diff options
Diffstat (limited to 'mesalib/src/glsl')
47 files changed, 2115 insertions, 589 deletions
diff --git a/mesalib/src/glsl/Makefile.am b/mesalib/src/glsl/Makefile.am index d0e5cd1d0..2bbad3d65 100644 --- a/mesalib/src/glsl/Makefile.am +++ b/mesalib/src/glsl/Makefile.am @@ -27,8 +27,7 @@ AM_CPPFLAGS = \ -I$(top_srcdir)/src/mesa/ \ -I$(top_srcdir)/src/glsl/glcpp \ -I$(top_srcdir)/src/gtest/include \ - $(DEFINES) \ - $(API_DEFINES) + $(DEFINES) AM_CFLAGS = $(VISIBILITY_CFLAGS) AM_CXXFLAGS = $(VISIBILITY_CXXFLAGS) diff --git a/mesalib/src/glsl/Makefile.sources b/mesalib/src/glsl/Makefile.sources index b5282a604..50bad85ad 100644 --- a/mesalib/src/glsl/Makefile.sources +++ b/mesalib/src/glsl/Makefile.sources @@ -16,6 +16,7 @@ LIBGLCPP_GENERATED_FILES = \ # libglsl LIBGLSL_FILES = \ + $(GLSL_SRCDIR)/ast_array_index.cpp \ $(GLSL_SRCDIR)/ast_expr.cpp \ $(GLSL_SRCDIR)/ast_function.cpp \ $(GLSL_SRCDIR)/ast_to_hir.cpp \ @@ -45,6 +46,7 @@ LIBGLSL_FILES = \ $(GLSL_SRCDIR)/ir_variable_refcount.cpp \ $(GLSL_SRCDIR)/linker.cpp \ $(GLSL_SRCDIR)/link_functions.cpp \ + $(GLSL_SRCDIR)/link_interface_blocks.cpp \ $(GLSL_SRCDIR)/link_uniforms.cpp \ $(GLSL_SRCDIR)/link_uniform_initializers.cpp \ $(GLSL_SRCDIR)/link_uniform_block_active_visitor.cpp \ @@ -62,12 +64,14 @@ LIBGLSL_FILES = \ $(GLSL_SRCDIR)/lower_mat_op_to_vec.cpp \ $(GLSL_SRCDIR)/lower_noise.cpp \ $(GLSL_SRCDIR)/lower_packed_varyings.cpp \ + $(GLSL_SRCDIR)/lower_named_interface_blocks.cpp \ $(GLSL_SRCDIR)/lower_packing_builtins.cpp \ $(GLSL_SRCDIR)/lower_texture_projection.cpp \ $(GLSL_SRCDIR)/lower_variable_index_to_cond_assign.cpp \ $(GLSL_SRCDIR)/lower_vec_index_to_cond_assign.cpp \ $(GLSL_SRCDIR)/lower_vec_index_to_swizzle.cpp \ $(GLSL_SRCDIR)/lower_vector.cpp \ + $(GLSL_SRCDIR)/lower_vector_insert.cpp \ $(GLSL_SRCDIR)/lower_output_reads.cpp \ $(GLSL_SRCDIR)/lower_ubo_reference.cpp \ $(GLSL_SRCDIR)/opt_algebraic.cpp \ @@ -81,6 +85,7 @@ LIBGLSL_FILES = \ $(GLSL_SRCDIR)/opt_dead_code_local.cpp \ $(GLSL_SRCDIR)/opt_dead_functions.cpp \ $(GLSL_SRCDIR)/opt_flatten_nested_if_blocks.cpp \ + $(GLSL_SRCDIR)/opt_flip_matrices.cpp \ $(GLSL_SRCDIR)/opt_function_inlining.cpp \ $(GLSL_SRCDIR)/opt_if_simplification.cpp \ $(GLSL_SRCDIR)/opt_noop_swizzle.cpp \ diff --git a/mesalib/src/glsl/ast.h b/mesalib/src/glsl/ast.h index f920759d6..f73265bf3 100644 --- a/mesalib/src/glsl/ast.h +++ b/mesalib/src/glsl/ast.h @@ -809,9 +809,9 @@ public: ast_compound_statement *body; }; -class ast_uniform_block : public ast_node { +class ast_interface_block : public ast_node { public: - ast_uniform_block(ast_type_qualifier layout, + ast_interface_block(ast_type_qualifier layout, const char *instance_name, ast_expression *array_size) : layout(layout), block_name(NULL), instance_name(instance_name), @@ -858,7 +858,17 @@ _mesa_ast_field_selection_to_hir(const ast_expression *expr, exec_list *instructions, struct _mesa_glsl_parse_state *state); +extern ir_rvalue * +_mesa_ast_array_index_to_hir(void *mem_ctx, + struct _mesa_glsl_parse_state *state, + ir_rvalue *array, ir_rvalue *idx, + YYLTYPE &loc, YYLTYPE &idx_loc); + void emit_function(_mesa_glsl_parse_state *state, ir_function *f); +extern void +check_builtin_array_max_size(const char *name, unsigned size, + YYLTYPE loc, struct _mesa_glsl_parse_state *state); + #endif /* AST_H */ diff --git a/mesalib/src/glsl/ast_array_index.cpp b/mesalib/src/glsl/ast_array_index.cpp new file mode 100644 index 000000000..4baeb6f9d --- /dev/null +++ b/mesalib/src/glsl/ast_array_index.cpp @@ -0,0 +1,190 @@ +/* + * Copyright © 2010 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. + */ + +#include "ast.h" +#include "glsl_types.h" +#include "ir.h" + +ir_rvalue * +_mesa_ast_array_index_to_hir(void *mem_ctx, + struct _mesa_glsl_parse_state *state, + ir_rvalue *array, ir_rvalue *idx, + YYLTYPE &loc, YYLTYPE &idx_loc) +{ + if (!array->type->is_error() + && !array->type->is_array() + && !array->type->is_matrix() + && !array->type->is_vector()) { + _mesa_glsl_error(& idx_loc, state, + "cannot dereference non-array / non-matrix / " + "non-vector"); + } + + if (!idx->type->is_error()) { + if (!idx->type->is_integer()) { + _mesa_glsl_error(& idx_loc, state, "array index must be integer type"); + } else if (!idx->type->is_scalar()) { + _mesa_glsl_error(& idx_loc, state, "array index must be scalar"); + } + } + + /* If the array index is a constant expression and the array has a + * declared size, ensure that the access is in-bounds. If the array + * index is not a constant expression, ensure that the array has a + * declared size. + */ + ir_constant *const const_index = idx->constant_expression_value(); + if (const_index != NULL && idx->type->is_integer()) { + const int idx = const_index->value.i[0]; + const char *type_name = "error"; + unsigned bound = 0; + + /* From page 24 (page 30 of the PDF) of the GLSL 1.50 spec: + * + * "It is illegal to declare an array with a size, and then + * later (in the same shader) index the same array with an + * integral constant expression greater than or equal to the + * declared size. It is also illegal to index an array with a + * negative constant expression." + */ + if (array->type->is_matrix()) { + if (array->type->row_type()->vector_elements <= idx) { + type_name = "matrix"; + bound = array->type->row_type()->vector_elements; + } + } else if (array->type->is_vector()) { + if (array->type->vector_elements <= idx) { + type_name = "vector"; + bound = array->type->vector_elements; + } + } else { + /* glsl_type::array_size() returns 0 for non-array types. This means + * that we don't need to verify that the type is an array before + * doing the bounds checking. + */ + if ((array->type->array_size() > 0) + && (array->type->array_size() <= idx)) { + type_name = "array"; + bound = array->type->array_size(); + } + } + + if (bound > 0) { + _mesa_glsl_error(& loc, state, "%s index must be < %u", + type_name, bound); + } else if (idx < 0) { + _mesa_glsl_error(& loc, state, "%s index must be >= 0", + type_name); + } + + if (array->type->is_array()) { + /* If the array is a variable dereference, it dereferences the + * whole array, by definition. Use this to get the variable. + * + * FINISHME: Should some methods for getting / setting / testing + * FINISHME: array access limits be added to ir_dereference? + */ + ir_variable *const v = array->whole_variable_referenced(); + if ((v != NULL) && (unsigned(idx) > v->max_array_access)) { + v->max_array_access = idx; + + /* Check whether this access will, as a side effect, implicitly + * cause the size of a built-in array to be too large. + */ + check_builtin_array_max_size(v->name, idx+1, loc, state); + } + } + } else if (const_index == NULL && array->type->is_array()) { + if (array->type->array_size() == 0) { + _mesa_glsl_error(&loc, state, "unsized array index must be constant"); + } else if (array->type->fields.array->is_interface()) { + /* Page 46 in section 4.3.7 of the OpenGL ES 3.00 spec says: + * + * "All indexes used to index a uniform block array must be + * constant integral expressions." + */ + _mesa_glsl_error(&loc, state, + "uniform block array index must be constant"); + } else { + /* whole_variable_referenced can return NULL if the array is a + * member of a structure. In this case it is safe to not update + * the max_array_access field because it is never used for fields + * of structures. + */ + ir_variable *v = array->whole_variable_referenced(); + if (v != NULL) + v->max_array_access = array->type->array_size() - 1; + } + + /* From page 23 (29 of the PDF) of the GLSL 1.30 spec: + * + * "Samplers aggregated into arrays within a shader (using square + * brackets [ ]) can only be indexed with integral constant + * expressions [...]." + * + * This restriction was added in GLSL 1.30. Shaders using earlier + * version of the language should not be rejected by the compiler + * front-end for using this construct. This allows useful things such + * as using a loop counter as the index to an array of samplers. If the + * loop in unrolled, the code should compile correctly. Instead, emit a + * warning. + */ + if (array->type->element_type()->is_sampler()) { + if (!state->is_version(130, 100)) { + if (state->es_shader) { + _mesa_glsl_warning(&loc, state, + "sampler arrays indexed with non-constant " + "expressions is optional in %s", + state->get_version_string()); + } else { + _mesa_glsl_warning(&loc, state, + "sampler arrays indexed with non-constant " + "expressions will be forbidden in GLSL 1.30 " + "and later"); + } + } else { + _mesa_glsl_error(&loc, state, + "sampler arrays indexed with non-constant " + "expressions is forbidden in GLSL 1.30 and " + "later"); + } + } + } + + /* After performing all of the error checking, generate the IR for the + * expression. + */ + if (array->type->is_array() + || array->type->is_matrix()) { + return new(mem_ctx) ir_dereference_array(array, idx); + } else if (array->type->is_vector()) { + return new(mem_ctx) ir_expression(ir_binop_vector_extract, array, idx); + } else if (array->type->is_error()) { + return array; + } else { + ir_rvalue *result = new(mem_ctx) ir_dereference_array(array, idx); + result->type = glsl_type::error_type; + + return result; + } +} diff --git a/mesalib/src/glsl/ast_function.cpp b/mesalib/src/glsl/ast_function.cpp index 26f72cf8e..00e0c05dd 100644 --- a/mesalib/src/glsl/ast_function.cpp +++ b/mesalib/src/glsl/ast_function.cpp @@ -165,10 +165,18 @@ verify_parameter_modes(_mesa_glsl_parse_state *state, actual->variable_referenced()->name); return false; } else if (!actual->is_lvalue()) { - _mesa_glsl_error(&loc, state, - "function parameter '%s %s' is not an lvalue", - mode, formal->name); - return false; + /* Even though ir_binop_vector_extract is not an l-value, let it + * slop through. generate_call will handle it correctly. + */ + ir_expression *const expr = ((ir_rvalue *) actual)->as_expression(); + if (expr == NULL + || expr->operation != ir_binop_vector_extract + || !expr->operands[0]->is_lvalue()) { + _mesa_glsl_error(&loc, state, + "function parameter '%s %s' is not an lvalue", + mode, formal->name); + return false; + } } } @@ -178,6 +186,93 @@ verify_parameter_modes(_mesa_glsl_parse_state *state, return true; } +static void +fix_parameter(void *mem_ctx, ir_rvalue *actual, const glsl_type *formal_type, + exec_list *before_instructions, exec_list *after_instructions, + bool parameter_is_inout) +{ + ir_expression *const expr = actual->as_expression(); + + /* If the types match exactly and the parameter is not a vector-extract, + * nothing needs to be done to fix the parameter. + */ + if (formal_type == actual->type + && (expr == NULL || expr->operation != ir_binop_vector_extract)) + return; + + /* To convert an out parameter, we need to create a temporary variable to + * hold the value before conversion, and then perform the conversion after + * the function call returns. + * + * This has the effect of transforming code like this: + * + * void f(out int x); + * float value; + * f(value); + * + * Into IR that's equivalent to this: + * + * void f(out int x); + * float value; + * int out_parameter_conversion; + * f(out_parameter_conversion); + * value = float(out_parameter_conversion); + * + * If the parameter is an ir_expression of ir_binop_vector_extract, + * additional conversion is needed in the post-call re-write. + */ + ir_variable *tmp = + new(mem_ctx) ir_variable(formal_type, "inout_tmp", ir_var_temporary); + + before_instructions->push_tail(tmp); + + /* If the parameter is an inout parameter, copy the value of the actual + * parameter to the new temporary. Note that no type conversion is allowed + * here because inout parameters must match types exactly. + */ + if (parameter_is_inout) { + /* Inout parameters should never require conversion, since that would + * require an implicit conversion to exist both to and from the formal + * parameter type, and there are no bidirectional implicit conversions. + */ + assert (actual->type == formal_type); + + ir_dereference_variable *const deref_tmp_1 = + new(mem_ctx) ir_dereference_variable(tmp); + ir_assignment *const assignment = + new(mem_ctx) ir_assignment(deref_tmp_1, actual); + before_instructions->push_tail(assignment); + } + + /* Replace the parameter in the call with a dereference of the new + * temporary. + */ + ir_dereference_variable *const deref_tmp_2 = + new(mem_ctx) ir_dereference_variable(tmp); + actual->replace_with(deref_tmp_2); + + + /* Copy the temporary variable to the actual parameter with optional + * type conversion applied. + */ + ir_rvalue *rhs = new(mem_ctx) ir_dereference_variable(tmp); + if (actual->type != formal_type) + rhs = convert_component(rhs, actual->type); + + ir_rvalue *lhs = actual; + if (expr != NULL && expr->operation == ir_binop_vector_extract) { + rhs = new(mem_ctx) ir_expression(ir_triop_vector_insert, + expr->operands[0]->type, + expr->operands[0]->clone(mem_ctx, NULL), + rhs, + expr->operands[1]->clone(mem_ctx, NULL)); + lhs = expr->operands[0]->clone(mem_ctx, NULL); + } + + ir_assignment *const assignment_2 = new(mem_ctx) ir_assignment(lhs, rhs); + after_instructions->push_tail(assignment_2); +} + /** * If a function call is generated, \c call_ir will point to it on exit. * Otherwise \c call_ir will be set to \c NULL. @@ -218,50 +313,10 @@ generate_call(exec_list *instructions, ir_function_signature *sig, break; } case ir_var_function_out: - if (actual->type != formal->type) { - /* To convert an out parameter, we need to create a - * temporary variable to hold the value before conversion, - * and then perform the conversion after the function call - * returns. - * - * This has the effect of transforming code like this: - * - * void f(out int x); - * float value; - * f(value); - * - * Into IR that's equivalent to this: - * - * void f(out int x); - * float value; - * int out_parameter_conversion; - * f(out_parameter_conversion); - * value = float(out_parameter_conversion); - */ - ir_variable *tmp = - new(ctx) ir_variable(formal->type, - "out_parameter_conversion", - ir_var_temporary); - instructions->push_tail(tmp); - ir_dereference_variable *deref_tmp_1 - = new(ctx) ir_dereference_variable(tmp); - ir_dereference_variable *deref_tmp_2 - = new(ctx) ir_dereference_variable(tmp); - ir_rvalue *converted_tmp - = convert_component(deref_tmp_1, actual->type); - ir_assignment *assignment - = new(ctx) ir_assignment(actual, converted_tmp); - post_call_conversions.push_tail(assignment); - actual->replace_with(deref_tmp_2); - } - break; case ir_var_function_inout: - /* Inout parameters should never require conversion, since that - * would require an implicit conversion to exist both to and - * from the formal parameter type, and there are no - * bidirectional implicit conversions. - */ - assert (actual->type == formal->type); + fix_parameter(ctx, actual, formal->type, + instructions, &post_call_conversions, + formal->mode == ir_var_function_inout); break; default: assert (!"Illegal formal parameter mode"); diff --git a/mesalib/src/glsl/ast_to_hir.cpp b/mesalib/src/glsl/ast_to_hir.cpp index 92065f5b7..e918adeef 100644 --- a/mesalib/src/glsl/ast_to_hir.cpp +++ b/mesalib/src/glsl/ast_to_hir.cpp @@ -672,6 +672,30 @@ do_assignment(exec_list *instructions, struct _mesa_glsl_parse_state *state, void *ctx = state; bool error_emitted = (lhs->type->is_error() || rhs->type->is_error()); + /* If the assignment LHS comes back as an ir_binop_vector_extract + * expression, move it to the RHS as an ir_triop_vector_insert. + */ + if (lhs->ir_type == ir_type_expression) { + ir_expression *const expr = lhs->as_expression(); + + if (unlikely(expr->operation == ir_binop_vector_extract)) { + ir_rvalue *new_rhs = + validate_assignment(state, 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, + expr->operands[0]->type, + expr->operands[0], + new_rhs, + expr->operands[1]); + lhs = expr->operands[0]->clone(ctx, NULL); + } + } + } + ir_variable *lhs_var = lhs->variable_referenced(); if (lhs_var) lhs_var->assigned = true; @@ -904,7 +928,7 @@ get_scalar_boolean_operand(exec_list *instructions, * If name refers to a builtin array whose maximum allowed size is less than * size, report an error and return true. Otherwise return false. */ -static bool +void check_builtin_array_max_size(const