blob: 0250d5fb17e125316baa37bfdb60019de56c54d8 [file] [log] [blame]
// Copyright 2020 The Dawn & Tint Authors
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// 1. Redistributions of source code must retain the above copyright notice, this
// list of conditions and the following disclaimer.
//
// 2. Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// 3. Neither the name of the copyright holder nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "src/tint/lang/wgsl/resolver/resolver.h"
#include <algorithm>
#include <cmath>
#include <iomanip>
#include <limits>
#include <string_view>
#include <utility>
#include "src/tint/lang/core/builtin_type.h"
#include "src/tint/lang/core/constant/scalar.h"
#include "src/tint/lang/core/fluent_types.h"
#include "src/tint/lang/core/texel_format.h"
#include "src/tint/lang/core/type/abstract_float.h"
#include "src/tint/lang/core/type/abstract_int.h"
#include "src/tint/lang/core/type/array.h"
#include "src/tint/lang/core/type/atomic.h"
#include "src/tint/lang/core/type/builtin_structs.h"
#include "src/tint/lang/core/type/depth_multisampled_texture.h"
#include "src/tint/lang/core/type/depth_texture.h"
#include "src/tint/lang/core/type/external_texture.h"
#include "src/tint/lang/core/type/input_attachment.h"
#include "src/tint/lang/core/type/memory_view.h"
#include "src/tint/lang/core/type/multisampled_texture.h"
#include "src/tint/lang/core/type/pointer.h"
#include "src/tint/lang/core/type/reference.h"
#include "src/tint/lang/core/type/sampled_texture.h"
#include "src/tint/lang/core/type/sampler.h"
#include "src/tint/lang/core/type/storage_texture.h"
#include "src/tint/lang/wgsl/ast/alias.h"
#include "src/tint/lang/wgsl/ast/assignment_statement.h"
#include "src/tint/lang/wgsl/ast/attribute.h"
#include "src/tint/lang/wgsl/ast/break_statement.h"
#include "src/tint/lang/wgsl/ast/builtin_value_name.h"
#include "src/tint/lang/wgsl/ast/call_statement.h"
#include "src/tint/lang/wgsl/ast/continue_statement.h"
#include "src/tint/lang/wgsl/ast/disable_validation_attribute.h"
#include "src/tint/lang/wgsl/ast/discard_statement.h"
#include "src/tint/lang/wgsl/ast/for_loop_statement.h"
#include "src/tint/lang/wgsl/ast/id_attribute.h"
#include "src/tint/lang/wgsl/ast/if_statement.h"
#include "src/tint/lang/wgsl/ast/input_attachment_index_attribute.h"
#include "src/tint/lang/wgsl/ast/internal_attribute.h"
#include "src/tint/lang/wgsl/ast/interpolate_attribute.h"
#include "src/tint/lang/wgsl/ast/loop_statement.h"
#include "src/tint/lang/wgsl/ast/return_statement.h"
#include "src/tint/lang/wgsl/ast/switch_statement.h"
#include "src/tint/lang/wgsl/ast/traverse_expressions.h"
#include "src/tint/lang/wgsl/ast/unary_op_expression.h"
#include "src/tint/lang/wgsl/ast/variable_decl_statement.h"
#include "src/tint/lang/wgsl/ast/while_statement.h"
#include "src/tint/lang/wgsl/ast/workgroup_attribute.h"
#include "src/tint/lang/wgsl/intrinsic/ctor_conv.h"
#include "src/tint/lang/wgsl/intrinsic/dialect.h"
#include "src/tint/lang/wgsl/resolver/incomplete_type.h"
#include "src/tint/lang/wgsl/resolver/uniformity.h"
#include "src/tint/lang/wgsl/resolver/unresolved_identifier.h"
#include "src/tint/lang/wgsl/sem/array.h"
#include "src/tint/lang/wgsl/sem/break_if_statement.h"
#include "src/tint/lang/wgsl/sem/builtin_attribute.h"
#include "src/tint/lang/wgsl/sem/builtin_enum_expression.h"
#include "src/tint/lang/wgsl/sem/call.h"
#include "src/tint/lang/wgsl/sem/for_loop_statement.h"
#include "src/tint/lang/wgsl/sem/function.h"
#include "src/tint/lang/wgsl/sem/function_expression.h"
#include "src/tint/lang/wgsl/sem/if_statement.h"
#include "src/tint/lang/wgsl/sem/index_accessor_expression.h"
#include "src/tint/lang/wgsl/sem/load.h"
#include "src/tint/lang/wgsl/sem/loop_statement.h"
#include "src/tint/lang/wgsl/sem/materialize.h"
#include "src/tint/lang/wgsl/sem/member_accessor_expression.h"
#include "src/tint/lang/wgsl/sem/module.h"
#include "src/tint/lang/wgsl/sem/statement.h"
#include "src/tint/lang/wgsl/sem/struct.h"
#include "src/tint/lang/wgsl/sem/switch_statement.h"
#include "src/tint/lang/wgsl/sem/type_expression.h"
#include "src/tint/lang/wgsl/sem/value_constructor.h"
#include "src/tint/lang/wgsl/sem/value_conversion.h"
#include "src/tint/lang/wgsl/sem/variable.h"
#include "src/tint/lang/wgsl/sem/while_statement.h"
#include "src/tint/utils/containers/reverse.h"
#include "src/tint/utils/containers/transform.h"
#include "src/tint/utils/containers/vector.h"
#include "src/tint/utils/macros/compiler.h"
#include "src/tint/utils/macros/defer.h"
#include "src/tint/utils/macros/scoped_assignment.h"
#include "src/tint/utils/math/math.h"
#include "src/tint/utils/text/string.h"
#include "src/tint/utils/text/string_stream.h"
#include "src/tint/utils/text/styled_text.h"
#include "src/tint/utils/text/text_style.h"
using namespace tint::core::fluent_types; // NOLINT
namespace tint::resolver {
namespace {
/// ICE() is a wrapper around TINT_ICE() that includes a prefixed source location
#define ICE(SOURCE) TINT_ICE() << SOURCE << (SOURCE.file ? ": " : "")
using CtorConvIntrinsic = wgsl::intrinsic::CtorConv;
using OverloadFlag = core::intrinsic::OverloadFlag;
constexpr int64_t kMaxArrayElementCount = 65536;
constexpr uint32_t kMaxStatementDepth = 127;
constexpr size_t kMaxNestDepthOfCompositeType = 255;
} // namespace
Resolver::Resolver(ProgramBuilder* builder,
const wgsl::AllowedFeatures& allowed_features,
wgsl::ValidationMode mode)
: b(*builder),
diagnostics_(builder->Diagnostics()),
const_eval_(builder->constants, diagnostics_),
intrinsic_table_{builder->Types(), builder->Symbols()},
sem_(builder),
validator_(builder,
sem_,
enabled_extensions_,
allowed_features_,
mode,
atomic_composite_info_,
valid_type_storage_layouts_),
allowed_features_(allowed_features) {}
Resolver::~Resolver() = default;
bool Resolver::Resolve() {
if (diagnostics_.ContainsErrors()) {
return false;
}
b.Sem().Reserve(b.LastAllocatedNodeID());
// Pre-allocate the marked bitset with the total number of AST nodes.
marked_.Resize(b.ASTNodes().Count());
if (!DependencyGraph::Build(b.AST(), diagnostics_, dependencies_)) {
return false;
}
bool result = ResolveInternal();
if (TINT_UNLIKELY(!result && !diagnostics_.ContainsErrors())) {
TINT_ICE() << "resolving failed, but no error was raised";
}
if (!validator_.Enables(b.AST().Enables())) {
return false;
}
// Create the semantic module. Don't be tempted to std::move() these, they're used below.
auto* mod = b.create<sem::Module>(dependencies_.ordered_globals, enabled_extensions_);
ApplyDiagnosticSeverities(mod);
b.Sem().SetModule(mod);
const bool disable_uniformity_analysis =
enabled_extensions_.Contains(wgsl::Extension::kChromiumDisableUniformityAnalysis);
if (result && !disable_uniformity_analysis) {
// Run the uniformity analysis, which requires a complete semantic module.
if (!AnalyzeUniformity(b, dependencies_)) {
return false;
}
}
return result;
}
bool Resolver::ResolveInternal() {
Mark(&b.AST());
// Process all module-scope declarations in dependency order.
Vector<const ast::DiagnosticControl*, 4> diagnostic_controls;
for (auto* decl : dependencies_.ordered_globals) {
Mark(decl);
if (!Switch<bool>(
decl, //
[&](const ast::DiagnosticDirective* d) {
diagnostic_controls.Push(&d->control);
return DiagnosticControl(d->control);
},
[&](const ast::Enable* e) { return Enable(e); },
[&](const ast::Requires* r) { return Requires(r); },
[&](const ast::TypeDecl* td) { return TypeDecl(td); },
[&](const ast::Function* func) { return Function(func); },
[&](const ast::Variable* var) { return GlobalVariable(var); },
[&](const ast::ConstAssert* ca) { return ConstAssert(ca); }, //
TINT_ICE_ON_NO_MATCH)) {
return false;
}
}
if (!AllocateOverridableConstantIds()) {
return false;
}
SetShadows();
if (!validator_.DiagnosticControls(diagnostic_controls, "directive")) {
return false;
}
if (!validator_.PipelineStages(entry_points_)) {
return false;
}
if (!validator_.ModuleScopeVarUsages(entry_points_)) {
return false;
}
bool result = true;
for (auto* node : b.ASTNodes().Objects()) {
if (TINT_UNLIKELY(!marked_[node->node_id.value])) {
ICE(node->source) << "AST node '" << node->TypeInfo().name
<< "' was not reached by the resolver\n"
<< "Pointer: " << node;
}
}
return result;
}
sem::Variable* Resolver::Variable(const ast::Variable* v, bool is_global) {
Mark(v->name);
return Switch(
v, //
[&](const ast::Var* var) { return Var(var, is_global); },
[&](const ast::Let* let) { return Let(let); },
[&](const ast::Override* override) { return Override(override); },
[&](const ast::Const* const_) { return Const(const_, is_global); }, //
TINT_ICE_ON_NO_MATCH);
}
sem::Variable* Resolver::Let(const ast::Let* v) {
auto* sem = b.create<sem::LocalVariable>(v, current_statement_);
sem->SetStage(core::EvaluationStage::kRuntime);
b.Sem().Add(v, sem);
// If the variable has a declared type, resolve it.
if (v->type) {
auto* ty = Type(v->type);
if (TINT_UNLIKELY(!ty)) {
return nullptr;
}
sem->SetType(ty);
}
for (auto* attribute : v->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::InternalAttribute* attr) -> bool { return InternalAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute,
StyledText{} << style::Keyword("let") << " declaration");
return false;
});
if (!ok) {
return nullptr;
}
}
if (TINT_UNLIKELY(!v->initializer)) {
AddError(v->source) << style::Keyword("let") << " declaration must have an initializer";
return nullptr;
}
auto* rhs = Load(Materialize(ValueExpression(v->initializer), sem->Type()));
if (TINT_UNLIKELY(!rhs)) {
return nullptr;
}
sem->SetInitializer(rhs);
// If the variable has no declared type, infer it from the RHS
if (!sem->Type()) {
sem->SetType(rhs->Type()->UnwrapRef()); // Implicit load of RHS
}
if (TINT_UNLIKELY(rhs && !validator_.VariableInitializer(v, sem->Type(), rhs))) {
return nullptr;
}
if (!ApplyAddressSpaceUsageToType(core::AddressSpace::kUndefined,
const_cast<core::type::Type*>(sem->Type()), v->source)) {
AddNote(v->source) << "while instantiating " << style::Keyword("let ")
<< style::Variable(v->name->symbol.NameView());
return nullptr;
}
return sem;
}
sem::Variable* Resolver::Override(const ast::Override* v) {
auto* sem = b.create<sem::GlobalVariable>(v);
b.Sem().Add(v, sem);
sem->SetStage(core::EvaluationStage::kOverride);
on_transitively_reference_global_.Push([&](const sem::GlobalVariable* ref) {
if (ref->Declaration()->Is<ast::Override>()) {
sem->AddTransitivelyReferencedOverride(ref);
}
});
TINT_DEFER(on_transitively_reference_global_.Pop());
// If the variable has a declared type, resolve it.
const core::type::Type* ty = nullptr;
if (v->type) {
ty = Type(v->type);
if (!ty) {
return nullptr;
}
}
// Does the variable have an initializer?
const sem::ValueExpression* init = nullptr;
if (v->initializer) {
// Note: RHS must be a const or override expression, which excludes references.
