diff options
Diffstat (limited to 'mesalib/src/glsl')
| -rw-r--r-- | mesalib/src/glsl/.gitignore | 1 | ||||
| -rw-r--r-- | mesalib/src/glsl/Makefile.sources | 1 | ||||
| -rw-r--r-- | mesalib/src/glsl/Makefile.template | 50 | ||||
| -rw-r--r-- | mesalib/src/glsl/ast_to_hir.cpp | 99 | ||||
| -rw-r--r-- | mesalib/src/glsl/builtin_compiler/.gitignore | 1 | ||||
| -rw-r--r-- | mesalib/src/glsl/glcpp/.gitignore | 2 | ||||
| -rw-r--r-- | mesalib/src/glsl/glsl_parser.yy | 1 | ||||
| -rw-r--r-- | mesalib/src/glsl/link_varyings.cpp | 1051 | ||||
| -rw-r--r-- | mesalib/src/glsl/link_varyings.h | 186 | ||||
| -rw-r--r-- | mesalib/src/glsl/linker.cpp | 1127 | ||||
| -rw-r--r-- | mesalib/src/glsl/linker.h | 9 | ||||
| -rw-r--r-- | mesalib/src/glsl/loop_controls.cpp | 8 | ||||
| -rw-r--r-- | mesalib/src/glsl/lower_packed_varyings.cpp | 95 |
13 files changed, 1414 insertions, 1217 deletions
diff --git a/mesalib/src/glsl/.gitignore b/mesalib/src/glsl/.gitignore index 068d4874b..4eb4fef9b 100644 --- a/mesalib/src/glsl/.gitignore +++ b/mesalib/src/glsl/.gitignore @@ -5,4 +5,3 @@ glsl_parser.h glsl_parser.output builtin_function.cpp glsl_test -/Makefile diff --git a/mesalib/src/glsl/Makefile.sources b/mesalib/src/glsl/Makefile.sources index d984c5ca7..2227c5ef5 100644 --- a/mesalib/src/glsl/Makefile.sources +++ b/mesalib/src/glsl/Makefile.sources @@ -46,6 +46,7 @@ LIBGLSL_FILES = \ $(GLSL_SRCDIR)/link_functions.cpp \ $(GLSL_SRCDIR)/link_uniforms.cpp \ $(GLSL_SRCDIR)/link_uniform_initializers.cpp \ + $(GLSL_SRCDIR)/link_varyings.cpp \ $(GLSL_SRCDIR)/loop_analysis.cpp \ $(GLSL_SRCDIR)/loop_controls.cpp \ $(GLSL_SRCDIR)/loop_unroll.cpp \ diff --git a/mesalib/src/glsl/Makefile.template b/mesalib/src/glsl/Makefile.template deleted file mode 100644 index 974987a0a..000000000 --- a/mesalib/src/glsl/Makefile.template +++ /dev/null @@ -1,50 +0,0 @@ -# src/glsl/Makefile.template - -# Template makefile for glsl libraries. -# -# Usage: -# The minimum that the including makefile needs to define -# is TOP, LIBNAME and one of of the *_SOURCES. -# -# Optional defines: -# LIBRARY_INCLUDES are appended to the list of includes directories. -# LIBRARY_DEFINES is not used for makedepend, but for compilation. - - -### Basic defines ### - -OBJECTS = $(C_SOURCES:.c=.o) - -INCLUDES = \ - -I. \ - $(LIBRARY_INCLUDES) - - -##### TARGETS ##### - -default: depend lib$(LIBNAME).a - -lib$(LIBNAME).a: $(OBJECTS) Makefile $(TOP)/src/glsl/Makefile.template - $(MKLIB) -o $(LIBNAME) -static $(OBJECTS) - -depend: $(C_SOURCES) - rm -f depend - touch depend - $(MKDEP) $(MKDEP_OPTIONS) $(INCLUDES) $(C_SOURCES) 2> /dev/null - -# Remove .o and backup files -clean: - rm -f $(OBJECTS) lib$(LIBNAME).a depend depend.bak - -# Dummy target -install: - @echo -n "" - - -##### RULES ##### - -.c.o: - $(CC) -c $(INCLUDES) $(CFLAGS) $(LIBRARY_DEFINES) $< -o $@ - --include depend - diff --git a/mesalib/src/glsl/ast_to_hir.cpp b/mesalib/src/glsl/ast_to_hir.cpp index 94b63f682..f934c8e2d 100644 --- a/mesalib/src/glsl/ast_to_hir.cpp +++ b/mesalib/src/glsl/ast_to_hir.cpp @@ -1910,12 +1910,36 @@ ast_type_specifier::glsl_type(const char **name, } +/** + * Determine whether a toplevel variable declaration declares a varying. This + * function operates by examining the variable's mode and the shader target, + * so it correctly identifies linkage variables regardless of whether they are + * declared using the deprecated "varying" syntax or the new "in/out" syntax. + * + * Passing a non-toplevel variable declaration (e.g. a function parameter) to + * this function will produce undefined results. + */ +static bool +is_varying_var(ir_variable *var, _mesa_glsl_parser_targets target) +{ + switch (target) { + case vertex_shader: + return var->mode == ir_var_out; + case fragment_shader: + return var->mode == ir_var_in; + default: + return var->mode == ir_var_out || var->mode == ir_var_in; + } +} + + static void apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual, ir_variable *var, struct _mesa_glsl_parse_state *state, YYLTYPE *loc, - bool ubo_qualifiers_valid) + bool ubo_qualifiers_valid, + bool is_parameter) { if (qual->flags.q.invariant) { if (var->used) { @@ -1944,27 +1968,6 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual, _mesa_glsl_shader_target_name(state->target)); } - /* From page 25 (page 31 of the PDF) of the GLSL 1.10 spec: - * - * "The varying qualifier can be used only with the data types - * float, vec2, vec3, vec4, mat2, mat3, and mat4, or arrays of - * these." - */ - if (qual->flags.q.varying) { - const glsl_type *non_array_type; - - if (var->type && var->type->is_array()) - non_array_type = var->type->fields.array; - else - non_array_type = var->type; - - if (non_array_type && non_array_type->base_type != GLSL_TYPE_FLOAT) { - var->type = glsl_type::error_type; - _mesa_glsl_error(loc, state, - "varying variables must be of base type float"); - } - } - /* If there is no qualifier that changes the mode of the variable, leave * the setting alone. */ @@ -1979,6 +1982,54 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual, else if (qual->flags.q.uniform) var->mode = ir_var_uniform; + if (!is_parameter && is_varying_var(var, state->target)) { + /* This variable is being used to link data between shader stages (in + * pre-glsl-1.30 parlance, it's a "varying"). Check that it has a type + * that is allowed for such purposes. + * + * From page 25 (page 31 of the PDF) of the GLSL 1.10 spec: + * + * "The varying qualifier can be used only with the data types + * float, vec2, vec3, vec4, mat2, mat3, and mat4, or arrays of + * these." + * + * This was relaxed in GLSL version 1.30 and GLSL ES version 3.00. From + * page 31 (page 37 of the PDF) of the GLSL 1.30 spec: + * + * "Fragment inputs can only be signed and unsigned integers and + * integer vectors, float, floating-point vectors, matrices, or + * arrays of these. Structures cannot be input. + * + * Similar text exists in the section on vertex shader outputs. + * + * Similar text exists in the GLSL ES 3.00 spec, except that the GLSL ES + * 3.00 spec claims to allow structs as well. However, this is likely + * an error, since section 11 of the spec ("Counting of Inputs and + * Outputs") enumerates all possible types of interstage linkage + * variables, and it does not mention structs. + */ + switch (var->type->get_scalar_type()->base_type) { + case GLSL_TYPE_FLOAT: + /* Ok in all GLSL versions */ + break; + case GLSL_TYPE_UINT: + case GLSL_TYPE_INT: + if (state->is_version(130, 300)) + break; + _mesa_glsl_error(loc, state, + "varying variables must be of base type float in %s", + state->get_version_string()); + break; + case GLSL_TYPE_STRUCT: + _mesa_glsl_error(loc, state, + "varying variables may not be of type struct"); + break; + default: + _mesa_glsl_error(loc, state, "illegal type for a varying variable"); + break; + } + } + if (state->all_invariant && (state->current_function == NULL)) { switch (state->target) { case vertex_shader: @@ -2626,7 +2677,7 @@ ast_declarator_list::hir(exec_list *instructions, } apply_type_qualifier_to_variable(& this->type->qualifier, var, state, - & loc, this->ubo_qualifiers_valid); + & loc, this->ubo_qualifiers_valid, false); if (this->type->qualifier.flags.q.invariant) { if ((state->target == vertex_shader) && !(var->mode == ir_var_out || @@ -3065,7 +3116,7 @@ ast_parameter_declarator::hir(exec_list *instructions, * for function parameters the default mode is 'in'. */ apply_type_qualifier_to_variable(& this->type->qualifier, var, state, & loc, - false); + false, true); /* From page 17 (page 23 of the PDF) of the GLSL 1.20 spec: * diff --git a/mesalib/src/glsl/builtin_compiler/.gitignore b/mesalib/src/glsl/builtin_compiler/.gitignore index 40c551baa..ea7d5de9c 100644 --- a/mesalib/src/glsl/builtin_compiler/.gitignore +++ b/mesalib/src/glsl/builtin_compiler/.gitignore @@ -3,4 +3,3 @@ glcpp-lex.c glcpp-parse.c glcpp-parse.h glcpp-parse.output -/Makefile diff --git a/mesalib/src/glsl/glcpp/.gitignore b/mesalib/src/glsl/glcpp/.gitignore index 64033e2aa..01d2888e7 100644 --- a/mesalib/src/glsl/glcpp/.gitignore +++ b/mesalib/src/glsl/glcpp/.gitignore @@ -3,5 +3,3 @@ glcpp-lex.c glcpp-parse.output glcpp-parse.c glcpp-parse.h -tests/*.out -/Makefile diff --git a/mesalib/src/glsl/glsl_parser.yy b/mesalib/src/glsl/glsl_parser.yy index d8494667b..88aae64d4 100644 --- a/mesalib/src/glsl/glsl_parser.yy +++ b/mesalib/src/glsl/glsl_parser.yy @@ -1542,7 +1542,6 @@ struct_declarator: void *ctx = state; $$ = new(ctx) ast_declaration($1, false, NULL, NULL); $$->set_location(yylloc); - state->symbols->add_variable(new(state) ir_variable(NULL, $1, ir_var_auto)); } | any_identifier '[' constant_expression ']' { diff --git a/mesalib/src/glsl/link_varyings.cpp b/mesalib/src/glsl/link_varyings.cpp new file mode 100644 index 000000000..5c27f231e --- /dev/null +++ b/mesalib/src/glsl/link_varyings.cpp @@ -0,0 +1,1051 @@ +/* + * Copyright © 2012 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +/** + * \file link_varyings.cpp + * + * Linker functions related specifically to linking varyings between shader + * stages. + */ + + +#include "main/mtypes.h" +#include "glsl_symbol_table.h" +#include "ir_optimization.h" +#include "linker.h" +#include "link_varyings.h" +#include "main/macros.h" + + +/** + * Validate that outputs from one stage match inputs of another + */ +bool +cross_validate_outputs_to_inputs(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer) +{ + glsl_symbol_table parameters; + /* FINISHME: Figure these out dynamically. */ + const char *const producer_stage = "vertex"; + const char *const consumer_stage = "fragment"; + + /* Find all shader outputs in the "producer" stage. + */ + foreach_list(node, producer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((var == NULL) || (var->mode != ir_var_out)) + continue; + + parameters.add_variable(var); + } + + + /* Find all shader inputs in the "consumer" stage. Any variables that have + * matching outputs already in the symbol table must have the same type and + * qualifiers. + */ + foreach_list(node, consumer->ir) { + ir_variable *const input = ((ir_instruction *) node)->as_variable(); + + /* FINISHME: For geometry shaders, this should also look for inout + * FINISHME: variables. + */ + if ((input == NULL) || (input->mode != ir_var_in)) + continue; + + ir_variable *const output = parameters.get_variable(input->name); + if (output != NULL) { + /* Check that the types match between stages. + */ + if (input->type != output->type) { + /* There is a bit of a special case for gl_TexCoord. This + * built-in is unsized by default. Applications