diff options
Diffstat (limited to 'mesalib/src/glsl')
27 files changed, 814 insertions, 358 deletions
diff --git a/mesalib/src/glsl/ast_array_index.cpp b/mesalib/src/glsl/ast_array_index.cpp index b457ec899..f7b5e8350 100644 --- a/mesalib/src/glsl/ast_array_index.cpp +++ b/mesalib/src/glsl/ast_array_index.cpp @@ -143,7 +143,7 @@ _mesa_ast_array_index_to_hir(void *mem_ctx, bound = array->type->vector_elements; } } else { - /* glsl_type::array_size() returns 0 for non-array types. This means + /* glsl_type::array_size() returns -1 for non-array types. This means * that we don't need to verify that the type is an array before * doing the bounds checking. */ @@ -165,7 +165,7 @@ _mesa_ast_array_index_to_hir(void *mem_ctx, if (array->type->is_array()) update_max_array_access(array, idx, &loc, state); } else if (const_index == NULL && array->type->is_array()) { - if (array->type->array_size() == 0) { + if (array->type->is_unsized_array()) { _mesa_glsl_error(&loc, state, "unsized array index must be constant"); } else if (array->type->fields.array->is_interface() && array->variable_referenced()->mode == ir_var_uniform) { diff --git a/mesalib/src/glsl/ast_function.cpp b/mesalib/src/glsl/ast_function.cpp index 02aad4f8f..2707522ef 100644 --- a/mesalib/src/glsl/ast_function.cpp +++ b/mesalib/src/glsl/ast_function.cpp @@ -732,21 +732,21 @@ process_array_constructor(exec_list *instructions, exec_list actual_parameters; const unsigned parameter_count = process_parameters(instructions, &actual_parameters, parameters, state); + bool is_unsized_array = constructor_type->is_unsized_array(); - if ((parameter_count == 0) - || ((constructor_type->length != 0) - && (constructor_type->length != parameter_count))) { - const unsigned min_param = (constructor_type->length == 0) - ? 1 : constructor_type->length; + if ((parameter_count == 0) || + (!is_unsized_array && (constructor_type->length != parameter_count))) { + const unsigned min_param = is_unsized_array + ? 1 : constructor_type->length; _mesa_glsl_error(loc, state, "array constructor must have %s %u " "parameter%s", - (constructor_type->length == 0) ? "at least" : "exactly", + is_unsized_array ? "at least" : "exactly", min_param, (min_param <= 1) ? "" : "s"); return ir_rvalue::error_value(ctx); } - if (constructor_type->length == 0) { + if (is_unsized_array) { constructor_type = glsl_type::get_array_instance(constructor_type->element_type(), parameter_count); diff --git a/mesalib/src/glsl/ast_to_hir.cpp b/mesalib/src/glsl/ast_to_hir.cpp index 3551a5956..f75e68ce1 100644 --- a/mesalib/src/glsl/ast_to_hir.cpp +++ b/mesalib/src/glsl/ast_to_hir.cpp @@ -675,8 +675,8 @@ shift_result_type(const struct glsl_type *type_a, */ ir_rvalue * validate_assignment(struct _mesa_glsl_parse_state *state, - const glsl_type *lhs_type, ir_rvalue *rhs, - bool is_initializer) + YYLTYPE loc, const glsl_type *lhs_type, + ir_rvalue *rhs, bool is_initializer) { /* If there is already some error in the RHS, just return it. Anything * else will lead to an avalanche of error message back to the user. @@ -689,16 +689,15 @@ validate_assignment(struct _mesa_glsl_parse_state *state, if (rhs->type == lhs_type) return rhs; - /* If the array element types are the same and the size of the LHS is zero, + /* If the array element types are the same and the LHS is unsized, * the assignment is okay for initializers embedded in variable * declarations. * * Note: Whole-array assignments are not permitted in GLSL 1.10, but this * is handled by ir_dereference::is_lvalue. */ - if (is_initializer && lhs_type->is_array() && rhs->type->is_array() - && (lhs_type->element_type() == rhs->type->element_type()) - && (lhs_type->array_size() == 0)) { + if (is_initializer && lhs_type->is_unsized_array() && rhs->type->is_array() + && (lhs_type->element_type() == rhs->type->element_type())) { return rhs; } @@ -708,6 +707,12 @@ validate_assignment(struct _mesa_glsl_parse_state *state, return rhs; } + _mesa_glsl_error(&loc, state, + "%s of type %s cannot be assigned to " + "variable of type %s", + is_initializer ? "initializer" : "value", + rhs->type->name, lhs_type->name); + return NULL; } @@ -738,10 +743,10 @@ do_assignment(exec_list *instructions, struct _mesa_glsl_parse_state *state, if (unlikely(expr->operation == ir_binop_vector_extract)) { ir_rvalue *new_rhs = - validate_assignment(state, lhs->type, rhs, is_initializer); + validate_assignment(state, lhs_loc, lhs->type, + rhs, is_initializer); if (new_rhs == NULL) { - _mesa_glsl_error(& lhs_loc, state, "type mismatch"); return lhs; } else { rhs = new(ctx) ir_expression(ir_triop_vector_insert, @@ -790,10 +795,8 @@ do_assignment(exec_list *instructions, struct _mesa_glsl_parse_state *state, } ir_rvalue *new_rhs = - validate_assignment(state, lhs->type, rhs, is_initializer); - if (new_rhs == NULL) { - _mesa_glsl_error(& lhs_loc, state, "type mismatch"); - } else { + validate_assignment(state, lhs_loc, lhs->type, rhs, is_initializer); + if (new_rhs != NULL) { rhs = new_rhs; /* If the LHS array was not declared with a size, it takes it size from @@ -801,7 +804,7 @@ do_assignment(exec_list *instructions, struct _mesa_glsl_parse_state *state, * dereference of a variable. Any other case would require that the LHS * is either not an l-value or not a whole array. */ - if (lhs->type->array_size() == 0) { + if (lhs->type->is_unsized_array()) { ir_dereference *const d = lhs->as_dereference(); assert(d != NULL); @@ -940,6 +943,7 @@ do_comparison(void *mem_ctx, int operation, ir_rvalue *op0, ir_rvalue *op1) case GLSL_TYPE_VOID: case GLSL_TYPE_SAMPLER: case GLSL_TYPE_INTERFACE: + case GLSL_TYPE_ATOMIC_UINT: /* I assume a comparison of a struct containing a sampler just * ignores the sampler present in the type. */ @@ -1234,6 +1238,10 @@ ast_expression::hir(exec_list *instructions, !state->check_version(120, 300, &loc, "array comparisons forbidden")) { error_emitted = true; + } else if ((op[0]->type->contains_opaque() || + op[1]->type->contains_opaque())) { + _mesa_glsl_error(&loc, state, "opaque type comparisons forbidden"); + error_emitted = true; } if (error_emitted) { @@ -2039,6 +2047,96 @@ interpret_interpolation_qualifier(const struct ast_type_qualifier *qual, static void +validate_explicit_location(const struct ast_type_qualifier *qual, + ir_variable *var, + struct _mesa_glsl_parse_state *state, + YYLTYPE *loc) +{ + bool fail = false; + + /* In the vertex shader only shader inputs can be given explicit + * locations. + * + * In the fragment shader only shader outputs can be given explicit + * locations. + */ + switch (state->target) { + case vertex_shader: + if (var->mode == ir_var_shader_in) { + if (!state->check_explicit_attrib_location_allowed(loc, var)) + return; + + break; + } + + fail = true; + break; + + case geometry_shader: + _mesa_glsl_error(loc, state, + "geometry shader variables cannot be given " + "explicit locations"); + return; + + case fragment_shader: + if (var->mode == ir_var_shader_out) { + if (!state->check_explicit_attrib_location_allowed(loc, var)) + return; + + break; + } + + fail = true; + break; + }; + + if (fail) { + _mesa_glsl_error(loc, state, + "%s cannot be given an explicit location in %s shader", + mode_string(var), + _mesa_glsl_shader_target_name(state->target)); + } else { + var->explicit_location = true; + + /* This bit of silliness is needed because invalid explicit locations + * are supposed to be flagged during linking. Small negative values + * biased by VERT_ATTRIB_GENERIC0 or FRAG_RESULT_DATA0 could alias + * built-in values (e.g., -16+VERT_ATTRIB_GENERIC0 = VERT_ATTRIB_POS). + * The linker needs to be able to differentiate these cases. This + * ensures that negative values stay negative. + */ + if (qual->location >= 0) { + var->location = (state->target == vertex_shader) + ? (qual->location + VERT_ATTRIB_GENERIC0) + : (qual->location + FRAG_RESULT_DATA0); + } else { + var->location = qual->location; + } + + if (qual->flags.q.explicit_index) { + /* From the GLSL 4.30 specification, section 4.4.2 (Output + * Layout Qualifiers): + * + * "It is also a compile-time error if a fragment shader + * sets a layout index to less than 0 or greater than 1." + * + * Older specifications don't mandate a behavior; we take + * this as a clarification and always generate the error. + */ + if (qual->index < 0 || qual->index > 1) { + _mesa_glsl_error(loc, state, + "explicit index may only be 0 or 1"); + } else { + var->explicit_index = true; + var->index = qual->index; + } + } + } + + return; +} + +static void apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual, ir_variable *var, struct _mesa_glsl_parse_state *state, @@ -2190,81 +2288,7 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual, } if (qual->flags.q.explicit_location) { - const bool global_scope = (state->current_function == NULL); - bool fail = false; - const char *string = ""; - - /* In the vertex shader only shader inputs can be given explicit - * locations. - * - * In the fragment shader only shader outputs can be given explicit - * locations. - */ - switch (state->target) { - case vertex_shader: - if (!global_scope || (var->mode != ir_var_shader_in)) { - fail = true; - string = "input"; - } - break; - - case geometry_shader: - _mesa_glsl_error(loc, state, - "geometry shader variables cannot be given " - "explicit locations"); - break; - - case fragment_shader: - if (!global_scope || (var->mode != ir_var_shader_out)) { - fail = true; - string = "output"; - } - break; - }; - - if (fail) { - _mesa_glsl_error(loc, state, - "only %s shader %s variables can be given an " - "explicit location", - _mesa_glsl_shader_target_name(state->target), - string); - } else { - var->explicit_location = true; - - /* This bit of silliness is needed because invalid explicit locations - * are supposed to be flagged during linking. Small negative values - * biased by VERT_ATTRIB_GENERIC0 or FRAG_RESULT_DATA0 could alias - * built-in values (e.g., -16+VERT_ATTRIB_GENERIC0 = VERT_ATTRIB_POS). - * The linker needs to be able to differentiate these cases. This - * ensures that negative values stay negative. - */ - if (qual->location >= 0) { - var->location = (state->target == vertex_shader) - ? (qual->location + VERT_ATTRIB_GENERIC0) - : (qual->location + FRAG_RESULT_DATA0); - } else { - var->location = qual->location; - } - - if (qual->flags.q.explicit_index) { - /* From the GLSL 4.30 specification, section 4.4.2 (Output - * Layout Qualifiers): - * - * "It is also a compile-time error if a fragment shader - * sets a layout index to less than 0 or greater than 1." - * - * Older specifications don't mandate a behavior; we take - * this as a clarification and always generate the error. - */ - if (qual->index < 0 || qual->index > 1) { - _mesa_glsl_error(loc, state, - "explicit index may only be 0 or 1"); - } else { - var->explicit_index = true; - var->index = qual->index; - } - } - } + validate_explicit_location(qual, var, state, loc); } else if (qual->flags.q.explicit_index) { _mesa_glsl_error(loc, state, "explicit index requires explicit location"); @@ -2403,8 +2427,7 @@ get_variable_being_redeclared(ir_variable *var, YYLTYPE loc, * later re-declare the same name as an array of the same * type and specify a size." */ - if ((earlier->type->array_size() == 0) - && var->type->is_array() + if (earlier->type->is_unsized_array() && var->type->is_array() && (var->type->element_type() == earlier->type->element_type())) { /* FINISHME: This doesn't match the qualifiers on the two * FINISHME: declarations. It's not 100% clear whether this is @@ -2547,7 +2570,8 @@ process_initializer(ir_variable *var, ast_declaration *decl, */ if (type->qualifier.flags.q.constant || type->qualifier.flags.q.uniform) { - ir_rvalue *new_rhs = validate_assignment(state, var->type, rhs, true); + ir_rvalue *new_rhs = validate_assignment(state, initializer_loc, + var->type, rhs, true); if (new_rhs != NULL) { rhs = new_rhs; @@ -2576,10 +2600,6 @@ process_initializer(ir_variable *var, ast_declaration *decl, var->constant_value = constant_value; } } else { - _mesa_glsl_error(&initializer_loc, state, - "initializer of type %s cannot be assigned to " - "variable of type %s", - rhs->type->name, var->type->name); if (var->type->is_numeric()) { /* Reduce cascading errors. */ var->constant_value = ir_constant::zero(state, var->type); @@ -2659,7 +2679,7 @@ handle_geometry_shader_input_decl(struct _mesa_glsl_parse_state *state, return; } - if (var->type->length == 0) { + if (var->type->is_unsized_array()) { /* Section 4.3.8.1 (Input Layout Qualifiers) of the GLSL 1.50 spec says: * * All geometry shader input unsized array declarations will be @@ -3308,7 +3328,7 @@ ast_declarator_list::hir(exec_list *instructions, const glsl_type *const t = (earlier == NULL) ? var->type : earlier->type; - if (t->is_array() && t->length == 0) + if (t->is_unsized_array()) /* Section 10.17 of the GLSL ES 1.00 specification states that * unsized array declarations have been removed from the language. * Arrays that are sized using an initializer are still explicitly @@ -3441,7 +3461,7 @@ ast_parameter_declarator::hir(exec_list *instructions, type = process_array_type(&loc, type, this->array_size, state); } - if (!type->is_error() && type->array_size() == 0) { + if (!type->is_error() && type->is_unsized_array()) { _mesa_glsl_error(&loc, state, "arrays passed as parameters must have " "a declared size"); type = glsl_type::error_type; @@ -3613,7 +3633,7 @@ ast_function::hir(exec_list *instructions, * "Arrays are allowed as arguments and as the return type. In both * cases, the array must be explicitly sized." */ - if (return_type->is_array() && return_type->length == 0) { + if (return_type->is_unsized_array()) { YYLTYPE loc = this->get_location(); _mesa_glsl_error(& loc, state, "function `%s' return type array must be explicitly " @@ -5099,10 +5119,8 @@ ast_gs_input_layout::hir(exec_list *instructions, /* Note: gl_PrimitiveIDIn has mode ir_var_shader_in, but it's not an * array; skip it. */ - if (!var->type->is_array()) - continue; - if (var->type->length == 0) { + if (var->type->is_unsized_array()) { if (var->max_array_access >= num_vertices) { _mesa_glsl_error(&loc, state, "this geometry shader input layout implies %u" diff --git a/mesalib/src/glsl/builtin_functions.cpp b/mesalib/src/glsl/builtin_functions.cpp index d40888d38..3fa0cb5ad 100644 --- a/mesalib/src/glsl/builtin_functions.cpp +++ b/mesalib/src/glsl/builtin_functions.cpp @@ -277,6 +277,17 @@ texture_gather(const _mesa_glsl_parse_state *state) state->ARB_gpu_shader5_enable; } +/* Only ARB_texture_gather but not GLSL 4.0 or ARB_gpu_shader5. + * used for relaxation of const offset requirements. + */ +static bool +texture_gather_only(const _mesa_glsl_parse_state *state) +{ + return !state->is_version(400, 0) && + !state->ARB_gpu_shader5_enable && + state->ARB_texture_gather_enable; +} + /* Desktop GL or OES_standard_derivatives + fragment shader only */ static bool fs_oes_derivatives(const _mesa_glsl_parse_state *state) @@ -315,6 +326,13 @@ tex3d_lod(const _mesa_glsl_parse_state *state) { return tex3d(state) && lod_exists_in_stage(state); } + +static bool +shader_atomic_counters(const _mesa_glsl_parse_state *state) +{ + return state->ARB_shader_atomic_counters_enable; +} + /** @} */ /******************************************************************************/ @@ -354,6 +372,7 @@ private: ir_variable *gl_Vertex; void create_shader(); + void create_intrinsics(); void create_builtins(); /** @@ -375,6 +394,14 @@ private: ir_expression *asin_expr(ir_variable *x); + /** + * Call function \param f with parameters specified as the linked + * list \param params of \c ir_variable objects. \param ret should + * point to the ir_variable that will hold the function return + * value, or be \c NULL if the function has void return type. + */ + ir_call *call(ir_function *f, ir_variable *ret, exec_list params); + /** Create a new function and add the given signatures. */ void add_function(const char *name, ...); @@ -495,6 +522,8 @@ private: #define TEX_PROJECT 1 #define TEX_OFFSET 2 #define TEX_COMPONENT 4 +#define TEX_OFFSET_NONCONST 8 +#define TEX_OFFSET_ARRAY 16 ir_function_signature *_texture(ir_texture_opcode opcode, builtin_available_predicate avail, @@ -534,6 +563,11 @@ private: B1(uaddCarry) B1(usubBorrow) B1(mulExtended) + + ir_function_signature *_atomic_intrinsic(builtin_available_predicate avail); + ir_function_signature *_atomic_op(const char *intrinsic, + builtin_available_predicate avail); + #undef B0 #undef B1 #undef B2 @@ -596,6 +630,7 @@ builtin_builder::initialize() mem_ctx = ralloc_context(NULL); create_shader(); + create_intrinsics(); create_builtins(); } @@ -633,6 +668,24 @@ builtin_builder::create_shader() /** @} */ /** + * Create ir_function and ir_function_signature objects for each + * intrinsic. + */ +void +builtin_builder::create_intrinsics() +{ + add_function("__intrinsic_atomic_read", + _atomic_intrinsic(shader_atomic_counters), + NULL); + add_function("__intrinsic_atomic_increment", + _atomic_intrinsic(shader_atomic_counters), + NULL); + add_function("__intrinsic_atomic_predecrement", + _atomic_intrinsic(shader_atomic_counters), + NULL); +} + +/** * Create ir_function