// So there's no need to load or unwrap references here.
ExprEvalStageConstraint constraint{core::EvaluationStage::kOverride,
"override initializer"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
init = Materialize(ValueExpression(v->initializer), ty);
if (TINT_UNLIKELY(!init)) {
return nullptr;
}
sem->SetInitializer(init);
// If the variable has no declared type, infer it from the initializer
if (!ty) {
ty = init->Type();
}
} else if (!ty) {
AddError(v->source) << "override declaration requires a type or initializer";
return nullptr;
}
sem->SetType(ty);
if (init && !validator_.VariableInitializer(v, ty, init)) {
return nullptr;
}
if (!ApplyAddressSpaceUsageToType(core::AddressSpace::kUndefined,
const_cast<core::type::Type*>(ty), v->source)) {
AddNote(v->source) << "while instantiating " << style::Keyword("override ")
<< style::Variable(v->name->symbol.NameView());
return nullptr;
}
for (auto* attribute : v->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::IdAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@id"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return false;
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source)
<< style::Attribute("@id") << " must be an " << style::Type("i32") << " or "
<< style::Type("u32") << " value";
return false;
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source)
<< style::Attribute("@id") << " value must be non-negative";
return false;
}
if (value > std::numeric_limits<decltype(OverrideId::value)>::max()) {
AddError(attr->source)
<< style::Attribute("@id") << " value must be between 0 and "
<< std::numeric_limits<decltype(OverrideId::value)>::max();
return false;
}
auto o = OverrideId{static_cast<decltype(OverrideId::value)>(value)};
sem->Attributes().override_id = o;
// Track the constant IDs that are specified in the shader.
override_ids_.Add(o, sem);
return true;
},
[&](Default) {
ErrorInvalidAttribute(attribute,
StyledText{} << style::Keyword("override") << " declaration");
return false;
});
if (!ok) {
return nullptr;
}
}
return sem;
}
sem::Variable* Resolver::Const(const ast::Const* c, bool is_global) {
sem::Variable* sem = nullptr;
sem::GlobalVariable* global = nullptr;
if (is_global) {
global = b.create<sem::GlobalVariable>(c);
sem = global;
} else {
sem = b.create<sem::LocalVariable>(c, current_statement_);
}
b.Sem().Add(c, sem);
for (auto* attribute : c->attributes) {
Mark(attribute);
bool ok =
Switch(attribute, //
[&](Default) {
ErrorInvalidAttribute(
attribute, StyledText{} << style::Keyword("const") << " declaration");
return false;
});
if (!ok) {
return nullptr;
}
}
if (TINT_UNLIKELY(!c->initializer)) {
AddError(c->source) << "'const' declaration must have an initializer";
return nullptr;
}
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "const initializer"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
const auto* init = ValueExpression(c->initializer);
if (TINT_UNLIKELY(!init)) {
return nullptr;
}
// Note: RHS must be a const expression, which excludes references.
// So there's no need to load or unwrap references here.
// If the variable has a declared type, resolve it.
const core::type::Type* ty = nullptr;
if (c->type) {
ty = Type(c->type);
if (TINT_UNLIKELY(!ty)) {
return nullptr;
}
}
if (ty) {
// If an explicit type was specified, materialize to that type
init = Materialize(init, ty);
if (TINT_UNLIKELY(!init)) {
return nullptr;
}
} else {
// If no type was specified, infer it from the RHS
ty = init->Type();
}
sem->SetInitializer(init);
sem->SetStage(core::EvaluationStage::kConstant);
sem->SetConstantValue(init->ConstantValue());
sem->SetType(ty);
if (!validator_.VariableInitializer(c, ty, init)) {
return nullptr;
}
if (!ApplyAddressSpaceUsageToType(core::AddressSpace::kUndefined,
const_cast<core::type::Type*>(ty), c->source)) {
AddNote(c->source) << "while instantiating 'const' " << c->name->symbol.NameView();
return nullptr;
}
return sem;
}
sem::Variable* Resolver::Var(const ast::Var* var, bool is_global) {
sem::Variable* sem = nullptr;
sem::GlobalVariable* global = nullptr;
if (is_global) {
global = b.create<sem::GlobalVariable>(var);
sem = global;
} else {
sem = b.create<sem::LocalVariable>(var, current_statement_);
}
sem->SetStage(core::EvaluationStage::kRuntime);
b.Sem().Add(var, sem);
if (is_global) {
on_transitively_reference_global_.Push([&](const sem::GlobalVariable* ref) {
if (ref->Declaration()->Is<ast::Override>()) {
global->AddTransitivelyReferencedOverride(ref);
}
});
}
TINT_DEFER({
if (is_global) {
on_transitively_reference_global_.Pop();
}
});
// If the variable has a declared type, resolve it.
const core::type::Type* storage_ty = nullptr;
if (auto ty = var->type) {
storage_ty = Type(ty);
if (TINT_UNLIKELY(!storage_ty)) {
return nullptr;
}
}
// Does the variable have a initializer?
if (var->initializer) {
ExprEvalStageConstraint constraint{
is_global ? core::EvaluationStage::kOverride : core::EvaluationStage::kRuntime,
"var initializer",
};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* init = Load(Materialize(ValueExpression(var->initializer), storage_ty));
if (TINT_UNLIKELY(!init)) {
return nullptr;
}
sem->SetInitializer(init);
// If the variable has no declared type, infer it from the RHS
if (!storage_ty) {
storage_ty = init->Type();
}
}
if (!storage_ty) {
AddError(var->source) << "var declaration requires a type or initializer";
return nullptr;
}
if (var->declared_address_space) {
auto space = AddressSpaceExpression(var->declared_address_space);
if (TINT_UNLIKELY(!space)) {
return nullptr;
}
sem->SetAddressSpace(space->Value());
} else {
// No declared address space. Infer from usage / type.
if (!is_global) {
sem->SetAddressSpace(core::AddressSpace::kFunction);
} else if (storage_ty->UnwrapRef()->is_handle()) {
// https://gpuweb.github.io/gpuweb/wgsl/#module-scope-variables
// If the store type is a texture type or a sampler type, then the
// variable declaration must not have a address space attribute. The
// address space will always be handle.
sem->SetAddressSpace(core::AddressSpace::kHandle);
}
}
if (!is_global && sem->AddressSpace() != core::AddressSpace::kFunction &&
validator_.IsValidationEnabled(var->attributes,
ast::DisabledValidation::kIgnoreAddressSpace)) {
AddError(var->source)
<< "function-scope 'var' declaration must use 'function' address space";
return nullptr;
}
if (var->declared_access) {
auto expr = AccessExpression(var->declared_access);
if (!expr) {
return nullptr;
}
sem->SetAccess(expr->Value());
} else {
sem->SetAccess(DefaultAccessForAddressSpace(sem->AddressSpace()));
}
sem->SetType(b.create<core::type::Reference>(sem->AddressSpace(), storage_ty, sem->Access()));
if (sem->Initializer() &&
!validator_.VariableInitializer(var, storage_ty, sem->Initializer())) {
return nullptr;
}
if (!ApplyAddressSpaceUsageToType(sem->AddressSpace(),
const_cast<core::type::Type*>(sem->Type()),
var->type ? var->type->source : var->source)) {
AddNote(var->source) << "while instantiating 'var' " << var->name->symbol.NameView();
return nullptr;
}
if (is_global) {
bool has_io_address_space = sem->AddressSpace() == core::AddressSpace::kIn ||
sem->AddressSpace() == core::AddressSpace::kOut;
std::optional<uint32_t> group, binding, input_attachment_index;
for (auto* attribute : var->attributes) {
Mark(attribute);
enum Status { kSuccess, kErrored, kInvalid };
auto res = Switch(
attribute, //
[&](const ast::BindingAttribute* attr) {
auto value = BindingAttribute(attr);
if (value != Success) {
return kErrored;
}
binding = value.Get();
return kSuccess;
},
[&](const ast::GroupAttribute* attr) {
auto value = GroupAttribute(attr);
if (value != Success) {
return kErrored;
}
group = value.Get();
return kSuccess;
},
[&](const ast::InputAttachmentIndexAttribute* attr) {
auto value = InputAttachmentIndexAttribute(attr);
if (value != Success) {
return kErrored;
}
input_attachment_index = value.Get();
return kSuccess;
},
[&](const ast::LocationAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
auto value = LocationAttribute(attr);
if (value != Success) {
return kErrored;
}
global->Attributes().location = value.Get();
return kSuccess;
},
[&](const ast::BlendSrcAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
auto value = BlendSrcAttribute(attr);
if (value != Success) {
return kErrored;
}
global->Attributes().blend_src = value.Get();
return kSuccess;
},
[&](const ast::ColorAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
auto value = ColorAttribute(attr);
if (value != Success) {
return kErrored;
}
global->Attributes().color = value.Get();
return kSuccess;
},
[&](const ast::BuiltinAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
return BuiltinAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](const ast::InterpolateAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
return InterpolateAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](const ast::InvariantAttribute* attr) {
if (!has_io_address_space) {
return kInvalid;
}
return InvariantAttribute(attr) ? kSuccess : kErrored;
},
[&](const ast::InternalAttribute* attr) {
return InternalAttribute(attr) ? kSuccess : kErrored;
},
[&](Default) { return kInvalid; });
switch (res) {
case kSuccess:
break;
case kErrored:
return nullptr;
case kInvalid:
ErrorInvalidAttribute(attribute,
StyledText{} << "module-scope " << style::Keyword("var"));
return nullptr;
}
}
if (group && binding) {
global->Attributes().binding_point = BindingPoint{group.value(), binding.value()};
}
if (input_attachment_index) {
global->Attributes().input_attachment_index = input_attachment_index;
}
} else {
for (auto* attribute : var->attributes) {
Mark(attribute);
bool ok = Switch(
attribute,
[&](const ast::InternalAttribute* attr) { return InternalAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(
attribute, StyledText{} << "function-scope " << style::Keyword("var"));
return false;
});
if (!ok) {
return nullptr;
}
}
}
return sem;
}
sem::Parameter* Resolver::Parameter(const ast::Parameter* param,
const ast::Function* func,
uint32_t index) {
Mark(param->name);
auto* sem = b.create<sem::Parameter>(param, index);
b.Sem().Add(param, sem);
auto add_note = [&] {
AddNote(param->source) << "while instantiating parameter "
<< param->name->symbol.NameView();
};
if (func->IsEntryPoint()) {
std::optional<uint32_t> group, binding;
for (auto* attribute : param->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::LocationAttribute* attr) {
auto value = LocationAttribute(attr);
if (TINT_UNLIKELY(value != Success)) {
return false;
}
sem->Attributes().location = value.Get();
return true;
},
[&](const ast::ColorAttribute* attr) {
auto value = ColorAttribute(attr);
if (TINT_UNLIKELY(value != Success)) {
return false;
}
sem->Attributes().color = value.Get();
return true;
},
[&](const ast::BuiltinAttribute* attr) {
return BuiltinAttribute(attr) == Success;
},
[&](const ast::InvariantAttribute* attr) -> bool {
return InvariantAttribute(attr);
},
[&](const ast::InterpolateAttribute* attr) {
return InterpolateAttribute(attr) == Success;
},
[&](const ast::InternalAttribute* attr) -> bool { return InternalAttribute(attr); },
[&](const ast::GroupAttribute* attr) {
if (validator_.IsValidationEnabled(
param->attributes, ast::DisabledValidation::kEntryPointParameter)) {
ErrorInvalidAttribute(attribute, StyledText{} << "function parameters");
return false;
}
auto value = GroupAttribute(attr);
if (TINT_UNLIKELY(value != Success)) {
return false;
}
group = value.Get();
return true;
},
[&](const ast::BindingAttribute* attr) -> bool {
if (validator_.IsValidationEnabled(
param->attributes, ast::DisabledValidation::kEntryPointParameter)) {
ErrorInvalidAttribute(attribute, StyledText{} << "function parameters");
return false;
}
auto value = BindingAttribute(attr);
if (TINT_UNLIKELY(value != Success)) {
return false;
}
binding = value.Get();
return true;
},
[&](Default) {
ErrorInvalidAttribute(attribute, StyledText{} << "function parameters");
return false;
});
if (!ok) {
return nullptr;
}
}
if (group && binding) {
sem->Attributes().binding_point = BindingPoint{group.value(), binding.value()};
}
} else {
for (auto* attribute : param->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::InternalAttribute* attr) -> bool { return InternalAttribute(attr); },
[&](Default) {
if (attribute->IsAnyOf<ast::LocationAttribute, ast::BuiltinAttribute,
ast::InvariantAttribute, ast::InterpolateAttribute>()) {
ErrorInvalidAttribute(
attribute, StyledText{} << "non-entry point function parameters");
} else {
ErrorInvalidAttribute(attribute, StyledText{} << "function parameters");
}
return false;
});
if (!ok) {
return nullptr;
}
}
}
if (!validator_.NoDuplicateAttributes(param->attributes)) {
return nullptr;
}
core::type::Type* ty = Type(param->type);
if (TINT_UNLIKELY(!ty)) {
return nullptr;
}
sem->SetType(ty);
if (!ApplyAddressSpaceUsageToType(core::AddressSpace::kUndefined, ty, param->type->source)) {
add_note();
return nullptr;
}
if (auto* ptr = ty->As<core::type::Pointer>()) {
// For MSL, we push module-scope variables into the entry point as pointer
// parameters, so we also need to handle their store type.