that variable + * access it must redeclare it with a size. There is some + * language in the GLSL spec that implies the fragment shader + * and vertex shader do not have to agree on this size. Other + * driver behave this way, and one or two applications seem to + * rely on it. + * + * Neither declaration needs to be modified here because the array + * sizes are fixed later when update_array_sizes is called. + * + * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: + * + * "Unlike user-defined varying variables, the built-in + * varying variables don't have a strict one-to-one + * correspondence between the vertex language and the + * fragment language." + */ + if (!output->type->is_array() + || (strncmp("gl_", output->name, 3) != 0)) { + linker_error(prog, + "%s shader output `%s' declared as type `%s', " + "but %s shader input declared as type `%s'\n", + producer_stage, output->name, + output->type->name, + consumer_stage, input->type->name); + return false; + } + } + + /* Check that all of the qualifiers match between stages. + */ + if (input->centroid != output->centroid) { + linker_error(prog, + "%s shader output `%s' %s centroid qualifier, " + "but %s shader input %s centroid qualifier\n", + producer_stage, + output->name, + (output->centroid) ? "has" : "lacks", + consumer_stage, + (input->centroid) ? "has" : "lacks"); + return false; + } + + if (input->invariant != output->invariant) { + linker_error(prog, + "%s shader output `%s' %s invariant qualifier, " + "but %s shader input %s invariant qualifier\n", + producer_stage, + output->name, + (output->invariant) ? "has" : "lacks", + consumer_stage, + (input->invariant) ? "has" : "lacks"); + return false; + } + + if (input->interpolation != output->interpolation) { + linker_error(prog, + "%s shader output `%s' specifies %s " + "interpolation qualifier, " + "but %s shader input specifies %s " + "interpolation qualifier\n", + producer_stage, + output->name, + output->interpolation_string(), + consumer_stage, + input->interpolation_string()); + return false; + } + } + } + + return true; +} + + +/** + * Initialize this object based on a string that was passed to + * glTransformFeedbackVaryings. If there is a parse error, the error is + * reported using linker_error(), and false is returned. + */ +bool +tfeedback_decl::init(struct gl_context *ctx, struct gl_shader_program *prog, + const void *mem_ctx, const char *input) +{ + /* We don't have to be pedantic about what is a valid GLSL variable name, + * because any variable with an invalid name can't exist in the IR anyway. + */ + + this->location = -1; + this->orig_name = input; + this->is_clip_distance_mesa = false; + this->skip_components = 0; + this->next_buffer_separator = false; + + if (ctx->Extensions.ARB_transform_feedback3) { + /* Parse gl_NextBuffer. */ + if (strcmp(input, "gl_NextBuffer") == 0) { + this->next_buffer_separator = true; + return true; + } + + /* Parse gl_SkipComponents. */ + if (strcmp(input, "gl_SkipComponents1") == 0) + this->skip_components = 1; + else if (strcmp(input, "gl_SkipComponents2") == 0) + this->skip_components = 2; + else if (strcmp(input, "gl_SkipComponents3") == 0) + this->skip_components = 3; + else if (strcmp(input, "gl_SkipComponents4") == 0) + this->skip_components = 4; + + if (this->skip_components) + return true; + } + + /* Parse a declaration. */ + const char *bracket = strrchr(input, '['); + + if (bracket) { + this->var_name = ralloc_strndup(mem_ctx, input, bracket - input); + if (sscanf(bracket, "[%u]", &this->array_subscript) != 1) { + linker_error(prog, "Cannot parse transform feedback varying %s", input); + return false; + } + this->is_subscripted = true; + } else { + this->var_name = ralloc_strdup(mem_ctx, input); + this->is_subscripted = false; + } + + /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this + * class must behave specially to account for the fact that gl_ClipDistance + * is converted from a float[8] to a vec4[2]. + */ + if (ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].LowerClipDistance && + strcmp(this->var_name, "gl_ClipDistance") == 0) { + this->is_clip_distance_mesa = true; + } + + return true; +} + + +/** + * Determine whether two tfeedback_decl objects refer to the same variable and + * array index (if applicable). + */ +bool +tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y) +{ + assert(x.is_varying() && y.is_varying()); + + if (strcmp(x.var_name, y.var_name) != 0) + return false; + if (x.is_subscripted != y.is_subscripted) + return false; + if (x.is_subscripted && x.array_subscript != y.array_subscript) + return false; + return true; +} + + +/** + * Assign a location for this tfeedback_decl object based on the location + * assignment in output_var. + * + * If an error occurs, the error is reported through linker_error() and false + * is returned. + */ +bool +tfeedback_decl::assign_location(struct gl_context *ctx, + struct gl_shader_program *prog, + ir_variable *output_var) +{ + assert(this->is_varying()); + + if (output_var->type->is_array()) { + /* Array variable */ + const unsigned matrix_cols = + output_var->type->fields.array->matrix_columns; + const unsigned vector_elements = + output_var->type->fields.array->vector_elements; + unsigned actual_array_size = this->is_clip_distance_mesa ? + prog->Vert.ClipDistanceArraySize : output_var->type->array_size(); + + if (this->is_subscripted) { + /* Check array bounds. */ + if (this->array_subscript >= actual_array_size) { + linker_error(prog, "Transform feedback varying %s has index " + "%i, but the array size is %u.", + this->orig_name, this->array_subscript, + actual_array_size); + return false; + } + if (this->is_clip_distance_mesa) { + this->location = + output_var->location + this->array_subscript / 4; + this->location_frac = this->array_subscript % 4; + } else { + unsigned fine_location + = output_var->location * 4 + output_var->location_frac; + unsigned array_elem_size = vector_elements * matrix_cols; + fine_location += array_elem_size * this->array_subscript; + this->location = fine_location / 4; + this->location_frac = fine_location % 4; + } + this->size = 1; + } else { + this->location = output_var->location; + this->location_frac = output_var->location_frac; + this->size = actual_array_size; + } + this->vector_elements = vector_elements; + this->matrix_columns = matrix_cols; + if (this->is_clip_distance_mesa) + this->type = GL_FLOAT; + else + this->type = output_var->type->fields.array->gl_type; + } else { + /* Regular variable (scalar, vector, or matrix) */ + if (this->is_subscripted) { + linker_error(prog, "Transform feedback varying %s requested, " + "but %s is not an array.", + this->orig_name, this->var_name); + return false; + } + this->location = output_var->location; + this->location_frac = output_var->location_frac; + this->size = 1; + this->vector_elements = output_var->type->vector_elements; + this->matrix_columns = output_var->type->matrix_columns; + this->type = output_var->type->gl_type; + } + + /* From GL_EXT_transform_feedback: + * A program will fail to link if: + * + * * the total number of components to capture in any varying + * variable in <varyings> is greater than the constant + * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the + * buffer mode is SEPARATE_ATTRIBS_EXT; + */ + if (prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS && + this->num_components() > + ctx->Const.MaxTransformFeedbackSeparateComponents) { + linker_error(prog, "Transform feedback varying %s exceeds " + "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.", + this->orig_name); + return false; + } + + return true; +} + + +unsigned +tfeedback_decl::get_num_outputs() const +{ + if (!this->is_varying()) { + return 0; + } + + return (this->num_components() + this->location_frac + 3)/4; +} + + +/** + * Update gl_transform_feedback_info to reflect this tfeedback_decl. + * + * If an error occurs, the error is reported through linker_error() and false + * is returned. + */ +bool +tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog, + struct gl_transform_feedback_info *info, + unsigned buffer, const unsigned max_outputs) const +{ + assert(!this->next_buffer_separator); + + /* Handle gl_SkipComponents. */ + if (this->skip_components) { + info->BufferStride[buffer] += this->skip_components; + return true; + } + + /* From GL_EXT_transform_feedback: + * A program will fail to link if: + * + * * the total number of components to capture is greater than + * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT + * and the buffer mode is INTERLEAVED_ATTRIBS_EXT. + */ + if (prog->TransformFeedback.BufferMode == GL_INTERLEAVED_ATTRIBS && + info->BufferStride[buffer] + this->num_components() > + ctx->Const.MaxTransformFeedbackInterleavedComponents) { + linker_error(prog, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS " + "limit has been exceeded."); + return false; + } + + unsigned location = this->location; + unsigned location_frac = this->location_frac; + unsigned num_components = this->num_components(); + while (num_components > 0) { + unsigned output_size = MIN2(num_components, 4 - location_frac); + assert(info->NumOutputs < max_outputs); + info->Outputs[info->NumOutputs].ComponentOffset = location_frac; + info->Outputs[info->NumOutputs].OutputRegister = location; + info->Outputs[info->NumOutputs].NumComponents = output_size; + info->Outputs[info->NumOutputs].OutputBuffer = buffer; + info->Outputs[info->NumOutputs].DstOffset = info->BufferStride[buffer]; + ++info->NumOutputs; + info->BufferStride[buffer] += output_size; + num_components -= output_size; + location++; + location_frac = 0; + } + + info->Varyings[info->NumVarying].Name = ralloc_strdup(prog, this->orig_name); + info->Varyings[info->NumVarying].Type = this->type; + info->Varyings[info->NumVarying].Size = this->size; + info->NumVarying++; + + return true; +} + + +ir_variable * +tfeedback_decl::find_output_var(gl_shader_program *prog, + gl_shader *producer) const +{ + const char *name = this->is_clip_distance_mesa + ? "gl_ClipDistanceMESA" : this->var_name; + ir_variable *var = producer->symbols->get_variable(name); + if (var && var->mode == ir_var_out) + return var; + + /* From GL_EXT_transform_feedback: + * A program will fail to link if: + * + * * any variable name specified in the <varyings> array is not + * declared as an output in the geometry shader (if present) or + * the vertex shader (if no geometry shader is present); + */ + linker_error(prog, "Transform feedback varying %s undeclared.", + this->orig_name); + return NULL; +} + + +/** + * Parse all the