char *name, unsigned size, YYLTYPE loc, struct _mesa_glsl_parse_state *state) { @@ -918,7 +942,6 @@ check_builtin_array_max_size(const char *name, unsigned size, _mesa_glsl_error(&loc, state, "`gl_TexCoord' array size cannot " "be larger than gl_MaxTextureCoords (%u)\n", state->Const.MaxTextureCoords); - return true; } else if (strcmp("gl_ClipDistance", name) == 0 && size > state->Const.MaxClipPlanes) { /* From section 7.1 (Vertex Shader Special Variables) of the @@ -933,9 +956,7 @@ check_builtin_array_max_size(const char *name, unsigned size, _mesa_glsl_error(&loc, state, "`gl_ClipDistance' array size cannot " "be larger than gl_MaxClipDistances (%u)\n", state->Const.MaxClipPlanes); - return true; } - return false; } /** @@ -1517,172 +1538,11 @@ ast_expression::hir(exec_list *instructions, op[0] = subexpressions[0]->hir(instructions, state); op[1] = subexpressions[1]->hir(instructions, state); - error_emitted = op[0]->type->is_error() || op[1]->type->is_error(); - - ir_rvalue *const array = op[0]; - - result = new(ctx) ir_dereference_array(op[0], op[1]); - - /* Do not use op[0] after this point. Use array. - */ - op[0] = NULL; - + result = _mesa_ast_array_index_to_hir(ctx, state, op[0], op[1], + loc, index_loc); - if (error_emitted) - break; - - if (!array->type->is_array() - && !array->type->is_matrix() - && !array->type->is_vector()) { - _mesa_glsl_error(& index_loc, state, - "cannot dereference non-array / non-matrix / " - "non-vector"); + if (result->type->is_error()) error_emitted = true; - } - - if (!op[1]->type->is_integer()) { - _mesa_glsl_error(& index_loc, state, - "array index must be integer type"); - error_emitted = true; - } else if (!op[1]->type->is_scalar()) { - _mesa_glsl_error(& index_loc, state, - "array index must be scalar"); - error_emitted = true; - } - - /* If the array index is a constant expression and the array has a - * declared size, ensure that the access is in-bounds. If the array - * index is not a constant expression, ensure that the array has a - * declared size. - */ - ir_constant *const const_index = op[1]->constant_expression_value(); - if (const_index != NULL) { - const int idx = const_index->value.i[0]; - const char *type_name; - unsigned bound = 0; - - if (array->type->is_matrix()) { - type_name = "matrix"; - } else if (array->type->is_vector()) { - type_name = "vector"; - } else { - type_name = "array"; - } - - /* From page 24 (page 30 of the PDF) of the GLSL 1.50 spec: - * - * "It is illegal to declare an array with a size, and then - * later (in the same shader) index the same array with an - * integral constant expression greater than or equal to the - * declared size. It is also illegal to index an array with a - * negative constant expression." - */ - if (array->type->is_matrix()) { - if (array->type->row_type()->vector_elements <= idx) { - bound = array->type->row_type()->vector_elements; - } - } else if (array->type->is_vector()) { - if (array->type->vector_elements <= idx) { - bound = array->type->vector_elements; - } - } else { - if ((array->type->array_size() > 0) - && (array->type->array_size() <= idx)) { - bound = array->type->array_size(); - } - } - - if (bound > 0) { - _mesa_glsl_error(& loc, state, "%s index must be < %u", - type_name, bound); - error_emitted = true; - } else if (idx < 0) { - _mesa_glsl_error(& loc, state, "%s index must be >= 0", - type_name); - error_emitted = true; - } - - if (array->type->is_array()) { - /* If the array is a variable dereference, it dereferences the - * whole array, by definition. Use this to get the variable. - * - * FINISHME: Should some methods for getting / setting / testing - * FINISHME: array access limits be added to ir_dereference? - */ - ir_variable *const v = array->whole_variable_referenced(); - if ((v != NULL) && (unsigned(idx) > v->max_array_access)) { - v->max_array_access = idx; - - /* Check whether this access will, as a side effect, implicitly - * cause the size of a built-in array to be too large. - */ - if (check_builtin_array_max_size(v->name, idx+1, loc, state)) - error_emitted = true; - } - } - } else if (array->type->array_size() == 0) { - _mesa_glsl_error(&loc, state, "unsized array index must be constant"); - } else if (array->type->is_array() - && array->type->fields.array->is_interface()) { - /* Page 46 in section 4.3.7 of the OpenGL ES 3.00 spec says: - * - * "All indexes used to index a uniform block array must be - * constant integral expressions." - */ - _mesa_glsl_error(&loc, state, - "uniform block array index must be constant"); - } else { - if (array->type->is_array()) { - /* whole_variable_referenced can return NULL if the array is a - * member of a structure. In this case it is safe to not update - * the max_array_access field because it is never used for fields - * of structures. - */ - ir_variable *v = array->whole_variable_referenced(); - if (v != NULL) - v->max_array_access = array->type->array_size() - 1; - } - } - - /* From page 23 (29 of the PDF) of the GLSL 1.30 spec: - * - * "Samplers aggregated into arrays within a shader (using square - * brackets [ ]) can only be indexed with integral constant - * expressions [...]." - * - * This restriction was added in GLSL 1.30. Shaders using earlier version - * of the language should not be rejected by the compiler front-end for - * using this construct. This allows useful things such as using a loop - * counter as the index to an array of samplers. If the loop in unrolled, - * the code should compile correctly. Instead, emit a warning. - */ - if (array->type->is_array() && - array->type->element_type()->is_sampler() && - const_index == NULL) { - - if (!state->is_version(130, 100)) { - if (state->es_shader) { - _mesa_glsl_warning(&loc, state, - "sampler arrays indexed with non-constant " - "expressions is optional in %s", - state->get_version_string()); - } else { - _mesa_glsl_warning(&loc, state, - "sampler arrays indexed with non-constant " - "expressions will be forbidden in GLSL 1.30 and " - "later"); - } - } else { - _mesa_glsl_error(&loc, state, - "sampler arrays indexed with non-constant " - "expressions is forbidden in GLSL 1.30 and " - "later"); - error_emitted = true; - } - } - - if (error_emitted) - result->type = glsl_type::error_type; break; } @@ -1842,6 +1702,9 @@ process_array_type(YYLTYPE *loc, const glsl_type *base, ast_node *array_size, { unsigned length = 0; + if (base == NULL) + return glsl_type::error_type; + /* From page 19 (page 25) of the GLSL 1.20 spec: * * "Only one-dimensional arrays may be declared." @@ -1894,7 +1757,8 @@ process_array_type(YYLTYPE *loc, const glsl_type *base, ast_node *array_size, "allowed in GLSL ES 1.00."); } - return glsl_type::get_array_instance(base, length); + const glsl_type *array_type = glsl_type::get_array_instance(base, length); + return array_type != NULL ? array_type : glsl_type::error_type; } @@ -3365,10 +3229,17 @@ ast_function::hir(exec_list *instructions, "match prototype", name); } - if (is_definition && sig->is_defined) { - YYLTYPE loc = this->get_location(); - - _mesa_glsl_error(& loc, state, "function `%s' redefined", name); + if (sig->is_defined) { + if (is_definition) { + YYLTYPE loc = this->get_location(); + _mesa_glsl_error(& loc, state, "function `%s' redefined", name); + } else { + /* We just encountered a prototype that exactly matches a + * function that's already been defined. This is redundant, + * and we should ignore it. + */ + return NULL; + } } } } else { @@ -4149,8 +4020,14 @@ ast_process_structure_or_interface_block(exec_list *instructions, * blocks. All other types, arrays, and structures * allowed for uniforms are allowed within a uniform * block." + * + * It should be impossible for decl_type to be NULL here. Cases that + * might naturally lead to decl_type being NULL, especially for the + * is_interface case, will have resulted in compilation having + * already halted due to a syntax error. */ - const struct glsl_type *field_type = decl_type; + const struct glsl_type *field_type = + decl_type != NULL ? decl_type : glsl_type::error_type; if (is_interface && field_type->contains_sampler()) { YYLTYPE loc = decl_list->get_location(); @@ -4173,12 +4050,15 @@ ast_process_structure_or_interface_block(exec_list *instructions, field_type = process_array_type(&loc, decl_type, decl->array_size, state); } - fields[i].type = (field_type != NULL) - ? field_type : glsl_type::error_type; + fields[i].type = field_type; fields[i].name = decl->identifier; if (qual->flags.q.row_major || qual->flags.q.column_major) { - if (!field_type->is_matrix() && !field_type->is_record()) { + if (!qual->flags.q.uniform) { + _mesa_glsl_error(&loc, state, + "row_major and column_major can only be " + "applied to uniform interface blocks."); + } else if (!field_type->is_matrix() && !field_type->is_record()) { _mesa_glsl_error(&loc, state, "uniform block layout qualifiers row_major and " "column_major can only be applied to matrix and " @@ -4187,6 +4067,12 @@ ast_process_structure_or_interface_block(exec_list *instructions, validate_matrix_layout_for_type(state, &loc, field_type); } + if (qual->flags.q.uniform && qual->has_interpolation()) { + _mesa_glsl_error(&loc, state, + "interpolation qualifiers cannot be used " + "with uniform interface blocks"); + } + if (field_type->is_matrix() || (field_type->is_array() && field_type->fields.array->is_matrix())) { fields[i].row_major = block_row_major; @@ -4244,12 +4130,12 @@ ast_struct_specifier::hir(exec_list *instructions, } ir_rvalue * -ast_uniform_block::hir(exec_list *instructions, - struct _mesa_glsl_parse_state *state) +ast_interface_block::hir(exec_list *instructions, + struct _mesa_glsl_parse_state *state) { YYLTYPE loc = this->get_location(); - /* The ast_uniform_block has a list of ast_declarator_lists. We + /* The ast_interface_block has a list of ast_declarator_lists. We * need to turn those into ir_variables with an association * with this uniform block. */ @@ -4276,16 +4162,32 @@ ast_uniform_block::hir(exec_list *instructions, true, block_row_major); + ir_variable_mode var_mode; + const char *iface_type_name; + if (this->layout.flags.q.in) { + var_mode = ir_var_shader_in; + iface_type_name = "in"; + } else if (this->layout.flags.q.out) { + var_mode = ir_var_shader_out; + iface_type_name = "out"; + } else if (this->layout.flags.q.uniform) { + var_mode = ir_var_uniform; + iface_type_name = "uniform"; + } else { + assert(!"interface block layout qualifier not found!"); + } + const glsl_type *block_type = glsl_type::get_interface_instance(fields, num_variables, packing, this->block_name); - if (!state->symbols->add_type(block_type->name, block_type)) { + if (!state->symbols->add_interface(block_type->name, block_type, var_mode)) { YYLTYPE loc = this->get_location(); - _mesa_glsl_error(&loc, state, "Uniform block name `%s' already taken in " - "the current scope.\n", this->block_name); + _mesa_glsl_error(&loc, state, "Interface block `%s' with type `%s' " + "already taken in the current scope.\n", + this->block_name, iface_type_name); } /* Since interface blocks cannot contain statements, it should be @@ -4309,11 +4211,11 @@ ast_uniform_block::hir(exec_list *instructions, var = new(state) ir_variable(block_array_type, this->instance_name, - ir_var_uniform); + var_mode); } else { var = new(state) ir_variable(block_type, this->instance_name, - ir_var_uniform); + var_mode); } var->interface_type = block_type; @@ -4329,7 +4231,7 @@ ast_uniform_block::hir(exec_list *instructions, ir_variable *var = new(state) ir_variable(fields[i].type, ralloc_strdup(state, fields[i].name), - ir_var_uniform); + var_mode); var->interface_type = block_type; state->symbols->add_variable(var); diff --git a/mesalib/src/glsl/builtin_variables.cpp b/mesalib/src/glsl/builtin_variables.cpp index b0c7a2035..4bb361c2e 100644 --- a/mesalib/src/glsl/builtin_variables.cpp +++ b/mesalib/src/glsl/builtin_variables.cpp @@ -39,6 +39,12 @@ generate_ARB_draw_instanced_variables(exec_list *, struct _mesa_glsl_parse_state *, bool, _mesa_glsl_parser_targets); +static void +generate_AMD_vertex_shader_layer_variables(exec_list *instructions, + struct _mesa_glsl_parse_state *state, + bool warn, + _mesa_glsl_parser_targets target); + struct builtin_variable { enum ir_variable_mode mode; int slot; @@ -535,7 +541,7 @@ generate_common_ES_uniforms(exec_list *instructions, add_builtin_constant(instructions, symtab, "gl_MaxTextureImageUnits", state->Const.MaxTextureImageUnits); add_builtin_constant(instructions, symtab, "gl_MaxFragmentUniformVectors", - state->Const.MaxFragmentUniformComponents); + state->Const.MaxFragmentUniformComponents / 4); add_uniform(instructions, symtab, "gl_DepthRange", state->symbols->get_type("gl_DepthRangeParameters")); @@ -818,6 +824,8 @@ generate_130_vs_variables(exec_list *instructions, "gl_ClipDistance", clip_distance_array_type, ir_var_shader_out, VARYING_SLOT_CLIP_DIST0); + generate_AMD_vertex_shader_layer_variables(instructions, state, false, + vertex_shader); } @@ -1020,6 +1028,29 @@ generate_ARB_draw_instanced_variables(exec_list *instructions, } } +static void +generate_AMD_vertex_shader_layer_variables(exec_list *instructions, + struct _mesa_glsl_parse_state *state, + bool warn, + _mesa_glsl_parser_targets target) +{ + /* gl_Layer is only available in the vertex shader for the + * AMD_vertex_shader_layer extension. It will also be available in the + * geometry shader when GLSL 1.50 is supported. + */ + if (target != vertex_shader) + return; + + if (state->AMD_vertex_shader_layer_enable) { + ir_variable *inst = + add_variable(instructions, state->symbols, + "gl_Layer", glsl_type::int_type, + ir_var_shader_out, VARYING_SLOT_LAYER); + + if (warn) + inst->warn_extension = "GL_AMD_vertex_shader_layer"; + } +} static void generate_ARB_shader_stencil_export_variables(exec_list *instructions, diff --git a/mesalib/src/glsl/glcpp/glcpp-parse.y b/mesalib/src/glsl/glcpp/glcpp-parse.y index 00edbbfbd..81ba04bcc 100644 --- a/mesalib/src/glsl/glcpp/glcpp-parse.y +++ b/mesalib/src/glsl/glcpp/glcpp-parse.y @@ -1236,6 +1236,12 @@ glcpp_parser_create (const struct gl_extensions *extensions, int api) if (extensions->ARB_texture_query_lod) add_builtin_define(parser, "GL_ARB_texture_query_lod", 1); + + if (extensions->ARB_gpu_shader5) + add_builtin_define(parser, "GL_ARB_gpu_shader5", 1); + + if (extensions->AMD_vertex_shader_layer) + add_builtin_define(parser, "GL_AMD_vertex_shader_layer", 1); } } diff --git a/mesalib/src/glsl/glcpp/pp.c b/mesalib/src/glsl/glcpp/pp.c index 789f7f941..7e1b6c689 100644 --- a/mesalib/src/glsl/glcpp/pp.c +++ b/mesalib/src/glsl/glcpp/pp.c @@ -97,8 +97,10 @@ remove_line_continuations(glcpp_parser_t *ctx, const char *shader) { ralloc_strncat(&clean, shader, newline - shader + 1); - while (collapsed_newlines--) + while (collapsed_newlines) { ralloc_strcat(&clean, "\n"); + collapsed_newlines--; + } shader = newline + 1; search_start = shader; } diff --git a/mesalib/src/glsl/glsl_parser.yy b/mesalib/src/glsl/glsl_parser.yy index f52ed9b0a..6e92c2651 100644 --- a/mesalib/src/glsl/glsl_parser.yy +++ b/mesalib/src/glsl/glsl_parser.yy @@ -79,7 +79,7 @@ static void yyerror(YYLTYPE *loc, _mesa_glsl_parse_state *st, const char *msg) ast_case_label_list *case_label_list; ast_case_statement *case_statement; ast_case_statement_list *case_statement_list; - ast_uniform_block *uniform_block; + ast_interface_block *interface_block; struct { ast_node *cond; @@ -115,7 +115,7 @@ static void yyerror(YYLTYPE *loc, _mesa_glsl_parse_state *st, const char *msg) %token STRUCT VOID_TOK WHILE %token <identifier> IDENTIFIER TYPE_IDENTIFIER NEW_IDENTIFIER %type <identifier> any_identifier -%type <uniform_block> instance_name_opt +%type <interface_block> instance_name_opt %token <real> FLOATCONSTANT %token <n> INTCONSTANT UINTCONSTANT BOOLCONSTANT %token <identifier> FIELD_SELECTION @@ -164,7 +164,8 @@ static void yyerror(YYLTYPE *loc, _mesa_glsl_parse_state *st, const char *msg) %type <type_qualifier> interpolation_qualifier %type <type_qualifier> layout_qualifier %type <type_qualifier> layout_qualifier_id_list layout_qualifier_id -%type <type_qualifier> uniform_block_layout_qualifier +%type <type_qualifier> interface_block_layout_qualifier +%type <type_qualifier> interface_qualifier %type <type_specifier> type_specifier %type <type_specifier> type_specifier_no_prec %type <type_specifier> type_specifier_nonarray @@ -223,8 +224,8 @@ static void yyerror(YYLTYPE *loc, _mesa_glsl_parse_state *st, const char *msg) %type <node> declaration %type <node> declaration_statement %type <node> jump_statement -%type <node> uniform_block -%type <uniform_block> basic_uniform_block +%type <node> interface_block +%type <interface_block> basic_interface_block %type <struct_specifier> struct_specifier %type <declarator_list> struct_declaration_list %type <declarator_list> struct_declaration @@ -784,7 +785,7 @@ declaration: $3->is_precision_statement = true; $$ = $3; } - | uniform_block + | interface_block { $$ = $1; } @@ -1140,7 +1141,7 @@ layout_qualifier_id: } } - /* See also uniform_block_layout_qualifier. */ + /* See also interface_block_layout_qualifier. */ if (!