if (!ApplyAddressSpaceUsageToType(ptr->AddressSpace(),
const_cast<core::type::Type*>(ptr->StoreType()),
param->source)) {
add_note();
return nullptr;
}
}
if (!validator_.Parameter(sem)) {
return nullptr;
}
return sem;
}
core::Access Resolver::DefaultAccessForAddressSpace(core::AddressSpace address_space) {
// https://gpuweb.github.io/gpuweb/wgsl/#storage-class
switch (address_space) {
case core::AddressSpace::kStorage:
case core::AddressSpace::kUniform:
case core::AddressSpace::kHandle:
return core::Access::kRead;
default:
break;
}
return core::Access::kReadWrite;
}
bool Resolver::AllocateOverridableConstantIds() {
constexpr size_t kLimit = std::numeric_limits<decltype(OverrideId::value)>::max();
// The next pipeline constant ID to try to allocate.
OverrideId next_id;
bool ids_exhausted = false;
auto increment_next_id = [&] {
if (next_id.value == kLimit) {
ids_exhausted = true;
} else {
next_id.value = next_id.value + 1;
}
};
// Allocate constant IDs in global declaration order, so that they are
// deterministic.
// TODO(crbug.com/tint/1192): If a transform changes the order or removes an
// unused constant, the allocation may change on the next Resolver pass.
for (auto* decl : b.AST().GlobalDeclarations()) {
auto* override = decl->As<ast::Override>();
if (!override) {
continue;
}
auto* sem = sem_.Get(override);
OverrideId id;
if (auto sem_id = sem->Attributes().override_id) {
id = *sem_id;
} else {
// No ID was specified, so allocate the next available ID.
while (!ids_exhausted && override_ids_.Contains(next_id)) {
increment_next_id();
}
if (ids_exhausted) {
AddError(decl->source)
<< "number of 'override' variables exceeded limit of " << kLimit;
return false;
}
id = next_id;
increment_next_id();
}
const_cast<sem::GlobalVariable*>(sem)->Attributes().override_id = id;
}
return true;
}
void Resolver::SetShadows() {
for (auto& it : dependencies_.shadows) {
CastableBase* shadowed = sem_.Get(it.value);
if (TINT_UNLIKELY(!shadowed)) {
ICE(it.value->source) << "AST node '" << it.value->TypeInfo().name
<< "' had no semantic info\n"
<< "Pointer: " << it.value;
}
Switch(
sem_.Get(it.key.Value()), //
[&](sem::LocalVariable* local) { local->SetShadows(shadowed); },
[&](sem::Parameter* param) { param->SetShadows(shadowed); });
}
}
sem::GlobalVariable* Resolver::GlobalVariable(const ast::Variable* v) {
auto* sem = As<sem::GlobalVariable>(Variable(v, /* is_global */ true));
if (!sem) {
return nullptr;
}
if (!validator_.NoDuplicateAttributes(v->attributes)) {
return nullptr;
}
if (!validator_.GlobalVariable(sem, override_ids_)) {
return nullptr;
}
return sem;
}
sem::Statement* Resolver::ConstAssert(const ast::ConstAssert* assertion) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "const assertion"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* expr = ValueExpression(assertion->condition);
if (!expr) {
return nullptr;
}
auto* cond = expr->ConstantValue();
if (auto* ty = cond->Type(); !ty->Is<core::type::Bool>()) {
AddError(assertion->condition->source)
<< "const assertion condition must be a bool, got '" << ty->FriendlyName() << "'";
return nullptr;
}
if (!cond->ValueAs<bool>()) {
AddError(assertion->source) << "const assertion failed";
return nullptr;
}
auto* sem = b.create<sem::Statement>(assertion, current_compound_statement_, current_function_);
b.Sem().Add(assertion, sem);
return sem;
}
sem::Function* Resolver::Function(const ast::Function* decl) {
Mark(decl->name);
auto* func = b.create<sem::Function>(decl);
b.Sem().Add(decl, func);
TINT_SCOPED_ASSIGNMENT(current_function_, func);
on_transitively_reference_global_.Push([&](const sem::GlobalVariable* ref) { //
func->AddDirectlyReferencedGlobal(ref);
});
TINT_DEFER(on_transitively_reference_global_.Pop());
validator_.DiagnosticFilters().Push();
TINT_DEFER(validator_.DiagnosticFilters().Pop());
for (auto* attribute : decl->attributes) {
Mark(attribute);
bool ok = Switch(
attribute,
[&](const ast::DiagnosticAttribute* attr) { return DiagnosticAttribute(attr); },
[&](const ast::StageAttribute* attr) { return StageAttribute(attr); },
[&](const ast::MustUseAttribute* attr) { return MustUseAttribute(attr); },
[&](const ast::WorkgroupAttribute* attr) {
auto value = WorkgroupAttribute(attr);
if (value != Success) {
return false;
}
func->SetWorkgroupSize(value.Get());
return true;
},
[&](const ast::InternalAttribute* attr) { return InternalAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute, StyledText{} << "functions");
return false;
});
if (!ok) {
return nullptr;
}
}
if (!validator_.NoDuplicateAttributes(decl->attributes)) {
return nullptr;
}
// Resolve all the parameters
uint32_t parameter_index = 0;
Hashmap<Symbol, Source, 8> parameter_names;
for (auto* param : decl->params) {
Mark(param);
{ // Check the parameter name is unique for the function
if (auto added = parameter_names.Add(param->name->symbol, param->source); !added) {
auto name = param->name->symbol.NameView();
AddError(param->source) << "redefinition of parameter '" << name << "'";
AddNote(added.value) << "previous definition is here";
return nullptr;
}
}
auto* p = Parameter(param, decl, parameter_index++);
if (!p) {
return nullptr;
}
func->AddParameter(p);
auto* p_ty = const_cast<core::type::Type*>(p->Type());
if (auto* str = p_ty->As<core::type::Struct>()) {
switch (decl->PipelineStage()) {
case ast::PipelineStage::kVertex:
str->AddUsage(core::type::PipelineStageUsage::kVertexInput);
break;
case ast::PipelineStage::kFragment:
str->AddUsage(core::type::PipelineStageUsage::kFragmentInput);
break;
case ast::PipelineStage::kCompute:
str->AddUsage(core::type::PipelineStageUsage::kComputeInput);
break;
case ast::PipelineStage::kNone:
break;
}
}
}
// Resolve the return type
core::type::Type* return_type = nullptr;
if (auto ty = decl->return_type) {
return_type = Type(ty);
if (!return_type) {
return nullptr;
}
} else {
return_type = b.create<core::type::Void>();
}
func->SetReturnType(return_type);
if (decl->IsEntryPoint()) {
// Determine if the return type has a location
bool permissive = validator_.IsValidationDisabled(
decl->attributes, ast::DisabledValidation::kEntryPointParameter) ||
validator_.IsValidationDisabled(
decl->attributes, ast::DisabledValidation::kFunctionParameter);
for (auto* attribute : decl->return_type_attributes) {
Mark(attribute);
enum Status { kSuccess, kErrored, kInvalid };
auto res = Switch(
attribute, //
[&](const ast::LocationAttribute* attr) {
auto value = LocationAttribute(attr);
if (value != Success) {
return kErrored;
}
func->SetReturnLocation(value.Get());
return kSuccess;
},
[&](const ast::BlendSrcAttribute* attr) {
if (!permissive) {
return kInvalid;
}
auto value = BlendSrcAttribute(attr);
if (value != Success) {
return kErrored;
}
func->SetReturnIndex(value.Get());
return kSuccess;
},
[&](const ast::BuiltinAttribute* attr) {
return BuiltinAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](const ast::InternalAttribute* attr) {
return InternalAttribute(attr) ? kSuccess : kErrored;
},
[&](const ast::InterpolateAttribute* attr) {
return InterpolateAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](const ast::InvariantAttribute* attr) {
return InvariantAttribute(attr) ? kSuccess : kErrored;
},
[&](const ast::BindingAttribute* attr) {
if (!permissive) {
return kInvalid;
}
return BindingAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](const ast::GroupAttribute* attr) {
if (!permissive) {
return kInvalid;
}
return GroupAttribute(attr) == Success ? kSuccess : kErrored;
},
[&](Default) { return kInvalid; });
switch (res) {
case kSuccess:
break;
case kErrored:
return nullptr;
case kInvalid:
ErrorInvalidAttribute(attribute, StyledText{} << "entry point return types");
return nullptr;
}
}
} else {
for (auto* attribute : decl->return_type_attributes) {
Mark(attribute);
bool ok =
Switch(attribute, //
[&](Default) {
ErrorInvalidAttribute(
attribute, StyledText{} << "non-entry point function return types");
return false;
});
if (!ok) {
return nullptr;
}
}
}
if (auto* str = return_type->As<core::type::Struct>()) {
if (!ApplyAddressSpaceUsageToType(core::AddressSpace::kUndefined, str,
decl->return_type->source)) {
AddNote(decl->return_type->source)
<< "while instantiating return type for " << decl->name->symbol.NameView();
return nullptr;
}
switch (decl->PipelineStage()) {
case ast::PipelineStage::kVertex:
str->AddUsage(core::type::PipelineStageUsage::kVertexOutput);
break;
case ast::PipelineStage::kFragment:
str->AddUsage(core::type::PipelineStageUsage::kFragmentOutput);
break;
case ast::PipelineStage::kCompute:
str->AddUsage(core::type::PipelineStageUsage::kComputeOutput);
break;
case ast::PipelineStage::kNone:
break;
}
}
ApplyDiagnosticSeverities(func);
if (decl->IsEntryPoint()) {
entry_points_.Push(func);
}
if (decl->body) {
Mark(decl->body);
if (TINT_UNLIKELY(current_compound_statement_)) {
ICE(decl->body->source)
<< "Resolver::Function() called with a current compound statement";
}
auto* body = StatementScope(decl->body, b.create<sem::FunctionBlockStatement>(func),
[&] { return Statements(decl->body->statements); });
if (!body) {
return nullptr;
}
func->Behaviors() = body->Behaviors();
if (func->Behaviors().Contains(sem::Behavior::kReturn)) {
// https://www.w3.org/TR/WGSL/#behaviors-rules
// We assign a behavior to each function: it is its body’s behavior
// (treating the body as a regular statement), with any "Return" replaced
// by "Next".