transform feedback declarations that were passed to + * glTransformFeedbackVaryings() and store them in tfeedback_decl objects. + * + * If an error occurs, the error is reported through linker_error() and false + * is returned. + */ +bool +parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog, + const void *mem_ctx, unsigned num_names, + char **varying_names, tfeedback_decl *decls) +{ + for (unsigned i = 0; i < num_names; ++i) { + if (!decls[i].init(ctx, prog, mem_ctx, varying_names[i])) + return false; + + if (!decls[i].is_varying()) + continue; + + /* From GL_EXT_transform_feedback: + * A program will fail to link if: + * + * * any two entries in the <varyings> array specify the same varying + * variable; + * + * We interpret this to mean "any two entries in the <varyings> array + * specify the same varying variable and array index", since transform + * feedback of arrays would be useless otherwise. + */ + for (unsigned j = 0; j < i; ++j) { + if (!decls[j].is_varying()) + continue; + + if (tfeedback_decl::is_same(decls[i], decls[j])) { + linker_error(prog, "Transform feedback varying %s specified " + "more than once.", varying_names[i]); + return false; + } + } + } + return true; +} + + +/** + * Store transform feedback location assignments into + * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls. + * + * If an error occurs, the error is reported through linker_error() and false + * is returned. + */ +bool +store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog, + unsigned num_tfeedback_decls, + tfeedback_decl *tfeedback_decls) +{ + bool separate_attribs_mode = + prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS; + + ralloc_free(prog->LinkedTransformFeedback.Varyings); + ralloc_free(prog->LinkedTransformFeedback.Outputs); + + memset(&prog->LinkedTransformFeedback, 0, + sizeof(prog->LinkedTransformFeedback)); + + prog->LinkedTransformFeedback.Varyings = + rzalloc_array(prog, + struct gl_transform_feedback_varying_info, + num_tfeedback_decls); + + unsigned num_outputs = 0; + for (unsigned i = 0; i < num_tfeedback_decls; ++i) + num_outputs += tfeedback_decls[i].get_num_outputs(); + + prog->LinkedTransformFeedback.Outputs = + rzalloc_array(prog, + struct gl_transform_feedback_output, + num_outputs); + + unsigned num_buffers = 0; + + if (separate_attribs_mode) { + /* GL_SEPARATE_ATTRIBS */ + for (unsigned i = 0; i < num_tfeedback_decls; ++i) { + if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback, + num_buffers, num_outputs)) + return false; + + num_buffers++; + } + } + else { + /* GL_INVERLEAVED_ATTRIBS */ + for (unsigned i = 0; i < num_tfeedback_decls; ++i) { + if (tfeedback_decls[i].is_next_buffer_separator()) { + num_buffers++; + continue; + } + + if (!tfeedback_decls[i].store(ctx, prog, + &prog->LinkedTransformFeedback, + num_buffers, num_outputs)) + return false; + } + num_buffers++; + } + + assert(prog->LinkedTransformFeedback.NumOutputs == num_outputs); + + prog->LinkedTransformFeedback.NumBuffers = num_buffers; + return true; +} + + +/** + * Data structure recording the relationship between outputs of one shader + * stage (the "producer") and inputs of another (the "consumer"). + */ +class varying_matches +{ +public: + varying_matches(bool disable_varying_packing); + ~varying_matches(); + void record(ir_variable *producer_var, ir_variable *consumer_var); + unsigned assign_locations(); + void store_locations(unsigned producer_base, unsigned consumer_base) const; + +private: + /** + * If true, this driver disables varying packing, so all varyings need to + * be aligned on slot boundaries, and take up a number of slots equal to + * their number of matrix columns times their array size. + */ + const bool disable_varying_packing; + + /** + * Enum representing the order in which varyings are packed within a + * packing class. + * + * Currently we pack vec4's first, then vec2's, then scalar values, then + * vec3's. This order ensures that the only vectors that are at risk of + * having to be "double parked" (split between two adjacent varying slots) + * are the vec3's. + */ + enum packing_order_enum { + PACKING_ORDER_VEC4, + PACKING_ORDER_VEC2, + PACKING_ORDER_SCALAR, + PACKING_ORDER_VEC3, + }; + + static unsigned compute_packing_class(ir_variable *var); + static packing_order_enum compute_packing_order(ir_variable *var); + static int match_comparator(const void *x_generic, const void *y_generic); + + /** + * Structure recording the relationship between a single producer output + * and a single consumer input. + */ + struct match { + /** + * Packing class for this varying, computed by compute_packing_class(). + */ + unsigned packing_class; + + /** + * Packing order for this varying, computed by compute_packing_order(). + */ + packing_order_enum packing_order; + unsigned num_components; + + /** + * The output variable in the producer stage. + */ + ir_variable *producer_var; + + /** + * The input variable in the consumer stage. + */ + ir_variable *consumer_var; + + /** + * The location which has been assigned for this varying. This is + * expressed in multiples of a float, with the first generic varying + * (i.e. the one referred to by VERT_RESULT_VAR0 or FRAG_ATTRIB_VAR0) + * represented by the value 0. + */ + unsigned generic_location; + } *matches; + + /** + * The number of elements in the \c matches array that are currently in + * use. + */ + unsigned num_matches; + + /** + * The number of elements that were set aside for the \c matches array when + * it was allocated. + */ + unsigned matches_capacity; +}; + + +varying_matches::varying_matches(bool disable_varying_packing) + : disable_varying_packing(disable_varying_packing) +{ + /* Note: this initial capacity is rather arbitrarily chosen to be large + * enough for many cases without wasting an unreasonable amount of space. + * varying_matches::record() will resize the array if there are more than + * this number of varyings. + */ + this->matches_capacity = 8; + this->matches = (match *) + malloc(sizeof(*this->matches) * this->matches_capacity); + this->num_matches = 0; +} + + +varying_matches::~varying_matches() +{ + free(this->matches); +} + + +/** + * Record the given producer/consumer variable pair in the list of variables + * that should later be assigned locations. + * + * It is permissible for \c consumer_var to be NULL (this happens if a + * variable is output by the producer and consumed by transform feedback, but + * not consumed by the consumer). + * + * If \c producer_var has already been paired up with a consumer_var, or + * producer_var is part of fixed pipeline functionality (and hence already has + * a location assigned), this function has no effect. + */ +void +varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var) +{ + if (!producer_var->is_unmatched_generic_inout) { + /* Either a location already exists for this variable (since it is part + * of fixed functionality), or it has already been recorded as part of a + * previous match. + */ + return; + } + + if (this->num_matches == this->matches_capacity) { + this->matches_capacity *= 2; + this->matches = (match *) + realloc(this->matches, + sizeof(*this->matches) * this->matches_capacity); + } + this->matches[this->num_matches].packing_class + = this->compute_packing_class(producer_var); + this->matches[this->num_matches].packing_order + = this->compute_packing_order(producer_var); + if (this->disable_varying_packing) { + unsigned slots = producer_var->type->is_array() + ? (producer_var->type->length + * producer_var->type->fields.array->matrix_columns) + : producer_var->type->matrix_columns; + this->matches[this->num_matches].num_components = 4 * slots; + } else { + this->matches[this->num_matches].num_components + = producer_var->type->component_slots(); + } + this->matches[this->num_matches].producer_var = producer_var; + this->matches[this->num_matches].consumer_var = consumer_var; + this->num_matches++; + producer_var->is_unmatched_generic_inout = 0; + if (consumer_var) + consumer_var->is_unmatched_generic_inout = 0; +} + + +/** + * Choose locations for all of the variable matches that were previously + * passed to varying_matches::record(). + */ +unsigned +varying_matches::assign_locations() +{ + /* Sort varying matches into an order that makes them easy to pack. */ + qsort(this->matches, this->num_matches, sizeof(*this->matches), + &varying_matches::match_comparator); + + unsigned generic_location = 0; + + for (unsigned i = 0; i < this->num_matches; i++) { + /* Advance to the next slot if this varying has a different packing + * class than the previous one, and we're not already on a slot + * boundary. + */ + if (i > 0 && + this->matches[i - 1].packing_class + != this->matches[i].packing_class) { + generic_location = ALIGN(generic_location, 4); + } + + this->matches[i].generic_location = generic_location; + + generic_location += this->matches[i].num_components; + } + + return (generic_location + 3) / 4; +} + + +/** + * Update the producer and consumer shaders to reflect the locations + * assignments that were made by varying_matches::assign_locations(). + */ +void +varying_matches::store_locations(unsigned producer_base, + unsigned consumer_base) const +{ + for (unsigned i = 0; i < this->num_matches; i++) { + ir_variable *producer_var = this->matches[i].producer_var; + ir_variable *consumer_var = this->matches[i].consumer_var; + unsigned generic_location = this->matches[i].generic_location; + unsigned slot = generic_location / 4; + unsigned offset = generic_location % 4; + + producer_var->location = producer_base + slot; + producer_var->location_frac = offset; + if (consumer_var) { + assert(consumer_var->location == -1); + consumer_var->location = consumer_base + slot; + consumer_var->location_frac = offset; + } + } +} + + +/** + * Compute the "packing class" of the given varying. This is an unsigned + * integer with the property that two variables in the same packing class can + * be safely backed into the same vec4. + */ +unsigned +varying_matches::compute_packing_class(ir_variable *var) +{ + /* Without help from the back-end, there is no way to pack together + * variables with different interpolation types, because + * lower_packed_varyings must choose exactly one interpolation type for + * each packed varying it creates. + * + * However, we can safely pack together floats, ints, and uints, because: + * + * - varyings of base type "int" and "uint" must use the "flat" + * interpolation type, which can only occur in GLSL 1.30 and above. + * + * - On platforms that support GLSL 1.30 and above, lower_packed_varyings + * can store flat floats as ints without losing any information (using + * the ir_unop_bitcast_* opcodes). + * + * Therefore, the packing class depends only on the interpolation type. + */ + unsigned packing_class = var->centroid ? 