$$.flags.i && state->ARB_uniform_buffer_object_enable) { if (strcmp($1, "std140") == 0) { $$.flags.q.std140 = 1; @@ -1211,15 +1212,15 @@ layout_qualifier_id: "identifier `%s' used\n", $1); } } - | uniform_block_layout_qualifier + | interface_block_layout_qualifier { $$ = $1; /* Layout qualifiers for ARB_uniform_buffer_object. */ - if (!state->ARB_uniform_buffer_object_enable) { + if ($$.flags.q.uniform && !state->ARB_uniform_buffer_object_enable) { _mesa_glsl_error(& @1, state, "#version 140 / GL_ARB_uniform_buffer_object " "layout qualifier `%s' is used\n", $1); - } else if (state->ARB_uniform_buffer_object_warn) { + } else if ($$.flags.q.uniform && state->ARB_uniform_buffer_object_warn) { _mesa_glsl_warning(& @1, state, "#version 140 / GL_ARB_uniform_buffer_object " "layout qualifier `%s' is used\n", $1); @@ -1232,7 +1233,7 @@ layout_qualifier_id: * most qualifiers. See the any_identifier path of * layout_qualifier_id for the others. */ -uniform_block_layout_qualifier: +interface_block_layout_qualifier: ROW_MAJOR { memset(& $$, 0, sizeof($$)); @@ -1893,14 +1894,14 @@ function_definition: ; /* layout_qualifieropt is packed into this rule */ -uniform_block: - basic_uniform_block +interface_block: + basic_interface_block { $$ = $1; } - | layout_qualifier basic_uniform_block + | layout_qualifier basic_interface_block { - ast_uniform_block *block = $2; + ast_interface_block *block = $2; if (!block->layout.merge_qualifier(& @1, state, $1)) { YYERROR; } @@ -1908,55 +1909,137 @@ uniform_block: } ; -basic_uniform_block: - UNIFORM NEW_IDENTIFIER '{' member_list '}' instance_name_opt ';' +basic_interface_block: + interface_qualifier NEW_IDENTIFIER '{' member_list '}' instance_name_opt ';' { - ast_uniform_block *const block = $6; + ast_interface_block *const block = $6; block->block_name = $2; block->declarations.push_degenerate_list_at_head(& $4->link); - if (!state->ARB_uniform_buffer_object_enable) { + if ($1.flags.q.uniform) { + if (!state->ARB_uniform_buffer_object_enable) { + _mesa_glsl_error(& @1, state, + "#version 140 / GL_ARB_uniform_buffer_object " + "required for defining uniform blocks\n"); + } else if (state->ARB_uniform_buffer_object_warn) { + _mesa_glsl_warning(& @1, state, + "#version 140 / GL_ARB_uniform_buffer_object " + "required for defining uniform blocks\n"); + } + } else { + if (state->es_shader || state->language_version < 150) { + _mesa_glsl_error(& @1, state, + "#version 150 required for using " + "interface blocks.\n"); + } + } + + /* From the GLSL 1.50.11 spec, section 4.3.7 ("Interface Blocks"): + * "It is illegal to have an input block in a vertex shader + * or an output block in a fragment shader" + */ + if ((state->target == vertex_shader) && $1.flags.q.in) { _mesa_glsl_error(& @1, state, - "#version 140 / GL_ARB_uniform_buffer_object " - "required for defining uniform blocks\n"); - } else if (state->ARB_uniform_buffer_object_warn) { - _mesa_glsl_warning(& @1, state, - "#version 140 / GL_ARB_uniform_buffer_object " - "required for defining uniform blocks\n"); + "`in' interface block is not allowed for " + "a vertex shader\n"); + } else if ((state->target == fragment_shader) && $1.flags.q.out) { + _mesa_glsl_error(& @1, state, + "`out' interface block is not allowed for " + "a fragment shader\n"); } /* Since block arrays require names, and both features are added in * the same language versions, we don't have to explicitly * version-check both things. */ - if (block->instance_name != NULL - && !(state->language_version == 300 && state->es_shader)) { - _mesa_glsl_error(& @1, state, - "#version 300 es required for using uniform " - "blocks with an instance name\n"); + if (block->instance_name != NULL) { + state->check_version(150, 300, & @1, "interface blocks with " + "an instance name are not allowed"); + } + + unsigned interface_type_mask; + struct ast_type_qualifier temp_type_qualifier; + + /* Get a bitmask containing only the in/out/uniform flags, allowing us + * to ignore other irrelevant flags like interpolation qualifiers. + */ + temp_type_qualifier.flags.i = 0; + temp_type_qualifier.flags.q.uniform = true; + temp_type_qualifier.flags.q.in = true; + temp_type_qualifier.flags.q.out = true; + interface_type_mask = temp_type_qualifier.flags.i; + + /* Get the block's interface qualifier. The interface_qualifier + * production rule guarantees that only one bit will be set (and + * it will be in/out/uniform). + */ + unsigned block_interface_qualifier = $1.flags.i; + + block->layout.flags.i |= block_interface_qualifier; + + foreach_list_typed (ast_declarator_list, member, link, &block->declarations) { + ast_type_qualifier& qualifier = member->type->qualifier; + if ((qualifier.flags.i & interface_type_mask) == 0) { + /* GLSLangSpec.1.50.11, 4.3.7 (Interface Blocks): + * "If no optional qualifier is used in a member declaration, the + * qualifier of the variable is just in, out, or uniform as declared + * by interface-qualifier." + */ + qualifier.flags.i |= block_interface_qualifier; + } else if ((qualifier.flags.i & interface_type_mask) != + block_interface_qualifier) { + /* GLSLangSpec.1.50.11, 4.3.7 (Interface Blocks): + * "If optional qualifiers are used, they can include interpolation + * and storage qualifiers and they must declare an input, output, + * or uniform variable consistent with the interface qualifier of + * the block." + */ + _mesa_glsl_error(& @1, state, + "uniform/in/out qualifier on " + "interface block member does not match " + "the interface block\n"); + } } $$ = block; } ; +interface_qualifier: + IN_TOK + { + memset(& $$, 0, sizeof($$)); + $$.flags.q.in = 1; + } + | OUT_TOK + { + memset(& $$, 0, sizeof($$)); + $$.flags.q.out = 1; + } + | UNIFORM + { + memset(& $$, 0, sizeof($$)); + $$.flags.q.uniform = 1; + } + ; + instance_name_opt: /* empty */ { - $$ = new(state) ast_uniform_block(*state->default_uniform_qualifier, + $$ = new(state) ast_interface_block(*state->default_uniform_qualifier, NULL, NULL); } | NEW_IDENTIFIER { - $$ = new(state) ast_uniform_block(*state->default_uniform_qualifier, + $$ = new(state) ast_interface_block(*state->default_uniform_qualifier, $1, NULL); } | NEW_IDENTIFIER '[' constant_expression ']' { - $$ = new(state) ast_uniform_block(*state->default_uniform_qualifier, + $$ = new(state) ast_interface_block(*state->default_uniform_qualifier, $1, $3); } @@ -1965,7 +2048,7 @@ instance_name_opt: _mesa_glsl_error(& @1, state, "instance block arrays must be explicitly sized\n"); - $$ = new(state) ast_uniform_block(*state->default_uniform_qualifier, + $$ = new(state) ast_interface_block(*state->default_uniform_qualifier, $1, NULL); } @@ -1984,41 +2067,28 @@ member_list: } ; -/* Specifying "uniform" inside of a uniform block is redundant. */ -uniformopt: - /* nothing */ - | UNIFORM - ; - member_declaration: - layout_qualifier uniformopt type_specifier struct_declarator_list ';' + fully_specified_type struct_declarator_list ';' { void *ctx = state; - ast_fully_specified_type *type = new(ctx) ast_fully_specified_type(); + ast_fully_specified_type *type = $1; type->set_location(yylloc); - type->qualifier = $1; - type->qualifier.flags.q.uniform = true; - type->specifier = $3; - $$ = new(ctx) ast_declarator_list(type); - $$->set_location(yylloc); - $$->ubo_qualifiers_valid = true; - - $$->declarations.push_degenerate_list_at_head(& $4->link); - } - | uniformopt type_specifier struct_declarator_list ';' - { - void *ctx = state; - ast_fully_specified_type *type = new(ctx) ast_fully_specified_type(); - type->set_location(yylloc); + if (type->qualifier.flags.q.attribute) { + _mesa_glsl_error(& @1, state, + "keyword 'attribute' cannot be used with " + "interface block member\n"); + } else if (type->qualifier.flags.q.varying) { + _mesa_glsl_error(& @1, state, + "keyword 'varying' cannot be used with " + "interface block member\n"); + } - type->qualifier.flags.q.uniform = true; - type->specifier = $2; $$ = new(ctx) ast_declarator_list(type); $$->set_location(yylloc); $$->ubo_qualifiers_valid = true; - $$->declarations.push_degenerate_list_at_head(& $3->link); + $$->declarations.push_degenerate_list_at_head(& $2->link); } ; diff --git a/mesalib/src/glsl/glsl_parser_extras.cpp b/mesalib/src/glsl/glsl_parser_extras.cpp index 099229410..c0dd71370 100644 --- a/mesalib/src/glsl/glsl_parser_extras.cpp +++ b/mesalib/src/glsl/glsl_parser_extras.cpp @@ -93,9 +93,9 @@ _mesa_glsl_parse_state::_mesa_glsl_parse_state(struct gl_context *_ctx, this->Const.MaxVertexAttribs = ctx->Const.VertexProgram.MaxAttribs; this->Const.MaxVertexUniformComponents = ctx->Const.VertexProgram.MaxUniformComponents; this->Const.MaxVaryingFloats = ctx->Const.MaxVarying * 4; - this->Const.MaxVertexTextureImageUnits = ctx->Const.MaxVertexTextureImageUnits; + this->Const.MaxVertexTextureImageUnits = ctx->Const.VertexProgram.MaxTextureImageUnits; this->Const.MaxCombinedTextureImageUnits = ctx->Const.MaxCombinedTextureImageUnits; - this->Const.MaxTextureImageUnits = ctx->Const.MaxTextureImageUnits; + this->Const.MaxTextureImageUnits = ctx->Const.FragmentProgram.MaxTextureImageUnits; this->Const.MaxFragmentUniformComponents = ctx->Const.FragmentProgram.MaxUniformComponents; this->Const.MinProgramTexelOffset = ctx->Const.MinProgramTexelOffset; this->Const.MaxProgramTexelOffset = ctx->Const.MaxProgramTexelOffset; @@ -468,6 +468,8 @@ static const _mesa_glsl_extension _mesa_glsl_supported_extensions[] = { EXT(ARB_shading_language_packing, true, false, true, true, false, ARB_shading_language_packing), EXT(ARB_texture_multisample, true, false, true, true, false, ARB_texture_multisample), EXT(ARB_texture_query_lod, false, false, true, true, false, ARB_texture_query_lod), + EXT(ARB_gpu_shader5, true, true, true, true, false, ARB_gpu_shader5), + EXT(AMD_vertex_shader_layer, true, false, false, true, false, AMD_vertex_shader_layer), }; #undef EXT @@ -1202,11 +1204,13 @@ ast_struct_specifier::ast_struct_specifier(const char *identifier, * \param max_unroll_iterations Maximum number of loop iterations to be * unrolled. Setting to 0 disables loop * unrolling. + * \param options The driver's preferred shader options. */ bool do_common_optimization(exec_list *ir, bool linked, bool uniform_locations_assigned, - unsigned max_unroll_iterations) + unsigned max_unroll_iterations, + const struct gl_shader_compiler_options *options) { GLboolean progress = GL_FALSE; @@ -1221,6 +1225,10 @@ do_common_optimization(exec_list *ir, bool linked, progress = opt_flatten_nested_if_blocks(ir) || progress; progress = do_copy_propagation(ir) || progress; progress = do_copy_propagation_elements(ir) || progress; + + if (options->PreferDP4 && !linked) + progress = opt_flip_matrices(ir) || progress; + if (linked) progress = do_dead_code(ir, uniform_locations_assigned) || progress; else @@ -1236,6 +1244,7 @@ do_common_optimization(exec_list *ir, bool linked, progress = do_algebraic(ir) || progress; progress = do_lower_jumps(ir) || progress; progress = do_vec_index_to_swizzle(ir) || progress; + progress = lower_vector_insert(ir, false) || progress; progress = do_swizzle_swizzle(ir) || progress; progress = do_noop_swizzle(ir) || progress; diff --git a/mesalib/src/glsl/glsl_parser_extras.h b/mesalib/src/glsl/glsl_parser_extras.h index 37a47cf9a..89c1a565e 100644 --- a/mesalib/src/glsl/glsl_parser_extras.h +++ b/mesalib/src/glsl/glsl_parser_extras.h @@ -288,6 +288,10 @@ struct _mesa_glsl_parse_state { bool ARB_texture_multisample_warn; bool ARB_texture_query_lod_enable; bool ARB_texture_query_lod_warn; + bool ARB_gpu_shader5_enable; + bool ARB_gpu_shader5_warn; + bool AMD_vertex_shader_layer_enable; + bool AMD_vertex_shader_layer_warn; /*@}*/ /** Extensions supported by the OpenGL implementation. */ diff --git a/mesalib/src/glsl/glsl_symbol_table.cpp b/mesalib/src/glsl/glsl_symbol_table.cpp index d303430bb..dcc3a1e02 100644 --- a/mesalib/src/glsl/glsl_symbol_table.cpp +++ b/mesalib/src/glsl/glsl_symbol_table.cpp @@ -45,13 +45,67 @@ public: ralloc_free(entry); } - symbol_table_entry(ir_variable *v) : v(v), f(0), t(0) {} - symbol_table_entry(ir_function *f) : v(0), f(f), t(0) {} - symbol_table_entry(const glsl_type *t) : v(0), f(0), t(t) {} + bool add_interface(const glsl_type *i, enum ir_variable_mode mode) + { + const glsl_type **dest; + + switch (mode) { + case ir_var_uniform: + dest = &ibu; + break; + case ir_var_shader_in: + dest = &ibi; + break; + case ir_var_shader_out: + dest = &ibo; + break; + default: + assert(!"Unsupported interface variable mode!"); + return false; + } + + if (*dest != NULL) { + return false; + } else { + *dest = i; + return true; + } + } + + const glsl_type *get_interface(enum ir_variable_mode mode) + { + switch (mode) { + case ir_var_uniform: + return ibu; + case ir_var_shader_in: + return ibi; + case ir_var_shader_out: + return ibo; + default: + assert(!"Unsupported interface variable mode!"); + return NULL; + } + } + + symbol_table_entry(ir_variable *v) : + v(v), f(0), t(0), ibu(0), ibi(0), ibo(0) {} + symbol_table_entry(ir_function *f) : + v(0), f(f), t(0), ibu(0), ibi(0), ibo(0) {} + symbol_table_entry(const glsl_type *t) : + v(0), f(0), t(t), ibu(0), ibi(0), ibo(0) {} + symbol_table_entry(const glsl_type *t, enum ir_variable_mode mode) : + v(0), f(0), t(0), ibu(0), ibi(0), ibo(0) + { + assert(t->is_interface()); + add_interface(t, mode); + } ir_variable *v; ir_function *f; const glsl_type *t; + const glsl_type *ibu; + const glsl_type *ibi; + const glsl_type *ibo; }; glsl_symbol_table::glsl_symbol_table() @@ -122,6 +176,23 @@ bool glsl_symbol_table::add_type(const char *name, const glsl_type *t) return _mesa_symbol_table_add_symbol(table, -1, name, entry) == 0; } +bool glsl_symbol_table::add_interface(const char *name, const glsl_type *i, + enum ir_variable_mode mode) +{ + assert(i->is_interface()); + symbol_table_entry *entry = get_entry(name); + if (entry == NULL) { + symbol_table_entry *entry = + new(mem_ctx) symbol_table_entry(i, mode); + bool add_interface_symbol_result = + _mesa_symbol_table_add_symbol(table, -1, name, entry) == 0; + assert(add_interface_symbol_result); + return add_interface_symbol_result; + } else { + return entry->add_interface(i, mode); + } +} + bool glsl_symbol_table::add_function(ir_function *f) { if (this->separate_function_namespace && name_declared_this_scope(f->name)) { @@ -156,6 +227,13 @@ const glsl_type *glsl_symbol_table::get_type(const char *name) return entry != NULL ? entry->t : NULL; } +const glsl_type *glsl_symbol_table::get_interface(const char *name, + enum ir_variable_mode mode) +{ + symbol_table_entry *entry = get_entry(name); + return entry != NULL ? entry->get_interface(mode) : NULL; +} + ir_function *glsl_symbol_table::get_function(const char *name) { symbol_table_entry *entry = get_entry(name); diff --git a/mesalib/src/glsl/glsl_symbol_table.h b/mesalib/src/glsl/glsl_symbol_table.h index 55baebf10..98084c990 100644 --- a/mesalib/src/glsl/glsl_symbol_table.h +++ b/mesalib/src/glsl/glsl_symbol_table.h @@ -104,6 +104,8 @@ public: bool add_variable(ir_variable *v); bool add_type(const char *name, const glsl_type *t); bool add_function(ir_function *f); + bool add_interface(const char *name, const glsl_type *i, + enum ir_variable_mode mode); /*@}*/ /** @@ -118,6 +120,8 @@ public: ir_variable *get_variable(const char *name); const glsl_type *get_type(const char *name); ir_function *get_function(const char *name); + const