From e8dbe430aa77d6a775e087938bd19002f2a39e18 Mon Sep 17 00:00:00 2001 From: Francisco Jerez Date: Mon, 25 Nov 2013 20:29:57 -0800 Subject: [PATCH] glsl: Add image built-in function generator. 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. v4: Cosmetic changes. Acked-by: Paul Berry --- src/glsl/builtin_functions.cpp | 251 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 251 insertions(+) diff --git a/src/glsl/builtin_functions.cpp b/src/glsl/builtin_functions.cpp index 4ff2212..2905a35 100644 --- a/src/glsl/builtin_functions.cpp +++ b/src/glsl/builtin_functions.cpp @@ -340,6 +340,13 @@ shader_trinary_minmax(const _mesa_glsl_parse_state *state) return state->AMD_shader_trinary_minmax_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); +} + /** @} */ /******************************************************************************/ @@ -413,6 +420,33 @@ 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_RETURNS_VOID = (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, ...); @@ -580,6 +614,20 @@ private: B1(max3) B1(mid3) + 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 @@ -694,6 +742,12 @@ 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); } /** @@ -2167,6 +2221,12 @@ builtin_builder::create_builtins() _mid3(glsl_type::uvec4_type), NULL); + add_image_functions(true); + + add_function("memoryBarrier", + _memory_barrier(shader_image_load_store), + NULL); + #undef F #undef FI #undef FIU @@ -2200,6 +2260,104 @@ 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]->sampler_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) +{ + const unsigned flags = (glsl ? IMAGE_FUNCTION_EMIT_STUB : 0); + + add_image_function(glsl ? "imageLoad" : "__intrinsic_image_load", + "__intrinsic_image_load", 0, + (flags | 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, + (flags | IMAGE_FUNCTION_RETURNS_VOID | + 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, flags); + + add_image_function(glsl ? "imageAtomicMin" : "__intrinsic_image_atomic_min", + "__intrinsic_image_atomic_min", 1, flags); + + add_image_function(glsl ? "imageAtomicMax" : "__intrinsic_image_atomic_max", + "__intrinsic_image_atomic_max", 1, flags); + + add_image_function(glsl ? "imageAtomicAnd" : "__intrinsic_image_atomic_and", + "__intrinsic_image_atomic_and", 1, flags); + + add_image_function(glsl ? "imageAtomicOr" : "__intrinsic_image_atomic_or", + "__intrinsic_image_atomic_or", 1, flags); + + add_image_function(glsl ? "imageAtomicXor" : "__intrinsic_image_atomic_xor", + "__intrinsic_image_atomic_xor", 1, flags); + + add_image_function((glsl ? "imageAtomicExchange" : + "__intrinsic_image_atomic_exchange"), + "__intrinsic_image_atomic_exchange", 1, flags); + + add_image_function((glsl ? "imageAtomicCompSwap" : + "__intrinsic_image_atomic_comp_swap"), + "__intrinsic_image_atomic_comp_swap", 2, flags); +} + ir_variable * builtin_builder::in_var(const glsl_type *type, const char *name) { @@ -4094,6 +4252,99 @@ builtin_builder::_mid3(const glsl_type *type) 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->sampler_type, + (flags & IMAGE_FUNCTION_HAS_VECTOR_DATA_TYPE ? 4 : 1), + 1); + const glsl_type *ret_type = (flags & IMAGE_FUNCTION_RETURNS_VOID ? + glsl_type::void_type : data_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->sampler_dimensionality == GLSL_SAMPLER_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->data.image.read_only = flags & IMAGE_FUNCTION_READ_ONLY; + image->data.image.write_only = flags & IMAGE_FUNCTION_WRITE_ONLY; + image->data.image.coherent = true; + image->data.image._volatile = true; + image->data.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_RETURNS_VOID) { + body.emit(call(f, NULL, sig->parameters)); + } else { + 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)); + } + + 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; +} + /** @} */ /******************************************************************************/ -- 2.7.4