func->Behaviors().Remove(sem::Behavior::kReturn);
func->Behaviors().Add(sem::Behavior::kNext);
}
}
if (!validator_.NoDuplicateAttributes(decl->return_type_attributes)) {
return nullptr;
}
auto stage = current_function_ ? current_function_->Declaration()->PipelineStage()
: ast::PipelineStage::kNone;
if (!validator_.Function(func, stage)) {
return nullptr;
}
// If this is an entry point, mark all transitively called functions as being
// used by this entry point.
if (decl->IsEntryPoint()) {
for (auto* f : func->TransitivelyCalledFunctions()) {
const_cast<sem::Function*>(f)->AddAncestorEntryPoint(func);
}
}
return func;
}
bool Resolver::Statements(VectorRef<const ast::Statement*> stmts) {
sem::Behaviors behaviors{sem::Behavior::kNext};
bool reachable = true;
for (auto* stmt : stmts) {
Mark(stmt);
auto* sem = Statement(stmt);
if (!sem) {
return false;
}
// s1 s2:(B1∖{Next}) ∪ B2
sem->SetIsReachable(reachable);
if (reachable) {
behaviors = (behaviors - sem::Behavior::kNext) + sem->Behaviors();
}
reachable = reachable && sem->Behaviors().Contains(sem::Behavior::kNext);
}
current_statement_->Behaviors() = behaviors;
if (!validator_.Statements(stmts)) {
return false;
}
return true;
}
sem::Statement* Resolver::Statement(const ast::Statement* stmt) {
return Switch(
stmt,
// Compound statements. These create their own sem::CompoundStatement
// bindings.
[&](const ast::BlockStatement* s) { return BlockStatement(s); },
[&](const ast::ForLoopStatement* s) { return ForLoopStatement(s); },
[&](const ast::LoopStatement* s) { return LoopStatement(s); },
[&](const ast::WhileStatement* s) { return WhileStatement(s); },
[&](const ast::IfStatement* s) { return IfStatement(s); },
[&](const ast::SwitchStatement* s) { return SwitchStatement(s); },
// Non-Compound statements
[&](const ast::AssignmentStatement* s) { return AssignmentStatement(s); },
[&](const ast::BreakStatement* s) { return BreakStatement(s); },
[&](const ast::BreakIfStatement* s) { return BreakIfStatement(s); },
[&](const ast::CallStatement* s) { return CallStatement(s); },
[&](const ast::CompoundAssignmentStatement* s) { return CompoundAssignmentStatement(s); },
[&](const ast::ContinueStatement* s) { return ContinueStatement(s); },
[&](const ast::DiscardStatement* s) { return DiscardStatement(s); },
[&](const ast::IncrementDecrementStatement* s) { return IncrementDecrementStatement(s); },
[&](const ast::ReturnStatement* s) { return ReturnStatement(s); },
[&](const ast::VariableDeclStatement* s) { return VariableDeclStatement(s); },
[&](const ast::ConstAssert* s) { return ConstAssert(s); },
// Error cases
[&](const ast::CaseStatement*) {
AddError(stmt->source) << "case statement can only be used inside a switch statement";
return nullptr;
},
[&](Default) {
AddError(stmt->source)
<< "unknown statement type: " << std::string(stmt->TypeInfo().name);
return nullptr;
});
}
sem::CaseStatement* Resolver::CaseStatement(const ast::CaseStatement* stmt,
const core::type::Type* ty) {
auto* sem = b.create<sem::CaseStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
sem->Selectors().reserve(stmt->selectors.Length());
for (auto* sel : stmt->selectors) {
Mark(sel);
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "case selector"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
const core::constant::Value* const_value = nullptr;
if (!sel->IsDefault()) {
// The sem statement was created in the switch when attempting to determine the
// common type.
auto* materialized = Materialize(sem_.GetVal(sel->expr), ty);
if (!materialized) {
return false;
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(sel->source) << "case selector must be an i32 or u32 value";
return false;
}
const_value = materialized->ConstantValue();
if (!const_value) {
AddError(sel->source) << "case selector must be a constant expression";
return false;
}
}
sem->Selectors().emplace_back(b.create<sem::CaseSelector>(sel, const_value));
}
Mark(stmt->body);
auto* body = BlockStatement(stmt->body);
if (!body) {
return false;
}
sem->SetBlock(body);
sem->Behaviors() = body->Behaviors();
return true;
});
}
sem::IfStatement* Resolver::IfStatement(const ast::IfStatement* stmt) {
auto* sem = b.create<sem::IfStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto* cond = Load(ValueExpression(stmt->condition));
if (!cond) {
return false;
}
sem->SetCondition(cond);
sem->Behaviors() = cond->Behaviors();
sem->Behaviors().Remove(sem::Behavior::kNext);
Mark(stmt->body);
auto* body = b.create<sem::BlockStatement>(stmt->body, current_compound_statement_,
current_function_);
if (!StatementScope(stmt->body, body, [&] { return Statements(stmt->body->statements); })) {
return false;
}
sem->Behaviors().Add(body->Behaviors());
if (stmt->else_statement) {
Mark(stmt->else_statement);
auto* else_sem = Statement(stmt->else_statement);
if (!else_sem) {
return false;
}
sem->Behaviors().Add(else_sem->Behaviors());
} else {
// https://www.w3.org/TR/WGSL/#behaviors-rules
// if statements without an else branch are treated as if they had an
// empty else branch (which adds Next to their behavior)
sem->Behaviors().Add(sem::Behavior::kNext);
}
return validator_.IfStatement(sem);
});
}
sem::BlockStatement* Resolver::BlockStatement(const ast::BlockStatement* stmt) {
auto* sem = b.create<sem::BlockStatement>(stmt->As<ast::BlockStatement>(),
current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] { return Statements(stmt->statements); });
}
sem::LoopStatement* Resolver::LoopStatement(const ast::LoopStatement* stmt) {
auto* sem = b.create<sem::LoopStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
Mark(stmt->body);
auto* body = b.create<sem::LoopBlockStatement>(stmt->body, current_compound_statement_,
current_function_);
return StatementScope(stmt->body, body, [&] {
if (!Statements(stmt->body->statements)) {
return false;
}
auto& behaviors = sem->Behaviors();
behaviors = body->Behaviors();
if (stmt->continuing) {
Mark(stmt->continuing);
auto* continuing = StatementScope(
stmt->continuing,
b.create<sem::LoopContinuingBlockStatement>(
stmt->continuing, current_compound_statement_, current_function_),
[&] { return Statements(stmt->continuing->statements); });
if (!continuing) {
return false;
}
behaviors.Add(continuing->Behaviors());
}
if (behaviors.Contains(sem::Behavior::kBreak)) { // Does the loop exit?
behaviors.Add(sem::Behavior::kNext);
} else {
behaviors.Remove(sem::Behavior::kNext);
}
behaviors.Remove(sem::Behavior::kBreak, sem::Behavior::kContinue);
return validator_.LoopStatement(sem);
});
});
}
sem::ForLoopStatement* Resolver::ForLoopStatement(const ast::ForLoopStatement* stmt) {
auto* sem =
b.create<sem::ForLoopStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto& behaviors = sem->Behaviors();
if (auto* initializer = stmt->initializer) {
Mark(initializer);
auto* init = Statement(initializer);
if (!init) {
return false;
}
behaviors.Add(init->Behaviors());
}
if (auto* cond_expr = stmt->condition) {
auto* cond = Load(ValueExpression(cond_expr));
if (!cond) {
return false;
}
sem->SetCondition(cond);
behaviors.Add(cond->Behaviors());
}
if (auto* continuing = stmt->continuing) {
Mark(continuing);
auto* cont = Statement(continuing);
if (!cont) {
return false;
}
behaviors.Add(cont->Behaviors());
}
Mark(stmt->body);
auto* body = b.create<sem::LoopBlockStatement>(stmt->body, current_compound_statement_,
current_function_);
if (!StatementScope(stmt->body, body, [&] { return Statements(stmt->body->statements); })) {
return false;
}
behaviors.Add(body->Behaviors());
if (stmt->condition || behaviors.Contains(sem::Behavior::kBreak)) { // Does the loop exit?
behaviors.Add(sem::Behavior::kNext);
} else {
behaviors.Remove(sem::Behavior::kNext);
}
behaviors.Remove(sem::Behavior::kBreak, sem::Behavior::kContinue);
return validator_.ForLoopStatement(sem);
});
}
sem::WhileStatement* Resolver::WhileStatement(const ast::WhileStatement* stmt) {
auto* sem = b.create<sem::WhileStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto& behaviors = sem->Behaviors();
auto* cond = Load(ValueExpression(stmt->condition));
if (!cond) {
return false;
}
sem->SetCondition(cond);
behaviors.Add(cond->Behaviors());
Mark(stmt->body);
auto* body = b.create<sem::LoopBlockStatement>(stmt->body, current_compound_statement_,
current_function_);
if (!StatementScope(stmt->body, body, [&] { return Statements(stmt->body->statements); })) {
return false;
}
behaviors.Add(body->Behaviors());
// Always consider the while as having a 'next' behaviour because it has
// a condition. We don't check if the condition will terminate but it isn't
// valid to have an infinite loop in a WGSL program, so a non-terminating
// condition is already an invalid program.
behaviors.Add(sem::Behavior::kNext);
behaviors.Remove(sem::Behavior::kBreak, sem::Behavior::kContinue);
return validator_.WhileStatement(sem);
});
}
sem::Expression* Resolver::Expression(const ast::Expression* root) {
Vector<const ast::Expression*, 64> sorted;
constexpr size_t kMaxExpressionDepth = 512U;
bool failed = false;
if (!ast::TraverseExpressions<ast::TraverseOrder::RightToLeft>(
root, [&](const ast::Expression* expr, size_t depth) {
if (depth > kMaxExpressionDepth) {
AddError(expr->source)
<< "reached max expression depth of " << kMaxExpressionDepth;
failed = true;
return ast::TraverseAction::Stop;
}
if (!Mark(expr)) {
failed = true;
return ast::TraverseAction::Stop;
}
if (auto* binary = expr->As<ast::BinaryExpression>();
binary && binary->IsLogical()) {
// Store potential const-eval short-circuit pair
logical_binary_lhs_to_parent_.Add(binary->lhs, binary);
}
sorted.Push(expr);
return ast::TraverseAction::Descend;
})) {
AddError(root->source) << "TraverseExpressions failed";
return nullptr;
}
if (failed) {
return nullptr;
}
for (auto* expr : tint::Reverse(sorted)) {
auto* sem_expr = Switch(
expr, //
[&](const ast::IndexAccessorExpression* array) { return IndexAccessor(array); },
[&](const ast::BinaryExpression* bin_op) { return Binary(bin_op); },
[&](const ast::CallExpression* call) { return Call(call); },
[&](const ast::IdentifierExpression* ident) { return Identifier(ident); },
[&](const ast::LiteralExpression* literal) { return Literal(literal); },
[&](const ast::MemberAccessorExpression* member) { return MemberAccessor(member); },
[&](const ast::UnaryOpExpression* unary) { return UnaryOp(unary); },
[&](const ast::PhonyExpression*) {
return b.create<sem::ValueExpression>(expr, b.create<core::type::Void>(),
core::EvaluationStage::kRuntime,
current_statement_,
/* constant_value */ nullptr,
/* has_side_effects */ false);
}, //
TINT_ICE_ON_NO_MATCH);
if (!sem_expr) {
return nullptr;
}
auto* val = sem_expr->As<sem::ValueExpression>();
if (val) {
if (auto* constraint = expr_eval_stage_constraint_.constraint) {
if (!validator_.EvaluationStage(val, expr_eval_stage_constraint_.stage,
constraint)) {
return nullptr;
}
}
}
b.Sem().Add(expr, sem_expr);
if (expr == root) {
return sem_expr;
}
// If we just processed the lhs of a constexpr logical binary expression, mark the rhs for
// short-circuiting.