glsl_type *get_interface(const char *name, + enum ir_variable_mode mode); /*@}*/ private: diff --git a/mesalib/src/glsl/glsl_types.cpp b/mesalib/src/glsl/glsl_types.cpp index 419761a7d..df9c5d36f 100644 --- a/mesalib/src/glsl/glsl_types.cpp +++ b/mesalib/src/glsl/glsl_types.cpp @@ -57,6 +57,7 @@ glsl_type::glsl_type(GLenum gl_type, length(0) { init_ralloc_type_ctx(); + assert(name != NULL); this->name = ralloc_strdup(this->mem_ctx, name); /* Neither dimension is zero or both dimensions are zero. */ @@ -75,6 +76,7 @@ glsl_type::glsl_type(GLenum gl_type, length(0) { init_ralloc_type_ctx(); + assert(name != NULL); this->name = ralloc_strdup(this->mem_ctx, name); memset(& fields, 0, sizeof(fields)); } @@ -91,6 +93,7 @@ glsl_type::glsl_type(const glsl_struct_field *fields, unsigned num_fields, unsigned int i; init_ralloc_type_ctx(); + assert(name != NULL); this->name = ralloc_strdup(this->mem_ctx, name); this->fields.structure = ralloc_array(this->mem_ctx, glsl_struct_field, length); @@ -114,6 +117,7 @@ glsl_type::glsl_type(const glsl_struct_field *fields, unsigned num_fields, unsigned int i; init_ralloc_type_ctx(); + assert(name != NULL); this->name = ralloc_strdup(this->mem_ctx, name); this->fields.structure = ralloc_array(this->mem_ctx, glsl_struct_field, length); diff --git a/mesalib/src/glsl/glsl_types.h b/mesalib/src/glsl/glsl_types.h index 2f3b19f51..31e3dd253 100644 --- a/mesalib/src/glsl/glsl_types.h +++ b/mesalib/src/glsl/glsl_types.h @@ -132,8 +132,7 @@ struct glsl_type { /** * Name of the data type * - * This may be \c NULL for anonymous structures, for arrays, or for - * function types. + * Will never be \c NULL. */ const char *name; diff --git a/mesalib/src/glsl/ir.cpp b/mesalib/src/glsl/ir.cpp index 05b77da2c..dad58deeb 100644 --- a/mesalib/src/glsl/ir.cpp +++ b/mesalib/src/glsl/ir.cpp @@ -399,6 +399,10 @@ ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1) this->type = op0->type; break; + case ir_binop_vector_extract: + this->type = op0->type->get_scalar_type(); + break; + default: assert(!"not reached: missing automatic type setup for ir_expression"); this->type = glsl_type::float_type; @@ -419,7 +423,7 @@ ir_expression::get_num_operands(ir_expression_operation op) if (op <= ir_last_triop) return 3; - if (op == ir_quadop_vector) + if (op <= ir_last_quadop) return 4; assert(false); @@ -477,6 +481,10 @@ static const char *const operator_strs[] = { "unpackHalf2x16", "unpackHalf2x16_split_x", "unpackHalf2x16_split_y", + "bitfield_reverse", + "bit_count", + "find_msb", + "find_lsb", "noise", "+", "-", @@ -504,8 +512,14 @@ static const char *const operator_strs[] = { "max", "pow", "packHalf2x16_split", + "bfm", "ubo_load", + "vector_extract", "lrp", + "bfi", + "bitfield_extract", + "vector_insert", + "bitfield_insert", "vector", }; diff --git a/mesalib/src/glsl/ir.h b/mesalib/src/glsl/ir.h index 0c3e39979..6d4150136 100644 --- a/mesalib/src/glsl/ir.h +++ b/mesalib/src/glsl/ir.h @@ -131,6 +131,7 @@ public: virtual class ir_swizzle * as_swizzle() { return NULL; } virtual class ir_constant * as_constant() { return NULL; } virtual class ir_discard * as_discard() { return NULL; } + virtual class ir_jump * as_jump() { return NULL; } /*@}*/ protected: @@ -273,7 +274,8 @@ enum ir_variable_mode { ir_var_function_inout, ir_var_const_in, /**< "in" param that must be a constant expression */ ir_var_system_value, /**< Ex: front-face, instance-id, etc. */ - ir_var_temporary /**< Temporary variable generated during compilation. */ + ir_var_temporary, /**< Temporary variable generated during compilation. */ + ir_var_mode_count /**< Number of variable modes */ }; /** @@ -1031,6 +1033,16 @@ enum ir_expression_operation { ir_unop_unpack_half_2x16_split_y, /*@}*/ + /** + * \name Bit operations, part of ARB_gpu_shader5. + */ + /*@{*/ + ir_unop_bitfield_reverse, + ir_unop_bit_count, + ir_unop_find_msb, + ir_unop_find_lsb, + /*@}*/ + ir_unop_noise, /** @@ -1107,6 +1119,15 @@ enum ir_expression_operation { /*@}*/ /** + * \name First half of a lowered bitfieldInsert() operation. + * + * \see lower_instructions::bitfield_insert_to_bfm_bfi + */ + /*@{*/ + ir_binop_bfm, + /*@}*/ + + /** * Load a value the size of a given GLSL type from a uniform block. * * operand0 is the ir_constant uniform block index in the linked shader. @@ -1115,20 +1136,55 @@ enum ir_expression_operation { ir_binop_ubo_load, /** + * Extract a scalar from a vector + * + * operand0 is the vector + * operand1 is the index of the field to read from operand0 + */ + ir_binop_vector_extract, + + /** * A sentinel marking the last of the binary operations. */ - ir_last_binop = ir_binop_ubo_load, + ir_last_binop = ir_binop_vector_extract, ir_triop_lrp, /** + * \name Second half of a lowered bitfieldInsert() operation. + * + * \see lower_instructions::bitfield_insert_to_bfm_bfi + */ + /*@{*/ + ir_triop_bfi, + /*@}*/ + + ir_triop_bitfield_extract, + + /** + * Generate a value with one field of a vector changed + * + * operand0 is the vector + * operand1 is the value to write into the vector result + * operand2 is the index in operand0 to be modified + */ + ir_triop_vector_insert, + + /** * A sentinel marking the last of the ternary operations. */ - ir_last_triop = ir_triop_lrp, + ir_last_triop = ir_triop_vector_insert, + + ir_quadop_bitfield_insert, ir_quadop_vector, /** + * A sentinel marking the last of the ternary operations. + */ + ir_last_quadop = ir_quadop_vector, + + /** * A sentinel marking the last of all operations. */ ir_last_opcode = ir_quadop_vector @@ -1297,6 +1353,12 @@ protected: { ir_type = ir_type_unset; } + +public: + virtual ir_jump *as_jump() + { + return this; + } }; class ir_return : public ir_jump { diff --git a/mesalib/src/glsl/ir_basic_block.cpp b/mesalib/src/glsl/ir_basic_block.cpp index 86e0cf795..2cbc682d4 100644 --- a/mesalib/src/glsl/ir_basic_block.cpp +++ b/mesalib/src/glsl/ir_basic_block.cpp @@ -77,7 +77,7 @@ void call_for_basic_blocks(exec_list *instructions, callback(leader, ir, data); leader = NULL; call_for_basic_blocks(&ir_loop->body_instructions, callback, data); - } else if (ir->as_return() || ir->as_call()) { + } else if (ir->as_jump() || ir->as_call()) { callback(leader, ir, data); leader = NULL; } else if ((ir_function = ir->as_function())) { diff --git a/mesalib/src/glsl/ir_constant_expression.cpp b/mesalib/src/glsl/ir_constant_expression.cpp index c09e56a3d..0a725b45b 100644 --- a/mesalib/src/glsl/ir_constant_expression.cpp +++ b/mesalib/src/glsl/ir_constant_expression.cpp @@ -391,9 +391,17 @@ ir_expression::constant_expression_value(struct hash_table *variable_context) } if (op[1] != NULL) - assert(op[0]->type->base_type == op[1]->type->base_type || - this->operation == ir_binop_lshift || - this->operation == ir_binop_rshift); + switch (this->operation) { + case ir_binop_lshift: + case ir_binop_rshift: + case ir_binop_vector_extract: + case ir_triop_bitfield_extract: + break; + + default: + assert(op[0]->type->base_type == op[1]->type->base_type); + break; + } bool op0_scalar = op[0]->type->is_scalar(); bool op1_scalar = op[1] != NULL && op[1]->type->is_scalar(); @@ -1230,6 +1238,29 @@ ir_expression::constant_expression_value(struct hash_table *variable_context) } break; + case ir_binop_vector_extract: { + const int c = CLAMP(op[1]->value.i[0], 0, + (int) op[0]->type->vector_elements - 1); + + switch (op[0]->type->base_type) { + case GLSL_TYPE_UINT: + data.u[0] = op[0]->value.u[c]; + break; + case GLSL_TYPE_INT: + data.i[0] = op[0]->value.i[c]; + break; + case GLSL_TYPE_FLOAT: + data.f[0] = op[0]->value.f[c]; + break; + case GLSL_TYPE_BOOL: + data.b[0] = op[0]->value.b[c]; + break; + default: + assert(0); + } + break; + } + case ir_binop_bit_xor: for (unsigned c = 0, c0 = 0, c1 = 0; c < components; @@ -1248,6 +1279,102 @@ ir_expression::constant_expression_value(struct hash_table *variable_context) } break; + case ir_unop_bitfield_reverse: + /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */ + for (unsigned c = 0; c < components; c++) { + unsigned int v = op[0]->value.u[c]; // input bits to be reversed + unsigned int r = v; // r will be reversed bits of v; first get LSB of v + int s = sizeof(v) * CHAR_BIT - 1; // extra shift needed at end + + for (v >>= 1; v; v >>= 1) { + r <<= 1; + r |= v & 1; + s--; + } + r <<= s; // shift when v's highest bits are zero + + data.u[c] = r; + } + break; + + case ir_unop_bit_count: + for (unsigned c = 0; c < components; c++) { + unsigned count = 0; + unsigned v = op[0]->value.u[c]; + + for (; v; count++) { + v &= v - 1; + } + data.u[c] = count; + } + break; + + case ir_unop_find_msb: + for (unsigned c = 0; c < components; c++) { + int v = op[0]->value.i[c]; + + if (v == 0 || (op[0]->type->base_type == GLSL_TYPE_INT && v == -1)) + data.i[c] = -1; + else { + int count = 0; + int top_bit = op[0]->type->base_type == GLSL_TYPE_UINT + ? 0 : v & (1 << 31); + + while (((v & (1 << 31)) == top_bit) && count != 32) { + count++; + v <<= 1; + } + + data.i[c] = 31 - count; + } + } + break; + + case ir_unop_find_lsb: + for (unsigned c = 0; c < components; c++) { + if (op[0]->value.i[c] == 0) + data.i[c] = -1; + else { + unsigned pos = 0; + unsigned v = op[0]->value.u[c]; + + for (; !(v & 1); v >>= 1) { + pos++; + } + data.u[c] = pos; + } + } + break; + + case ir_triop_bitfield_extract: { + int offset = op[1]->value.i[0]; + int bits = op[2]->value.i[0]; + + for (unsigned c = 0; c < components; c++) { + if (bits == 0) + data.u[c] = 0; + else if (offset < 0 || bits < 0) + data.u[c] = 0; /* Undefined, per spec. */ + else if (offset + bits > 32) + data.u[c] = 0; /* Undefined, per spec. */ + else { + if (op[0]->type->base_type == GLSL_TYPE_INT) { + /* int so that the right shift will sign-extend. */ + int value = op[0]->value.i[c]; + value <<= 32 - bits - offset; + value >>= 32 - bits; + data.i[c] = value; + } else { + unsigned value = op[0]->value.u[c]; + value <<= 32 - bits - offset; + value >>= 32 - bits; + data.u[c] = value; + } + } + } + break; + } + case ir_triop_lrp: { assert(op[0]->type->base_type == GLSL_TYPE_FLOAT); assert(op[1]->type->base_type == GLSL_TYPE_FLOAT); @@ -1261,6 +1388,58 @@ ir_expression::constant_expression_value(struct hash_table *variable_context) break; } + case ir_triop_vector_insert: { + const unsigned idx = op[2]->value.u[0]; + + memcpy(&data, &op[0]->value, sizeof(data)); + + switch (this->type->base_type) { + case GLSL_TYPE_INT: + data.i[idx] = op[1]->value.i[0]; + break; + case GLSL_TYPE_UINT: + data.u[idx] = op[1]->value.u[0]; + break; + case GLSL_TYPE_FLOAT: + data.f[idx] = op[1]->value.f[0]; + break; + case GLSL_TYPE_BOOL: + data.b[idx] = op[1]->value.b[0]; + break; + default: + assert(!"Should not get here."); + break; + } + break; + } + + case ir_quadop_bitfield_insert: { + int offset = op[2]->value.i[0]; + int bits = op[3]->value.i[0]; + + for (unsigned c = 0; c < components; c++) { + if (bits == 0) + data.u[c] = op[0]->value.u[c]; + else if (offset < 0 || bits < 0) + data.u[c] = 0; /* Undefined, per spec. */ + else if (offset + bits > 32) + data.u[c] = 0; /* Undefined, per spec. */ + else { + unsigned insert_mask = ((1 << bits) - 1) << offset; + + unsigned insert = op[1]->value.u[c]; + insert <<= offset; + insert &= insert_mask; + + unsigned base = op[0]->value.u[c]; + base &= ~insert_mask; + + data.u[c] = base | insert; + } + } + break; + } + case ir_quadop_vector: for (unsigned c = 0; c < this->type->vector_elements; c++) { switch (this->type->base_type) { diff --git a/mesalib/src/glsl/ir_optimization.h b/mesalib/src/glsl/ir_optimization.h index a8885d722..d38d5e303 100644 --- a/mesalib/src/glsl/ir_optimization.h +++ b/mesalib/src/glsl/ir_optimization.h @@ -37,6 +37,7 @@ #define MOD_TO_FRACT 0x20 #define INT_DIV_TO_MUL_RCP 0x40 #define LRP_TO_ARITH 0x80 +#define BITFIELD_INSERT_TO_BFM_BFI 0x100 /** * \see class lower_packing_builtins_visitor @@ -65,7 +66,8 @@ enum lower_packing_builtins_op { bool do_common_optimization(exec_list *ir, bool linked, bool uniform_locations_assigned, - unsigned max_unroll_iterations); + unsigned max_unroll_iterations, + const struct gl_shader_compiler_options *options); bool do_algebraic(exec_list *instructions); bool do_constant_folding(exec_list *instructions); @@ -78,6 +80,7 @@ bool do_dead_code(exec_list *instructions, bool uniform_locations_assigned); bool do_dead_code_local(exec_list *instructions); bool do_dead_code_unlinked(exec_list *instructions); bool do_dead_functions(exec_list *instructions); +bool opt_flip_matrices(exec_list *instructions); bool do_function_inlining(exec_list *instructions); bool do_lower_jumps(exec_list *instructions, bool pull_out_jumps = true, bool lower_sub_return = true, bool lower_main_return = false, bool lower_continue = false, bool lower_break = false); bool do_lower_texture_projection(exec_list *instructions); @@ -106,6 +109,8 @@ void lower_ubo_reference(struct gl_shader *shader, exec_list *instructions); void lower_packed_varyings(void *mem_ctx, unsigned location_base, unsigned locations_used, ir_variable_mode mode, gl_shader *shader); +bool lower_vector_insert(exec_list *instructions, bool lower_nonconstant_index); +void lower_named_interface_blocks(void *mem_ctx, gl_shader *shader); bool optimize_redundant_jumps(exec_list *instructions); bool optimize_split_arrays(exec_list *instructions, bool linked); diff --git a/mesalib/src/glsl/ir_print_visitor.cpp b/mesalib/src/glsl/ir_print_visitor.cpp index 597d2813f..f01019c98 100644 --- a/mesalib/src/glsl/ir_print_visitor.cpp +++ b/mesalib/src/glsl/ir_print_visitor.cpp @@ -24,6 +24,7 @@ #include "ir_print_visitor.h" #include "glsl_types.h" #include "glsl_parser_extras.h" +#include "main/macros.h" #include "program/hash_table.h" static void print_type(const glsl_type *t); @@ -149,7 +150,9 @@ void ir_print_visitor::visit(ir_variable *ir) const char *const mode[] = { "", "uniform ", "shader_in ", "shader_out ", "in ", "out ", "inout ", "const_in ", "sys ", "temporary " }; - const char *const interp[] = { "", "flat", "noperspective" }; + STATIC_ASSERT(ARRAY_SIZE(mode) == ir_var_mode_count); + const char *const interp[] = { "", "smooth", "flat", "noperspective" }; + STATIC_ASSERT(ARRAY_SIZE(interp) == INTERP_QUALIFIER_COUNT); printf("(%s%s%s%s) ", cent, inv, mode[ir->mode], interp[ir->interpolation]); diff --git a/mesalib/src/glsl/ir_reader.cpp b/mesalib/src/glsl/ir_reader.cpp index 16fdc41b4..b3667124f 100644 --- a/mesalib/src/glsl/ir_reader.cpp +++ b/mesalib/src/glsl/ir_reader.cpp @@ -676,16 +676,18 @@ ir_reader::read_expression(s_expression *expr) { s_expression *s_type; s_symbol *s_op; - s_expression *s_arg[3]; + s_expression *s_arg[4] = {NULL}; s_pattern pat[] = { "expression", s_type, s_op, s_arg[0] }; if (!PARTIAL_MATCH(expr, pat)) { ir_read_error(expr, "expected (expression <type> <operator> " - "<operand> [<operand>])"); + "<operand> [<operand>] [<operand>] [<operand>])"); return NULL; } s_arg[1] = (s_expression *) s_arg[0]->next; // may be tail sentinel s_arg[2] = (s_expression *) s_arg[1]->next; // may be tail sentinel or NULL + if (s_arg[2]) + s_arg[3] = (s_expression *) s_arg[2]->next; // may be tail sentinel or NULL const glsl_type *type = read_type(s_type); if (type == NULL) @@ -709,7 +711,7 @@ ir_reader::read_expression(s_expression *expr) return NULL; } - ir_rvalue *arg[3] = {NULL, NULL, NULL}; + ir_rvalue *arg[4] = {NULL}; for (int i = 0; i < num_operands; i++) { arg[i] = read_rvalue(s_arg[i]); if (arg[i] == NULL) { @@ -718,7 +720,7 @@ ir_reader::read_expression(s_expression *expr) } } - return new(mem_ctx) ir_expression(op, type, arg[0], arg[1], arg[2]); + return new(mem_ctx) ir_expression(op, type, arg[0], arg[1], arg[2], arg[3]); } ir_swizzle * diff --git a/mesalib/src/glsl/ir_uniform.h b/mesalib/src/glsl/ir_uniform.h index 30e6f260d..8198c4819 100644 --- a/mesalib/src/glsl/ir_uniform.h +++ b/mesalib/src/glsl/ir_uniform.h @@ -99,15 +99,24 @@ struct gl_uniform_storage { */ bool initialized; - /** - * Base sampler index - * - * If \c ::base_type is \c GLSL_TYPE_SAMPLER, this represents the index of - * this sampler. If \c ::array_elements is not zero, the array will use - * sampler indexes \c ::sampler through \c ::sampler + \c ::array_elements - * - 1, inclusive. - */ - uint8_t sampler; + struct { + /** + * Base sampler index + * + * If \c ::base_type