_DATA_TYPE |
                       IMAGE_FUNCTION_WRITE_ONLY));

   add_image_function("imageAtomicAdd", "__intrinsic_image_atomic_add", 1,
                      base_flags);

   add_image_function("imageAtomicMin", "__intrinsic_image_atomic_min", 1,
                      base_flags);

   add_image_function("imageAtomicMax", "__intrinsic_image_atomic_max", 1,
                      base_flags);

   add_image_function("imageAtomicAnd", "__intrinsic_image_atomic_and", 1,
                      base_flags);

   add_image_function("imageAtomicOr", "__intrinsic_image_atomic_or", 1,
                      base_flags);

   add_image_function("imageAtomicXor", "__intrinsic_image_atomic_xor", 1,
                      base_flags);

   add_image_function("imageAtomicExchange",
                      "__intrinsic_image_atomic_exchange", 1,
                      base_flags);

   add_image_function("imageAtomicCompSwap",
                      "__intrinsic_image_atomic_comp_swap", 2,
                      base_flags);

which IMHO is a lot more compact and easy to read.

With those changes, the patch is:

Reviewed-by: Paul Berry <stereotype441 at gmail.com>



> +}
> +
>  ir_variable *
>  builtin_builder::in_var(const glsl_type *type, const char *name)
>  {
> @@ -3991,6 +4165,99 @@ builtin_builder::_atomic_op(const char *intrinsic,
>     return sig;
>  }
>
> +ir_function_signature *
> +builtin_builder::_image_prototype(const glsl_type *image_type,
> +                                  const char *intrinsic_name,
> +                                  unsigned num_arguments,
> +                                  unsigned flags)
> +{
> +   const glsl_type *data_type = glsl_type::get_instance(
> +      image_type->fields.image.type,
> +      (flags & IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE ? 4 : 1),
> +      1);
> +   const glsl_type *ret_type = (flags & IMAGE_FUNCTION_HAS_RETURN ?
> data_type :
> +                                glsl_type::void_type);
> +
> +   /* Addressing arguments that are always present. */
> +   ir_variable *image = in_var(image_type, "image");
> +   ir_variable *coord = in_var(
> +      glsl_type::ivec(image_type->coordinate_components()), "coord");
> +
> +   ir_function_signature *sig = new_sig(
> +      ret_type, shader_image_load_store, 2, image, coord);
> +
> +   /* Sample index for multisample images. */
> +   if (image_type->fields.image.dimension == GLSL_IMAGE_DIM_MS)
> +      sig->parameters.push_tail(in_var(glsl_type::int_type, "sample"));
> +
> +   /* Data arguments. */
> +   for (unsigned i = 0; i < num_arguments; ++i)
> +      sig->parameters.push_tail(in_var(data_type,
> +                                       ralloc_asprintf(NULL, "arg%d",
> i)));
> +
> +   /* Set the maximal set of qualifiers allowed for this image
> +    * built-in.  Function calls with arguments having fewer
> +    * qualifiers than present in the prototype are allowed by the
> +    * spec, but not with more, i.e. this will make the compiler
> +    * accept everything that needs to be accepted, and reject cases
> +    * like loads from write-only or stores to read-only images.
> +    */
> +   image->image.read_only = flags & IMAGE_FUNCTION_READ_ONLY;
> +   image->image.write_only = flags & IMAGE_FUNCTION_WRITE_ONLY;
> +   image->image.coherent = true;
> +   image->image._volatile = true;
> +   image->image._restrict = true;
> +
> +   return sig;
> +}
> +
> +ir_function_signature *
> +builtin_builder::_image(const glsl_type *image_type,
> +                        const char *intrinsic_name,
> +                        unsigned num_arguments,
> +                        unsigned flags)
> +{
> +   ir_function_signature *sig = _image_prototype(image_type,
> intrinsic_name,
> +                                                 num_arguments, flags);
> +
> +   if (flags & IMAGE_FUNCTION_EMIT_STUB) {
> +      ir_factory body(&sig->body, mem_ctx);
> +      ir_function *f = shader->symbols->get_function(intrinsic_name);
> +
> +      if (flags & IMAGE_FUNCTION_HAS_RETURN) {
> +         ir_variable *ret_val =
> +            body.make_temp(sig->return_type, "_ret_val");
> +         body.emit(call(f, ret_val, sig->parameters));
> +         body.emit(ret(ret_val));
> +      } else {
> +         body.emit(call(f, NULL, sig->parameters));
> +      }
> +
> +      sig->is_defined = true;
> +
> +   } else {
> +      sig->is_intrinsic = true;
> +   }
> +
> +   return sig;
> +}
> +
> +ir_function_signature *
> +builtin_builder::_memory_barrier_intrinsic(builtin_available_predicate
> avail)
> +{
> +   MAKE_INTRINSIC(glsl_type::void_type, avail, 0);
> +   return sig;
> +}
> +
> +ir_function_signature *
> +builtin_builder::_memory_barrier(builtin_available_predicate avail)
> +{
> +   MAKE_SIG(glsl_type::void_type, avail, 0);
> +
> body.emit(call(shader->symbols->get_function("__intrinsic_memory_barrier"),
> +                  NULL, sig->parameters));
> +   return sig;
> +}
> +
>  /** @} */
>
>
>  /******************************************************************************/
> --
> 1.8.5.1
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20131217/51705322/attachment-0001.html>


More information about the mesa-dev mailing list