and ir_function_signature objects for each built-in. * * Contains a list of every available built-in. @@ -1909,16 +1962,80 @@ builtin_builder::create_builtins() _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::samplerCubeArray_type, glsl_type::vec4_type, TEX_COMPONENT), _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isamplerCubeArray_type, glsl_type::vec4_type, TEX_COMPONENT), _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usamplerCubeArray_type, glsl_type::vec4_type, TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DShadow_type, glsl_type::vec2_type), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArrayShadow_type, glsl_type::vec3_type), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::samplerCubeShadow_type, glsl_type::vec3_type), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::samplerCubeArrayShadow_type, glsl_type::vec4_type), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRectShadow_type, glsl_type::vec2_type), NULL); add_function("textureGatherOffset", - _texture(ir_tg4, texture_gather, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET), - _texture(ir_tg4, texture_gather, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET), - _texture(ir_tg4, texture_gather, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET), + _texture(ir_tg4, texture_gather_only, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET), + _texture(ir_tg4, texture_gather_only, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET), + _texture(ir_tg4, texture_gather_only, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET), + + _texture(ir_tg4, texture_gather_only, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), + _texture(ir_tg4, texture_gather_only, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), + _texture(ir_tg4, texture_gather_only, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DShadow_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArrayShadow_type, glsl_type::vec3_type, TEX_OFFSET_NONCONST), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRectShadow_type, glsl_type::vec2_type, TEX_OFFSET_NONCONST), + NULL); + + add_function("textureGatherOffsets", + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2D_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY), - _texture(ir_tg4, texture_gather, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), - _texture(ir_tg4, texture_gather, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), - _texture(ir_tg4, texture_gather, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DArray_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::ivec4_type, glsl_type::isampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + _texture(ir_tg4, gpu_shader5, glsl_type::uvec4_type, glsl_type::usampler2DRect_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY | TEX_COMPONENT), + + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DShadow_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DArrayShadow_type, glsl_type::vec3_type, TEX_OFFSET_ARRAY), + _texture(ir_tg4, gpu_shader5, glsl_type::vec4_type, glsl_type::sampler2DRectShadow_type, glsl_type::vec2_type, TEX_OFFSET_ARRAY), NULL); F(dFdx) @@ -1974,6 +2091,20 @@ builtin_builder::create_builtins() _mulExtended(glsl_type::uvec3_type), _mulExtended(glsl_type::uvec4_type), NULL); + + add_function("atomicCounter", + _atomic_op("__intrinsic_atomic_read", + shader_atomic_counters), + NULL); + add_function("atomicCounterIncrement", + _atomic_op("__intrinsic_atomic_increment", + shader_atomic_counters), + NULL); + add_function("atomicCounterDecrement", + _atomic_op("__intrinsic_atomic_predecrement", + shader_atomic_counters), + NULL); + #undef F #undef FI #undef FIU @@ -1994,8 +2125,6 @@ builtin_builder::add_function(const char *name, ...) if (sig == NULL) break; - sig->is_defined = true; - if (false) { exec_list stuff; stuff.push_tail(sig); @@ -2093,7 +2222,13 @@ builtin_builder::new_sig(const glsl_type *return_type, #define MAKE_SIG(return_type, avail, ...) \ ir_function_signature *sig = \ new_sig(return_type, avail, __VA_ARGS__); \ - ir_factory body(&sig->body, mem_ctx); + ir_factory body(&sig->body, mem_ctx); \ + sig->is_defined = true; + +#define MAKE_INTRINSIC(return_type, avail, ...) \ + ir_function_signature *sig = \ + new_sig(return_type, avail, __VA_ARGS__); \ + sig->is_intrinsic = true; ir_function_signature * builtin_builder::unop(builtin_available_predicate avail, @@ -2185,6 +2320,26 @@ builtin_builder::asin_expr(ir_variable *x) mul(abs(x), imm(-0.03102955f)))))))))); } +ir_call * +builtin_builder::call(ir_function *f, ir_variable *ret, exec_list params) +{ + exec_list actual_params; + + foreach_iter(exec_list_iterator, it, params) { + ir_variable *var = ((ir_instruction *)it.get())->as_variable(); + actual_params.push_tail(var_ref(var)); + } + + ir_function_signature *sig = + f->exact_matching_signature(NULL, &actual_params); + if (!sig) + return NULL; + + ir_dereference_variable *deref = + (sig->return_type->is_void() ? NULL : var_ref(ret)); + + return new(mem_ctx) ir_call(sig, deref, &actual_params); +} ir_function_signature * builtin_builder::_asin(const glsl_type *type) @@ -3347,11 +3502,21 @@ builtin_builder::_texture(ir_texture_opcode opcode, if (flags & TEX_PROJECT) tex->projector = swizzle(P, coord_type->vector_elements - 1, 1); - /* The shadow comparitor is normally in the Z component, but a few types - * have sufficiently large coordinates that it's in W. - */ - if (sampler_type->sampler_shadow) - tex->shadow_comparitor = swizzle(P, MAX2(coord_size, SWIZZLE_Z), 1); + if (sampler_type->sampler_shadow) { + if (opcode == ir_tg4) { + /* gather has refz as a separate parameter, immediately after the + * coordinate + */ + ir_variable *refz = in_var(glsl_type::float_type, "refz"); + sig->parameters.push_tail(refz); + tex->shadow_comparitor = var_ref(refz); + } else { + /* The shadow comparitor is normally in the Z component, but a few types + * have sufficiently large coordinates that it's in W. + */ + tex->shadow_comparitor = swizzle(P, MAX2(coord_size, SWIZZLE_Z), 1); + } + } if (opcode == ir_txl) { ir_variable *lod = in_var(glsl_type::float_type, "lod"); @@ -3367,14 +3532,23 @@ builtin_builder::_texture(ir_texture_opcode opcode, tex->lod_info.grad.dPdy = var_ref(dPdy); } - if (flags & TEX_OFFSET) { + if (flags & (TEX_OFFSET | TEX_OFFSET_NONCONST)) { int offset_size = coord_size - (sampler_type->sampler_array ? 1 : 0); ir_variable *offset = - new(mem_ctx) ir_variable(glsl_type::ivec(offset_size), "offset", ir_var_const_in); + new(mem_ctx) ir_variable(glsl_type::ivec(offset_size), "offset", + (flags & TEX_OFFSET) ? ir_var_const_in : ir_var_function_in); sig->parameters.push_tail(offset); tex->offset = var_ref(offset); } + if (flags & TEX_OFFSET_ARRAY) { + ir_variable *offsets = + new(mem_ctx) ir_variable(glsl_type::get_array_instance(glsl_type::ivec2_type, 4), + "offsets", ir_var_const_in); + sig->parameters.push_tail(offsets); + tex->offset = var_ref(offsets); + } + if (opcode == ir_tg4) { if (flags & TEX_COMPONENT) { ir_variable *component = @@ -3793,6 +3967,29 @@ builtin_builder::_mulExtended(const glsl_type *type) return sig; } + +ir_function_signature * +builtin_builder::_atomic_intrinsic(builtin_available_predicate avail) +{ + ir_variable *counter = in_var(glsl_type::atomic_uint_type, "counter"); + MAKE_INTRINSIC(glsl_type::uint_type, avail, 1, counter); + return sig; +} + +ir_function_signature * +builtin_builder::_atomic_op(const char *intrinsic, + builtin_available_predicate avail) +{ + ir_variable *counter = in_var(glsl_type::atomic_uint_type, "atomic_counter"); + MAKE_SIG(glsl_type::uint_type, avail, 1, counter); + + ir_variable *retval = body.make_temp(glsl_type::uint_type, "atomic_retval"); + body.emit(call(shader->symbols->get_function(intrinsic), retval, + sig->parameters)); + body.emit(ret(retval)); + return sig; +} + /** @} */ /******************************************************************************/ diff --git a/mesalib/src/glsl/builtin_type_macros.h b/mesalib/src/glsl/builtin_type_macros.h index fec38da12..263fd83ff 100644 --- a/mesalib/src/glsl/builtin_type_macros.h +++ b/mesalib/src/glsl/builtin_type_macros.h @@ -110,6 +110,8 @@ DECL_TYPE(sampler2DRectShadow, GL_SAMPLER_2D_RECT_SHADOW, GLSL_SAMPLER DECL_TYPE(samplerExternalOES, GL_SAMPLER_EXTERNAL_OES, GLSL_SAMPLER_DIM_EXTERNAL, 0, 0, GLSL_TYPE_FLOAT) +DECL_TYPE(atomic_uint, GL_UNSIGNED_INT_ATOMIC_COUNTER, GLSL_TYPE_ATOMIC_UINT, 1, 1) + STRUCT_TYPE(gl_DepthRangeParameters) STRUCT_TYPE(gl_PointParameters) STRUCT_TYPE(gl_MaterialParameters) diff --git a/mesalib/src/glsl/builtin_types.cpp