is \c GLSL_TYPE_SAMPLER, this represents the index + * of this sampler. If \c ::array_elements is not zero, the array will + * use sampler indices \c ::sampler through \c ::sampler + + * \c ::array_elements - 1, inclusive. + * + * Note that the index may be different in each shader stage. + */ + uint8_t index; + + /** + * Whether this sampler is used in this shader stage. + */ + bool active; + } sampler[MESA_SHADER_TYPES]; /** * Storage used by the driver for the uniform diff --git a/mesalib/src/glsl/ir_validate.cpp b/mesalib/src/glsl/ir_validate.cpp index 699c192cd..ce96f6855 100644 --- a/mesalib/src/glsl/ir_validate.cpp +++ b/mesalib/src/glsl/ir_validate.cpp @@ -69,6 +69,8 @@ public: virtual ir_visitor_status visit_leave(ir_expression *ir); virtual ir_visitor_status visit_leave(ir_swizzle *ir); + virtual ir_visitor_status visit_enter(class ir_dereference_array *); + virtual ir_visitor_status visit_enter(ir_assignment *ir); virtual ir_visitor_status visit_enter(ir_call *ir); @@ -102,6 +104,33 @@ ir_validate::visit(ir_dereference_variable *ir) } ir_visitor_status +ir_validate::visit_enter(class ir_dereference_array *ir) +{ + if (!ir->array->type->is_array() && !ir->array->type->is_matrix()) { + printf("ir_dereference_array @ %p does not specify an array or a " + "matrix\n", + (void *) ir); + ir->print(); + printf("\n"); + abort(); + } + + if (!ir->array_index->type->is_scalar()) { + printf("ir_dereference_array @ %p does not have scalar index: %s\n", + (void *) ir, ir->array_index->type->name); + abort(); + } + + if (!ir->array_index->type->is_integer()) { + printf("ir_dereference_array @ %p does not have integer index: %s\n", + (void *) ir, ir->array_index->type->name); + abort(); + } + + return visit_continue; +} + +ir_visitor_status ir_validate::visit_enter(ir_if *ir) { if (ir->condition->type != glsl_type::bool_type) { @@ -361,6 +390,19 @@ ir_validate::visit_leave(ir_expression *ir) assert(ir->operands[0]->type == glsl_type::uint_type); break; + case ir_unop_bitfield_reverse: + assert(ir->operands[0]->type == ir->type); + assert(ir->type->is_integer()); + break; + + case ir_unop_bit_count: + case ir_unop_find_msb: + case ir_unop_find_lsb: + assert(ir->operands[0]->type->vector_elements == ir->type->vector_elements); + assert(ir->operands[0]->type->is_integer()); + assert(ir->type->base_type == GLSL_TYPE_INT); + break; + case ir_unop_noise: /* XXX what can we assert here? */ break; @@ -461,6 +503,12 @@ ir_validate::visit_leave(ir_expression *ir) assert(ir->operands[1]->type == glsl_type::float_type); break; + case ir_binop_bfm: + assert(ir->type->is_integer()); + assert(ir->operands[0]->type->is_integer()); + assert(ir->operands[1]->type->is_integer()); + break; + case ir_binop_ubo_load: assert(ir->operands[0]->as_constant()); assert(ir->operands[0]->type == glsl_type::uint_type); @@ -468,12 +516,46 @@ ir_validate::visit_leave(ir_expression *ir) assert(ir->operands[1]->type == glsl_type::uint_type); break; + case ir_binop_vector_extract: + assert(ir->operands[0]->type->is_vector()); + assert(ir->operands[1]->type->is_scalar() + && ir->operands[1]->type->is_integer()); + break; + case ir_triop_lrp: assert(ir->operands[0]->type->base_type == GLSL_TYPE_FLOAT); assert(ir->operands[0]->type == ir->operands[1]->type); assert(ir->operands[2]->type == ir->operands[0]->type || ir->operands[2]->type == glsl_type::float_type); break; + case ir_triop_bfi: + assert(ir->operands[0]->type->is_integer()); + assert(ir->operands[1]->type == ir->operands[2]->type); + assert(ir->operands[1]->type == ir->type); + break; + + case ir_triop_bitfield_extract: + assert(ir->operands[0]->type == ir->type); + assert(ir->operands[1]->type == glsl_type::int_type); + assert(ir->operands[2]->type == glsl_type::int_type); + break; + + case ir_triop_vector_insert: + assert(ir->operands[0]->type->is_vector()); + assert(ir->operands[1]->type->is_scalar()); + assert(ir->operands[0]->type->base_type == ir->operands[1]->type->base_type); + assert(ir->operands[2]->type->is_scalar() + && ir->operands[2]->type->is_integer()); + assert(ir->type == ir->operands[0]->type); + break; + + case ir_quadop_bitfield_insert: + assert(ir->operands[0]->type == ir->type); + assert(ir->operands[1]->type == ir->type); + assert(ir->operands[2]->type == glsl_type::int_type); + assert(ir->operands[3]->type == glsl_type::int_type); + break; + case ir_quadop_vector: /* The vector operator collects some number of scalars and generates a * vector from them. diff --git a/mesalib/src/glsl/link_interface_blocks.cpp b/mesalib/src/glsl/link_interface_blocks.cpp new file mode 100644 index 000000000..b91860d03 --- /dev/null +++ b/mesalib/src/glsl/link_interface_blocks.cpp @@ -0,0 +1,110 @@ +/* + * Copyright © 2013 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_interface_blocks.cpp + * Linker support for GLSL's interface blocks. + */ + +#include "ir.h" +#include "glsl_symbol_table.h" +#include "linker.h" +#include "main/macros.h" + +bool +validate_intrastage_interface_blocks(const gl_shader **shader_list, + unsigned num_shaders) +{ + glsl_symbol_table interfaces; + + for (unsigned int i = 0; i < num_shaders; i++) { + if (shader_list[i] == NULL) + continue; + + foreach_list(node, shader_list[i]->ir) { + ir_variable *var = ((ir_instruction *) node)->as_variable(); + if (!var) + continue; + + const glsl_type *iface_type = var->interface_type; + + if (iface_type == NULL) + continue; + + const glsl_type *old_iface_type = + interfaces.get_interface(iface_type->name, + (enum ir_variable_mode) var->mode); + + if (old_iface_type == NULL) { + /* This is the first time we've seen the interface, so save + * it into our symbol table. + */ + interfaces.add_interface(iface_type->name, iface_type, + (enum ir_variable_mode) var->mode); + } else if (old_iface_type != iface_type) { + return false; + } + } + } + + return true; +} + +bool +validate_interstage_interface_blocks(const gl_shader *producer, + const gl_shader *consumer) +{ + glsl_symbol_table interfaces; + + /* Add non-output interfaces from the consumer to the symbol table. */ + foreach_list(node, consumer->ir) { + ir_variable *var = ((ir_instruction *) node)->as_variable(); + if (!var || !var->interface_type || var->mode == ir_var_shader_out) + continue; + + interfaces.add_interface(var->interface_type->name, + var->interface_type, + (enum ir_variable_mode) var->mode); + } + + /* Verify that the producer's interfaces match. */ + foreach_list(node, producer->ir) { + ir_variable *var = ((ir_instruction *) node)->as_variable(); + if (!var || !var->interface_type || var->mode == ir_var_shader_in) + continue; + + enum ir_variable_mode consumer_mode = + var->mode == ir_var_uniform ? ir_var_uniform : ir_var_shader_in; + const glsl_type *expected_type = + interfaces.get_interface(var->interface_type->name, consumer_mode); + + /* The consumer doesn't use this output block. Ignore it. */ + if (expected_type == NULL) + continue; + + if (var->interface_type != expected_type) + return false; + } + + return true; +} diff --git a/mesalib/src/glsl/link_uniform_initializers.cpp b/mesalib/src/glsl/link_uniform_initializers.cpp index 836a360fa..54d9bf1f5 100644 --- a/mesalib/src/glsl/link_uniform_initializers.cpp +++ b/mesalib/src/glsl/link_uniform_initializers.cpp @@ -138,8 +138,16 @@ set_uniform_initializer(void *mem_ctx, gl_shader_program *prog, } if (base_type == GLSL_TYPE_SAMPLER) { - for (unsigned int i = 0; i < storage->array_elements; i++) { - prog->SamplerUnits[storage->sampler + i] = storage->storage[i].i; + for (int sh = 0; sh < MESA_SHADER_TYPES; sh++) { + gl_shader *shader = prog->_LinkedShaders[sh]; + + if (shader && storage->sampler[sh].active) { + for (unsigned i = 0; i < storage->array_elements; i++) { + unsigned index = storage->sampler[sh].index + i; + + shader->SamplerUnits[index] = storage->storage[i].i; + } + } } } } else { @@ -148,8 +156,17 @@ set_uniform_initializer(void *mem_ctx, gl_shader_program *prog, val->type->base_type, val->type->components()); - if (storage->type->is_sampler()) - prog->SamplerUnits[storage->sampler] = storage->storage[0].i; + if (storage->type->is_sampler()) { + for (int sh = 0; sh < MESA_SHADER_TYPES; sh++) { + gl_shader *shader = prog->_LinkedShaders[sh]; + + if (shader && storage->sampler[sh].active) { + unsigned index = storage->sampler[sh].index; + + shader->SamplerUnits[index] = storage->storage[0].i; + } + } + } } storage->initialized = true; diff --git a/mesalib/src/glsl/link_uniforms.cpp b/mesalib/src/glsl/link_uniforms.cpp index efbb60c3a..65c06903a 100644 --- a/mesalib/src/glsl/link_uniforms.cpp +++ b/mesalib/src/glsl/link_uniforms.cpp @@ -263,15 +263,19 @@ public: parcel_out_uniform_storage(struct string_to_uint_map *map, struct gl_uniform_storage *uniforms, union gl_constant_value *values) - : map(map), uniforms(uniforms), next_sampler(0), values(values) + : map(map), uniforms(uniforms), values(values) { - memset(this->targets, 0, sizeof(this->targets)); } - void start_shader() + void start_shader(gl_shader_type shader_type) { + assert(shader_type < MESA_SHADER_TYPES); + this->shader_type = shader_type; + this->shader_samplers_used = 0; this->shader_shadow_samplers = 0; + this->next_sampler = 0; + memset(this->targets, 0, sizeof(this->targets)); } void set_and_process(struct gl_shader_program *prog, @@ -335,8 +339,37 @@ public: int ubo_block_index; int ubo_byte_offset; bool ubo_row_major; + gl_shader_type shader_type; private: + void handle_samplers(const glsl_type *base_type, + struct gl_uniform_storage *uniform) + { + if (base_type->is_sampler()) { + uniform->sampler[shader_type].index = this->next_sampler; + uniform->sampler[shader_type].active = true; + + /* Increment the sampler by 1 for non-arrays and by the number of + * array elements for arrays. + */ + this->next_sampler += + MAX2(1, uniform->array_elements); + + const gl_texture_index target = base_type->sampler_index(); + const unsigned shadow = base_type->sampler_shadow; + for (unsigned i = uniform->sampler[shader_type].index; + i < MIN2(this->next_sampler, MAX_SAMPLERS); + i++) { + this->targets[i] = target; + this->shader_samplers_used |= 1U << i; + this->shader_shadow_samplers |= shadow << i; + } + } else { + uniform->sampler[shader_type].index = ~0; + uniform->sampler[shader_type].active = false; + } + } + virtual void visit_field(const glsl_type *type, const char *name, bool row_major) { @@ -354,31 +387,6 @@ private: if (!found) return; - /* If there is already storage associated with this uniform, it means - * that it was set while processing an earlier shader stage. For - * example, we may be processing the uniform in the fragment shader, but - * the uniform was already processed in the vertex shader. - */ - if (this->uniforms[id].storage != NULL) { - /* If the uniform already has storage set from another shader stage, - * mark the samplers used for this shader stage. - */ - if (type->contains_sampler()) { - const unsigned count = MAX2(1, this->uniforms[id].array_elements); - const unsigned shadow = (type->is_array()) - ? type->fields.array->sampler_shadow : type->sampler_shadow; - - for (unsigned i = 0; i < count; i++) { - const unsigned s = this->uniforms[id].sampler + i; - - this->shader_samplers_used |= 1U << s; - this->shader_shadow_samplers |= shadow << s; - } - } - - return; - } - const glsl_type *base_type; if (type->is_array()) { this->uniforms[id].array_elements = type->length; @@ -388,26 +396,16 @@ private: base_type = type; } - if (base_type->is_sampler()) { - this->uniforms[id].sampler = this->next_sampler; + /* This assigns sampler uniforms to sampler units. */ + handle_samplers(base_type, &this->uniforms[id]); - /* Increment the sampler by 1 for non-arrays and by the number of - * array elements for arrays. - */ - this->next_sampler += MAX2(1, this->uniforms[id].array_elements); - - const gl_texture_index target = base_type->sampler_index(); - const unsigned shadow = base_type->sampler_shadow; - for (unsigned i = this->uniforms[id].sampler - ; i < MIN2(this->next_sampler, MAX_SAMPLERS) - ; i++) { - this->targets[i] = target; - this->shader_samplers_used |= 1U << i; - this->shader_shadow_samplers |= shadow << i; - } - - } else { - this->uniforms[id].sampler = ~0; + /* If there is already storage associated with this uniform, it means + * that it was set while processing an earlier shader stage. For + * example, we may be processing the uniform in the fragment shader, but + * the uniform was already processed in the vertex shader. + */ + if (this->uniforms[id].storage != NULL) { + return; } this->uniforms[id].name = ralloc_strdup(this->uniforms, name); @@ -633,17 +631,6 @@ link_assign_uniform_locations(struct gl_shader_program *prog) prog->UniformHash = new string_to_uint_map; } - /* Uniforms that lack an initializer in the shader code have an initial - * value of zero. This includes sampler uniforms. - * - * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says: - * - * "The link time initial value is either the value of the variable's - * initializer, if present, or 0 if no initializer is present. Sampler - * types cannot have initializers." - */ - memset(prog->SamplerUnits, 0, sizeof(prog->SamplerUnits)); - /* First pass: Count the uniform resources used by the user-defined * uniforms. While this happens, each active uniform will have an index * assigned to it. @@ -656,6 +643,18 @@ link_assign_uniform_locations(struct gl_shader_program *prog) if (prog->_LinkedShaders[i] == NULL) continue; + /* Uniforms that lack an initializer in the shader code have an initial + * value of zero. This includes sampler uniforms. + * + * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says: + * + * "The link time initial value is either the value of the variable's + * initializer, if present, or 0 if no initializer is present. Sampler + * types cannot have initializers." + */ + memset(prog->_LinkedShaders[i]->SamplerUnits, 0, + sizeof(prog->_LinkedShaders[i]->SamplerUnits)); + link_update_uniform_buffer_variables(prog->_LinkedShaders[i]); /* Reset various per-shader target counts. @@ -706,9 +705,7 @@ link_assign_uniform_locations(struct gl_shader_program *prog) if (prog->_LinkedShaders[i] == NULL) continue; - /* Reset various per-shader target counts. - */ - parcel.start_shader(); + parcel.start_shader((gl_shader_type)i); foreach_list(node, prog->_LinkedShaders[i]->ir) { ir_variable *const var = ((ir_instruction *) node)->as_variable(); @@ -726,10 +723,11 @@ link_assign_uniform_locations(struct gl_shader_program *prog) prog->_LinkedShaders[i]->active_samplers = parcel.shader_samplers_used; prog->_LinkedShaders[i]->shadow_samplers = parcel.shader_shadow_samplers; - } - assert(sizeof(prog->SamplerTargets) == sizeof(parcel.targets)); - memcpy(prog->SamplerTargets, parcel.targets, sizeof(prog->SamplerTargets)); + STATIC_ASSERT(sizeof(prog->_LinkedShaders[i]->SamplerTargets) == sizeof(parcel.targets)); + memcpy(prog->_LinkedShaders[i]->SamplerTargets, parcel.targets, + sizeof(prog->_LinkedShaders[i]->SamplerTargets)); + } #ifndef NDEBUG for (unsigned i = 0; i < num_user_uniforms; i++) { diff --git a/mesalib/src/glsl/link_varyings.cpp b/mesalib/src/glsl/link_varyings.cpp index 04c9fdd7c..34e3440d6 100644 --- a/mesalib/src/glsl/link_varyings.cpp +++ b/mesalib/src/glsl/link_varyings.cpp @@ -541,7 +541,7 @@ store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog, class varying_matches { public: - varying_matches(bool disable_varying_packing); + varying_matches(bool disable_varying_packing, bool consumer_is_fs); ~varying_matches(); void record(ir_variable *producer_var, ir_variable *consumer_var); unsigned assign_locations(); @@ -621,11 +621,15 @@ private: * it was allocated. */ unsigned matches_capacity; + + const bool consumer_is_fs; }; -varying_matches::varying_matches(bool disable_varying_packing) - : disable_varying_packing(disable_varying_packing) +varying_matches::varying_matches(bool disable_varying_packing, + bool consumer_is_fs) + : disable_varying_packing(disable_varying_packing), + consumer_is_fs(consumer_is_fs) { /* Note: this initial capacity is rather arbitrarily chosen to be large * enough for many cases without wasting an unreasonable amount of space. @@ -656,6 +660,10 @@ varying_matches::~varying_matches() * 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. + * + * Note: as a side effect this function may change the interpolation type of + * \c producer_var, but only when the change couldn't possibly affect + * rendering. */ void varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var) @@ -668,6 +676,25 @@ varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var) return; } + if ((consumer_var == NULL && producer_var->type->contains_integer()) || + !consumer_is_fs) { + /* Since this varying is not being consumed by the fragment shader, its + * interpolation type varying cannot possibly affect rendering. Also, + * this variable is non-flat and is (or contains) an integer. + * + * lower_packed_varyings requires all integer varyings to flat, + * regardless of where they appear. We can trivially satisfy that + * requirement by changing the interpolation type to flat here. + */ + producer_var->centroid = false; + producer_var->interpolation = INTERP_QUALIFIER_FLAT; + + if (consumer_var) { + consumer_var->centroid = false; + consumer_var->interpolation = INTERP_QUALIFIER_FLAT; + } + } + if (this->num_matches == this->matches_capacity) { this->matches_capacity *= 2; this->matches = (match *) @@ -960,11 +987,14 @@ assign_varying_locations(struct gl_context *ctx, { const unsigned producer_base = VARYING_SLOT_VAR0; const unsigned consumer_base = VARYING_SLOT_VAR0; - varying_matches matches(ctx->Const.DisableVaryingPacking); + varying_matches matches(ctx->Const.DisableVaryingPacking, + consumer && consumer->Type == GL_FRAGMENT_SHADER); hash_table *tfeedback_candidates = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare); hash_table *consumer_inputs = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare); + hash_table *consumer_interface_inputs + = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare); /* Operate in a total of three passes. * @@ -983,8 +1013,17 @@ assign_varying_locations(struct gl_context *ctx, ((ir_instruction *) node)->as_variable(); if ((input_var != NULL) && (input_var->mode == ir_var_shader_in)) { - hash_table_insert(consumer_inputs, input_var, - ralloc_strdup(mem_ctx, input_var->name)); + if (input_var->interface_type != NULL) { + char *const iface_field_name = + ralloc_asprintf(mem_ctx, "%s.