if (val && val->ConstantValue()) {
if (auto binary = logical_binary_lhs_to_parent_.Get(expr)) {
const bool lhs_is_true = val->ConstantValue()->ValueAs<bool>();
if (((*binary)->IsLogicalAnd() && !lhs_is_true) ||
((*binary)->IsLogicalOr() && lhs_is_true)) {
// Mark entire expression tree to not const-evaluate
auto r = ast::TraverseExpressions( //
(*binary)->rhs, [&](const ast::Expression* e) {
not_evaluated_.Add(e);
return ast::TraverseAction::Descend;
});
if (!r) {
AddError(root->source) << "TraverseExpressions failed";
return nullptr;
}
}
}
}
}
ICE(root->source) << "Expression() did not find root node";
}
sem::ValueExpression* Resolver::ValueExpression(const ast::Expression* expr) {
return sem_.AsValueExpression(Expression(expr));
}
sem::TypeExpression* Resolver::TypeExpression(const ast::Expression* expr) {
return sem_.AsTypeExpression(Expression(expr));
}
sem::FunctionExpression* Resolver::FunctionExpression(const ast::Expression* expr) {
return sem_.AsFunctionExpression(Expression(expr));
}
core::type::Type* Resolver::Type(const ast::Expression* ast) {
Vector<const sem::GlobalVariable*, 4> referenced_overrides;
on_transitively_reference_global_.Push([&](const sem::GlobalVariable* ref) {
if (ref->Declaration()->Is<ast::Override>()) {
referenced_overrides.Push(ref);
}
});
TINT_DEFER(on_transitively_reference_global_.Pop());
auto* type_expr = TypeExpression(ast);
if (TINT_UNLIKELY(!type_expr)) {
return nullptr;
}
auto* type = const_cast<core::type::Type*>(type_expr->Type());
if (TINT_UNLIKELY(!type)) {
return nullptr;
}
if (auto* arr = type->As<sem::Array>()) {
for (auto* ref : referenced_overrides) {
arr->AddTransitivelyReferencedOverride(ref);
}
}
return type;
}
sem::BuiltinEnumExpression<core::AddressSpace>* Resolver::AddressSpaceExpression(
const ast::Expression* expr) {
auto address_space_expr = sem_.AsAddressSpace(Expression(expr));
if (TINT_UNLIKELY(!address_space_expr)) {
return nullptr;
}
if (TINT_UNLIKELY(
address_space_expr->Value() == core::AddressSpace::kPixelLocal &&
!enabled_extensions_.Contains(wgsl::Extension::kChromiumExperimentalPixelLocal))) {
AddError(expr->source) << "'pixel_local' address space requires the '"
<< wgsl::Extension::kChromiumExperimentalPixelLocal
<< "' extension enabled";
return nullptr;
}
return address_space_expr;
}
sem::BuiltinEnumExpression<core::TexelFormat>* Resolver::TexelFormatExpression(
const ast::Expression* expr) {
return sem_.AsTexelFormat(Expression(expr));
}
sem::BuiltinEnumExpression<core::Access>* Resolver::AccessExpression(const ast::Expression* expr) {
return sem_.AsAccess(Expression(expr));
}
sem::BuiltinEnumExpression<core::InterpolationSampling>* Resolver::InterpolationSampling(
const ast::Expression* expr) {
return sem_.AsInterpolationSampling(Expression(expr));
}
sem::BuiltinEnumExpression<core::InterpolationType>* Resolver::InterpolationType(
const ast::Expression* expr) {
return sem_.AsInterpolationType(Expression(expr));
}
void Resolver::RegisterStore(const sem::ValueExpression* expr) {
Switch(
expr->RootIdentifier(),
[&](const sem::GlobalVariable* global) {
alias_analysis_infos_[current_function_].module_scope_writes.Add(global, expr);
},
[&](const sem::Parameter* param) {
alias_analysis_infos_[current_function_].parameter_writes.Add(param);
});
}
void Resolver::RegisterLoad(const sem::ValueExpression* expr) {
Switch(
expr->RootIdentifier(),
[&](const sem::GlobalVariable* global) {
alias_analysis_infos_[current_function_].module_scope_reads.Add(global, expr);
},
[&](const sem::Parameter* param) {
alias_analysis_infos_[current_function_].parameter_reads.Add(param);
});
}
bool Resolver::AliasAnalysis(const sem::Call* call) {
auto* target = call->Target()->As<sem::Function>();
if (!target) {
return true;
}
if (validator_.IsValidationDisabled(target->Declaration()->attributes,
ast::DisabledValidation::kIgnorePointerAliasing)) {
return true;
}
// Helper to generate an aliasing error diagnostic.
struct Alias {
const sem::ValueExpression* expr; // the "other expression"
enum { Argument, ModuleScope } type; // the type of the "other" expression
std::string access; // the access performed for the "other" expression
};
auto make_error = [&](const sem::ValueExpression* arg, Alias&& var) {
AddError(arg->Declaration()->source) << "invalid aliased pointer argument";
switch (var.type) {
case Alias::Argument:
AddNote(var.expr->Declaration()->source)
<< "aliases with another argument passed here";
break;
case Alias::ModuleScope: {
auto* func = var.expr->Stmt()->Function();
auto func_name = func->Declaration()->name->symbol.NameView();
AddNote(var.expr->Declaration()->source)
<< "aliases with module-scope variable " << var.access << " in '" << func_name
<< "'";
break;
}
}
return false;
};
auto& args = call->Arguments();
auto& target_info = alias_analysis_infos_[target];
auto& caller_info = alias_analysis_infos_[current_function_];
// Track the set of root identifiers that are read and written by arguments passed in this
// call.
Hashmap<const sem::Variable*, const sem::ValueExpression*, 4> arg_reads;
Hashmap<const sem::Variable*, const sem::ValueExpression*, 4> arg_writes;
for (size_t i = 0; i < args.Length(); i++) {
auto* arg = args[i];
if (!arg->Type()->Is<core::type::Pointer>()) {
continue;
}
auto* root = arg->RootIdentifier();
if (target_info.parameter_writes.Contains(target->Parameters()[i])) {
// Arguments that are written to can alias with any other argument or module-scope
// variable access.
if (auto write = arg_writes.Get(root)) {
return make_error(arg, {*write, Alias::Argument, "write"});
}
if (auto read = arg_reads.Get(root)) {
return make_error(arg, {*read, Alias::Argument, "read"});
}
if (auto read = target_info.module_scope_reads.Get(root)) {
return make_error(arg, {*read, Alias::ModuleScope, "read"});
}
if (auto write = target_info.module_scope_writes.Get(root)) {
return make_error(arg, {*write, Alias::ModuleScope, "write"});
}
arg_writes.Add(root, arg);
// Propagate the write access to the caller.
Switch(
root,
[&](const sem::GlobalVariable* global) {
caller_info.module_scope_writes.Add(global, arg);
},
[&](const sem::Parameter* param) { caller_info.parameter_writes.Add(param); });
} else if (target_info.parameter_reads.Contains(target->Parameters()[i])) {
// Arguments that are read from can alias with arguments or module-scope variables
// that are written to.
if (auto write = arg_writes.Get(root)) {
return make_error(arg, {*write, Alias::Argument, "write"});
}
if (auto write = target_info.module_scope_writes.Get(root)) {
return make_error(arg, {*write, Alias::ModuleScope, "write"});
}
arg_reads.Add(root, arg);
// Propagate the read access to the caller.
Switch(
root,
[&](const sem::GlobalVariable* global) {
caller_info.module_scope_reads.Add(global, arg);
},
[&](const sem::Parameter* param) { caller_info.parameter_reads.Add(param); });
}
}
// Propagate module-scope variable uses to the caller.
for (auto read : target_info.module_scope_reads) {
caller_info.module_scope_reads.Add(read.key, read.value);
}
for (auto write : target_info.module_scope_writes) {
caller_info.module_scope_writes.Add(write.key, write.value);
}
return true;
}
const core::type::Type* Resolver::ConcreteType(const core::type::Type* ty,
const core::type::Type* target_ty,
const Source& source) {
auto i32 = [&] { return b.create<core::type::I32>(); };
auto f32 = [&] { return b.create<core::type::F32>(); };
auto i32v = [&](uint32_t width) { return b.create<core::type::Vector>(i32(), width); };
auto f32v = [&](uint32_t width) { return b.create<core::type::Vector>(f32(), width); };
auto f32m = [&](uint32_t columns, uint32_t rows) {
return b.create<core::type::Matrix>(f32v(rows), columns);
};
return Switch(
ty, //
[&](const core::type::AbstractInt*) { return target_ty ? target_ty : i32(); },
[&](const core::type::AbstractFloat*) { return target_ty ? target_ty : f32(); },
[&](const core::type::Vector* v) {
return Switch(
v->type(), //
[&](const core::type::AbstractInt*) {
return target_ty ? target_ty : i32v(v->Width());
},
[&](const core::type::AbstractFloat*) {
return target_ty ? target_ty : f32v(v->Width());
});
},
[&](const core::type::Matrix* m) {
return Switch(m->type(), //
[&](const core::type::AbstractFloat*) {
return target_ty ? target_ty : f32m(m->columns(), m->rows());
});
},
[&](const sem::Array* a) -> const core::type::Type* {
const core::type::Type* target_el_ty = nullptr;
if (auto* target_arr_ty = As<sem::Array>(target_ty)) {
target_el_ty = target_arr_ty->ElemType();
}
if (auto* el_ty = ConcreteType(a->ElemType(), target_el_ty, source)) {
return Array(source, source, source, el_ty, a->Count(), /* explicit_stride */ 0);
}
return nullptr;
},
[&](const core::type::Struct* s) -> const core::type::Type* {
if (auto tys = s->ConcreteTypes(); !tys.IsEmpty()) {
return target_ty ? target_ty : tys[0];
}
return nullptr;
});
}
const sem::ValueExpression* Resolver::Load(const sem::ValueExpression* expr) {
if (!expr) {
// Allow for Load(ValueExpression(blah)), where failures pass through Load()
return nullptr;
}
if (!expr->Type()->Is<core::type::Reference>()) {
// Expression is not a reference type, so cannot be loaded. Just return expr.
return expr;
}
auto* load = b.create<sem::Load>(expr, current_statement_, expr->Stage());
load->Behaviors() = expr->Behaviors();
b.Sem().Replace(expr->Declaration(), load);
// Register the load for the alias analysis.