b/mesalib/src/glsl/builtin_types.cpp index 1a5e5a190..92e386057 100644 --- a/mesalib/src/glsl/builtin_types.cpp +++ b/mesalib/src/glsl/builtin_types.cpp @@ -203,6 +203,8 @@ const static struct builtin_type_versions { T(sampler2DRectShadow, 140, 999) T(struct_gl_DepthRangeParameters, 110, 100) + + T(atomic_uint, 420, 999) }; const glsl_type *const deprecated_types[] = { @@ -284,5 +286,9 @@ _mesa_glsl_initialize_types(struct _mesa_glsl_parse_state *state) if (state->OES_texture_3D_enable) { add_type(symbols, glsl_type::sampler3D_type); } + + if (state->ARB_shader_atomic_counters_enable) { + add_type(symbols, glsl_type::atomic_uint_type); + } } /** @} */ diff --git a/mesalib/src/glsl/builtin_variables.cpp b/mesalib/src/glsl/builtin_variables.cpp index 018daf67f..7a3505ace 100644 --- a/mesalib/src/glsl/builtin_variables.cpp +++ b/mesalib/src/glsl/builtin_variables.cpp @@ -638,6 +638,21 @@ builtin_variable_generator::generate_constants() */ add_const("gl_MaxTextureCoords", state->Const.MaxTextureCoords); } + + if (state->ARB_shader_atomic_counters_enable) { + add_const("gl_MaxVertexAtomicCounters", + state->Const.MaxVertexAtomicCounters); + add_const("gl_MaxGeometryAtomicCounters", + state->Const.MaxGeometryAtomicCounters); + add_const("gl_MaxFragmentAtomicCounters", + state->Const.MaxFragmentAtomicCounters); + add_const("gl_MaxCombinedAtomicCounters", + state->Const.MaxCombinedAtomicCounters); + add_const("gl_MaxAtomicCounterBindings", + state->Const.MaxAtomicBufferBindings); + add_const("gl_MaxTessControlAtomicCounters", 0); + add_const("gl_MaxTessEvaluationAtomicCounters", 0); + } } diff --git a/mesalib/src/glsl/glcpp/glcpp-parse.y b/mesalib/src/glsl/glcpp/glcpp-parse.y index 02100ab0c..86f3cd5aa 100644 --- a/mesalib/src/glsl/glcpp/glcpp-parse.y +++ b/mesalib/src/glsl/glcpp/glcpp-parse.y @@ -1254,6 +1254,9 @@ glcpp_parser_create (const struct gl_extensions *extensions, int api) if (extensions->ARB_texture_gather) add_builtin_define(parser, "GL_ARB_texture_gather", 1); + + if (extensions->ARB_shader_atomic_counters) + add_builtin_define(parser, "GL_ARB_shader_atomic_counters", 1); } } diff --git a/mesalib/src/glsl/glsl_parser.yy b/mesalib/src/glsl/glsl_parser.yy index 0a0708e95..14420f8a3 100644 --- a/mesalib/src/glsl/glsl_parser.yy +++ b/mesalib/src/glsl/glsl_parser.yy @@ -66,14 +66,8 @@ static bool match_layout_qualifier(const char *s1, const char *s2, */ if (state->es_shader) return strcmp(s1, s2); - else { -#if defined(_MSC_VER) - /* MSVC doesn't have a strcasecmp() function; instead it has _stricmp. */ - return _stricmp(s1, s2); -#else + else return strcasecmp(s1, s2); -#endif - } } %} @@ -1308,29 +1302,25 @@ layout_qualifier_id: { memset(& $$, 0, sizeof($$)); - if (state->has_explicit_attrib_location()) { - if (match_layout_qualifier("location", $1, state) == 0) { - $$.flags.q.explicit_location = 1; + if (match_layout_qualifier("location", $1, state) == 0) { + $$.flags.q.explicit_location = 1; - if ($3 >= 0) { - $$.location = $3; - } else { - _mesa_glsl_error(& @3, state, - "invalid location %d specified", $3); - YYERROR; - } + if ($3 >= 0) { + $$.location = $3; + } else { + _mesa_glsl_error(& @3, state, "invalid location %d specified", $3); + YYERROR; } + } - if (match_layout_qualifier("index", $1, state) == 0) { - $$.flags.q.explicit_index = 1; + if (match_layout_qualifier("index", $1, state) == 0) { + $$.flags.q.explicit_index = 1; - if ($3 >= 0) { - $$.index = $3; - } else { - _mesa_glsl_error(& @3, state, - "invalid index %d specified", $3); - YYERROR; - } + if ($3 >= 0) { + $$.index = $3; + } else { + _mesa_glsl_error(& @3, state, "invalid index %d specified", $3); + YYERROR; } } diff --git a/mesalib/src/glsl/glsl_parser_extras.cpp b/mesalib/src/glsl/glsl_parser_extras.cpp index be17109e1..77e8816c4 100644 --- a/mesalib/src/glsl/glsl_parser_extras.cpp +++ b/mesalib/src/glsl/glsl_parser_extras.cpp @@ -118,6 +118,12 @@ _mesa_glsl_parse_state::_mesa_glsl_parse_state(struct gl_context *_ctx, this->Const.MaxGeometryTotalOutputComponents = ctx->Const.MaxGeometryTotalOutputComponents; this->Const.MaxGeometryUniformComponents = ctx->Const.GeometryProgram.MaxUniformComponents; + this->Const.MaxVertexAtomicCounters = ctx->Const.VertexProgram.MaxAtomicCounters; + this->Const.MaxGeometryAtomicCounters = ctx->Const.GeometryProgram.MaxAtomicCounters; + this->Const.MaxFragmentAtomicCounters = ctx->Const.FragmentProgram.MaxAtomicCounters; + this->Const.MaxCombinedAtomicCounters = ctx->Const.MaxCombinedAtomicCounters; + this->Const.MaxAtomicBufferBindings = ctx->Const.MaxAtomicBufferBindings; + this->current_function = NULL; this->toplevel_ir = NULL; this->found_return = false; @@ -533,6 +539,7 @@ static const _mesa_glsl_extension _mesa_glsl_supported_extensions[] = { EXT(AMD_vertex_shader_layer, true, false, AMD_vertex_shader_layer), EXT(EXT_shader_integer_mix, true, true, EXT_shader_integer_mix), EXT(ARB_texture_gather, true, false, ARB_texture_gather), + EXT(ARB_shader_atomic_counters, true, false, ARB_shader_atomic_counters), }; #undef EXT diff --git a/mesalib/src/glsl/glsl_parser_extras.h b/mesalib/src/glsl/glsl_parser_extras.h index a67438412..f22dac355 100644 --- a/mesalib/src/glsl/glsl_parser_extras.h +++ b/mesalib/src/glsl/glsl_parser_extras.h @@ -69,6 +69,10 @@ typedef struct YYLTYPE { # define YYLTYPE_IS_DECLARED 1 # define YYLTYPE_IS_TRIVIAL 1 +extern void _mesa_glsl_error(YYLTYPE *locp, _mesa_glsl_parse_state *state, + const char *fmt, ...); + + struct _mesa_glsl_parse_state { _mesa_glsl_parse_state(struct gl_context *_ctx, GLenum target, void *mem_ctx); @@ -121,6 +125,22 @@ struct _mesa_glsl_parse_state { return check_version(130, 300, locp, "bit-wise operations are forbidden"); } + bool check_explicit_attrib_location_allowed(YYLTYPE *locp, + const ir_variable *var) + { + if (!this->has_explicit_attrib_location()) { + const char *const requirement = this->es_shader + ? "GLSL ES 300" + : "GL_ARB_explicit_attrib_location extension or GLSL 330"; + + _mesa_glsl_error(locp, this, "%s explicit location requires %s", + mode_string(var), requirement); + return false; + } + + return true; + } + bool has_explicit_attrib_location() const { return ARB_explicit_attrib_location_enable || is_version(330, 300); @@ -229,6 +249,13 @@ struct _mesa_glsl_parse_state { unsigned MaxGeometryOutputVertices; unsigned MaxGeometryTotalOutputComponents; unsigned MaxGeometryUniformComponents; + + /* ARB_shader_atomic_counters */ + unsigned MaxVertexAtomicCounters; + unsigned MaxGeometryAtomicCounters; + unsigned MaxFragmentAtomicCounters; + unsigned MaxCombinedAtomicCounters; + unsigned MaxAtomicBufferBindings; } Const; /** @@ -325,6 +352,8 @@ struct _mesa_glsl_parse_state { bool ARB_shading_language_420pack_warn; bool EXT_shader_integer_mix_enable; bool EXT_shader_integer_mix_warn; + bool ARB_shader_atomic_counters_enable; + bool ARB_shader_atomic_counters_warn; /*@}*/ /** Extensions supported by the OpenGL implementation. */ @@ -363,9 +392,6 @@ do { \ (Current).source = 0; \ } while (0) -extern void _mesa_glsl_error(YYLTYPE *locp, _mesa_glsl_parse_state *state, - const char *fmt, ...); - /** * Emit a warning to the shader log * diff --git a/mesalib/src/glsl/glsl_types.cpp b/mesalib/src/glsl/glsl_types.cpp index bc8d87f5f..f74013096 100644 --- a/mesalib/src/glsl/glsl_types.cpp +++ b/mesalib/src/glsl/glsl_types.cpp @@ -168,6 +168,24 @@ glsl_type::contains_integer() const } } +bool +glsl_type::contains_opaque() const { + switch (base_type) { + case GLSL_TYPE_SAMPLER: + case GLSL_TYPE_ATOMIC_UINT: + return true; + case GLSL_TYPE_ARRAY: + return element_type()->contains_opaque(); + case GLSL_TYPE_STRUCT: + for (unsigned int i = 0; i < length; i++) { + if (fields.structure[i].type->contains_opaque()) + return true; + } + return false; + default: + return false; + } +} gl_texture_index glsl_type::sampler_index() const @@ -601,6 +619,7 @@ glsl_type::component_slots() const return this->length * this->fields.array->component_slots(); case GLSL_TYPE_SAMPLER: + case GLSL_TYPE_ATOMIC_UINT: case GLSL_TYPE_VOID: case GLSL_TYPE_ERROR: break; @@ -889,6 +908,7 @@ glsl_type::count_attribute_slots() const return this->length * this->fields.array->count_attribute_slots(); case GLSL_TYPE_SAMPLER: + case GLSL_TYPE_ATOMIC_UINT: case GLSL_TYPE_VOID: case GLSL_TYPE_ERROR: break; diff --git a/mesalib/src/glsl/glsl_types.h b/mesalib/src/glsl/glsl_types.h index 4b5b6efb3..fdb1f3a6e 100644 --- a/mesalib/src/glsl/glsl_types.h +++ b/mesalib/src/glsl/glsl_types.h @@ -53,6 +53,7 @@ enum glsl_base_type { GLSL_TYPE_FLOAT, GLSL_TYPE_BOOL, GLSL_TYPE_SAMPLER, + GLSL_TYPE_ATOMIC_UINT, GLSL_TYPE_STRUCT, GLSL_TYPE_INTERFACE, GLSL_TYPE_ARRAY, @@ -441,6 +442,32 @@ struct glsl_type { } /** + * Return the amount of atomic counter storage required for a type. + */ + unsigned atomic_size() const + { + if (base_type == GLSL_TYPE_ATOMIC_UINT) + return ATOMIC_COUNTER_SIZE; + else if (is_array()) + return length * element_type()->atomic_size(); + else + return 0; + } + + /** + * Return whether a type contains any atomic counters. + */ + bool contains_atomic() const + { + return atomic_size(); + } + + /** + * Return whether a type contains any opaque types. + */ + bool contains_opaque() const; + + /** * Query the full type of a matrix row * * \return @@ -468,7 +495,6 @@ struct glsl_type { : error_type; } - /** * Get the type of a structure field * @@ -478,13 +504,11 @@ struct glsl_type { */ const glsl_type *field_type(const char *name) const; - /** * Get the location of a filed within a record type */ int field_index(const char *name) const; - /** * Query the number of elements in an array type * @@ -499,6 +523,14 @@ struct glsl_type { } /** + * Query whether the array size for all dimensions has been declared. + */ + bool is_unsized_array() const + { + return is_array() && length == 0; + } + + /** * Return the number of coordinate components needed for this sampler type. * * This is based purely on the sampler's dimensionality. For example, this diff --git a/mesalib/src/glsl/hir_field_selection.cpp b/mesalib/src/glsl/hir_field_selection.cpp index 08be74365..1e92c89ae 100644 --- a/mesalib/src/glsl/hir_field_selection.cpp +++ b/mesalib/src/glsl/hir_field_selection.cpp @@ -72,7 +72,7 @@ _mesa_ast_field_selection_to_hir(const ast_expression *expr, _mesa_glsl_error(&loc, state, "length method takes no arguments"); if (op->type->is_array()) { - if (op->type->array_size() == 0) + if (op->type->is_unsized_array()) _mesa_glsl_error(&loc, state, "length called on unsized array"); result = new(ctx) ir_constant(op->type->array_size()); diff --git a/mesalib/src/glsl/ir.cpp b/mesalib/src/glsl/ir.cpp index c682e3ed5..1b4973612 100644 --- a/mesalib/src/glsl/ir.cpp +++ b/mesalib/src/glsl/ir.cpp @@ -1586,7 +1586,7 @@ ir_variable::ir_variable(const struct glsl_type *type, const char *name, ir_variable_mode mode) : max_array_access(0), max_ifc_array_access(NULL), read_only(false), centroid(false), invariant(false), - mode(mode), interpolation(INTERP_QUALIFIER_NONE) + mode(mode), interpolation(INTERP_QUALIFIER_NONE), atomic() { this->ir_type = ir_type_variable; this->type = type; @@ -1647,8 +1647,8 @@ ir_variable::determine_interpolation_mode(bool flat_shade) ir_function_signature::ir_function_signature(const glsl_type *return_type, builtin_available_predicate b) - : return_type(return_type), is_defined(false), builtin_avail(b), - _function(NULL) + : return_type(return_type), is_defined(false), is_intrinsic(false), + builtin_avail(b), _function(NULL) { this->ir_type = ir_type_function_signature; this->origin = NULL; @@ -1891,3 +1891,46 @@ vertices_per_prim(GLenum prim) return 3; } } + +/** + * Generate a string describing the mode of a variable + */ +const char * +mode_string(const ir_variable *var) +{ + switch (var->mode) { + case ir_var_auto: + return (var->read_only) ? "global constant" : "global variable"; + + case ir_var_uniform: + return "uniform"; + + case ir_var_shader_in: + return "shader input"; + + case ir_var_shader_out: + return "shader output"; + + case ir_var_function_in: + case ir_var_const_in: + return "function input"; + + case ir_var_function_out: + return "function output"; + + case ir_var_function_inout: + return "function inout"; + + case ir_var_system_value: + return "shader input"; + + case ir_var_temporary: + return "compiler temporary"; + + case ir_var_mode_count: + break; + } + + assert(!"Should not get here."); + return "invalid variable"; +} diff --git a/mesalib/src/glsl/ir.h b/mesalib/src/glsl/ir.h index 8d5bec9c1..5b30fe59b 100644 --- a/mesalib/src/glsl/ir.h +++ b/mesalib/src/glsl/ir.h @@ -639,6 +639,14 @@ public: int binding; /** + * Location an atomic counter is stored at. + */ + struct { + unsigned buffer_index; + unsigned offset; + } atomic; + + /** * Built-in state that backs this uniform * * Once set at variable creation, \c state_slots must remain invariant. @@ -779,6 +787,12 @@ public: /** Whether or not this function signature is a built-in. */ bool is_builtin() const; + /** + * Whether or not this function is an intrinsic to be implemented + * by the driver. + */ + bool is_intrinsic; + /** Whether or not a built-in is available for this shader. */ bool is_builtin_available(const _mesa_glsl_parse_state *state) const; @@ -2292,6 +2306,9 @@ extern char * prototype_string(const glsl_type *return_type, const char *name, exec_list *parameters); +const char * +mode_string(const ir_variable *var); + extern "C" { #endif /* __cplusplus */ diff --git a/mesalib/src/glsl/ir_clone.cpp b/mesalib/src/glsl/ir_clone.cpp index 105f9063a..b0f173a62 100644 --- a/mesalib/src/glsl/ir_clone.cpp +++ b/mesalib/src/glsl/ir_clone.cpp @@ -57,6 +57,8 @@ ir_variable::clone(void *mem_ctx, struct hash_table *ht) const var->location = this->location; var->index = this->index; var->binding = this->binding; + var->atomic.buffer_index = this->atomic.buffer_index; + var->atomic.offset = this->atomic.offset; var->warn_extension = this->warn_extension; var->origin_upper_left = this->origin_upper_left; var->pixel_center_integer = this->pixel_center_integer; @@ -395,6 +397,7 @@ ir_constant::clone(void *mem_ctx, struct hash_table *ht) const } case GLSL_TYPE_SAMPLER: + case GLSL_TYPE_ATOMIC_UINT: case GLSL_TYPE_VOID: case GLSL_TYPE_ERROR: case GLSL_TYPE_INTERFACE: diff --git a/mesalib/src/glsl/ir_uniform.h b/mesalib/src/glsl/ir_uniform.h index 8198c4819..13faab7c0 100644 --- a/mesalib/src/glsl/ir_uniform.h +++ b/mesalib/src/glsl/ir_uniform.h @@ -166,6 +166,13 @@ struct gl_uniform_storage { bool row_major; /** @} */ + + /** + * Index within gl_shader_program::AtomicBuffers[] of the atomic + * counter buffer this uniform is stored in, or -1 if this is not + * an atomic counter. + */ + int atomic_buffer_index; }; #ifdef __cplusplus diff --git a/mesalib/src/glsl/link_functions.cpp b/mesalib/src/glsl/link_functions.cpp index fd8009998..68aa62032 100644 --- a/mesalib/src/glsl/link_functions.cpp +++ b/mesalib/src/glsl/link_functions.cpp @@ -155,14 +155,17 @@ public: linked_sig->replace_parameters(&formal_parameters); - foreach_list_const(node, &sig->body) { - const ir_instruction *const original = (ir_instruction *) node; + if (sig->is_defined) { + foreach_list_const(node, &sig->body) { + const ir_instruction *const original = (ir_instruction *) node; - ir_instruction *copy = original->clone(linked, ht); - linked_sig->body.push_tail(copy); + ir_instruction *copy = original->clone(linked, ht); + linked_sig->body.push_tail(copy); + } + + linked_sig->is_defined = true; } - linked_sig->is_defined = true; hash_table_dtor(ht); /* Patch references inside the function to things outside the function @@ -307,7 +310,8 @@ find_matching_signature(const char *name, const exec_list *actual_parameters, ir_function_signature *sig = f->matching_signature(NULL, actual_parameters); - if ((sig == NULL) || !sig->is_defined) + if ((sig == NULL) || + (!sig->is_defined && !sig->is_intrinsic)) continue; /* If this function expects to bind to a built-in function and the diff --git a/mesalib/src/glsl/link_uniform_initializers.cpp b/mesalib/src/glsl/link_uniform_initializers.cpp index 3f6671047..786aaf0b4 100644 --- a/mesalib/src/glsl/link_uniform_initializers.cpp +++ b/mesalib/src/glsl/link_uniform_initializers.cpp @@ -69,6 +69,7 @@ copy_constant_to_storage(union gl_constant_value *storage, break; case GLSL_TYPE_ARRAY: case GLSL_TYPE_STRUCT: + case GLSL_TYPE_ATOMIC_UINT: case GLSL_TYPE_INTERFACE: case GLSL_TYPE_VOID: case GLSL_TYPE_ERROR: diff --git a/mesalib/src/glsl/link_uniforms.cpp b/mesalib/src/glsl/link_uniforms.cpp index ea71b3063..0a15739c2 100644 --- a/mesalib/src/glsl/link_uniforms.cpp +++ b/mesalib/src/glsl/link_uniforms.cpp @@ -513,6 +513,7 @@ private: this->uniforms[id].num_driver_storage = 0; this->uniforms[id].driver_storage = NULL; this->uniforms[id].storage = this->values; + this->uniforms[id].atomic_buffer_index = -1; if (this->ubo_block_index != -1) { this->uniforms[id].block_index = this->ubo_block_index; diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index d8f655c39..49bb142a8 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -558,29 +558,6 @@ validate_geometry_shader_executable(struct gl_shader_program *prog, /** - * Generate a string describing the