%s", + input_var->interface_type->name, + input_var->name); + hash_table_insert(consumer_interface_inputs, input_var, + iface_field_name); + } else { + hash_table_insert(consumer_inputs, input_var, + ralloc_strdup(mem_ctx, input_var->name)); + } } } } @@ -998,8 +1037,19 @@ assign_varying_locations(struct gl_context *ctx, tfeedback_candidate_generator g(mem_ctx, tfeedback_candidates); g.process(output_var); - ir_variable *input_var = - (ir_variable *) hash_table_find(consumer_inputs, output_var->name); + ir_variable *input_var; + if (output_var->interface_type != NULL) { + char *const iface_field_name = + ralloc_asprintf(mem_ctx, "%s.%s", + output_var->interface_type->name, + output_var->name); + input_var = + (ir_variable *) hash_table_find(consumer_interface_inputs, + iface_field_name); + } else { + input_var = + (ir_variable *) hash_table_find(consumer_inputs, output_var->name); + } if (input_var && input_var->mode != ir_var_shader_in) input_var = NULL; @@ -1019,6 +1069,7 @@ assign_varying_locations(struct gl_context *ctx, if (matched_candidate == NULL) { hash_table_dtor(tfeedback_candidates); hash_table_dtor(consumer_inputs); + hash_table_dtor(consumer_interface_inputs); return false; } @@ -1036,12 +1087,14 @@ assign_varying_locations(struct gl_context *ctx, if (!tfeedback_decls[i].assign_location(ctx, prog)) { hash_table_dtor(tfeedback_candidates); hash_table_dtor(consumer_inputs); + hash_table_dtor(consumer_interface_inputs); return false; } } hash_table_dtor(tfeedback_candidates); hash_table_dtor(consumer_inputs); + hash_table_dtor(consumer_interface_inputs); if (ctx->Const.DisableVaryingPacking) { /* Transform feedback code assumes varyings are packed, so if the driver diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp index 2b30d2b65..982fe46bd 100644 --- a/mesalib/src/glsl/linker.cpp +++ b/mesalib/src/glsl/linker.cpp @@ -938,6 +938,12 @@ link_intrastage_shaders(void *mem_ctx, if (!cross_validate_globals(prog, shader_list, num_shaders, false)) return NULL; + /* Check that interface blocks defined in multiple shaders are consistent. + */ + if (!validate_intrastage_interface_blocks((const gl_shader **)shader_list, + num_shaders)) + return NULL; + /* Check that uniform blocks between shaders for a stage agree. */ const int num_uniform_blocks = link_uniform_blocks(mem_ctx, prog, shader_list, num_shaders, @@ -1512,15 +1518,15 @@ check_resources(struct gl_context *ctx, struct gl_shader_program *prog) }; const unsigned max_samplers[MESA_SHADER_TYPES] = { - ctx->Const.MaxVertexTextureImageUnits, - ctx->Const.MaxTextureImageUnits, - ctx->Const.MaxGeometryTextureImageUnits + ctx->Const.VertexProgram.MaxTextureImageUnits, + ctx->Const.FragmentProgram.MaxTextureImageUnits, + ctx->Const.GeometryProgram.MaxTextureImageUnits }; const unsigned max_uniform_components[MESA_SHADER_TYPES] = { ctx->Const.VertexProgram.MaxUniformComponents, ctx->Const.FragmentProgram.MaxUniformComponents, - 0 /* FINISHME: Geometry shaders. */ + ctx->Const.GeometryProgram.MaxUniformComponents }; const unsigned max_uniform_blocks[MESA_SHADER_TYPES] = { @@ -1722,6 +1728,12 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) if (prog->_LinkedShaders[i] == NULL) continue; + if (!validate_interstage_interface_blocks(prog->_LinkedShaders[prev], + prog->_LinkedShaders[i])) { + linker_error(prog, "interface block mismatch between shader stages\n"); + goto done; + } + if (!cross_validate_outputs_to_inputs(prog, prog->_LinkedShaders[prev], prog->_LinkedShaders[i])) @@ -1733,6 +1745,12 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) prog->LinkStatus = true; } + + for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { + if (prog->_LinkedShaders[i] != NULL) + lower_named_interface_blocks(mem_ctx, prog->_LinkedShaders[i]); + } + /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do * it before optimization because we want most of the checks to get * dropped thanks to constant propagation. @@ -1767,7 +1785,7 @@ link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) unsigned max_unroll = ctx->ShaderCompilerOptions[i].MaxUnrollIterations; - while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, false, max_unroll)) + while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, false, max_unroll, &ctx->ShaderCompilerOptions[i])) ; } diff --git a/mesalib/src/glsl/linker.h b/mesalib/src/glsl/linker.h index f1ce50ace..2fe2410c2 100644 --- a/mesalib/src/glsl/linker.h +++ b/mesalib/src/glsl/linker.h @@ -60,6 +60,14 @@ link_uniform_blocks(void *mem_ctx, unsigned num_shaders, struct gl_uniform_block **blocks_ret); +bool +validate_intrastage_interface_blocks(const gl_shader **shader_list, + unsigned num_shaders); + +bool +validate_interstage_interface_blocks(const gl_shader *producer, + const gl_shader *consumer); + /** * Class for processing all of the leaf fields of a variable that corresponds * to a program resource. diff --git a/mesalib/src/glsl/lower_clip_distance.cpp b/mesalib/src/glsl/lower_clip_distance.cpp index 643807de8..d6cf94438 100644 --- a/mesalib/src/glsl/lower_clip_distance.cpp +++ b/mesalib/src/glsl/lower_clip_distance.cpp @@ -46,10 +46,11 @@ */ #include "glsl_symbol_table.h" -#include "ir_hierarchical_visitor.h" +#include "ir_rvalue_visitor.h" #include "ir.h" +#include "program/prog_instruction.h" /* For WRITEMASK_* */ -class lower_clip_distance_visitor : public ir_hierarchical_visitor { +class lower_clip_distance_visitor : public ir_rvalue_visitor { public: lower_clip_distance_visitor() : progress(false), old_clip_distance_var(NULL), @@ -59,11 +60,14 @@ public: virtual ir_visitor_status visit(ir_variable *); void create_indices(ir_rvalue*, ir_rvalue *&, ir_rvalue *&); - virtual ir_visitor_status visit_leave(ir_dereference_array *); virtual ir_visitor_status visit_leave(ir_assignment *); void visit_new_assignment(ir_assignment *ir); virtual ir_visitor_status visit_leave(ir_call *); + virtual void handle_rvalue(ir_rvalue **rvalue); + + void fix_lhs(ir_assignment *); + bool progress; /** @@ -173,35 +177,70 @@ lower_clip_distance_visitor::create_indices(ir_rvalue *old_index, } -/** - * Replace any expression that indexes into the gl_ClipDistance array with an - * expression that indexes into one of the vec4's in gl_ClipDistanceMESA and - * accesses the appropriate component. - */ -ir_visitor_status -lower_clip_distance_visitor::visit_leave(ir_dereference_array *ir) +void +lower_clip_distance_visitor::handle_rvalue(ir_rvalue **rv) { /* If the gl_ClipDistance var hasn't been declared yet, then * there's no way this deref can refer to it. */ - if (!this->old_clip_distance_var) - return visit_continue; + if (!this->old_clip_distance_var || *rv == NULL) + return; + + ir_dereference_array *const array_deref = (*rv)->as_dereference_array(); + if (array_deref == NULL) + return; - ir_dereference_variable *old_var_ref = ir->array->as_dereference_variable(); + /* Replace any expression that indexes into the gl_ClipDistance array + * with an expression that indexes into one of the vec4's in + * gl_ClipDistanceMESA and accesses the appropriate component. + */ + ir_dereference_variable *old_var_ref = + array_deref->array->as_dereference_variable(); if (old_var_ref && old_var_ref->var == this->old_clip_distance_var) { this->progress = true; ir_rvalue *array_index; ir_rvalue *swizzle_index; - this->create_indices(ir->array_index, array_index, swizzle_index); - void *mem_ctx = ralloc_parent(ir); - ir->array = new(mem_ctx) ir_dereference_array( - this->new_clip_distance_var, array_index); - ir->array_index = swizzle_index; - } + this->create_indices(array_deref->array_index, array_index, swizzle_index); + void *mem_ctx = ralloc_parent(array_deref); - return visit_continue; + ir_dereference_array *const ClipDistanceMESA_deref = + new(mem_ctx) ir_dereference_array(this->new_clip_distance_var, + array_index); + + ir_expression *const expr = + new(mem_ctx) ir_expression(ir_binop_vector_extract, + ClipDistanceMESA_deref, + swizzle_index); + + *rv = expr; + } } +void +lower_clip_distance_visitor::fix_lhs(ir_assignment *ir) +{ + if (ir->lhs->ir_type == ir_type_expression) { + void *mem_ctx = ralloc_parent(ir); + ir_expression *const expr = (ir_expression *) ir->lhs; + + /* The expression must be of the form: + * + * (vector_extract gl_ClipDistanceMESA[i], j). + */ + assert(expr->operation == ir_binop_vector_extract); + assert(expr->operands[0]->ir_type == ir_type_dereference_array); + assert(expr->operands[0]->type == glsl_type::vec4_type); + + ir_dereference *const new_lhs = (ir_dereference *) expr->operands[0]; + ir->rhs = new(mem_ctx) ir_expression(ir_triop_vector_insert, + glsl_type::vec4_type, + new_lhs->clone(mem_ctx, NULL), + ir->rhs, + expr->operands[1]); + ir->set_lhs(new_lhs); + ir->write_mask = WRITEMASK_XYZW; + } +} /** * Replace any assignment having gl_ClipDistance (undereferenced) as its LHS @@ -223,29 +262,50 @@ lower_clip_distance_visitor::visit_leave(ir_assignment *ir) * each of them. * * Note: to unroll into element-by-element assignments, we need to make - * clones of the LHS and RHS. This is only safe if the LHS and RHS are - * side-effect free. Fortunately, we know that they are, because the - * only kind of rvalue that can have side effects is an ir_call, and - * ir_calls only appear (a) as a statement on their own, or (b) as the - * RHS of an assignment that stores the result of the call in a - * temporary variable. + * clones of the LHS and RHS. This is safe because expressions and + * l-values are side-effect free. */ void *ctx = ralloc_parent(ir); int array_size = this->old_clip_distance_var->type->array_size(); for (int i = 0; i < array_size; ++i) { ir_dereference_array *new_lhs = new(ctx) ir_dereference_array( ir->lhs->clone(ctx, NULL), new(ctx) ir_constant(i)); - new_lhs->accept(this); ir_dereference_array *new_rhs = new(ctx) ir_dereference_array( ir->rhs->clone(ctx, NULL), new(ctx) ir_constant(i)); - new_rhs->accept(this); - this->base_ir->insert_before( - new(ctx) ir_assignment(new_lhs, new_rhs)); + this->handle_rvalue((ir_rvalue **) &new_rhs); + + /* Handle the LHS after creating the new assignment. This must + * happen in this order because handle_rvalue may replace the old LHS + * with an ir_expression of ir_binop_vector_extract. Since this is + * not a valide l-value, this will cause an assertion in the + * ir_assignment constructor to fail. + * + * If this occurs, replace the mangled LHS with a dereference of the + * vector, and replace the RHS with an ir_triop_vector_insert. + */ + ir_assignment *const assign = new(ctx) ir_assignment(new_lhs, new_rhs); + this->handle_rvalue((ir_rvalue **) &assign->lhs); + this->fix_lhs(assign); + + this->base_ir->insert_before(assign); } ir->remove(); + + return visit_continue; } - return visit_continue; + /* Handle the LHS as if it were an r-value. Normally + * rvalue_visit(ir_assignment *) only visits the RHS, but we need to lower + * expressions in the LHS as well. + * + * This may cause the LHS to get replaced with an ir_expression of + * ir_binop_vector_extract. If this occurs, replace it with a dereference + * of the vector, and replace the RHS with an ir_triop_vector_insert. + */ + handle_rvalue((ir_rvalue **)&ir->lhs); + this->fix_lhs(ir); + + return rvalue_visit(ir); } @@ -330,7 +390,7 @@ lower_clip_distance_visitor::visit_leave(ir_call *ir) } } - return visit_continue; + return rvalue_visit(ir); } diff --git a/mesalib/src/glsl/lower_instructions.cpp b/mesalib/src/glsl/lower_instructions.cpp index 1ce7b7c9d..d32ec80d6 100644 --- a/mesalib/src/glsl/lower_instructions.cpp +++ b/mesalib/src/glsl/lower_instructions.cpp @@ -38,6 +38,7 @@ * - LOG_TO_LOG2 * - MOD_TO_FRACT * - LRP_TO_ARITH + * - BITFIELD_INSERT_TO_BFM_BFI * * SUB_TO_ADD_NEG: * --------------- @@ -84,6 +85,15 @@ * LRP_TO_ARITH: * ------------- * Converts ir_triop_lrp to (op0 * (1.0f - op2)) + (op1 * op2). + * + * BITFIELD_INSERT_TO_BFM_BFI: + * --------------------------- + * Breaks ir_quadop_bitfield_insert into ir_binop_bfm (bitfield mask) and + * ir_triop_bfi (bitfield insert). + * + * Many GPUs implement the bitfieldInsert() built-in from ARB_gpu_shader_5 + * with a pair of instructions. + * */ #include "main/core.h" /* for M_LOG2E */ @@ -114,6 +124,7 @@ private: void pow_to_exp2(ir_expression *); void log_to_log2(ir_expression *); void lrp_to_arith(ir_expression *); + void bitfield_insert_to_bfm_bfi(ir_expression *); }; /** @@ -298,6 +309,29 @@ lower_instructions_visitor::lrp_to_arith(ir_expression *ir) this->progress = true; } +void +lower_instructions_visitor::bitfield_insert_to_bfm_bfi(ir_expression *ir) +{ + /* Translates + * ir_quadop_bitfield_insert base insert offset bits + * into + * ir_triop_bfi (ir_binop_bfm bits offset) insert base + */ + + ir_rvalue *base_expr = ir->operands[0]; + + ir->operation = ir_triop_bfi; + ir->operands[0] = new(ir) ir_expression(ir_binop_bfm, + ir->type->get_base_type(), + ir->operands[3], + ir->operands[2]); + /* ir->operands[1] is still the value to insert. */ + ir->operands[2] = base_expr; + ir->operands[3] = NULL; + + this->progress = true; +} + ir_visitor_status lower_instructions_visitor::visit_leave(ir_expression *ir) { @@ -339,6 +373,11 @@ lower_instructions_visitor::visit_leave(ir_expression *ir) lrp_to_arith(ir); break; + case ir_quadop_bitfield_insert: + if (lowering(BITFIELD_INSERT_TO_BFM_BFI)) + bitfield_insert_to_bfm_bfi(ir); + break; + default: return visit_continue; } diff --git a/mesalib/src/glsl/lower_named_interface_blocks.cpp b/mesalib/src/glsl/lower_named_interface_blocks.cpp new file mode 100644 index 000000000..eba667a8b --- /dev/null +++ b/mesalib/src/glsl/lower_named_interface_blocks.cpp @@ -0,0 +1,238 @@ +/* + * Copyright (c) 2013 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 lower_named_interface_blocks.cpp + * + * This lowering pass converts all interface blocks with instance names + * into interface blocks without an instance name. + * + * For example, the following shader: + * + * out block { + * float block_var; + * } inst_name; + * + * main() + * { + * inst_name.block_var = 0.0; + * } + * + * Is rewritten to: + * + * out block { + * float block_var; + * }; + * + * main() + * { + * block_var = 0.0; + * } + * + * This takes place after the shader code has already been verified with + * the interface name in place. + * + * The linking phase will use the interface block name rather than the + * interface's instance name when linking interfaces. + * + * This modification to the ir allows our currently existing dead code + * elimination to work with interface blocks without changes. + */ + +#include "glsl_symbol_table.h" +#include "ir.h" +#include "ir_optimization.h" +#include "ir_rvalue_visitor.h" +#include "program/hash_table.h" + +class flatten_named_interface_blocks_declarations : public ir_rvalue_visitor +{ +public: + void * const mem_ctx; + hash_table *interface_namespace; + + flatten_named_interface_blocks_declarations(void *mem_ctx) + : mem_ctx(mem_ctx) + { + } + + void run(exec_list *instructions); + + virtual ir_visitor_status visit_leave(ir_assignment *); + virtual void handle_rvalue(ir_rvalue **rvalue); +}; + +void +flatten_named_interface_blocks_declarations::run(exec_list *instructions) +{ + interface_namespace = hash_table_ctor(0, hash_table_string_hash, + hash_table_string_compare); + + /* First pass: adjust instance block variables with an instance name + * to not have an instance name. + * + * The interface block variables are stored in the interface_namespace + * hash table so they can be used in the second pass. + */ + foreach_list_safe(node, instructions) { + ir_variable *var = ((ir_instruction *) node)->as_variable(); + if (!var || !var->is_interface_instance()) + continue; + + /* It should be possible to handle uniforms during this pass, + * but, this will require changes to the other uniform block + * support code. + */ + if (var->mode == ir_var_uniform) + continue; + + const glsl_type * iface_t = var->type; + const glsl_type * array_t = NULL; + exec_node *insert_pos = var; + + if (iface_t->is_array()) { + array_t = iface_t; + iface_t = array_t->fields.array; + } + + assert (iface_t->is_interface()); + + for (unsigned i = 0; i < iface_t->length; i++) { + const char * field_name = iface_t->fields.structure[i].name; + char *iface_field_name = + ralloc_asprintf(mem_ctx, "%s.