1 : 0; + packing_class *= 4; + packing_class += var->interpolation; + return packing_class; +} + + +/** + * Compute the "packing order" of the given varying. This is a sort key we + * use to determine when to attempt to pack the given varying relative to + * other varyings in the same packing class. + */ +varying_matches::packing_order_enum +varying_matches::compute_packing_order(ir_variable *var) +{ + const glsl_type *element_type = var->type; + + /* FINISHME: Support for "varying" records in GLSL 1.50. */ + while (element_type->base_type == GLSL_TYPE_ARRAY) { + element_type = element_type->fields.array; + } + + switch (element_type->vector_elements) { + case 1: return PACKING_ORDER_SCALAR; + case 2: return PACKING_ORDER_VEC2; + case 3: return PACKING_ORDER_VEC3; + case 4: return PACKING_ORDER_VEC4; + default: + assert(!"Unexpected value of vector_elements"); + return PACKING_ORDER_VEC4; + } +} + + +/** + * Comparison function passed to qsort() to sort varyings by packing_class and + * then by packing_order. + */ +int +varying_matches::match_comparator(const void *x_generic, const void *y_generic) +{ + const match *x = (const match *) x_generic; + const match *y = (const match *) y_generic; + + if (x->packing_class != y->packing_class) + return x->packing_class - y->packing_class; + return x->packing_order - y->packing_order; +} + + +/** + * Is the given variable a varying variable to be counted against the + * limit in ctx->Const.MaxVarying? + * This includes variables such as texcoords, colors and generic + * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord. + */ +static bool +is_varying_var(GLenum shaderType, const ir_variable *var) +{ + /* Only fragment shaders will take a varying variable as an input */ + if (shaderType == GL_FRAGMENT_SHADER && + var->mode == ir_var_in) { + switch (var->location) { + case FRAG_ATTRIB_WPOS: + case FRAG_ATTRIB_FACE: + case FRAG_ATTRIB_PNTC: + return false; + default: + return true; + } + } + return false; +} + + +/** + * Assign locations for all variables that are produced in one pipeline stage + * (the "producer") and consumed in the next stage (the "consumer"). + * + * Variables produced by the producer may also be consumed by transform + * feedback. + * + * \param num_tfeedback_decls is the number of declarations indicating + * variables that may be consumed by transform feedback. + * + * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects + * representing the result of parsing the strings passed to + * glTransformFeedbackVaryings(). assign_location() will be called for + * each of these objects that matches one of the outputs of the + * producer. + * + * When num_tfeedback_decls is nonzero, it is permissible for the consumer to + * be NULL. In this case, varying locations are assigned solely based on the + * requirements of transform feedback. + */ +bool +assign_varying_locations(struct gl_context *ctx, + void *mem_ctx, + struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer, + unsigned num_tfeedback_decls, + tfeedback_decl *tfeedback_decls) +{ + /* FINISHME: Set dynamically when geometry shader support is added. */ + const unsigned producer_base = VERT_RESULT_VAR0; + const unsigned consumer_base = FRAG_ATTRIB_VAR0; + varying_matches matches(ctx->Const.DisableVaryingPacking); + + /* Operate in a total of three passes. + * + * 1. Assign locations for any matching inputs and outputs. + * + * 2. Mark output variables in the producer that do not have locations as + * not being outputs. This lets the optimizer eliminate them. + * + * 3. Mark input variables in the consumer that do not have locations as + * not being inputs. This lets the optimizer eliminate them. + */ + + foreach_list(node, producer->ir) { + ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); + + if ((output_var == NULL) || (output_var->mode != ir_var_out)) + continue; + + ir_variable *input_var = + consumer ? consumer->symbols->get_variable(output_var->name) : NULL; + + if (input_var && input_var->mode != ir_var_in) + input_var = NULL; + + if (input_var) { + matches.record(output_var, input_var); + } + } + + for (unsigned i = 0; i < num_tfeedback_decls; ++i) { + if (!tfeedback_decls[i].is_varying()) + continue; + + ir_variable *output_var + = tfeedback_decls[i].find_output_var(prog, producer); + + if (output_var == NULL) + return false; + + if (output_var->is_unmatched_generic_inout) { + matches.record(output_var, NULL); + } + } + + const unsigned slots_used = matches.assign_locations(); + matches.store_locations(producer_base, consumer_base); + + for (unsigned i = 0; i < num_tfeedback_decls; ++i) { + if (!tfeedback_decls[i].is_varying()) + continue; + + ir_variable *output_var + = tfeedback_decls[i].find_output_var(prog, producer); + + if (!tfeedback_decls[i].assign_location(ctx, prog, output_var)) + return false; + } + + if (ctx->Const.DisableVaryingPacking) { + /* Transform feedback code assumes varyings are packed, so if the driver + * has disabled varying packing, make sure it does not support transform + * feedback. + */ + assert(!ctx->Extensions.EXT_transform_feedback); + } else { + lower_packed_varyings(mem_ctx, producer_base, slots_used, ir_var_out, + producer); + if (consumer) { + lower_packed_varyings(mem_ctx, consumer_base, slots_used, ir_var_in, + consumer); + } + } + + unsigned varying_vectors = 0; + + if (consumer) { + foreach_list(node, consumer->ir) { + ir_variable *const var = ((ir_instruction *) node)->as_variable(); + + if ((var == NULL) || (var->mode != ir_var_in)) + continue; + + if (var->is_unmatched_generic_inout) { + if (prog->Version <= 120) { + /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: + * + * Only those varying variables used (i.e. read) in + * the fragment shader executable must be written to + * by the vertex shader executable; declaring + * superfluous varying variables in a vertex shader is + * permissible. + * + * We interpret this text as meaning that the VS must + * write the variable for the FS to read it. See + * "glsl1-varying read but not written" in piglit. + */ + + linker_error(prog, "fragment shader varying %s not written " + "by vertex shader\n.", var->name); + } + + /* An 'in' variable is only really a shader input if its + * value is written by the previous stage. + */ + var->mode = ir_var_auto; + } else if (is_varying_var(consumer->Type, var)) { + /* The packing rules are used for vertex shader inputs are also + * used for fragment shader inputs. + */ + varying_vectors += count_attribute_slots(var->type); + } + } + } + + if (ctx->API == API_OPENGLES2 || prog->IsES) { + if (varying_vectors > ctx->Const.MaxVarying) { + if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { + linker_warning(prog, "shader uses too many varying vectors " + "(%u > %u), but the driver will try to optimize " + "them out; this is non-portable out-of-spec " + "behavior\n", + varying_vectors, ctx->Const.MaxVarying); + } else { + linker_error(prog, "shader uses too many varying vectors " + "(%u > %u)\n", + varying_vectors, ctx->Const.MaxVarying); + return false; + } + } + } else { + const unsigned float_components = varying_vectors * 4; + if (float_components > ctx->Const.MaxVarying * 4) { + if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { + linker_warning(prog, "shader uses too many varying components " + "(%u > %u), but the driver will try to optimize " + "them out; this is non-portable out-of-spec " + "behavior\n", + float_components, ctx->Const.MaxVarying * 4); + } else { + linker_error(prog, "shader uses too many varying components " + "(%u > %u)\n", + float_components, ctx->Const.MaxVarying * 4); + return false; + } + } + } + + return true; +} diff --git a/mesalib/src/glsl/link_varyings.h b/mesalib/src/glsl/link_varyings.h new file mode 100644 index 000000000..057e4b219 --- /dev/null +++ b/mesalib/src/glsl/link_varyings.h @@ -0,0 +1,186 @@ +/* + * Copyright © 2012 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#pragma once +#ifndef GLSL_LINK_VARYINGS_H +#define GLSL_LINK_VARYINGS_H + +/** + * \file link_varyings.h + * + * Linker functions related specifically to linking varyings between shader + * stages. + */ + + +#include "main/glheader.h" + + +struct gl_shader_program; +struct gl_shader; +class ir_variable; + + +/** + * Data structure tracking information about a transform feedback declaration + * during linking. + */ +class tfeedback_decl +{ +public: + bool init(struct gl_context *ctx, struct gl_shader_program *prog, + const void *mem_ctx, const char *input); + static bool is_same(const tfeedback_decl &x, const tfeedback_decl &y); + bool assign_location(struct gl_context *ctx, struct gl_shader_program *prog, + ir_variable *output_var); + unsigned get_num_outputs() const; + bool store(struct gl_context *ctx, struct gl_shader_program *prog, + struct gl_transform_feedback_info *info, unsigned buffer, + const unsigned max_outputs) const; + ir_variable *find_output_var(gl_shader_program *prog, + gl_shader *producer) const; + + bool is_next_buffer_separator() const + { + return this->next_buffer_separator; + } + + bool is_varying() const + { + return !this->next_buffer_separator && !this->skip_components; + } + + /** + * The total number of varying components taken up by this variable. Only + * valid if assign_location() has been called. + */ + unsigned num_components() const + { + if (this->is_clip_distance_mesa) + return this->size; + else + return this->vector_elements * this->matrix_columns * this->size; + } + +private: + /** + * The name that was supplied to glTransformFeedbackVaryings. Used for + * error reporting and glGetTransformFeedbackVarying(). + */ + const char *orig_name; + + /** + * The name of the variable, parsed from orig_name. + */ + const char *var_name; + + /** + * True if the declaration in orig_name represents an array. + */ + bool is_subscripted; + + /** + * If is_subscripted is true, the subscript that was specified in orig_name. + */ + unsigned array_subscript; + + /** + * True if the variable is gl_ClipDistance and the driver lowers + * gl_ClipDistance to gl_ClipDistanceMESA. + */ + bool is_clip_distance_mesa; + + /** + * The vertex shader output location that the linker assigned for this + * variable. -1 if a location hasn't been assigned yet. + */ + int location; + + /** + * If non-zero, then this variable may be packed along with other variables + * into a single