RegisterLoad(expr);
return load;
}
const sem::ValueExpression* Resolver::Materialize(
const sem::ValueExpression* expr,
const core::type::Type* target_type /* = nullptr */) {
if (!expr) {
// Allow for Materialize(ValueExpression(blah)), where failures pass through Materialize()
return nullptr;
}
auto* decl = expr->Declaration();
auto* concrete_ty = ConcreteType(expr->Type(), target_type, decl->source);
if (!concrete_ty) {
return expr; // Does not require materialization
}
auto* src_ty = expr->Type();
if (!validator_.Materialize(concrete_ty, src_ty, decl->source)) {
return nullptr;
}
const core::constant::Value* materialized_val = nullptr;
if (!not_evaluated_.Contains(decl)) {
auto expr_val = expr->ConstantValue();
if (TINT_UNLIKELY(!expr_val)) {
ICE(decl->source) << "Materialize(" << decl->TypeInfo().name
<< ") called on expression with no constant value";
}
auto val = const_eval_.Convert(concrete_ty, expr_val, decl->source);
if (val != Success) {
// Convert() has already failed and raised an diagnostic error.
return nullptr;
}
materialized_val = val.Get();
if (TINT_UNLIKELY(!materialized_val)) {
ICE(decl->source) << "ConvertValue(" << expr_val->Type()->FriendlyName() << " -> "
<< concrete_ty->FriendlyName() << ") returned invalid value";
}
}
auto* m = b.create<sem::Materialize>(expr, current_statement_, concrete_ty, materialized_val);
m->Behaviors() = expr->Behaviors();
b.Sem().Replace(decl, m);
return m;
}
template <size_t N>
bool Resolver::MaybeMaterializeAndLoadArguments(Vector<const sem::ValueExpression*, N>& args,
const sem::CallTarget* target) {
for (size_t i = 0, n = std::min(args.Length(), target->Parameters().Length()); i < n; i++) {
const auto* param_ty = target->Parameters()[i]->Type();
if (ShouldMaterializeArgument(param_ty)) {
auto* materialized = Materialize(args[i], param_ty);
if (!materialized) {
return false;
}
args[i] = materialized;
}
if (!param_ty->Is<core::type::Reference>()) {
auto* load = Load(args[i]);
if (!load) {
return false;
}
args[i] = load;
}
}
return true;
}
bool Resolver::ShouldMaterializeArgument(const core::type::Type* parameter_ty) const {
const auto* param_el_ty = parameter_ty->DeepestElement();
return param_el_ty && !param_el_ty->Is<core::type::AbstractNumeric>();
}
bool Resolver::Convert(const core::constant::Value*& c,
const core::type::Type* target_ty,
const Source& source) {
auto r = const_eval_.Convert(target_ty, c, source);
if (r != Success) {
return false;
}
c = r.Get();
return true;
}
template <size_t N>
tint::Result<Vector<const core::constant::Value*, N>> Resolver::ConvertArguments(
const Vector<const sem::ValueExpression*, N>& args,
const sem::CallTarget* target) {
auto const_args = tint::Transform(args, [](auto* arg) { return arg->ConstantValue(); });
for (size_t i = 0, n = std::min(args.Length(), target->Parameters().Length()); i < n; i++) {
if (!Convert(const_args[i], target->Parameters()[i]->Type(),
args[i]->Declaration()->source)) {
return Failure{};
}
}
return const_args;
}
sem::ValueExpression* Resolver::IndexAccessor(const ast::IndexAccessorExpression* expr) {
auto* idx = Load(Materialize(sem_.GetVal(expr->index)));
if (!idx) {
return nullptr;
}
const auto* obj = sem_.GetVal(expr->object);
if (idx->Stage() != core::EvaluationStage::kConstant) {
// If the index is non-constant, then the resulting expression is non-constant, so we'll
// have to materialize the object. For example, consider:
// vec2(1, 2)[runtime-index]
obj = Materialize(obj);
}
if (!obj) {
return nullptr;
}
auto* object_ty = obj->Type();
auto* const memory_view = object_ty->As<core::type::MemoryView>();
const core::type::Type* storage_ty = object_ty->UnwrapRef();
if (memory_view) {
if (memory_view->Is<core::type::Pointer>() &&
!allowed_features_.features.count(wgsl::LanguageFeature::kPointerCompositeAccess)) {
AddError(expr->source)
<< "pointer composite access requires the pointer_composite_access language "
"feature, which is not allowed in the current environment";
return nullptr;
}
storage_ty = memory_view->StoreType();
}
auto* ty = Switch(
storage_ty, //
[&](const sem::Array* arr) { return arr->ElemType(); },
[&](const core::type::Vector* vec) { return vec->type(); },
[&](const core::type::Matrix* mat) {
return b.create<core::type::Vector>(mat->type(), mat->rows());
},
[&](Default) {
AddError(expr->source) << "cannot index type '" << sem_.TypeNameOf(storage_ty) << "'";
return nullptr;
});
if (ty == nullptr) {
return nullptr;
}
auto* idx_ty = idx->Type()->UnwrapRef();
if (!idx_ty->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(idx->Declaration()->source)
<< "index must be of type 'i32' or 'u32', found: '" << sem_.TypeNameOf(idx_ty) << "'";
return nullptr;
}
// If we're extracting from a memory view, we return a reference.
if (memory_view) {
ty =
b.create<core::type::Reference>(memory_view->AddressSpace(), ty, memory_view->Access());
}
const core::constant::Value* val = nullptr;
auto stage = core::EarliestStage(obj->Stage(), idx->Stage());
if (not_evaluated_.Contains(expr)) {
stage = core::EvaluationStage::kNotEvaluated;
} else {
if (auto* idx_val = idx->ConstantValue()) {
auto res = const_eval_.Index(obj->ConstantValue(), obj->Type(), idx_val,
idx->Declaration()->source);
if (res != Success) {
return nullptr;
}
val = res.Get();
}
}
bool has_side_effects = idx->HasSideEffects() || obj->HasSideEffects();
auto* sem = b.create<sem::IndexAccessorExpression>(expr, ty, stage, obj, idx,
current_statement_, std::move(val),
has_side_effects, obj->RootIdentifier());
sem->Behaviors() = idx->Behaviors() + obj->Behaviors();
return sem;
}
sem::Call* Resolver::Call(const ast::CallExpression* expr) {
// A CallExpression can resolve to one of:
// * A function call.
// * A builtin call.
// * A value constructor.
// * A value conversion.
auto* target = sem_.Get(expr->target);
if (TINT_UNLIKELY(!target)) {
return nullptr;
}
// Resolve all of the arguments, their types and the set of behaviors.
Vector<const sem::ValueExpression*, 8> args;
args.Reserve(expr->args.Length());
auto args_stage = core::EvaluationStage::kConstant;
sem::Behaviors arg_behaviors;
for (size_t i = 0; i < expr->args.Length(); i++) {
auto* arg = sem_.GetVal(expr->args[i]);
if (!arg) {
return nullptr;
}
args.Push(arg);
args_stage = core::EarliestStage(args_stage, arg->Stage());
arg_behaviors.Add(arg->Behaviors());
}
arg_behaviors.Remove(sem::Behavior::kNext);
// Did any arguments have side effects?
bool has_side_effects =
std::any_of(args.begin(), args.end(), [](auto* e) { return e->HasSideEffects(); });
// ctor_or_conv is a helper for building either a sem::ValueConstructor or
// sem::ValueConversion call for a CtorConvIntrinsic with an optional template argument type.
auto ctor_or_conv = [&](CtorConvIntrinsic ty,
VectorRef<const core::type::Type*> template_args) -> sem::Call* {
auto arg_tys = tint::Transform(args, [](auto* arg) { return arg->Type()->UnwrapRef(); });
auto match = intrinsic_table_.Lookup(ty, template_args, arg_tys, args_stage);
if (match != Success) {
AddError(expr->source) << match.Failure();
return nullptr;
}
auto overload_stage = match->const_eval_fn ? core::EvaluationStage::kConstant
: core::EvaluationStage::kRuntime;
sem::CallTarget* target_sem = nullptr;
// Is this overload a constructor or conversion?
if (match->info->flags.Contains(OverloadFlag::kIsConstructor)) {
// Type constructor
target_sem = constructors_.GetOrAdd(match.Get(), [&] {
auto params = Transform(match->parameters, [&](auto& p, size_t i) {
return b.create<sem::Parameter>(nullptr, static_cast<uint32_t>(i), p.type,
p.usage);
});
return b.create<sem::ValueConstructor>(match->return_type, std::move(params),
overload_stage);
});
} else {
// Type conversion
target_sem = converters_.GetOrAdd(match.Get(), [&] {
auto* param = b.create<sem::Parameter>(nullptr, 0u, match->parameters[0].type,
match->parameters[0].usage);
return b.create<sem::ValueConversion>(match->return_type, param, overload_stage);
});
}
if (!MaybeMaterializeAndLoadArguments(args, target_sem)) {
return nullptr;
}
const core::constant::Value* value = nullptr;
auto stage = core::EarliestStage(overload_stage, args_stage);
if (not_evaluated_.Contains(expr)) {
stage = core::EvaluationStage::kNotEvaluated;
}
if (stage == core::EvaluationStage::kConstant) {
auto const_args = ConvertArguments(args, target_sem);
if (const_args != Success) {
return nullptr;
}
auto const_eval_fn = match->const_eval_fn;
auto r = (const_eval_.*const_eval_fn)(target_sem->ReturnType(), const_args.Get(),
expr->source);
if (r != Success) {
return nullptr;
}
value = r.Get();
}
return b.create<sem::Call>(expr, target_sem, stage, std::move(args), current_statement_,
value, has_side_effects);
};
// arr_or_str_init is a helper for building a sem::ValueConstructor for an array or structure
// constructor call target.
auto arr_or_str_init = [&](const core::type::Type* ty,
const sem::CallTarget* call_target) -> sem::Call* {
auto stage = args_stage; // The evaluation stage of the call
const core::constant::Value* value = nullptr; // The constant value for the call
if (not_evaluated_.Contains(expr)) {
stage = core::EvaluationStage::kNotEvaluated;
}
if (stage == core::EvaluationStage::kConstant) {
auto const_args = ConvertArguments(args, call_target);
if (const_args != Success) {
return nullptr;
}
auto r = const_eval_.ArrayOrStructCtor(ty, std::move(const_args.Get()));
if (r != Success) {
return nullptr;
}
value = r.Get();
if (!value) {
// Constant evaluation failed.
// Can happen for expressions that will fail validation (later).