%s", + iface_t->name, field_name); + + ir_variable *found_var = + (ir_variable *) hash_table_find(interface_namespace, + iface_field_name); + if (!found_var) { + ir_variable *new_var; + if (array_t == NULL) { + char *var_name = + ralloc_strdup(mem_ctx, iface_t->fields.structure[i].name); + new_var = + new(mem_ctx) ir_variable(iface_t->fields.structure[i].type, + var_name, + (ir_variable_mode) var->mode); + } else { + const glsl_type *new_array_type = + glsl_type::get_array_instance( + iface_t->fields.structure[i].type, + array_t->length); + char *var_name = + ralloc_asprintf(mem_ctx, "%s[%d]", + iface_t->fields.structure[i].name, + array_t->length); + new_var = + new(mem_ctx) ir_variable(new_array_type, + var_name, + (ir_variable_mode) var->mode); + } + + new_var->interface_type = iface_t; + hash_table_insert(interface_namespace, new_var, + iface_field_name); + insert_pos->insert_after(new_var); + insert_pos = new_var; + } + } + var->remove(); + } + + /* Second pass: visit all ir_dereference_record instances, and if they + * reference an interface block, then flatten the refererence out. + */ + visit_list_elements(this, instructions); + hash_table_dtor(interface_namespace); + interface_namespace = NULL; +} + +ir_visitor_status +flatten_named_interface_blocks_declarations::visit_leave(ir_assignment *ir) +{ + ir_dereference_record *lhs_rec = ir->lhs->as_dereference_record(); + if (lhs_rec) { + ir_rvalue *lhs_rec_tmp = lhs_rec; + handle_rvalue(&lhs_rec_tmp); + if (lhs_rec_tmp != lhs_rec) { + ir->set_lhs(lhs_rec_tmp); + } + } + return rvalue_visit(ir); +} + +void +flatten_named_interface_blocks_declarations::handle_rvalue(ir_rvalue **rvalue) +{ + if (*rvalue == NULL) + return; + + ir_dereference_record *ir = (*rvalue)->as_dereference_record(); + if (ir == NULL) + return; + + ir_variable *var = ir->variable_referenced(); + + if (!var->is_interface_instance()) + return; + + /* It should be possible to handle uniforms during this pass, + * but, this will require changes to the other uniform block + * support code. + */ + if (var->mode == ir_var_uniform) + return; + + if (var->interface_type != NULL) { + char *iface_field_name = + ralloc_asprintf(mem_ctx, "%s.%s", var->interface_type->name, + ir->field); + /* Find the variable in the set of flattened interface blocks */ + ir_variable *found_var = + (ir_variable *) hash_table_find(interface_namespace, + iface_field_name); + assert(found_var); + + ir_dereference_variable *deref_var = + new(mem_ctx) ir_dereference_variable(found_var); + + ir_dereference_array *deref_array = + ir->record->as_dereference_array(); + if (deref_array != NULL) { + *rvalue = + new(mem_ctx) ir_dereference_array(deref_var, + deref_array->array_index); + } else { + *rvalue = deref_var; + } + } +} + +void +lower_named_interface_blocks(void *mem_ctx, gl_shader *shader) +{ + flatten_named_interface_blocks_declarations v_decl(mem_ctx); + v_decl.run(shader->ir); +} + diff --git a/mesalib/src/glsl/lower_packed_varyings.cpp b/mesalib/src/glsl/lower_packed_varyings.cpp index b4cc5cd0d..cdf2289b4 100644 --- a/mesalib/src/glsl/lower_packed_varyings.cpp +++ b/mesalib/src/glsl/lower_packed_varyings.cpp @@ -178,6 +178,14 @@ lower_packed_varyings_visitor::run(exec_list *instructions) !this->needs_lowering(var)) continue; + /* This lowering pass is only capable of packing floats and ints + * together when their interpolation mode is "flat". Therefore, to be + * safe, caller should ensure that integral varyings always use flat + * interpolation, even when this is not required by GLSL. + */ + assert(var->interpolation == INTERP_QUALIFIER_FLAT || + !var->type->contains_integer()); + /* Change the old varying into an ordinary global. */ var->mode = ir_var_auto; diff --git a/mesalib/src/glsl/lower_vec_index_to_cond_assign.cpp b/mesalib/src/glsl/lower_vec_index_to_cond_assign.cpp index f85875f49..880859688 100644 --- a/mesalib/src/glsl/lower_vec_index_to_cond_assign.cpp +++ b/mesalib/src/glsl/lower_vec_index_to_cond_assign.cpp @@ -52,7 +52,12 @@ public: progress = false; } - ir_rvalue *convert_vec_index_to_cond_assign(ir_rvalue *val); + ir_rvalue *convert_vec_index_to_cond_assign(void *mem_ctx, + ir_rvalue *orig_vector, + ir_rvalue *orig_index, + const glsl_type *type); + + ir_rvalue *convert_vector_extract_to_cond_assign(ir_rvalue *ir); virtual ir_visitor_status visit_enter(ir_expression *); virtual ir_visitor_status visit_enter(ir_swizzle *); @@ -65,24 +70,16 @@ public: }; ir_rvalue * -ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(ir_rvalue *ir) +ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(void *mem_ctx, + ir_rvalue *orig_vector, + ir_rvalue *orig_index, + const glsl_type *type) { - ir_dereference_array *orig_deref = ir->as_dereference_array(); ir_assignment *assign, *value_assign; ir_variable *index, *var, *value; ir_dereference *deref, *deref_value; unsigned i; - if (!orig_deref) - return ir; - - if (orig_deref->array->type->is_matrix() || - orig_deref->array->type->is_array()) - return ir; - - void *mem_ctx = ralloc_parent(ir); - - assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT); exec_list list; @@ -92,19 +89,19 @@ ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(ir_rvalue ir_var_temporary); list.push_tail(index); deref = new(base_ir) ir_dereference_variable(index); - assign = new(base_ir) ir_assignment(deref, orig_deref->array_index, NULL); + assign = new(base_ir) ir_assignment(deref, orig_index, NULL); list.push_tail(assign); /* Store the value inside a temp, thus avoiding matrixes duplication */ - value = new(base_ir) ir_variable(orig_deref->array->type, "vec_value_tmp", - ir_var_temporary); + value = new(base_ir) ir_variable(orig_vector->type, "vec_value_tmp", + ir_var_temporary); list.push_tail(value); deref_value = new(base_ir) ir_dereference_variable(value); - value_assign = new(base_ir) ir_assignment(deref_value, orig_deref->array); + value_assign = new(base_ir) ir_assignment(deref_value, orig_vector); list.push_tail(value_assign); /* Temporary where we store whichever value we swizzle out. */ - var = new(base_ir) ir_variable(ir->type, "vec_index_tmp_v", + var = new(base_ir) ir_variable(type, "vec_index_tmp_v", ir_var_temporary); list.push_tail(var); @@ -113,13 +110,14 @@ ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(ir_rvalue */ ir_rvalue *const cond_deref = compare_index_block(&list, index, 0, - orig_deref->array->type->vector_elements, + orig_vector->type->vector_elements, mem_ctx); /* Generate a conditional move of each vector element to the temp. */ - for (i = 0; i < orig_deref->array->type->vector_elements; i++) { + for (i = 0; i < orig_vector->type->vector_elements; i++) { ir_rvalue *condition_swizzle = - new(base_ir) ir_swizzle(cond_deref->clone(ir, NULL), i, 0, 0, 0, 1); + new(base_ir) ir_swizzle(cond_deref->clone(mem_ctx, NULL), + i, 0, 0, 0, 1); /* Just clone the rest of the deref chain when trying to get at the * underlying variable. @@ -142,13 +140,27 @@ ir_vec_index_to_cond_assign_visitor::convert_vec_index_to_cond_assign(ir_rvalue return new(base_ir) ir_dereference_variable(var); } +ir_rvalue * +ir_vec_index_to_cond_assign_visitor::convert_vector_extract_to_cond_assign(ir_rvalue *ir) +{ + ir_expression *const expr = ir->as_expression(); + + if (expr == NULL || expr->operation != ir_binop_vector_extract) + return ir; + + return convert_vec_index_to_cond_assign(ralloc_parent(ir), + expr->operands[0], + expr->operands[1], + ir->type); +} + ir_visitor_status ir_vec_index_to_cond_assign_visitor::visit_enter(ir_expression *ir) { unsigned int i; for (i = 0; i < ir->get_num_operands(); i++) { - ir->operands[i] = convert_vec_index_to_cond_assign(ir->operands[i]); + ir->operands[i] = convert_vector_extract_to_cond_assign(ir->operands[i]); } return visit_continue; @@ -161,7 +173,7 @@ ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir) * the result of indexing a vector is. But maybe at some point we'll end up * using swizzling of scalars for vector construction. */ - ir->val = convert_vec_index_to_cond_assign(ir->val); + ir->val = convert_vector_extract_to_cond_assign(ir->val); return visit_continue; } @@ -169,91 +181,12 @@ ir_vec_index_to_cond_assign_visitor::visit_enter(ir_swizzle *ir) ir_visitor_status ir_vec_index_to_cond_assign_visitor::visit_leave(ir_assignment *ir) { - ir_variable *index, *var; - ir_dereference_variable *deref; - ir_assignment *assign; - unsigned i; - - ir->rhs = convert_vec_index_to_cond_assign(ir->rhs); - if (ir->condition) - ir->condition = convert_vec_index_to_cond_assign(ir->condition); - - /* Last, handle the LHS */ - ir_dereference_array *orig_deref = ir->lhs->as_dereference_array(); - - if (!orig_deref || - orig_deref->array->type->is_matrix() || - orig_deref->array->type->is_array()) - return visit_continue; - - void *mem_ctx = ralloc_parent(ir); - - assert(orig_deref->array_index->type->base_type == GLSL_TYPE_INT); - - exec_list list; - - /* Store the index to a temporary to avoid reusing its tree. */ - index = new(ir) ir_variable(glsl_type::int_type, "vec_index_tmp_i", - ir_var_temporary); - list.push_tail(index); - deref = new(ir) ir_dereference_variable(index); - assign = new(ir) ir_assignment(deref, orig_deref->array_index, NULL); - list.push_tail(assign); - - /* Store the RHS to a temporary to avoid reusing its tree. */ - var = new(ir) ir_variable(ir->rhs->type, "vec_index_tmp_v", - ir_var_temporary); - list.push_tail(var); - deref = new(ir) ir_dereference_variable(var); - assign = new(ir) ir_assignment(deref, ir->rhs, NULL); - list.push_tail(assign); - - /* Generate a single comparison condition "mask" for all of the components - * in the vector. - */ - ir_rvalue *const cond_deref = - compare_index_block(&list, index, 0, - orig_deref->array->type->vector_elements, - mem_ctx); + ir->rhs = convert_vector_extract_to_cond_assign(ir->rhs); - /* Generate a conditional move of each vector element to the temp. */ - for (i = 0; i < orig_deref->array->type->vector_elements; i++) { - ir_rvalue *condition_swizzle = - new(ir) ir_swizzle(cond_deref->clone(ir, NULL), i, 0, 0, 0, 1); - - - /* Just clone the rest of the deref chain when trying to get at the - * underlying variable. - */ - ir_rvalue *swizzle = - new(ir) ir_swizzle(orig_deref->array->clone(mem_ctx, NULL), - i, 0, 0, 0, 1); - - deref = new(ir) ir_dereference_variable(var); - assign = new(ir) ir_assignment(swizzle, deref, condition_swizzle); - list.push_tail(assign); + if (ir->condition) { + ir->condition = convert_vector_extract_to_cond_assign(ir->condition); } - /* If the original assignment has a condition, respect that original - * condition! This is acomplished by wrapping the new conditional - * assignments in an if-statement that uses the original condition. - */ - if (ir->condition != NULL) { - /* No need to clone the condition because the IR that it hangs on is - * going to be removed from the instruction sequence. - */ - ir_if *if_stmt = new(mem_ctx) ir_if(ir->condition); - - list.move_nodes_to(&if_stmt->then_instructions); - ir->insert_before(if_stmt); - } else { - ir->insert_before(&list); - } - - ir->remove(); - - this->progress = true; - return visit_continue; } @@ -262,7 +195,7 @@ ir_vec_index_to_cond_assign_visitor::visit_enter(ir_call *ir) { foreach_iter(exec_list_iterator, iter, *ir) { ir_rvalue *param = (ir_rvalue *)iter.get(); - ir_rvalue *new_param = convert_vec_index_to_cond_assign(param); + ir_rvalue *new_param = convert_vector_extract_to_cond_assign(param); if (new_param != param) { param->replace_with(new_param); @@ -276,7 +209,7 @@ ir_visitor_status ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir) { if (ir->value) { - ir->value = convert_vec_index_to_cond_assign(ir->value); + ir->value = convert_vector_extract_to_cond_assign(ir->value); } return visit_continue; @@ -285,7 +218,7 @@ ir_vec_index_to_cond_assign_visitor::visit_enter(ir_return *ir) ir_visitor_status ir_vec_index_to_cond_assign_visitor::visit_enter(ir_if *ir) { - ir->condition = convert_vec_index_to_cond_assign(ir->condition); + ir->condition = convert_vector_extract_to_cond_assign(ir->condition); return visit_continue; } diff --git a/mesalib/src/glsl/lower_vec_index_to_swizzle.cpp b/mesalib/src/glsl/lower_vec_index_to_swizzle.cpp index 264d6dc07..d5ad692c2 100644 --- a/mesalib/src/glsl/lower_vec_index_to_swizzle.cpp +++ b/mesalib/src/glsl/lower_vec_index_to_swizzle.cpp @@ -46,7 +46,7 @@ public: progress = false; } - ir_rvalue *convert_vec_index_to_swizzle(ir_rvalue *val); + ir_rvalue *convert_vector_extract_to_swizzle(ir_rvalue *val); virtual ir_visitor_status visit_enter(ir_expression *); virtual ir_visitor_status visit_enter(ir_swizzle *); @@ -59,20 +59,14 @@ public: }; ir_rvalue * -ir_vec_index_to_swizzle_visitor::convert_vec_index_to_swizzle(ir_rvalue *ir) +ir_vec_index_to_swizzle_visitor::convert_vector_extract_to_swizzle(ir_rvalue *ir) { - ir_dereference_array *deref = ir->as_dereference_array(); - ir_constant *ir_constant; - - if (!deref) - return ir; - - if (deref->array->type->is_matrix() || deref->array->type->is_array()) + ir_expression *const expr = ir->as_expression(); + if (expr == NULL || expr->operation != ir_binop_vector_extract) return ir; - assert(deref->array_index->type->base_type == GLSL_TYPE_INT); - ir_constant = deref->array_index->constant_expression_value(); - if (!ir_constant) + ir_constant *const idx = expr->operands[1]->constant_expression_value(); + if (idx == NULL) return ir; void *ctx = ralloc_parent(ir); @@ -92,10 +86,10 @@ ir_vec_index_to_swizzle_visitor::convert_vec_index_to_swizzle(ir_rvalue *ir) * The ir_swizzle constructor gets angry if the index is negative or too * large. For simplicity sake, just clamp the index to [0, size-1]. */ - const int i = MIN2(MAX2(ir_constant->value.i[0], 0), - ((int) deref->array->type->vector_elements - 1)); + const int i = CLAMP(idx->value.i[0], 0, + (int) expr->operands[0]->type->vector_elements - 1); - return new(ctx) ir_swizzle(deref->array, i, 0, 0, 0, 1); + return