varying slot, so this offset should be applied when + * accessing components. For example, an offset of 1 means that the x + * component of this variable is actually stored in component y of the + * location specified by \c location. + * + * Only valid if location != -1. + */ + unsigned location_frac; + + /** + * If location != -1, the number of vector elements in this variable, or 1 + * if this variable is a scalar. + */ + unsigned vector_elements; + + /** + * If location != -1, the number of matrix columns in this variable, or 1 + * if this variable is not a matrix. + */ + unsigned matrix_columns; + + /** Type of the varying returned by glGetTransformFeedbackVarying() */ + GLenum type; + + /** + * If location != -1, the size that should be returned by + * glGetTransformFeedbackVarying(). + */ + unsigned size; + + /** + * How many components to skip. If non-zero, this is + * gl_SkipComponents{1,2,3,4} from ARB_transform_feedback3. + */ + unsigned skip_components; + + /** + * Whether this is gl_NextBuffer from ARB_transform_feedback3. + */ + bool next_buffer_separator; +}; + + +bool +cross_validate_outputs_to_inputs(struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer); + +bool +parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog, + const void *mem_ctx, unsigned num_names, + char **varying_names, tfeedback_decl *decls); + +bool +store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog, + unsigned num_tfeedback_decls, + tfeedback_decl *tfeedback_decls); + +bool +assign_varying_locations(struct gl_context *ctx, + void *mem_ctx, + struct gl_shader_program *prog, + gl_shader *producer, gl_shader *consumer, + unsigned num_tfeedback_decls, + tfeedback_decl *tfeedback_decls); + +#endif /* GLSL_LINK_VARYINGS_H */ diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index be08156e0..70f3d5bed 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -70,14 +70,13 @@ #include "program.h" #include "program/hash_table.h" #include "linker.h" +#include "link_varyings.h" #include "ir_optimization.h" extern "C" { #include "main/shaderobj.h" } -#define ALIGN(value, alignment) (((value) + alignment - 1) & ~(alignment - 1)) - /** * Visitor that determines whether or not a variable is ever written. */ @@ -651,126 +650,6 @@ interstage_cross_validate_uniform_blocks(struct gl_shader_program *prog) return true; } -/** - * Validate that outputs from one stage match inputs of another - */ -bool -cross_validate_outputs_to_inputs(struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer) -{ - glsl_symbol_table parameters; - /* FINISHME: Figure these out dynamically. */ - const char *const producer_stage = "vertex"; - const char *const consumer_stage = "fragment"; - - /* Find all shader outputs in the "producer" stage. - */ - foreach_list(node, producer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((var == NULL) || (var->mode != ir_var_out)) - continue; - - parameters.add_variable(var); - } - - - /* Find all shader inputs in the "consumer" stage. Any variables that have - * matching outputs already in the symbol table must have the same type and - * qualifiers. - */ - foreach_list(node, consumer->ir) { - ir_variable *const input = ((ir_instruction *) node)->as_variable(); - - /* FINISHME: For geometry shaders, this should also look for inout - * FINISHME: variables. - */ - if ((input == NULL) || (input->mode != ir_var_in)) - continue; - - ir_variable *const output = parameters.get_variable(input->name); - if (output != NULL) { - /* Check that the types match between stages. - */ - if (input->type != output->type) { - /* There is a bit of a special case for gl_TexCoord. This - * built-in is unsized by default. Applications that variable - * access it must redeclare it with a size. There is some - * language in the GLSL spec that implies the fragment shader - * and vertex shader do not have to agree on this size. Other - * driver behave this way, and one or two applications seem to - * rely on it. - * - * Neither declaration needs to be modified here because the array - * sizes are fixed later when update_array_sizes is called. - * - * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: - * - * "Unlike user-defined varying variables, the built-in - * varying variables don't have a strict one-to-one - * correspondence between the vertex language and the - * fragment language." - */ - if (!output->type->is_array() - || (strncmp("gl_", output->name, 3) != 0)) { - linker_error(prog, - "%s shader output `%s' declared as type `%s', " - "but %s shader input declared as type `%s'\n", - producer_stage, output->name, - output->type->name, - consumer_stage, input->type->name); - return false; - } - } - - /* Check that all of the qualifiers match between stages. - */ - if (input->centroid != output->centroid) { - linker_error(prog, - "%s shader output `%s' %s centroid qualifier, " - "but %s shader input %s centroid qualifier\n", - producer_stage, - output->name, - (output->centroid) ? "has" : "lacks", - consumer_stage, - (input->centroid) ? "has" : "lacks"); - return false; - } - - if (input->invariant != output->invariant) { - linker_error(prog, - "%s shader output `%s' %s invariant qualifier, " - "but %s shader input %s invariant qualifier\n", - producer_stage, - output->name, - (output->invariant) ? "has" : "lacks", - consumer_stage, - (input->invariant) ? "has" : "lacks"); - return false; - } - - if (input->interpolation != output->interpolation) { - linker_error(prog, - "%s shader output `%s' specifies %s " - "interpolation qualifier, " - "but %s shader input specifies %s " - "interpolation qualifier\n", - producer_stage, - output->name, - output->interpolation_string(), - consumer_stage, - input->interpolation_string()); - return false; - } - } - } - - return true; -} - /** * Populates a shaders symbol table with all global declarations @@ -1528,1010 +1407,6 @@ demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) /** - * Data structure tracking information about a transform feedback declaration - * during linking. - */ -class tfeedback_decl -{ -public: - bool init(struct gl_context *ctx, struct gl_shader_program *prog, - const void *mem_ctx, const char *input); - static bool is_same(const tfeedback_decl &x, const tfeedback_decl &y); - bool assign_location(struct gl_context *ctx, struct gl_shader_program *prog, - ir_variable *output_var); - unsigned get_num_outputs() const; - bool store(struct gl_context *ctx, struct gl_shader_program *prog, - struct gl_transform_feedback_info *info, unsigned buffer, - const unsigned max_outputs) const; - ir_variable *find_output_var(gl_shader_program *prog, - gl_shader *producer) const; - - bool is_next_buffer_separator() const - { - return this->next_buffer_separator; - } - - bool is_varying() const - { - return !this->next_buffer_separator && !this->skip_components; - } - - /** - * The total number of varying components taken up by this variable. Only - * valid if assign_location() has been called. - */ - unsigned num_components() const - { - if (this->is_clip_distance_mesa) - return this->size; - else - return this->vector_elements * this->matrix_columns * this->size; - } - -private: - /** - * The name that was supplied to glTransformFeedbackVaryings. Used for - * error reporting and glGetTransformFeedbackVarying(). - */ - const char *orig_name; - - /** - * The name of the variable, parsed from orig_name. - */ - const char *var_name; - - /** - * True if the declaration in orig_name represents an array. - */ - bool is_subscripted; - - /** - * If is_subscripted is true, the subscript that was specified in orig_name. - */ - unsigned array_subscript; - - /** - * True if the variable is gl_ClipDistance and the driver lowers - * gl_ClipDistance to gl_ClipDistanceMESA. - */ - bool is_clip_distance_mesa; - - /** - * The vertex shader output location that the linker assigned for this - * variable. -1 if a location hasn't been assigned yet. - */ - int location; - - /** - * If non-zero, then this variable may be packed along with other variables - * into a single varying slot, so this offset should be applied when - * accessing components. For example, an offset of 1 means that the x - * component of this variable is actually stored in component y of the - * location specified by \c location. - * - * Only valid if location != -1. - */ - unsigned location_frac; - - /** - * If location != -1, the number of vector elements in this variable, or 1 - * if this variable is a scalar. - */ - unsigned vector_elements; - - /** - * If location != -1, the number of matrix columns in this variable, or 1 - * if this variable is not a matrix. - */ - unsigned matrix_columns; - - /** Type of the varying returned by glGetTransformFeedbackVarying() */ - GLenum type; - - /** - * If location != -1, the size that should be returned by - * glGetTransformFeedbackVarying(). - */ - unsigned size; - - /** - * How many components to skip. If non-zero, this is - * gl_SkipComponents{1,2,3,4} from ARB_transform_feedback3. - */ - unsigned skip_components; - - /** - * Whether this is gl_NextBuffer from ARB_transform_feedback3. - */ - bool next_buffer_separator; -}; - - -/** - * Initialize this object based on a string that was passed to - * glTransformFeedbackVaryings. If there is a parse error, the error is - * reported using linker_error(), and false is returned. - */ -bool -tfeedback_decl::init(struct gl_context *ctx, struct gl_shader_program *prog, - const void *mem_ctx, const char *input) -{ - /* We don't have to be pedantic about what is a valid GLSL variable name, - * because any variable with an invalid name can't exist in the IR anyway. - */ - - this->location = -1; - this->orig_name = input; - this->is_clip_distance_mesa = false; - this->skip_components = 0; - this->next_buffer_separator = false; - - if (ctx->Extensions.ARB_transform_feedback3) { - /* Parse gl_NextBuffer. */ - if (strcmp(input, "gl_NextBuffer") == 0) { - this->next_buffer_separator = true; - return true; - } - - /* Parse gl_SkipComponents. */ - if (strcmp(input, "gl_SkipComponents1") == 0) - this->skip_components = 1; - else if (strcmp(input, "gl_SkipComponents2") == 0) - this->skip_components = 2; - else if (strcmp(input, "gl_SkipComponents3") == 0) - this->skip_components = 3; - else if (strcmp(input, "gl_SkipComponents4") == 0) - this->skip_components = 4; - - if (this->skip_components) - return true; - } - - /* Parse a declaration. */ - const char *bracket = strrchr(input, '['); - - if (bracket) { - this->var_name = ralloc_strndup(mem_ctx, input, bracket - input); - if (sscanf(bracket, "[%u]", &this->array_subscript) != 