// Use the kRuntime EvaluationStage, as kConstant will trigger an assertion in
// the sem::ValueExpression constructor, which checks that kConstant is paired
// with a constant value.
stage = core::EvaluationStage::kRuntime;
}
}
return b.create<sem::Call>(expr, call_target, stage, std::move(args), current_statement_,
value, has_side_effects);
};
auto ty_init_or_conv = [&](const core::type::Type* type) {
return Switch(
type, //
[&](const core::type::I32*) { return ctor_or_conv(CtorConvIntrinsic::kI32, Empty); },
[&](const core::type::U32*) { return ctor_or_conv(CtorConvIntrinsic::kU32, Empty); },
[&](const core::type::F16*) {
return validator_.CheckF16Enabled(expr->source)
? ctor_or_conv(CtorConvIntrinsic::kF16, Empty)
: nullptr;
},
[&](const core::type::F32*) { return ctor_or_conv(CtorConvIntrinsic::kF32, Empty); },
[&](const core::type::Bool*) { return ctor_or_conv(CtorConvIntrinsic::kBool, Empty); },
[&](const core::type::Vector* v) {
if (v->Packed()) {
TINT_ASSERT(v->Width() == 3u);
return ctor_or_conv(CtorConvIntrinsic::kPackedVec3, Vector{v->type()});
}
return ctor_or_conv(wgsl::intrinsic::VectorCtorConv(v->Width()), Vector{v->type()});
},
[&](const core::type::Matrix* m) {
return ctor_or_conv(wgsl::intrinsic::MatrixCtorConv(m->columns(), m->rows()),
Vector{m->type()});
},
[&](const sem::Array* arr) -> sem::Call* {
auto* call_target = array_ctors_.GetOrAdd(
ArrayConstructorSig{{arr, args.Length(), args_stage}},
[&]() -> sem::ValueConstructor* {
auto params = tint::Transform(args, [&](auto, size_t i) {
return b.create<sem::Parameter>(nullptr, // declaration
static_cast<uint32_t>(i), // index
arr->ElemType());
});
return b.create<sem::ValueConstructor>(arr, std::move(params), args_stage);
});
if (TINT_UNLIKELY(!MaybeMaterializeAndLoadArguments(args, call_target))) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.ArrayConstructor(expr, arr))) {
return nullptr;
}
return arr_or_str_init(arr, call_target);
},
[&](const core::type::Struct* str) -> sem::Call* {
auto* call_target = struct_ctors_.GetOrAdd(
StructConstructorSig{{str, args.Length(), args_stage}},
[&]() -> sem::ValueConstructor* {
Vector<sem::Parameter*, 8> params;
params.Resize(std::min(args.Length(), str->Members().Length()));
for (size_t i = 0, n = params.Length(); i < n; i++) {
params[i] =
b.create<sem::Parameter>(nullptr, // declaration
static_cast<uint32_t>(i), // index
str->Members()[i]->Type()); // type
}
return b.create<sem::ValueConstructor>(str, std::move(params), args_stage);
});
if (TINT_UNLIKELY(!MaybeMaterializeAndLoadArguments(args, call_target))) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.StructureInitializer(expr, str))) {
return nullptr;
}
return arr_or_str_init(str, call_target);
},
[&](Default) {
AddError(expr->source) << "type is not constructible";
return nullptr;
});
};
auto incomplete_type = [&](const IncompleteType* t) -> sem::Call* {
// A type without template arguments.
// Examples: vec3(...), array(...)
switch (t->builtin) {
case core::BuiltinType::kVec2:
return ctor_or_conv(CtorConvIntrinsic::kVec2, Empty);
case core::BuiltinType::kVec3:
return ctor_or_conv(CtorConvIntrinsic::kVec3, Empty);
case core::BuiltinType::kVec4:
return ctor_or_conv(CtorConvIntrinsic::kVec4, Empty);
case core::BuiltinType::kMat2X2:
return ctor_or_conv(CtorConvIntrinsic::kMat2x2, Empty);
case core::BuiltinType::kMat2X3:
return ctor_or_conv(CtorConvIntrinsic::kMat2x3, Empty);
case core::BuiltinType::kMat2X4:
return ctor_or_conv(CtorConvIntrinsic::kMat2x4, Empty);
case core::BuiltinType::kMat3X2:
return ctor_or_conv(CtorConvIntrinsic::kMat3x2, Empty);
case core::BuiltinType::kMat3X3:
return ctor_or_conv(CtorConvIntrinsic::kMat3x3, Empty);
case core::BuiltinType::kMat3X4:
return ctor_or_conv(CtorConvIntrinsic::kMat3x4, Empty);
case core::BuiltinType::kMat4X2:
return ctor_or_conv(CtorConvIntrinsic::kMat4x2, Empty);
case core::BuiltinType::kMat4X3:
return ctor_or_conv(CtorConvIntrinsic::kMat4x3, Empty);
case core::BuiltinType::kMat4X4:
return ctor_or_conv(CtorConvIntrinsic::kMat4x4, Empty);
case core::BuiltinType::kArray: {
auto el_count =
b.create<core::type::ConstantArrayCount>(static_cast<uint32_t>(args.Length()));
auto arg_tys =
tint::Transform(args, [](auto* arg) { return arg->Type()->UnwrapRef(); });
auto el_ty = core::type::Type::Common(arg_tys);
if (TINT_UNLIKELY(!el_ty)) {
AddError(expr->source)
<< "cannot infer common array element type from constructor arguments";
Hashset<const core::type::Type*, 8> types;
for (size_t i = 0; i < args.Length(); i++) {
if (types.Add(args[i]->Type())) {
AddNote(args[i]->Declaration()->source)
<< "argument " << i << " is of type '"
<< sem_.TypeNameOf(args[i]->Type()) << "'";
}
}
return nullptr;
}
auto* arr = Array(expr->source, expr->source, expr->source, el_ty, el_count,
/* explicit_stride */ 0);
if (TINT_UNLIKELY(!arr)) {
return nullptr;
}
return ty_init_or_conv(arr);
}
default: {
TINT_ICE() << "unhandled IncompleteType builtin: " << t->builtin;
}
}
};
auto* call = Switch(
target, //
[&](const sem::FunctionExpression* fn_expr) {
return FunctionCall(expr, const_cast<sem::Function*>(fn_expr->Function()),
std::move(args), arg_behaviors);
},
[&](const sem::TypeExpression* ty_expr) {
return Switch(
ty_expr->Type(), //
[&](const IncompleteType* t) -> sem::Call* {
auto* ctor = incomplete_type(t);
if (TINT_UNLIKELY(!ctor)) {
return nullptr;
}
// Replace incomplete type with resolved type
const_cast<sem::TypeExpression*>(ty_expr)->SetType(ctor->Type());
return ctor;
},
[&](Default) { return ty_init_or_conv(ty_expr->Type()); });
},
[&](const sem::BuiltinEnumExpression<wgsl::BuiltinFn>* fn_expr) {
return BuiltinCall(expr, fn_expr->Value(), args);
},
[&](Default) {
sem_.ErrorUnexpectedExprKind(target, "call target");
return nullptr;
});
if (!call) {
return nullptr;
}
return validator_.Call(call, current_statement_) ? call : nullptr;
}
template <size_t N>
sem::Call* Resolver::BuiltinCall(const ast::CallExpression* expr,
wgsl::BuiltinFn fn,
Vector<const sem::ValueExpression*, N>& args) {
auto arg_stage = core::EvaluationStage::kConstant;
for (auto* arg : args) {
arg_stage = core::EarliestStage(arg_stage, arg->Stage());
}
Vector<const core::type::Type*, 1> tmpl_args;
if (auto* tmpl = expr->target->identifier->As<ast::TemplatedIdentifier>()) {
for (auto* arg : tmpl->arguments) {
auto* arg_ty = sem_.AsTypeExpression(sem_.Get(arg));
if (TINT_UNLIKELY(!arg_ty)) {
return nullptr;
}
tmpl_args.Push(arg_ty->Type());
}
}
auto arg_tys = tint::Transform(args, [](auto* arg) { return arg->Type()->UnwrapRef(); });
auto overload = intrinsic_table_.Lookup(fn, tmpl_args, arg_tys, arg_stage);
if (overload != Success) {
AddError(expr->source) << overload.Failure();
return nullptr;
}
// De-duplicate builtins that are identical.
auto* target = builtins_.GetOrAdd(std::make_pair(overload.Get(), fn), [&] {
auto params = Transform(overload->parameters, [&](auto& p, size_t i) {
return b.create<sem::Parameter>(nullptr, static_cast<uint32_t>(i), p.type, p.usage);
});
sem::PipelineStageSet supported_stages;
auto flags = overload->info->flags;
if (flags.Contains(OverloadFlag::kSupportsVertexPipeline)) {
supported_stages.Add(ast::PipelineStage::kVertex);
}
if (flags.Contains(OverloadFlag::kSupportsFragmentPipeline)) {
supported_stages.Add(ast::PipelineStage::kFragment);
}
if (flags.Contains(OverloadFlag::kSupportsComputePipeline)) {
supported_stages.Add(ast::PipelineStage::kCompute);
}
auto eval_stage = overload->const_eval_fn ? core::EvaluationStage::kConstant
: core::EvaluationStage::kRuntime;
return b.create<sem::BuiltinFn>(fn, overload->return_type, std::move(params), eval_stage,
supported_stages, *overload->info);
});
if (fn == wgsl::BuiltinFn::kTintMaterialize) {
args[0] = Materialize(args[0]);
if (!args[0]) {
return nullptr;
}
} else {
// Materialize arguments if the parameter type is not abstract
if (!MaybeMaterializeAndLoadArguments(args, target)) {
return nullptr;
}
}
if (target->IsDeprecated()) {
AddWarning(expr->source) << "use of deprecated builtin";
}
// If the builtin is @const, and all arguments have constant values, evaluate the builtin
// now.