mode of a variable - */ -static const char * -mode_string(const ir_variable *var) -{ - switch (var->mode) { - case ir_var_auto: - return (var->read_only) ? "global constant" : "global variable"; - - case ir_var_uniform: return "uniform"; - case ir_var_shader_in: return "shader input"; - case ir_var_shader_out: return "shader output"; - - case ir_var_const_in: - case ir_var_temporary: - default: - assert(!"Should not get here."); - return "invalid variable"; - } -} - - -/** * Perform validation of global variables used across multiple shaders */ void @@ -1108,7 +1085,7 @@ private: */ static void fixup_type(const glsl_type **type, unsigned max_array_access) { - if ((*type)->is_array() && (*type)->length == 0) { + if ((*type)->is_unsized_array()) { *type = glsl_type::get_array_instance((*type)->fields.array, max_array_access + 1); assert(*type != NULL); @@ -1123,7 +1100,7 @@ private: { for (unsigned i = 0; i < type->length; i++) { const glsl_type *elem_type = type->fields.structure[i].type; - if (elem_type->is_array() && elem_type->length == 0) + if (elem_type->is_unsized_array()) return true; } return false; diff --git a/mesalib/src/glsl/opt_algebraic.cpp b/mesalib/src/glsl/opt_algebraic.cpp index 37b2f02c6..a07e153ae 100644 --- a/mesalib/src/glsl/opt_algebraic.cpp +++ b/mesalib/src/glsl/opt_algebraic.cpp @@ -32,8 +32,11 @@ #include "ir_visitor.h" #include "ir_rvalue_visitor.h" #include "ir_optimization.h" +#include "ir_builder.h" #include "glsl_types.h" +using namespace ir_builder; + namespace { /** @@ -194,7 +197,6 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) { ir_constant *op_const[4] = {NULL, NULL, NULL, NULL}; ir_expression *op_expr[4] = {NULL, NULL, NULL, NULL}; - ir_expression *temp; unsigned int i; assert(ir->get_num_operands() <= 4); @@ -217,12 +219,7 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) switch (op_expr[0]->operation) { case ir_unop_abs: case ir_unop_neg: - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_abs, - ir->type, - op_expr[0]->operands[0], - NULL); - return swizzle_if_required(ir, temp); + return abs(op_expr[0]->operands[0]); default: break; } @@ -233,8 +230,7 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) break; if (op_expr[0]->operation == ir_unop_neg) { - this->progress = true; - return swizzle_if_required(ir, op_expr[0]->operands[0]); + return op_expr[0]->operands[0]; } break; @@ -261,7 +257,6 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) } if (new_op != ir_unop_logic_not) { - this->progress = true; return new(mem_ctx) ir_expression(new_op, ir->type, op_expr[0]->operands[0], @@ -272,14 +267,10 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) } case ir_binop_add: - if (is_vec_zero(op_const[0])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[1]); - } - if (is_vec_zero(op_const[1])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[0]); - } + if (is_vec_zero(op_const[0])) + return ir->operands[1]; + if (is_vec_zero(op_const[1])) + return ir->operands[0]; /* Reassociate addition of constants so that we can do constant * folding. @@ -291,50 +282,25 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) break; case ir_binop_sub: - if (is_vec_zero(op_const[0])) { - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_neg, - ir->operands[1]->type, - ir->operands[1], - NULL); - return swizzle_if_required(ir, temp); - } - if (is_vec_zero(op_const[1])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[0]); - } + if (is_vec_zero(op_const[0])) + return neg(ir->operands[1]); + if (is_vec_zero(op_const[1])) + return ir->operands[0]; break; case ir_binop_mul: - if (is_vec_one(op_const[0])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[1]); - } - if (is_vec_one(op_const[1])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[0]); - } + if (is_vec_one(op_const[0])) + return ir->operands[1]; + if (is_vec_one(op_const[1])) + return ir->operands[0]; - if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1])) { - this->progress = true; + if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1])) return ir_constant::zero(ir, ir->type); - } - if (is_vec_negative_one(op_const[0])) { - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_neg, - ir->operands[1]->type, - ir->operands[1], - NULL); - return swizzle_if_required(ir, temp); - } - if (is_vec_negative_one(op_const[1])) { - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_neg, - ir->operands[0]->type, - ir->operands[0], - NULL); - return swizzle_if_required(ir, temp); - } + + if (is_vec_negative_one(op_const[0])) + return neg(ir->operands[1]); + if (is_vec_negative_one(op_const[1])) + return neg(ir->operands[0]); /* Reassociate multiplication of constants so that we can do @@ -349,26 +315,20 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) case ir_binop_div: if (is_vec_one(op_const[0]) && ir->type->base_type == GLSL_TYPE_FLOAT) { - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_rcp, + return new(mem_ctx) ir_expression(ir_unop_rcp, ir->operands[1]->type, ir->operands[1], NULL); - return swizzle_if_required(ir, temp); - } - if (is_vec_one(op_const[1])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[0]); } + if (is_vec_one(op_const[1])) + return ir->operands[0]; break; case ir_binop_dot: - if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1])) { - this->progress = true; + if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1])) return ir_constant::zero(mem_ctx, ir->type); - } + if (is_vec_basis(op_const[0])) { - this->progress = true; unsigned component = 0; for (unsigned c = 0; c < op_const[0]->type->vector_elements; c++) { if (op_const[0]->value.f[c] == 1.0) @@ -377,7 +337,6 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) return new(mem_ctx) ir_swizzle(ir->operands[1], component, 0, 0, 0, 1); } if (is_vec_basis(op_const[1])) { - this->progress = true; unsigned component = 0; for (unsigned c = 0; c < op_const[1]->type->vector_elements; c++) { if (op_const[1]->value.f[c] == 1.0) @@ -387,46 +346,52 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) } break; + case ir_binop_rshift: + case ir_binop_lshift: + /* 0 >> x == 0 */ + if (is_vec_zero(op_const[0])) + return ir->operands[0]; + /* x >> 0 == x */ + if (is_vec_zero(op_const[1])) + return ir->operands[0]; + break; + case ir_binop_logic_and: /* FINISHME: Also simplify (a && a) to (a). */ if (is_vec_one(op_const[0])) { - this->progress = true; return ir->operands[1]; } else if (is_vec_one(op_const[1])) { - this->progress = true; return ir->operands[0]; } else if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1])) { - this->progress = true; return ir_constant::zero(mem_ctx, ir->type); + } else if (op_expr[0] && op_expr[0]->operation == ir_unop_logic_not && + op_expr[1] && op_expr[1]->operation == ir_unop_logic_not) { + /* De Morgan's Law: + * (not A) and (not B) === not (A or B) + */ + return logic_not(logic_or(op_expr[0]->operands[0], + op_expr[1]->operands[0])); } break; case ir_binop_logic_xor: /* FINISHME: Also simplify (a ^^ a) to (false). */ if (is_vec_zero(op_const[0])) { - this->progress = true; return ir->operands[1]; } else if (is_vec_zero(op_const[1])) { - this->progress = true; return ir->operands[0]; } else if (is_vec_one(op_const[0])) { - this->progress = true; - return new(mem_ctx) ir_expression(ir_unop_logic_not, ir->type, - ir->operands[1], NULL); + return logic_not(ir->operands[1]); } else if (is_vec_one(op_const[1])) { - this->progress = true; - return new(mem_ctx) ir_expression(ir_unop_logic_not, ir->type, - ir->operands[0], NULL); + return logic_not(ir->operands[0]); } break; case ir_binop_logic_or: /* FINISHME: Also simplify (a || a) to (a). */ if (is_vec_zero(op_const[0])) { - this->progress = true; return ir->operands[1]; } else if (is_vec_zero(op_const[1])) { - this->progress = true; return ir->operands[0]; } else if (is_vec_one(op_const[0]) || is_vec_one(op_const[1])) { ir_constant_data data; @@ -434,16 +399,20 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) for (unsigned i = 0; i < 16; i++) data.b[i] = true; - this->progress = true; return new(mem_ctx) ir_constant(ir->type, &data); + } else if (op_expr[0] && op_expr[0]->operation == ir_unop_logic_not && + op_expr[1] && op_expr[1]->operation == ir_unop_logic_not) { + /* De Morgan's Law: + * (not A) or (not B) === not (A and B) + */ + return logic_not(logic_and(op_expr[0]->operands[0], + op_expr[1]->operands[0])); } break; case ir_unop_rcp: - if (op_expr[0] && op_expr[0]->operation == ir_unop_rcp) { - this->progress = true; + if (op_expr[0] && op_expr[0]->operation == ir_unop_rcp) return op_expr[0]->operands[0]; - } /* FINISHME: We