1) { - linker_error(prog, "Cannot parse transform feedback varying %s", input); - return false; - } - this->is_subscripted = true; - } else { - this->var_name = ralloc_strdup(mem_ctx, input); - this->is_subscripted = false; - } - - /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this - * class must behave specially to account for the fact that gl_ClipDistance - * is converted from a float[8] to a vec4[2]. - */ - if (ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].LowerClipDistance && - strcmp(this->var_name, "gl_ClipDistance") == 0) { - this->is_clip_distance_mesa = true; - } - - return true; -} - - -/** - * Determine whether two tfeedback_decl objects refer to the same variable and - * array index (if applicable). - */ -bool -tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y) -{ - assert(x.is_varying() && y.is_varying()); - - if (strcmp(x.var_name, y.var_name) != 0) - return false; - if (x.is_subscripted != y.is_subscripted) - return false; - if (x.is_subscripted && x.array_subscript != y.array_subscript) - return false; - return true; -} - - -/** - * Assign a location for this tfeedback_decl object based on the location - * assignment in output_var. - * - * If an error occurs, the error is reported through linker_error() and false - * is returned. - */ -bool -tfeedback_decl::assign_location(struct gl_context *ctx, - struct gl_shader_program *prog, - ir_variable *output_var) -{ - assert(this->is_varying()); - - if (output_var->type->is_array()) { - /* Array variable */ - const unsigned matrix_cols = - output_var->type->fields.array->matrix_columns; - const unsigned vector_elements = - output_var->type->fields.array->vector_elements; - unsigned actual_array_size = this->is_clip_distance_mesa ? - prog->Vert.ClipDistanceArraySize : output_var->type->array_size(); - - if (this->is_subscripted) { - /* Check array bounds. */ - if (this->array_subscript >= actual_array_size) { - linker_error(prog, "Transform feedback varying %s has index " - "%i, but the array size is %u.", - this->orig_name, this->array_subscript, - actual_array_size); - return false; - } - if (this->is_clip_distance_mesa) { - this->location = - output_var->location + this->array_subscript / 4; - this->location_frac = this->array_subscript % 4; - } else { - unsigned fine_location - = output_var->location * 4 + output_var->location_frac; - unsigned array_elem_size = vector_elements * matrix_cols; - fine_location += array_elem_size * this->array_subscript; - this->location = fine_location / 4; - this->location_frac = fine_location % 4; - } - this->size = 1; - } else { - this->location = output_var->location; - this->location_frac = output_var->location_frac; - this->size = actual_array_size; - } - this->vector_elements = vector_elements; - this->matrix_columns = matrix_cols; - if (this->is_clip_distance_mesa) - this->type = GL_FLOAT; - else - this->type = output_var->type->fields.array->gl_type; - } else { - /* Regular variable (scalar, vector, or matrix) */ - if (this->is_subscripted) { - linker_error(prog, "Transform feedback varying %s requested, " - "but %s is not an array.", - this->orig_name, this->var_name); - return false; - } - this->location = output_var->location; - this->location_frac = output_var->location_frac; - this->size = 1; - this->vector_elements = output_var->type->vector_elements; - this->matrix_columns = output_var->type->matrix_columns; - this->type = output_var->type->gl_type; - } - - /* From GL_EXT_transform_feedback: - * A program will fail to link if: - * - * * the total number of components to capture in any varying - * variable in <varyings> is greater than the constant - * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the - * buffer mode is SEPARATE_ATTRIBS_EXT; - */ - if (prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS && - this->num_components() > - ctx->Const.MaxTransformFeedbackSeparateComponents) { - linker_error(prog, "Transform feedback varying %s exceeds " - "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.", - this->orig_name); - return false; - } - - return true; -} - - -unsigned -tfeedback_decl::get_num_outputs() const -{ - if (!this->is_varying()) { - return 0; - } - - return (this->num_components() + this->location_frac + 3)/4; -} - - -/** - * Update gl_transform_feedback_info to reflect this tfeedback_decl. - * - * If an error occurs, the error is reported through linker_error() and false - * is returned. - */ -bool -tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog, - struct gl_transform_feedback_info *info, - unsigned buffer, const unsigned max_outputs) const -{ - assert(!this->next_buffer_separator); - - /* Handle gl_SkipComponents. */ - if (this->skip_components) { - info->BufferStride[buffer] += this->skip_components; - return true; - } - - /* From GL_EXT_transform_feedback: - * A program will fail to link if: - * - * * the total number of components to capture is greater than - * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT - * and the buffer mode is INTERLEAVED_ATTRIBS_EXT. - */ - if (prog->TransformFeedback.BufferMode == GL_INTERLEAVED_ATTRIBS && - info->BufferStride[buffer] + this->num_components() > - ctx->Const.MaxTransformFeedbackInterleavedComponents) { - linker_error(prog, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS " - "limit has been exceeded."); - return false; - } - - unsigned location = this->location; - unsigned location_frac = this->location_frac; - unsigned num_components = this->num_components(); - while (num_components > 0) { - unsigned output_size = MIN2(num_components, 4 - location_frac); - assert(info->NumOutputs < max_outputs); - info->Outputs[info->NumOutputs].ComponentOffset = location_frac; - info->Outputs[info->NumOutputs].OutputRegister = location; - info->Outputs[info->NumOutputs].NumComponents = output_size; - info->Outputs[info->NumOutputs].OutputBuffer = buffer; - info->Outputs[info->NumOutputs].DstOffset = info->BufferStride[buffer]; - ++info->NumOutputs; - info->BufferStride[buffer] += output_size; - num_components -= output_size; - location++; - location_frac = 0; - } - - info->Varyings[info->NumVarying].Name = ralloc_strdup(prog, this->orig_name); - info->Varyings[info->NumVarying].Type = this->type; - info->Varyings[info->NumVarying].Size = this->size; - info->NumVarying++; - - return true; -} - - -ir_variable * -tfeedback_decl::find_output_var(gl_shader_program *prog, - gl_shader *producer) const -{ - const char *name = this->is_clip_distance_mesa - ? "gl_ClipDistanceMESA" : this->var_name; - ir_variable *var = producer->symbols->get_variable(name); - if (var && var->mode == ir_var_out) - return var; - - /* From GL_EXT_transform_feedback: - * A program will fail to link if: - * - * * any variable name specified in the <varyings> array is not - * declared as an output in the geometry shader (if present) or - * the vertex shader (if no geometry shader is present); - */ - linker_error(prog, "Transform feedback varying %s undeclared.", - this->orig_name); - return NULL; -} - - -/** - * Parse all the transform feedback declarations that were passed to - * glTransformFeedbackVaryings() and store them in tfeedback_decl objects. - * - * If an error occurs, the error is reported through linker_error() and false - * is returned. - */ -static bool -parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog, - const void *mem_ctx, unsigned num_names, - char **varying_names, tfeedback_decl *decls) -{ - for (unsigned i = 0; i < num_names; ++i) { - if (!decls[i].init(ctx, prog, mem_ctx, varying_names[i])) - return false; - - if (!decls[i].is_varying()) - continue; - - /* From GL_EXT_transform_feedback: - * A program will fail to link if: - * - * * any two entries in the <varyings> array specify the same varying - * variable; - * - * We interpret this to mean "any two entries in the <varyings> array - * specify the same varying variable and array index", since transform - * feedback of arrays would be useless otherwise. - */ - for (unsigned j = 0; j < i; ++j) { - if (!decls[j].is_varying()) - continue; - - if (tfeedback_decl::is_same(decls[i], decls[j])) { - linker_error(prog, "Transform feedback varying %s specified " - "more than once.", varying_names[i]); - return false; - } - } - } - return true; -} - - -/** - * Data structure recording the relationship between outputs of one shader - * stage (the "producer") and inputs of another (the "consumer"). - */ -class varying_matches -{ -public: - varying_matches(bool disable_varying_packing); - ~varying_matches(); - void record(ir_variable *producer_var, ir_variable *consumer_var); - unsigned assign_locations(); - void store_locations(unsigned producer_base, unsigned consumer_base) const; - -private: - /** - * If true, this driver disables varying packing, so all varyings need to - * be aligned on slot boundaries, and take up a number of slots equal to - * their number of matrix columns times their array size. - */ - const bool disable_varying_packing; - - /** - * Enum representing the order in which varyings are packed within a - * packing class. - * - * Currently we pack vec4's first, then vec2's, then scalar values, then - * vec3's. This order ensures that the only vectors that are at risk of - * having to be "double parked" (split between two adjacent varying slots) - * are the vec3's. - */ - enum packing_order_enum { - PACKING_ORDER_VEC4, - PACKING_ORDER_VEC2, - PACKING_ORDER_SCALAR, - PACKING_ORDER_VEC3, - }; - - static unsigned compute_packing_class(ir_variable *var); - static packing_order_enum compute_packing_order(ir_variable *var); - static int match_comparator(const void *x_generic, const void *y_generic); - - /** - * Structure recording the relationship between a single producer output - * and a single consumer input. - */ - struct match { - /** - * Packing class for this varying, computed by compute_packing_class(). - */ - unsigned packing_class; - - /** - * Packing order for this varying, computed by compute_packing_order(). - */ - packing_order_enum packing_order; - unsigned num_components; - - /** - * The output variable in the producer stage. - */ - ir_variable *producer_var; - - /** - * The input variable in the consumer stage. - */ - ir_variable *consumer_var; - - /** - * The location which has been assigned for this varying. This is - * expressed in multiples of a float, with the first generic varying - * (i.e. the one referred to by VERT_RESULT_VAR0 or FRAG_ATTRIB_VAR0) - * represented by the value 0. - */ - unsigned generic_location; - } *matches; - - /** - * The number of elements in the \c matches array that are currently in - * use. - */ - unsigned num_matches; - - /** - * The number of elements that were set aside for the \c matches array when - * it was allocated. - */ - unsigned matches_capacity; -}; - - -varying_matches::varying_matches(bool disable_varying_packing) - : disable_varying_packing(disable_varying_packing) -{ - /* Note: this initial capacity is rather arbitrarily chosen to be large - * enough for many cases without wasting an unreasonable amount of space. - * varying_matches::record() will resize the array if there are more than - * this number of varyings. - */ - this->matches_capacity = 8; - this->matches = (match *) - malloc(sizeof(*this->matches) * this->matches_capacity); - this->num_matches = 0; -} - - -varying_matches::~varying_matches() -{ - free(this->matches); -} - - -/** - * Record the given producer/consumer variable pair in the list of variables - * that should later be assigned locations. - * - * It is permissible for \c consumer_var to be NULL (this happens if a - * variable is output by the producer and consumed by transform feedback, but - * not consumed by the consumer). - * - * If \c producer_var has already been paired up with a consumer_var, or - * producer_var is part of fixed pipeline functionality (and hence already has - * a location assigned), this function has no effect. - */ -void -varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var) -{ - if (!producer_var->is_unmatched_generic_inout) { - /* Either a location already exists for this variable (since it is part - * of fixed functionality), or it has already been recorded as part of a - * previous match. - */ - return; - } - - if (this->num_matches == this->matches_capacity) { - this->matches_capacity *= 2; - this->matches = (match *) - realloc(this->matches, - sizeof(*this->matches) * this->matches_capacity); - } - this->matches[this->num_matches].packing_class - = this->compute_packing_class(producer_var); - this->matches[this->num_matches].packing_order - = this->compute_packing_order(producer_var); - if (this->disable_varying_packing) { - unsigned slots = producer_var->type->is_array() - ? (producer_var->type->length - * producer_var->type->fields.array->matrix_columns) - : producer_var->type->matrix_columns; - this->matches[this->num_matches].num_components = 4 * slots; - } else { - this->matches[this->num_matches].num_components - = producer_var->type->component_slots(); - } - this->matches[this->num_matches].producer_var = producer_var; - this->matches[this->num_matches].consumer_var = consumer_var; - this->num_matches++; - producer_var->is_unmatched_generic_inout = 0; - if (consumer_var) - consumer_var->is_unmatched_generic_inout = 0; -} - - -/** - * Choose locations for all of the variable matches that were previously - * passed to varying_matches::record(). - */ -unsigned -varying_matches::assign_locations() -{ - /* Sort varying matches into an order that makes them easy to pack. */ - qsort(this->matches, this->num_matches, sizeof(*this->matches), - &varying_matches::match_comparator); - - unsigned generic_location = 0; - - for (unsigned i = 0; i < this->num_matches; i++) { - /* Advance to the next slot if this varying has a different packing - * class than the previous one, and we're not already on a slot - * boundary. - */ - if (i > 0 && - this->matches[i - 1].packing_class - != this->matches[i].packing_class) { - generic_location = ALIGN(generic_location, 4); - } - - this->matches[i].generic_location = generic_location; - - generic_location += this->matches[i].num_components; - } - - return (generic_location + 3) / 4; -} - - -/** - * Update the producer and consumer shaders to reflect the locations - * assignments that were made by varying_matches::assign_locations(). - */ -void -varying_matches::store_locations(unsigned producer_base, - unsigned consumer_base) const -{ - for (unsigned i = 0; i < this->num_matches; i++) { - ir_variable *producer_var = this->matches[i].producer_var; - ir_variable *consumer_var = this->matches[i].consumer_var; - unsigned generic_location = this->matches[i].generic_location; - unsigned slot = generic_location / 4; - unsigned offset = generic_location % 4; - - producer_var->location = producer_base + slot; - producer_var->location_frac = offset; - if (consumer_var) { - assert(consumer_var->location == -1); - consumer_var->location = consumer_base + slot; - consumer_var->location_frac = offset; - } - } -} - - -/** - * Compute the "packing class" of the given varying. This is an unsigned - * integer with the property that two variables in the same packing class can - * be safely backed into the same vec4. - */ -unsigned -varying_matches::compute_packing_class(ir_variable *var) -{ - /* In this initial implementation we conservatively assume that variables - * can only be packed if their base type (float/int/uint/bool) matches and - * their interpolation and centroid qualifiers match. - * - * TODO: relax these restrictions when the driver back-end permits. - */ - unsigned packing_class = var->centroid ? 1 : 0; - packing_class *= 4; - packing_class += var->interpolation; - packing_class *= GLSL_TYPE_ERROR; - packing_class += var->type->get_scalar_type()->base_type; - return packing_class; -} - - -/** - * Compute the "packing order" of the given varying. This is a sort key we - * use to determine when to attempt to pack the given varying relative to - * other varyings in the same packing class. - */ -varying_matches::packing_order_enum -varying_matches::compute_packing_order(ir_variable *var) -{ - const glsl_type *element_type = var->type; - - /* FINISHME: Support for "varying" records in GLSL 1.50. */ - while (element_type->base_type == GLSL_TYPE_ARRAY) { - element_type = element_type->fields.array; - } - - switch (element_type->vector_elements) { - case 1: return PACKING_ORDER_SCALAR; - case 2: return PACKING_ORDER_VEC2; - case 3: return PACKING_ORDER_VEC3; - case 4: return PACKING_ORDER_VEC4; - default: - assert(!"Unexpected value of vector_elements"); - return PACKING_ORDER_VEC4; - } -} - - -/** - * Comparison function passed to qsort() to sort varyings by packing_class and - * then by packing_order. - */ -int -varying_matches::match_comparator(const void *x_generic, const void *y_generic) -{ - const match *x = (const match *) x_generic; - const match *y = (const match *) y_generic; - - if (x->packing_class != y->packing_class) - return x->packing_class - y->packing_class; - return x->packing_order - y->packing_order; -} - - -/** - * Is the given variable a varying variable to be counted against the - * limit in ctx->Const.MaxVarying? - * This includes variables such as texcoords, colors and generic - * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord. - */ -static bool -is_varying_var(GLenum shaderType, const ir_variable *var) -{ - /* Only fragment shaders will take a varying variable as an input */ - if (shaderType == GL_FRAGMENT_SHADER && - var->mode == ir_var_in) { - switch (var->location) { - case FRAG_ATTRIB_WPOS: - case FRAG_ATTRIB_FACE: - case FRAG_ATTRIB_PNTC: - return false; - default: - return true; - } - } - return false; -} - - -/** - * Assign locations for all variables that are produced in one pipeline stage - * (the "producer") and consumed in the next stage (the "consumer"). - * - * Variables produced by the producer may also be consumed by transform - * feedback. - * - * \param num_tfeedback_decls is the number of declarations indicating - * variables that may be consumed by transform feedback. - * - * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects - * representing the result of parsing the strings passed to - * glTransformFeedbackVaryings(). assign_location() will be called for - * each of these objects that matches one of the outputs of the - * producer. - * - * When num_tfeedback_decls is nonzero, it is permissible for the consumer to - * be NULL. In this case, varying locations are assigned solely based on the - * requirements of transform feedback. - */ -bool -assign_varying_locations(struct gl_context *ctx, - void *mem_ctx, - struct gl_shader_program *prog, - gl_shader *producer, gl_shader *consumer, - unsigned num_tfeedback_decls, - tfeedback_decl *tfeedback_decls) -{ - /* FINISHME: Set dynamically when geometry shader support is added. */ - const unsigned producer_base = VERT_RESULT_VAR0; - const unsigned consumer_base = FRAG_ATTRIB_VAR0; - varying_matches matches(ctx->Const.DisableVaryingPacking); - - /* Operate in a total of three passes. - * - * 1. Assign locations for any matching inputs and outputs. - * - * 2. Mark output variables in the producer that do not have locations as - * not being outputs. This lets the optimizer eliminate them. - * - * 3. Mark input variables in the consumer that do not have locations as - * not being inputs. This lets the optimizer eliminate them. - */ - - foreach_list(node, producer->ir) { - ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); - - if ((output_var == NULL) || (output_var->mode != ir_var_out)) - continue; - - ir_variable *input_var = - consumer ? consumer->symbols->get_variable(output_var->name) : NULL; - - if (input_var && input_var->mode != ir_var_in) - input_var = NULL; - - if (input_var) { - matches.record(output_var, input_var); - } - } - - for (unsigned i = 0; i < num_tfeedback_decls; ++i) { - if (!tfeedback_decls[i].is_varying()) - continue; - - ir_variable *output_var - = tfeedback_decls[i].find_output_var(prog, producer); - - if (output_var == NULL) - return false; - - if (output_var->is_unmatched_generic_inout) { - matches.record(output_var, NULL); - } - } - - const unsigned slots_used = matches.assign_locations(); - matches.store_locations(producer_base, consumer_base); - - for (unsigned i = 0; i < num_tfeedback_decls; ++i) { - if (!tfeedback_decls[i].is_varying()) - continue; - - ir_variable *output_var - = tfeedback_decls[i].find_output_var(prog, producer); - - if (!tfeedback_decls[i].assign_location(ctx, prog, output_var)) - return false; - } - - if (ctx->Const.DisableVaryingPacking) { - /* Transform feedback code assumes varyings are packed, so if the driver - * has disabled varying packing, make sure it does not support transform - * feedback. - */ - assert(!ctx->Extensions.EXT_transform_feedback); - } else { - lower_packed_varyings(mem_ctx, producer_base, slots_used, ir_var_out, - producer); - if (consumer) { - lower_packed_varyings(mem_ctx, consumer_base, slots_used, ir_var_in, - consumer); - } - } - - unsigned varying_vectors = 0; - - if (consumer) { - foreach_list(node, consumer->ir) { - ir_variable *const var = ((ir_instruction *) node)->as_variable(); - - if ((var == NULL) || (var->mode != ir_var_in)) - continue; - - if (var->is_unmatched_generic_inout) { - if (prog->Version <= 120) { - /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: - * - * Only those varying variables used (i.e. read) in - * the fragment shader executable must be written to - * by the vertex shader executable; declaring - * superfluous varying variables in a vertex shader is - * permissible. - * - * We interpret this text as meaning that the VS must - * write the variable for the FS to read it. See - * "glsl1-varying read but not written" in piglit. - */ - - linker_error(prog, "fragment shader varying %s not written " - "by vertex shader\n.", var->name); - } - - /* An 'in' variable is only really a shader input if its - * value is written by the previous stage. - */ - var->mode = ir_var_auto; - } else if (is_varying_var(consumer->Type, var)) { - /* The packing rules are used for vertex shader inputs are also - * used for fragment shader inputs. - */ - varying_vectors += count_attribute_slots(var->type); - } - } - } - - if (ctx->API == API_OPENGLES2 || prog->IsES) { - if (varying_vectors > ctx->Const.MaxVarying) { - if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { - linker_warning(prog, "shader uses too many varying vectors " - "(%u > %u), but the driver will try to optimize " - "them out; this is non-portable out-of-spec " - "behavior\n", - varying_vectors, ctx->Const.MaxVarying); - } else { - linker_error(prog, "shader uses too many varying vectors " - "(%u > %u)\n", - varying_vectors, ctx->Const.MaxVarying); - return false; - } - } - } else { - const unsigned float_components = varying_vectors * 4; - if (float_components > ctx->Const.MaxVarying * 4) { - if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { - linker_warning(prog, "shader uses too many varying components " - "(%u > %u), but the driver will try to optimize " - "them out; this is non-portable out-of-spec " - "behavior\n", - float_components, ctx->Const.MaxVarying * 4); - } else { - linker_error(prog, "shader uses too many varying components " - "(%u > %u)\n", - float_components, ctx->Const.MaxVarying * 4); - return false; - } - } - } - - return true; -} - - -/** - * Store transform feedback location assignments into - * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls. - * - * If an error occurs, the error is reported through linker_error() and false - * is returned. - */ -static bool -store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog, - unsigned num_tfeedback_decls, - tfeedback_decl *tfeedback_decls) -{ - bool separate_attribs_mode = - prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS; - - ralloc_free(prog->LinkedTransformFeedback.Varyings); - ralloc_free(prog->LinkedTransformFeedback.Outputs); - - memset(&prog->LinkedTransformFeedback, 0, - sizeof(prog->LinkedTransformFeedback)); - - prog->LinkedTransformFeedback.Varyings = - rzalloc_array(prog, - struct gl_transform_feedback_varying_info, - num_tfeedback_decls); - - unsigned num_outputs = 0; - for (unsigned i = 0; i < num_tfeedback_decls; ++i) - num_outputs += tfeedback_decls[i].get_num_outputs(); - - prog->LinkedTransformFeedback.Outputs = - rzalloc_array(prog, - struct gl_transform_feedback_output, - num_outputs); - - unsigned num_buffers = 0; - - if (separate_attribs_mode) { - /* GL_SEPARATE_ATTRIBS */ - for (unsigned i = 0; i < num_tfeedback_decls; ++i) { - if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback, - num_buffers, num_outputs)) - return false; - - num_buffers++; - } - } - else { - /* GL_INVERLEAVED_ATTRIBS */ - for (unsigned i = 0; i < num_tfeedback_decls; ++i) { - if (tfeedback_decls[i].is_next_buffer_separator()) { - num_buffers++; - continue; - } - - if (!tfeedback_decls[i].store(ctx, prog, - &prog->LinkedTransformFeedback, - num_buffers, num_outputs)) - return false; - } - num_buffers++; - } - - assert(prog->LinkedTransformFeedback.NumOutputs == num_outputs); - - prog->LinkedTransformFeedback.NumBuffers = num_buffers; - return true; -} - -/** * Store the gl_FragDepth layout in the gl_shader_program struct. */ static void diff --git a/mesalib/src/glsl/linker.h b/mesalib/src/glsl/linker.h index 7d2e98a15..67c7f3488 100644 --- a/mesalib/src/glsl/linker.h +++ b/mesalib/src/glsl/linker.h @@ -91,4 +91,13 @@ private: void recursion(const glsl_type *t, char **name, size_t name_length); }; +void +linker_error(gl_shader_program *prog, const char *fmt, ...); + +void +linker_warning(gl_shader_program *prog, const char *fmt, ...); + +unsigned +count_attribute_slots(const glsl_type *t); + #endif /* GLSL_LINKER_H */ diff --git a/mesalib/src/glsl/loop_controls.cpp b/mesalib/src/glsl/loop_controls.cpp index 9acbadc50..79c820436 100644 --- a/mesalib/src/glsl/loop_controls.cpp +++ b/mesalib/src/glsl/loop_controls.cpp @@ -222,10 +222,10 @@ loop_control_visitor::visit_leave(ir_loop *ir) limit = cond->operands[0]->as_constant(); switch (cmp) { - case ir_binop_less: cmp = ir_binop_gequal; break; - case ir_binop_greater: cmp = ir_binop_lequal; break; - case ir_binop_lequal: cmp = ir_binop_greater; break; - case ir_binop_gequal: cmp = ir_binop_less; break; + case ir_binop_less: cmp = ir_binop_greater; break; + case ir_binop_greater: cmp = ir_binop_less; break; + case ir_binop_lequal: cmp = ir_binop_gequal; break; + case ir_binop_gequal: cmp = ir_binop_lequal; break; default: assert(!"Should not get here."); } } diff --git a/mesalib/src/glsl/lower_packed_varyings.cpp b/mesalib/src/glsl/lower_packed_varyings.cpp index 09c551c4e..9e7f274b7 100644 --- a/mesalib/src/glsl/lower_packed_varyings.cpp +++ b/mesalib/src/glsl/lower_packed_varyings.cpp @@ -66,6 +66,10 @@ * performance. However, hopefully in most cases the performance loss will * either be absorbed by a later optimization pass, or it will be offset by * memory bandwidth savings (because fewer varyings are used). + * + * This lowering pass also packs flat floats, ints, and uints together, by + * using ivec4 as the base type of flat "varyings", and using appropriate + * casts to convert floats and uints into ints. */ #include "glsl_symbol_table.h" @@ -90,6 +94,8 @@ public: void run(exec_list *instructions); private: + ir_assignment *bitwise_assign_pack(ir_rvalue *lhs, ir_rvalue *rhs); + ir_assignment *bitwise_assign_unpack(ir_rvalue *lhs, ir_rvalue *rhs); unsigned lower_rvalue(ir_rvalue *rvalue, unsigned fine_location, ir_variable *unpacked_var, const char *name); unsigned lower_arraylike(ir_rvalue *rvalue, unsigned array_size, @@ -181,6 +187,75 @@ lower_packed_varyings_visitor::run(exec_list *instructions) } } + +/** + * Make an ir_assignment from \c rhs to \c lhs, performing appropriate + * bitcasts if necessary to match up types. + * + * This function is called when packing varyings. + */ +ir_assignment * +lower_packed_varyings_visitor::bitwise_assign_pack(ir_rvalue *lhs, + ir_rvalue *rhs) +{ + if (lhs->type->base_type != rhs->type->base_type) { + /* Since we only mix types in flat varyings, and we always store flat + * varyings as type ivec4, we need only produce conversions from (uint + * or float) to int. + */ + assert(lhs->type->base_type == GLSL_TYPE_INT); + switch (rhs->type->base_type) { + case GLSL_TYPE_UINT: + rhs = new(this->mem_ctx) + ir_expression(ir_unop_u2i, lhs->type, rhs); + break; + case GLSL_TYPE_FLOAT: + rhs = new(this->mem_ctx) + ir_expression(ir_unop_bitcast_f2i, lhs->type, rhs); + break; + default: + assert(!"Unexpected type conversion while lowering varyings"); + break; + } + } + return new(this->mem_ctx) ir_assignment(lhs, rhs); +} + + +/** + * Make an ir_assignment from \c rhs to \c lhs, performing appropriate + * bitcasts if necessary to match up types. + * + * This function is called when unpacking varyings. + */ +ir_assignment * +lower_packed_varyings_visitor::bitwise_assign_unpack(ir_rvalue *lhs, + ir_rvalue *rhs) +{ + if (lhs->type->base_type != rhs->type->base_type) { + /* Since we only mix types in flat varyings, and we always store flat + * varyings as type ivec4, we need only produce conversions from int to + * (uint or float). + */ + assert(rhs->type->base_type == GLSL_TYPE_INT); + switch (lhs->type->base_type) { + case GLSL_TYPE_UINT: + rhs = new(this->mem_ctx) + ir_expression(ir_unop_i2u, lhs->type, rhs); + break; + case GLSL_TYPE_FLOAT: + rhs = new(this->mem_ctx) + ir_expression(ir_unop_bitcast_i2f, lhs->type, rhs); + break; + default: + assert(!"Unexpected type conversion while lowering varyings"); + break; + } + } + return new(this->mem_ctx) ir_assignment(lhs, rhs); +} + + /** * Recursively pack or unpack the given varying (or portion of a varying) by * traversing all of its constituent vectors. @@ -262,12 +337,12 @@ lower_packed_varyings_visitor::lower_rvalue(ir_rvalue *rvalue, ir_swizzle *swizzle = new(this->mem_ctx) ir_swizzle(packed_deref, swizzle_values, components); if (this->mode == ir_var_out) { - ir_assignment *assignment = new(this->mem_ctx) - ir_assignment(swizzle, rvalue); + ir_assignment *assignment + = this->bitwise_assign_pack(swizzle, rvalue); this->main_instructions->push_tail(assignment); } else { - ir_assignment *assignment = new(this->mem_ctx) - ir_assignment(rvalue, swizzle); + ir_assignment *assignment + = this->bitwise_assign_unpack(rvalue, swizzle); this->main_instructions->push_head(assignment); } return fine_location + components; @@ -306,8 +381,9 @@ lower_packed_varyings_visitor::lower_arraylike(ir_rvalue *rvalue, * If no packed varying has been created for the given varying location yet, * create it and add it to the shader before returning it. * - * The newly created varying inherits its base type (float, uint, or int) and - * interpolation parameters from \c unpacked_var. + * The newly created varying inherits its interpolation parameters from \c + * unpacked_var. Its base type is ivec4 if we are lowering a flat varying, + * vec4 otherwise. */ ir_variable * lower_packed_varyings_visitor::get_packed_varying(unsigned location, @@ -318,8 +394,11 @@ lower_packed_varyings_visitor::get_packed_varying(unsigned location, assert(slot < locations_used); if (this->packed_varyings[slot] == NULL) { char *packed_name = ralloc_asprintf(this->mem_ctx, "packed:%s", name); - const glsl_type *packed_type = glsl_type::get_instance( - unpacked_var->type->get_scalar_type()->base_type, 4, 1); + const glsl_type *packed_type; + if (unpacked_var->interpolation == INTERP_QUALIFIER_FLAT) + packed_type = glsl_type::ivec4_type; + else + packed_type = glsl_type::vec4_type; ir_variable *packed_var = new(this->mem_ctx) ir_variable(packed_type, packed_name, this->mode); packed_var->centroid = unpacked_var->centroid; |