new(ctx) ir_swizzle(expr->operands[0], i, 0, 0, 0, 1); } ir_visitor_status @@ -104,7 +98,7 @@ ir_vec_index_to_swizzle_visitor::visit_enter(ir_expression *ir) unsigned int i; for (i = 0; i < ir->get_num_operands(); i++) { - ir->operands[i] = convert_vec_index_to_swizzle(ir->operands[i]); + ir->operands[i] = convert_vector_extract_to_swizzle(ir->operands[i]); } return visit_continue; @@ -117,7 +111,7 @@ ir_vec_index_to_swizzle_visitor::visit_enter(ir_swizzle *ir) * the result of indexing a vector is. But maybe at some point we'll end up * using swizzling of scalars for vector construction. */ - ir->val = convert_vec_index_to_swizzle(ir->val); + ir->val = convert_vector_extract_to_swizzle(ir->val); return visit_continue; } @@ -125,8 +119,7 @@ ir_vec_index_to_swizzle_visitor::visit_enter(ir_swizzle *ir) ir_visitor_status ir_vec_index_to_swizzle_visitor::visit_enter(ir_assignment *ir) { - ir->set_lhs(convert_vec_index_to_swizzle(ir->lhs)); - ir->rhs = convert_vec_index_to_swizzle(ir->rhs); + ir->rhs = convert_vector_extract_to_swizzle(ir->rhs); return visit_continue; } @@ -136,7 +129,7 @@ ir_vec_index_to_swizzle_visitor::visit_enter(ir_call *ir) { foreach_iter(exec_list_iterator, iter, *ir) { ir_rvalue *param = (ir_rvalue *)iter.get(); - ir_rvalue *new_param = convert_vec_index_to_swizzle(param); + ir_rvalue *new_param = convert_vector_extract_to_swizzle(param); if (new_param != param) { param->replace_with(new_param); @@ -150,7 +143,7 @@ ir_visitor_status ir_vec_index_to_swizzle_visitor::visit_enter(ir_return *ir) { if (ir->value) { - ir->value = convert_vec_index_to_swizzle(ir->value); + ir->value = convert_vector_extract_to_swizzle(ir->value); } return visit_continue; @@ -159,7 +152,7 @@ ir_vec_index_to_swizzle_visitor::visit_enter(ir_return *ir) ir_visitor_status ir_vec_index_to_swizzle_visitor::visit_enter(ir_if *ir) { - ir->condition = convert_vec_index_to_swizzle(ir->condition); + ir->condition = convert_vector_extract_to_swizzle(ir->condition); return visit_continue; } diff --git a/mesalib/src/glsl/lower_vector_insert.cpp b/mesalib/src/glsl/lower_vector_insert.cpp new file mode 100644 index 000000000..0e640cc32 --- /dev/null +++ b/mesalib/src/glsl/lower_vector_insert.cpp @@ -0,0 +1,139 @@ +/* + * Copyright © 2013 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. + */ +#include "ir.h" +#include "ir_builder.h" +#include "ir_rvalue_visitor.h" +#include "ir_optimization.h" + +using namespace ir_builder; + +class vector_insert_visitor : public ir_rvalue_visitor { +public: + vector_insert_visitor(bool lower_nonconstant_index) + : progress(false), lower_nonconstant_index(lower_nonconstant_index) + { + factory.instructions = &factory_instructions; + } + + virtual ~vector_insert_visitor() + { + assert(factory_instructions.is_empty()); + } + + virtual void handle_rvalue(ir_rvalue **rv); + + ir_factory factory; + exec_list factory_instructions; + bool progress; + bool lower_nonconstant_index; +}; + + +void +vector_insert_visitor::handle_rvalue(ir_rvalue **rv) +{ + if (*rv == NULL || (*rv)->ir_type != ir_type_expression) + return; + + ir_expression *const expr = (ir_expression *) *rv; + + if (likely(expr->operation != ir_triop_vector_insert)) + return; + + factory.mem_ctx = ralloc_parent(expr); + + ir_constant *const idx = expr->operands[2]->constant_expression_value(); + if (idx != NULL) { + /* Replace (vector_insert (vec) (scalar) (index)) with a dereference of + * a new temporary. The new temporary gets assigned as + * + * t = vec + * t.mask = scalar + * + * where mask is the component selected by index. + */ + ir_variable *const temp = + factory.make_temp(expr->operands[0]->type, "vec_tmp"); + + const int mask = 1 << idx->value.i[0]; + + factory.emit(assign(temp, expr->operands[0])); + factory.emit(assign(temp, expr->operands[1], mask)); + + this->progress = true; + *rv = new(factory.mem_ctx) ir_dereference_variable(temp); + } else if (this->lower_nonconstant_index) { + /* Replace (vector_insert (vec) (scalar) (index)) with a dereference of + * a new temporary. The new temporary gets assigned as + * + * t = vec + * if (index == 0) + * t.x = scalar + * if (index == 1) + * t.y = scalar + * if (index == 2) + * t.z = scalar + * if (index == 3) + * t.w = scalar + */ + ir_variable *const temp = + factory.make_temp(expr->operands[0]->type, "vec_tmp"); + + ir_variable *const src_temp = + factory.make_temp(expr->operands[1]->type, "src_temp"); + + factory.emit(assign(temp, expr->operands[0])); + factory.emit(assign(src_temp, expr->operands[1])); + + for (unsigned i = 0; i < expr->type->vector_elements; i++) { + ir_constant *const cmp_index = + new(factory.mem_ctx) ir_constant(int(i)); + + ir_variable *const cmp_result = + factory.make_temp(glsl_type::bool_type, "index_condition"); + + factory.emit(assign(cmp_result, + equal(expr->operands[2]->clone(factory.mem_ctx, + NULL), + cmp_index))); + + factory.emit(if_tree(cmp_result, + assign(temp, src_temp, WRITEMASK_X << i))); + } + + this->progress = true; + *rv = new(factory.mem_ctx) ir_dereference_variable(temp); + } + + base_ir->insert_before(factory.instructions); +} + +bool +lower_vector_insert(exec_list *instructions, bool lower_nonconstant_index) +{ + vector_insert_visitor v(lower_nonconstant_index); + + visit_list_elements(&v, instructions); + + return v.progress; +} diff --git a/mesalib/src/glsl/main.cpp b/mesalib/src/glsl/main.cpp index 7989680a2..768415169 100644 --- a/mesalib/src/glsl/main.cpp +++ b/mesalib/src/glsl/main.cpp @@ -185,9 +185,11 @@ compile_shader(struct gl_context *ctx, struct gl_shader *shader) /* Optimization passes */ if (!state->error && !shader->ir->is_empty()) { + const struct gl_shader_compiler_options *opts = + &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)]; bool progress; do { - progress = do_common_optimization(shader->ir, false, false, 32); + progress = do_common_optimization(shader->ir, false, false, 32, opts); } while (progress); validate_ir_tree(shader->ir); diff --git a/mesalib/src/glsl/opt_algebraic.cpp b/mesalib/src/glsl/opt_algebraic.cpp index 70e016d22..d706a6ad1 100644 --- a/mesalib/src/glsl/opt_algebraic.cpp +++ b/mesalib/src/glsl/opt_algebraic.cpp @@ -186,12 +186,12 @@ ir_algebraic_visitor::swizzle_if_required(ir_expression *expr, ir_rvalue * ir_algebraic_visitor::handle_expression(ir_expression *ir) { - ir_constant *op_const[3] = {NULL, NULL, NULL}; - ir_expression *op_expr[3] = {NULL, NULL, NULL}; + 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() <= 3); + assert(ir->get_num_operands() <= 4); for (i = 0; i < ir->get_num_operands(); i++) { if (ir->operands[i]->type->is_matrix()) return ir; diff --git a/mesalib/src/glsl/opt_flip_matrices.cpp b/mesalib/src/glsl/opt_flip_matrices.cpp new file mode 100644 index 000000000..497513fe8 --- /dev/null +++ b/mesalib/src/glsl/opt_flip_matrices.cpp @@ -0,0 +1,122 @@ +/* + * Copyright © 2013 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 opt_flip_matrices.cpp + * + * Convert (matrix * vector) operations to (vector * matrixTranspose), + * which can be done using dot products rather than multiplies and adds. + * On some hardware, this is more efficient. + * + * This currently only does the conversion for built-in matrices which + * already have transposed equivalents. Namely, gl_ModelViewProjectionMatrix + * and gl_TextureMatrix. + */ +#include "ir.h" +#include "ir_optimization.h" +#include "main/macros.h" + +namespace { +class matrix_flipper : public ir_hierarchical_visitor { +public: + matrix_flipper(exec_list *instructions) + { + progress = false; + mvp_transpose = NULL; + texmat_transpose = NULL; + + foreach_list(n, instructions) { + ir_instruction *ir = (ir_instruction *) n; + ir_variable *var = ir->as_variable(); + if (!var) + continue; + if (strcmp(var->name, "gl_ModelViewProjectionMatrixTranspose") == 0) + mvp_transpose = var; + if (strcmp(var->name, "gl_TextureMatrixTranspose") == 0) + texmat_transpose = var; + } + } + + ir_visitor_status visit_enter(ir_expression *ir); + + bool progress; + +private: + ir_variable *mvp_transpose; + ir_variable *texmat_transpose; +}; +} + +ir_visitor_status +matrix_flipper::visit_enter(ir_expression *ir) +{ + if (ir->operation != ir_binop_mul || + !ir->operands[0]->type->is_matrix() || + !ir->operands[1]->type->is_vector()) + return visit_continue; + + ir_variable *mat_var = ir->operands[0]->variable_referenced(); + if (!mat_var) + return visit_continue; + + if (mvp_transpose && + strcmp(mat_var->name, "gl_ModelViewProjectionMatrix") == 0) { + ir_dereference_variable *deref = ir->operands[0]->as_dereference_variable(); + assert(deref && deref->var == mat_var); + + void *mem_ctx = ralloc_parent(ir); + + ir->operands[0] = ir->operands[1]; + ir->operands[1] = new(mem_ctx) ir_dereference_variable(mvp_transpose); + + progress = true; + } else if (texmat_transpose && + strcmp(mat_var->name, "gl_TextureMatrix") == 0) { + ir_dereference_array *array_ref = ir->operands[0]->as_dereference_array(); + assert(array_ref != NULL); + ir_dereference_variable *var_ref = array_ref->array->as_dereference_variable(); + assert(var_ref && var_ref->var == mat_var); + + ir->operands[0] = ir->operands[1]; + ir->operands[1] = array_ref; + + var_ref->var = texmat_transpose; + + texmat_transpose->max_array_access = + MAX2(texmat_transpose->max_array_access, mat_var->max_array_access); + + progress = true; + } + + return visit_continue; +} + +bool +opt_flip_matrices(struct exec_list *instructions) +{ + matrix_flipper v(instructions); + + visit_list_elements(&v, instructions); + + return v.progress; +} diff --git a/mesalib/src/glsl/opt_if_simplification.cpp b/mesalib/src/glsl/opt_if_simplification.cpp index e1a23d94a..db59b131d 100644 --- a/mesalib/src/glsl/opt_if_simplification.cpp +++ b/mesalib/src/glsl/opt_if_simplification.cpp @@ -25,7 +25,8 @@ * \file opt_if_simplification.cpp * * Moves constant branches of if statements out to the surrounding - * instruction stream. + * instruction stream, and inverts if conditionals to avoid empty + * "then" blocks. */ #include "ir.h" @@ -101,6 +102,30 @@ ir_if_simplification_visitor::visit_leave(ir_if *ir) } ir->remove(); this->made_progress = true; + return visit_continue; + } + + /* Turn: + * + * if (cond) { + * } else { + * do_work(); + * } + * + * into : + * + * if (!cond) + * do_work(); + * + * which avoids control flow for "else" (which is usually more + * expensive than normal operations), and the "not" can usually be + * folded into the generation of "cond" anyway. + */ + if (ir->then_instructions.is_empty()) { + ir->condition = new(ralloc_parent(ir->condition)) + ir_expression(ir_unop_logic_not, ir->condition); + ir->else_instructions.move_nodes_to(&ir->then_instructions); + this->made_progress = true; } return visit_continue; diff --git a/mesalib/src/glsl/program.h b/mesalib/src/glsl/program.h index 46ce9dccc..6a76d4d54 100644 --- a/mesalib/src/glsl/program.h +++ b/mesalib/src/glsl/program.h @@ -16,9 +16,10 @@ * 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 - * BRIAN PAUL 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. + * 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. */ #include "main/core.h" diff --git a/mesalib/src/glsl/ralloc.c b/mesalib/src/glsl/ralloc.c index 59e71c48b..e79dad764 100644 --- a/mesalib/src/glsl/ralloc.c +++ b/mesalib/src/glsl/ralloc.c @@ -107,9 +107,13 @@ void * ralloc_size(const void *ctx, size_t size) { void *block = calloc(1, size + sizeof(ralloc_header)); + ralloc_header *info; + ralloc_header *parent; - ralloc_header *info = (ralloc_header *) block; - ralloc_header *parent = ctx != NULL ? get_header(ctx) : NULL; + if (unlikely(block == NULL)) + return NULL; + info = (ralloc_header *) block; + parent = ctx != NULL ? get_header(ctx) : NULL; add_child(parent, info); diff --git a/mesalib/src/glsl/standalone_scaffolding.cpp b/mesalib/src/glsl/standalone_scaffolding.cpp index 0c1f52f48..11cd6cdc0 100644 --- a/mesalib/src/glsl/standalone_scaffolding.cpp +++ b/mesalib/src/glsl/standalone_scaffolding.cpp @@ -104,6 +104,7 @@ void initialize_context_to_defaults(struct gl_context *ctx, gl_api api) ctx->Extensions.ARB_texture_cube_map_array = true; ctx->Extensions.ARB_texture_multisample = true; ctx->Extensions.ARB_texture_query_lod = true; + ctx->Extensions.ARB_gpu_shader5 = true; ctx->Const.GLSLVersion = 120; @@ -116,10 +117,22 @@ void initialize_context_to_defaults(struct gl_context *ctx, gl_api api) ctx->Const.VertexProgram.MaxUniformComponents = 512; ctx->Const.MaxVarying = 8; /* == gl_MaxVaryingFloats / 4 */ - ctx->Const.MaxVertexTextureImageUnits = 0; + ctx->Const.VertexProgram.MaxTextureImageUnits = 0; ctx->Const.MaxCombinedTextureImageUnits = 2; - ctx->Const.MaxTextureImageUnits = 2; + ctx->Const.FragmentProgram.MaxTextureImageUnits = 2; ctx->Const.FragmentProgram.MaxUniformComponents = 64; ctx->Const.MaxDrawBuffers = 1; + + /* Set up default shader compiler options. */ + struct gl_shader_compiler_options options; + memset(&options, 0, sizeof(options)); + options.MaxUnrollIterations = 32; + options.MaxIfDepth = UINT_MAX; + + /* Default pragma settings */ + options.DefaultPragmas.Optimize = true; + + for (int sh = 0; sh < MESA_SHADER_TYPES; ++sh) + memcpy(&ctx->ShaderCompilerOptions[sh], &options, sizeof(options)); } diff --git a/mesalib/src/glsl/standalone_scaffolding.h b/mesalib/src/glsl/standalone_scaffolding.h index 0a2cde8a3..7afb1c313 100644 --- a/mesalib/src/glsl/standalone_scaffolding.h +++ b/mesalib/src/glsl/standalone_scaffolding.h @@ -31,6 +31,7 @@ #ifndef STANDALONE_SCAFFOLDING_H #define STANDALONE_SCAFFOLDING_H +#include <assert.h> #include "main/mtypes.h" extern "C" void @@ -47,6 +48,22 @@ extern "C" void _mesa_shader_debug(struct gl_context *ctx, GLenum type, GLuint *id, const char *msg, int len); +static inline gl_shader_type +_mesa_shader_type_to_index(GLenum v) +{ + switch (v) { + case GL_VERTEX_SHADER: + return MESA_SHADER_VERTEX; + case GL_FRAGMENT_SHADER: + return MESA_SHADER_FRAGMENT; + case GL_GEOMETRY_SHADER: + return MESA_SHADER_GEOMETRY; + default: + assert(!"bad value in _mesa_shader_type_to_index()"); + return MESA_SHADER_TYPES; + } +} + /** * Initialize the given gl_context structure to a reasonable set of * defaults representing the minimum capabilities required by the diff --git a/mesalib/src/glsl/test_optpass.cpp b/mesalib/src/glsl/test_optpass.cpp index 117b0b006..fc10cbbde 100644 --- a/mesalib/src/glsl/test_optpass.cpp +++ b/mesalib/src/glsl/test_optpass.cpp @@ -54,7 +54,8 @@ static string read_stdin_to_eof() } static GLboolean -do_optimization(struct exec_list *ir, const char *optimization) +do_optimization(struct exec_list *ir, const char *optimization, + const struct gl_shader_compiler_options *options) { int int_0; int int_1; @@ -64,7 +65,7 @@ do_optimization(struct exec_list *ir, const char *optimization) if (sscanf(optimization, "do_common_optimization ( %d , %d ) ", &int_0, &int_1) == 2) { - return do_common_optimization(ir, int_0 != 0, false, int_1); + return do_common_optimization(ir, int_0 != 0, false, int_1, options); } else if (strcmp(optimization, "do_algebraic") == 0) { return do_algebraic(ir); } else if (strcmp(optimization, "do_constant_folding") == 0) { @@ -141,7 +142,8 @@ do_optimization(struct exec_list *ir, const char *optimization) static GLboolean do_optimization_passes(struct exec_list *ir, char **optimizations, - int num_optimizations, bool quiet) + int num_optimizations, bool quiet, + const struct gl_shader_compiler_options *options) { GLboolean overall_progress = false; @@ -150,7 +152,7 @@ do_optimization_passes(struct exec_list *ir, char **optimizations, if (!quiet) { printf("*** Running optimization %s...", optimization); } - GLboolean progress = do_optimization(ir, optimization); + GLboolean progress = do_optimization(ir, optimization, options); if (!quiet) { printf("%s\n", progress ? "progress" : "no progress"); } @@ -240,9 +242,11 @@ int test_optpass(int argc, char **argv) /* Optimization passes */ if (!state->error) { GLboolean progress; + const struct gl_shader_compiler_options *options = + &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader_type)]; do { progress = do_optimization_passes(shader->ir, &argv[optind], - argc - optind, quiet != 0); + argc - optind, quiet != 0, options); } while (loop && progress); } |