should do rcp(rsq(x)) -> sqrt(x) for some * backends, except that some backends will have done sqrt -> @@ -452,12 +421,7 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) /* As far as we know, all backends are OK with rsq. */ if (op_expr[0] && op_expr[0]->operation == ir_unop_sqrt) { - this->progress = true; - temp = new(mem_ctx) ir_expression(ir_unop_rsq, - op_expr[0]->operands[0]->type, - op_expr[0]->operands[0], - NULL); - return swizzle_if_required(ir, temp); + return rsq(op_expr[0]->operands[0]); } break; @@ -465,11 +429,9 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir) case ir_triop_lrp: /* Operands are (x, y, a). */ if (is_vec_zero(op_const[2])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[0]); + return ir->operands[0]; } else if (is_vec_one(op_const[2])) { - this->progress = true; - return swizzle_if_required(ir, ir->operands[1]); + return ir->operands[1]; } break; @@ -490,7 +452,17 @@ ir_algebraic_visitor::handle_rvalue(ir_rvalue **rvalue) if (!expr || expr->operation == ir_quadop_vector) return; - *rvalue = handle_expression(expr); + ir_rvalue *new_rvalue = handle_expression(expr); + if (new_rvalue == *rvalue) + return; + + /* If the expr used to be some vec OP scalar returning a vector, and the + * optimization gave us back a scalar, we still need to turn it into a + * vector. + */ + *rvalue = swizzle_if_required(expr, new_rvalue); + + this->progress = true; } bool diff --git a/mesalib/src/glsl/opt_array_splitting.cpp b/mesalib/src/glsl/opt_array_splitting.cpp index 34ac836ae..c7c5f6712 100644 --- a/mesalib/src/glsl/opt_array_splitting.cpp +++ b/mesalib/src/glsl/opt_array_splitting.cpp @@ -132,7 +132,7 @@ ir_array_reference_visitor::get_variable_entry(ir_variable *var) /* If the array hasn't been sized yet, we can't split it. After * linking, this should be resolved. */ - if (var->type->is_array() && var->type->length == 0) + if (var->type->is_unsized_array()) return NULL; foreach_iter(exec_list_iterator, iter, this->variable_list) { diff --git a/mesalib/src/glsl/opt_dead_builtin_varyings.cpp b/mesalib/src/glsl/opt_dead_builtin_varyings.cpp index 7e8cd4372..b336bc0a8 100644 --- a/mesalib/src/glsl/opt_dead_builtin_varyings.cpp +++ b/mesalib/src/glsl/opt_dead_builtin_varyings.cpp @@ -42,9 +42,11 @@ * If any texture coordinate slots can be eliminated, the gl_TexCoord array is * broken down into separate vec4 variables with locations equal to * VARYING_SLOT_TEX0 + i. + * + * The same is done for the gl_FragData fragment shader output. */ -#include "main/imports.h" /* for snprintf */ +#include "main/core.h" /* for snprintf and ARRAY_SIZE */ #include "ir.h" #include "ir_rvalue_visitor.h" #include "ir_optimization.h" @@ -60,10 +62,14 @@ namespace { class varying_info_visitor : public ir_hierarchical_visitor { public: /* "mode" can be either ir_var_shader_in or ir_var_shader_out */ - varying_info_visitor(ir_variable_mode mode) + varying_info_visitor(ir_variable_mode mode, bool find_frag_outputs = false) : lower_texcoord_array(true), texcoord_array(NULL), texcoord_usage(0), + find_frag_outputs(find_frag_outputs), + lower_fragdata_array(true), + fragdata_array(NULL), + fragdata_usage(0), color_usage(0), tfeedback_color_usage(0), fog(NULL), @@ -79,8 +85,27 @@ public: { ir_variable *var = ir->variable_referenced(); - if (var && var->mode == this->mode && - var->location == VARYING_SLOT_TEX0) { + if (!var || var->mode != this->mode) + return visit_continue; + + if (this->find_frag_outputs && var->location == FRAG_RESULT_DATA0) { + this->fragdata_array = var; + + ir_constant *index = ir->array_index->as_constant(); + if (index == NULL) { + /* This is variable indexing. */ + this->fragdata_usage |= (1 << var->type->array_size()) - 1; + this->lower_fragdata_array = false; + } + else { + this->fragdata_usage |= 1 << index->get_uint_component(0); + } + + /* Don't visit the leaves of ir_dereference_array. */ + return visit_continue_with_parent; + } + + if (!this->find_frag_outputs && var->location == VARYING_SLOT_TEX0) { this->texcoord_array = var; ir_constant *index = ir->array_index->as_constant(); @@ -105,8 +130,17 @@ public: { ir_variable *var = ir->variable_referenced(); - if (var->mode == this->mode && var->type->is_array() && - var->location == VARYING_SLOT_TEX0) { + if (var->mode != this->mode || !var->type->is_array()) + return visit_continue; + + if (this->find_frag_outputs && var->location == FRAG_RESULT_DATA0) { + /* This is a whole array dereference. */ + this->fragdata_usage |= (1 << var->type->array_size()) - 1; + this->lower_fragdata_array = false; + return visit_continue; + } + + if (!this->find_frag_outputs && var->location == VARYING_SLOT_TEX0) { /* This is a whole array dereference like "gl_TexCoord = x;", * there's probably no point in lowering that. */ @@ -121,6 +155,10 @@ public: if (var->mode != this->mode) return visit_continue; + /* Nothing to do here for fragment outputs. */ + if (this->find_frag_outputs) + return visit_continue; + /* Handle colors and fog. */ switch (var->location) { case VARYING_SLOT_COL0: @@ -185,12 +223,20 @@ public: if (!this->texcoord_array) { this->lower_texcoord_array = false; } + if (!this->fragdata_array) { + this->lower_fragdata_array = false; + } } bool lower_texcoord_array; ir_variable *texcoord_array; unsigned texcoord_usage; /* bitmask */ + bool find_frag_outputs; /* false if it's looking for varyings */ + bool lower_fragdata_array; + ir_variable *fragdata_array; + unsigned fragdata_usage; /* bitmask */ + ir_variable *color[2]; ir_variable *backcolor[2]; unsigned color_usage; /* bitmask */ @@ -222,6 +268,7 @@ public: { void *const ctx = ir; + memset(this->new_fragdata, 0, sizeof(this->new_fragdata)); memset(this->new_texcoord, 0, sizeof(this->new_texcoord)); memset(this->new_color, 0, sizeof(this->new_color)); memset(this->new_backcolor, 0, sizeof(this->new_backcolor)); @@ -236,31 +283,16 @@ public: * occurences of gl_TexCoord will be replaced with. */ if (info->lower_texcoord_array) { - for (int i = MAX_TEXTURE_COORD_UNITS-1; i >= 0; i--) { - if (info->texcoord_usage & (1 << i)) { - char name[32]; - - if (!(external_texcoord_usage & (1 << i))) { - /* This varying is unused in the next stage. Declare - * a temporary instead of an output. */ - snprintf(name, 32, "gl_%s_TexCoord%i_dummy", mode_str, i); - this->new_texcoord[i] = - new (ctx) ir_variable(glsl_type::vec4_type, name, - ir_var_temporary); - } - else { - snprintf(name, 32, "gl_%s_TexCoord%i", mode_str, i); - this->new_texcoord[i] = - new(ctx) ir_variable(glsl_type::vec4_type, name, - info->mode); - this->new_texcoord[i]->location = VARYING_SLOT_TEX0 + i; - this->new_texcoord[i]->explicit_location = true; - this->new_texcoord[i]->explicit_index = 0; - } - - ir->head->insert_before(new_texcoord[i]); - } - } + prepare_array(ir, this->new_texcoord, ARRAY_SIZE(this->new_texcoord), + VARYING_SLOT_TEX0, "TexCoord", mode_str, + info->texcoord_usage, external_texcoord_usage); + } + + /* Handle gl_FragData in the same way like gl_TexCoord. */ + if (info->lower_fragdata_array) { + prepare_array(ir, this->new_fragdata, ARRAY_SIZE(this->new_fragdata), + FRAG_RESULT_DATA0, "FragData", mode_str, + info->fragdata_usage, (1 << MAX_DRAW_BUFFERS) - 1); } /* Create dummy variables which will replace set-but-unused color and @@ -301,6 +333,41 @@ public: visit_list_elements(this, ir); } + void prepare_array(exec_list *ir, + struct ir_variable **new_var, + int max_elements, unsigned start_location, + const char *var_name, const char *mode_str, + unsigned usage, unsigned external_usage) + { + void *const ctx = ir; + + for (int i = max_elements-1; i >= 0; i--) { + if (usage & (1 << i)) { + char name[32]; + + if (!(external_usage & (1 << i))) { + /* This varying is unused in the next stage. Declare + * a temporary instead of an output. */ + snprintf(name, 32, "gl_%s_%s%i_dummy", mode_str, var_name, i); + new_var[i] = + new (ctx) ir_variable(glsl_type::vec4_type, name, + ir_var_temporary); + } + else { + snprintf(name, 32, "gl_%s_%s%i", mode_str, var_name, i); + new_var[i] = + new(ctx) ir_variable(glsl_type::vec4_type, name, + this->info->mode); + new_var[i]->location = start_location + i; + new_var[i]->explicit_location = true; + new_var[i]->explicit_index = 0; + } + + ir->head->insert_before(new_var[i]); + } + } + } + virtual ir_visitor_status visit(ir_variable *var) { /* Remove the gl_TexCoord array. */ @@ -309,6 +376,12 @@ public: var->remove(); } + /* Remove the gl_FragData array. */ + if (this->info->lower_fragdata_array && + var == this->info->fragdata_array) { + var->remove(); + } + /* Replace set-but-unused color and fog outputs with dummy variables. */ for (int i = 0; i < 2; i++) { if (var == this->info->color[i] && this->new_color[i]) { @@ -350,6 +423,19 @@ public: } } + /* Same for gl_FragData. */ + if (this->info->lower_fragdata_array) { + /* gl_FragData[i] occurence */ + ir_dereference_array *const da = (*rvalue)->as_dereference_array(); + + if (da && da->variable_referenced() == this->info->fragdata_array) { + unsigned i = da->array_index->as_constant()->get_uint_component(0); + + *rvalue = new(ctx) ir_dereference_variable(this->new_fragdata[i]); + return; + } + } + /* Replace set-but-unused color and fog outputs with dummy variables. */ ir_dereference_variable *const dv = (*rvalue)->as_dereference_variable(); if (!dv) @@ -392,6 +478,7 @@ public: private: const varying_info_visitor *info; + ir_variable *new_fragdata[MAX_DRAW_BUFFERS]; ir_variable *new_texcoord[MAX_TEXTURE_COORD_UNITS]; ir_variable *new_color[2]; ir_variable *new_backcolor[2]; @@ -408,6 +495,15 @@ lower_texcoord_array(exec_list *ir, const varying_info_visitor *info) 1 | 2, true); } +static void +lower_fragdata_array(exec_list *ir) +{ + varying_info_visitor info(ir_var_shader_out, true); + info.get(ir, 0, NULL); + + replace_varyings_visitor(ir, &info, 0, 0, 0); +} + void do_dead_builtin_varyings(struct gl_context *ctx, @@ -415,8 +511,13 @@ do_dead_builtin_varyings(struct gl_context *ctx, unsigned num_tfeedback_decls, tfeedback_decl *tfeedback_decls) { - /* This optimization has no effect with the core context and GLES2, because - * the built-in varyings we're eliminating here are not available there. + /* Lower the gl_FragData array to separate variables. */ + if (consumer && consumer->Type == GL_FRAGMENT_SHADER) { + lower_fragdata_array(consumer->ir); + } + + /* Lowering of built-in varyings has no effect with the core context and + * GLES2, because they are not available there. * * EXT_separate_shader_objects doesn't allow this optimization, * because a program object can be bound partially (e.g. only one diff --git a/mesalib/src/glsl/opt_function_inlining.cpp b/mesalib/src/glsl/opt_function_inlining.cpp index 0733d5180..f8033a095 100644 --- a/mesalib/src/glsl/opt_function_inlining.cpp +++ b/mesalib/src/glsl/opt_function_inlining.cpp @@ -35,9 +35,9 @@ #include "program/hash_table.h" static void -do_sampler_replacement(exec_list *instructions, - ir_variable *sampler, - ir_dereference *deref); +do_variable_replacement(exec_list *instructions, + ir_variable *orig, + ir_dereference *repl); namespace { @@ -123,11 +123,11 @@ ir_call::generate_inline(ir_instruction *next_ir) ir_rvalue *param = (ir_rvalue *) param_iter.get(); /* Generate a new variable for the parameter. */ - if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) { - /* For samplers, we want the inlined sampler references - * referencing the passed in sampler variable, since that - * will have the location information, which an assignment of - * a sampler wouldn't. Fix it up below. + if (sig_param->type->contains_opaque()) { + /* For opaque types, we want the inlined variable references + * referencing the passed in variable, since that will have + * the location information, which an assignment of an opaque + * variable wouldn't. Fix it up below. */ parameters[i] = NULL; } else { @@ -169,8 +169,8 @@ ir_call::generate_inline(ir_instruction *next_ir) visit_tree(new_ir, replace_return_with_assignment, this->return_deref); } - /* If any samplers were passed in, replace any deref of the sampler - * with a deref of the sampler argument. + /* If any opaque types were passed in, replace any deref of the + * opaque variable with a deref of the argument. */ param_iter = this->actual_parameters.iterator(); sig_param_iter = this->callee->parameters.iterator(); @@ -178,11 +178,11 @@ ir_call::generate_inline(ir_instruction *next_ir) ir_instruction *const param = (ir_instruction *) param_iter.get(); ir_variable *sig_param = (ir_variable *) sig_param_iter.get(); - if (sig_param->type->base_type == GLSL_TYPE_SAMPLER) { + if (sig_param->type->contains_opaque()) { ir_dereference *deref = param->as_dereference(); assert(deref); - do_sampler_replacement(&new_instructions, sig_param, deref); + do_variable_replacement(&new_instructions, sig_param, deref); } param_iter.next(); sig_param_iter.next(); @@ -268,23 +268,23 @@ ir_function_inlining_visitor::visit_enter(ir_call *ir) /** - * Replaces references to the "sampler" variable with a clone of "deref." + * Replaces references to the "orig" variable with a clone of "repl." * - * From the spec, samplers can appear in the tree as function + * From the spec, opaque types can appear in the tree as function * (non-out) parameters and as the result of array indexing and * structure field selection. In our builtin implementation, they * also appear in the sampler field of an ir_tex instruction. */ -class ir_sampler_replacement_visitor : public ir_hierarchical_visitor { +class ir_variable_replacement_visitor : public ir_hierarchical_visitor { public: - ir_sampler_replacement_visitor(ir_variable *sampler, ir_dereference *deref) + ir_variable_replacement_visitor(ir_variable *orig, ir_dereference *repl) { - this->sampler = sampler; - this->deref = deref; + this->orig = orig; + this->repl = repl; } - virtual ~ir_sampler_replacement_visitor() + virtual ~ir_variable_replacement_visitor() { } @@ -296,21 +296,21 @@ public: void replace_deref(ir_dereference **deref); void replace_rvalue(ir_rvalue **rvalue); - ir_variable *sampler; - ir_dereference *deref; + ir_variable *orig; + ir_dereference *repl; }; void -ir_sampler_replacement_visitor::replace_deref(ir_dereference **deref) +ir_variable_replacement_visitor::replace_deref(ir_dereference **deref) { ir_dereference_variable *deref_var = (*deref)->as_dereference_variable(); - if (deref_var && deref_var->var == this->sampler) { - *deref = this->deref->clone(ralloc_parent(*deref), NULL); + if (deref_var && deref_var->var == this->orig) { + *deref = this->repl->clone(ralloc_parent(*deref), NULL); } } void -ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue) +ir_variable_replacement_visitor::replace_rvalue(ir_rvalue **rvalue) { if (!*rvalue) return; @@ -325,7 +325,7 @@ ir_sampler_replacement_visitor::replace_rvalue(ir_rvalue **rvalue) } ir_visitor_status -ir_sampler_replacement_visitor::visit_leave(ir_texture *ir) +ir_variable_replacement_visitor::visit_leave(ir_texture *ir) { replace_deref(&ir->sampler); @@ -333,21 +333,21 @@ ir_sampler_replacement_visitor::visit_leave(ir_texture *ir) } ir_visitor_status -ir_sampler_replacement_visitor::visit_leave(ir_dereference_array *ir) +ir_variable_replacement_visitor::visit_leave(ir_dereference_array *ir) { replace_rvalue(&ir->array); return visit_continue; } ir_visitor_status -ir_sampler_replacement_visitor::visit_leave(ir_dereference_record *ir) +ir_variable_replacement_visitor::visit_leave(ir_dereference_record *ir) { replace_rvalue(&ir->record); return visit_continue; } ir_visitor_status -ir_sampler_replacement_visitor::visit_leave(ir_call *ir) +ir_variable_replacement_visitor::visit_leave(ir_call *ir) { foreach_iter(exec_list_iterator, iter, *ir) { ir_rvalue *param = (ir_rvalue *)iter.get(); @@ -362,11 +362,11 @@ ir_sampler_replacement_visitor::visit_leave(ir_call *ir) } static void -do_sampler_replacement(exec_list *instructions, - ir_variable *sampler, - ir_dereference *deref) +do_variable_replacement(exec_list *instructions, + ir_variable *orig, + ir_dereference *repl) { - ir_sampler_replacement_visitor v(sampler, deref); + ir_variable_replacement_visitor v(orig, repl); visit_list_elements(&v, instructions); } diff --git a/mesalib/src/glsl/ralloc.h b/mesalib/src/glsl/ralloc.h index 31682d515..4581a7a4e 100644 --- a/mesalib/src/glsl/ralloc.h +++ b/mesalib/src/glsl/ralloc.h @@ -415,15 +415,29 @@ bool ralloc_vasprintf_append(char **str, const char *fmt, va_list args); * which is more idiomatic in C++ than calling ralloc. */ #define DECLARE_RALLOC_CXX_OPERATORS(TYPE) \ +private: \ + static void _ralloc_destructor(void *p) \ + { \ + reinterpret_cast<TYPE *>(p)->~TYPE(); \ + } \ +public: \ static void* operator new(size_t size, void *mem_ctx) \ { \ void *p = ralloc_size(mem_ctx, size); \ assert(p != NULL); \ + if (!HAS_TRIVIAL_DESTRUCTOR(TYPE)) \ + ralloc_set_destructor(p, _ralloc_destructor); \ return p; \ } \ \ static void operator delete(void *p) \ { \ + /* The object's destructor is guaranteed to have already been \ + * called by the delete operator at this point -- Make sure it's \ + * not called again. \ + */ \ + if (!HAS_TRIVIAL_DESTRUCTOR(TYPE)) \ + ralloc_set_destructor(p, NULL); \ ralloc_free(p); \ } |