const core::constant::Value* value = nullptr;
auto stage = core::EarliestStage(arg_stage, target->Stage());
if (not_evaluated_.Contains(expr)) {
stage = core::EvaluationStage::kNotEvaluated;
}
if (stage == core::EvaluationStage::kConstant) {
auto const_args = ConvertArguments(args, target);
if (const_args != Success) {
return nullptr;
}
auto const_eval_fn = overload->const_eval_fn;
auto r = (const_eval_.*const_eval_fn)(target->ReturnType(), const_args.Get(), expr->source);
if (r != Success) {
return nullptr;
}
value = r.Get();
}
bool has_side_effects =
target->HasSideEffects() ||
std::any_of(args.begin(), args.end(), [](auto* e) { return e->HasSideEffects(); });
auto* call = b.create<sem::Call>(expr, target, stage, std::move(args), current_statement_,
value, has_side_effects);
if (current_function_) {
current_function_->AddDirectlyCalledBuiltin(target);
current_function_->AddDirectCall(call);
}
if (!validator_.RequiredFeaturesForBuiltinFn(call)) {
return nullptr;
}
if (IsTexture(fn)) {
if (!validator_.TextureBuiltinFn(call)) {
return nullptr;
}
CollectTextureSamplerPairs(target, call->Arguments());
}
switch (fn) {
case wgsl::BuiltinFn::kWorkgroupUniformLoad:
if (!validator_.WorkgroupUniformLoad(call)) {
return nullptr;
}
RegisterLoad(args[0]);
break;
case wgsl::BuiltinFn::kSubgroupBroadcast:
if (!validator_.SubgroupBroadcast(call)) {
return nullptr;
}
break;
case wgsl::BuiltinFn::kAtomicLoad:
RegisterLoad(args[0]);
break;
case wgsl::BuiltinFn::kAtomicStore:
RegisterStore(args[0]);
break;
case wgsl::BuiltinFn::kAtomicAdd:
case wgsl::BuiltinFn::kAtomicSub:
case wgsl::BuiltinFn::kAtomicMax:
case wgsl::BuiltinFn::kAtomicMin:
case wgsl::BuiltinFn::kAtomicAnd:
case wgsl::BuiltinFn::kAtomicOr:
case wgsl::BuiltinFn::kAtomicXor:
case wgsl::BuiltinFn::kAtomicExchange:
case wgsl::BuiltinFn::kAtomicCompareExchangeWeak:
RegisterLoad(args[0]);
RegisterStore(args[0]);
break;
default:
break;
}
if (!validator_.BuiltinCall(call)) {
return nullptr;
}
return call;
}
core::type::Type* Resolver::BuiltinType(core::BuiltinType builtin_ty,
const ast::Identifier* ident) {
auto check_no_tmpl_args = [&](core::type::Type* ty) -> core::type::Type* {
return TINT_LIKELY(CheckNotTemplated("type", ident)) ? ty : nullptr;
};
switch (builtin_ty) {
case core::BuiltinType::kBool:
return check_no_tmpl_args(b.create<core::type::Bool>());
case core::BuiltinType::kI32:
return check_no_tmpl_args(I32());
case core::BuiltinType::kU32:
return check_no_tmpl_args(U32());
case core::BuiltinType::kF16:
return check_no_tmpl_args(F16(ident));
case core::BuiltinType::kF32:
return check_no_tmpl_args(b.create<core::type::F32>());
case core::BuiltinType::kVec2:
return VecT(ident, builtin_ty, 2);
case core::BuiltinType::kVec3:
return VecT(ident, builtin_ty, 3);
case core::BuiltinType::kVec4:
return VecT(ident, builtin_ty, 4);
case core::BuiltinType::kMat2X2:
return MatT(ident, builtin_ty, 2, 2);
case core::BuiltinType::kMat2X3:
return MatT(ident, builtin_ty, 2, 3);
case core::BuiltinType::kMat2X4:
return MatT(ident, builtin_ty, 2, 4);
case core::BuiltinType::kMat3X2:
return MatT(ident, builtin_ty, 3, 2);
case core::BuiltinType::kMat3X3:
return MatT(ident, builtin_ty, 3, 3);
case core::BuiltinType::kMat3X4:
return MatT(ident, builtin_ty, 3, 4);
case core::BuiltinType::kMat4X2:
return MatT(ident, builtin_ty, 4, 2);
case core::BuiltinType::kMat4X3:
return MatT(ident, builtin_ty, 4, 3);
case core::BuiltinType::kMat4X4:
return MatT(ident, builtin_ty, 4, 4);
case core::BuiltinType::kMat2X2F:
return check_no_tmpl_args(Mat(ident, F32(), 2u, 2u));
case core::BuiltinType::kMat2X3F:
return check_no_tmpl_args(Mat(ident, F32(), 2u, 3u));
case core::BuiltinType::kMat2X4F:
return check_no_tmpl_args(Mat(ident, F32(), 2u, 4u));
case core::BuiltinType::kMat3X2F:
return check_no_tmpl_args(Mat(ident, F32(), 3u, 2u));
case core::BuiltinType::kMat3X3F:
return check_no_tmpl_args(Mat(ident, F32(), 3u, 3u));
case core::BuiltinType::kMat3X4F:
return check_no_tmpl_args(Mat(ident, F32(), 3u, 4u));
case core::BuiltinType::kMat4X2F:
return check_no_tmpl_args(Mat(ident, F32(), 4u, 2u));
case core::BuiltinType::kMat4X3F:
return check_no_tmpl_args(Mat(ident, F32(), 4u, 3u));
case core::BuiltinType::kMat4X4F:
return check_no_tmpl_args(Mat(ident, F32(), 4u, 4u));
case core::BuiltinType::kMat2X2H:
return check_no_tmpl_args(Mat(ident, F16(ident), 2u, 2u));
case core::BuiltinType::kMat2X3H:
return check_no_tmpl_args(Mat(ident, F16(ident), 2u, 3u));
case core::BuiltinType::kMat2X4H:
return check_no_tmpl_args(Mat(ident, F16(ident), 2u, 4u));
case core::BuiltinType::kMat3X2H:
return check_no_tmpl_args(Mat(ident, F16(ident), 3u, 2u));
case core::BuiltinType::kMat3X3H:
return check_no_tmpl_args(Mat(ident, F16(ident), 3u, 3u));
case core::BuiltinType::kMat3X4H:
return check_no_tmpl_args(Mat(ident, F16(ident), 3u, 4u));
case core::BuiltinType::kMat4X2H:
return check_no_tmpl_args(Mat(ident, F16(ident), 4u, 2u));
case core::BuiltinType::kMat4X3H:
return check_no_tmpl_args(Mat(ident, F16(ident), 4u, 3u));
case core::BuiltinType::kMat4X4H:
return check_no_tmpl_args(Mat(ident, F16(ident), 4u, 4u));
case core::BuiltinType::kVec2F:
return check_no_tmpl_args(Vec(ident, F32(), 2u));
case core::BuiltinType::kVec3F:
return check_no_tmpl_args(Vec(ident, F32(), 3u));
case core::BuiltinType::kVec4F:
return check_no_tmpl_args(Vec(ident, F32(), 4u));
case core::BuiltinType::kVec2H:
return check_no_tmpl_args(Vec(ident, F16(ident), 2u));
case core::BuiltinType::kVec3H:
return check_no_tmpl_args(Vec(ident, F16(ident), 3u));
case core::BuiltinType::kVec4H:
return check_no_tmpl_args(Vec(ident, F16(ident), 4u));
case core::BuiltinType::kVec2I:
return check_no_tmpl_args(Vec(ident, I32(), 2u));
case core::BuiltinType::kVec3I:
return check_no_tmpl_args(Vec(ident, I32(), 3u));
case core::BuiltinType::kVec4I:
return check_no_tmpl_args(Vec(ident, I32(), 4u));
case core::BuiltinType::kVec2U:
return check_no_tmpl_args(Vec(ident, U32(), 2u));
case core::BuiltinType::kVec3U:
return check_no_tmpl_args(Vec(ident, U32(), 3u));
case core::BuiltinType::kVec4U:
return check_no_tmpl_args(Vec(ident, U32(), 4u));
case core::BuiltinType::kArray:
return Array(ident);
case core::BuiltinType::kAtomic:
return Atomic(ident);
case core::BuiltinType::kPtr:
return Ptr(ident);
case core::BuiltinType::kSampler:
return check_no_tmpl_args(
b.create<core::type::Sampler>(core::type::SamplerKind::kSampler));
case core::BuiltinType::kSamplerComparison:
return check_no_tmpl_args(
b.create<core::type::Sampler>(core::type::SamplerKind::kComparisonSampler));
case core::BuiltinType::kTexture1D:
return SampledTexture(ident, core::type::TextureDimension::k1d);
case core::BuiltinType::kTexture2D:
return SampledTexture(ident, core::type::TextureDimension::k2d);
case core::BuiltinType::kTexture2DArray:
return SampledTexture(ident, core::type::TextureDimension::k2dArray);
case core::BuiltinType::kTexture3D:
return SampledTexture(ident, core::type::TextureDimension::k3d);
case core::BuiltinType::kTextureCube:
return SampledTexture(ident, core::type::TextureDimension::kCube);
case core::BuiltinType::kTextureCubeArray:
return SampledTexture(ident, core::type::TextureDimension::kCubeArray);
case core::BuiltinType::kTextureDepth2D:
return check_no_tmpl_args(
b.create<core::type::DepthTexture>(core::type::TextureDimension::k2d));
case core::BuiltinType::kTextureDepth2DArray:
return check_no_tmpl_args(
b.create<core::type::DepthTexture>(core::type::TextureDimension::k2dArray));
case core::BuiltinType::kTextureDepthCube:
return check_no_tmpl_args(
b.create<core::type::DepthTexture>(core::type::TextureDimension::kCube));
case core::BuiltinType::kTextureDepthCubeArray:
return check_no_tmpl_args(
b.create<core::type::DepthTexture>(core::type::TextureDimension::kCubeArray));
case core::BuiltinType::kTextureDepthMultisampled2D:
return check_no_tmpl_args(
b.create<core::type::DepthMultisampledTexture>(core::type::TextureDimension::k2d));
case core::BuiltinType::kTextureExternal:
return check_no_tmpl_args(b.create<core::type::ExternalTexture>());
case core::BuiltinType::kTextureMultisampled2D:
return MultisampledTexture(ident, core::type::TextureDimension::k2d);
case core::BuiltinType::kTextureStorage1D:
return StorageTexture(ident, core::type::TextureDimension::k1d);
case core::BuiltinType::kTextureStorage2D:
return StorageTexture(ident, core::type::TextureDimension::k2d);
case core::BuiltinType::kTextureStorage2DArray:
return StorageTexture(ident, core::type::TextureDimension::k2dArray);
case core::BuiltinType::kTextureStorage3D:
return StorageTexture(ident, core::type::TextureDimension::k3d);
case core::BuiltinType::kInputAttachment:
return InputAttachment(ident);
case core::BuiltinType::kPackedVec3:
return PackedVec3T(ident);
case core::BuiltinType::kAtomicCompareExchangeResultI32:
return core::type::CreateAtomicCompareExchangeResult(b.Types(), b.Symbols(), I32());
case core::BuiltinType::kAtomicCompareExchangeResultU32:
return core::type::CreateAtomicCompareExchangeResult(b.Types(), b.Symbols(), U32());
case core::BuiltinType::kFrexpResultAbstract:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), AF());
case core::BuiltinType::kFrexpResultF16:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), F16(ident));
case core::BuiltinType::kFrexpResultF32:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), F32());
case core::BuiltinType::kFrexpResultVec2Abstract:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, AF(), 2));
case core::BuiltinType::kFrexpResultVec2F16:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 2));
case core::BuiltinType::kFrexpResultVec2F32:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F32(), 2));
case core::BuiltinType::kFrexpResultVec3Abstract:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, AF(), 3));
case core::BuiltinType::kFrexpResultVec3F16:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 3));
case core::BuiltinType::kFrexpResultVec3F32:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F32(), 3));
case core::BuiltinType::kFrexpResultVec4Abstract:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, AF(), 4));
case core::BuiltinType::kFrexpResultVec4F16:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 4));
case core::BuiltinType::kFrexpResultVec4F32:
return core::type::CreateFrexpResult(b.Types(), b.Symbols(), Vec(ident, F32(), 4));
case core::BuiltinType::kModfResultAbstract:
return core::type::CreateModfResult(b.Types(), b.Symbols(), AF());
case core::BuiltinType::kModfResultF16:
return core::type::CreateModfResult(b.Types(), b.Symbols(), F16(ident));
case core::BuiltinType::kModfResultF32:
return core::type::CreateModfResult(b.Types(), b.Symbols(), F32());
case core::BuiltinType::kModfResultVec2Abstract:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, AF(), 2));
case core::BuiltinType::kModfResultVec2F16:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 2));
case core::BuiltinType::kModfResultVec2F32:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F32(), 2));
case core::BuiltinType::kModfResultVec3Abstract:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, AF(), 3));
case core::BuiltinType::kModfResultVec3F16:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 3));
case core::BuiltinType::kModfResultVec3F32:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F32(), 3));
case core::BuiltinType::kModfResultVec4Abstract:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, AF(), 4));
case core::BuiltinType::kModfResultVec4F16:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F16(ident), 4));
case core::BuiltinType::kModfResultVec4F32:
return core::type::CreateModfResult(b.Types(), b.Symbols(), Vec(ident, F32(), 4));
case core::BuiltinType::kUndefined:
break;
}
ICE(ident->source) << " unhandled builtin type '" << ident->symbol.NameView() << "'";
}
core::type::AbstractFloat* Resolver::AF() {
return b.create<core::type::AbstractFloat>();
}
core::type::F32* Resolver::F32() {
return b.create<core::type::F32>();
}
core::type::I32* Resolver::I32() {
return b.create<core::type::I32>();
}
core::type::U32* Resolver::U32() {
return b.create<core::type::U32>();
}
core::type::F16* Resolver::F16(const ast::Identifier* ident) {
return validator_.CheckF16Enabled(ident->source) ? b.create<core::type::F16>() : nullptr;
}
core::type::Vector* Resolver::Vec(const ast::Identifier* ident, core::type::Type* el, uint32_t n) {
if (TINT_UNLIKELY(!el)) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.Vector(el, ident->source))) {
return nullptr;
}
return b.create<core::type::Vector>(el, n);
}
core::type::Type* Resolver::VecT(const ast::Identifier* ident,
core::BuiltinType builtin,
uint32_t n) {
auto* tmpl_ident = ident->As<ast::TemplatedIdentifier>();
if (!tmpl_ident) {
// 'vecN' has no template arguments, so return an incomplete type.
return b.create<IncompleteType>(builtin);
}
if (TINT_UNLIKELY(!CheckTemplatedIdentifierArgs(tmpl_ident, 1))) {
return nullptr;
}
auto* ty = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!ty)) {
return nullptr;
}
return Vec(ident, const_cast<core::type::Type*>(ty), n);
}
core::type::Matrix* Resolver::Mat(const ast::Identifier* ident,
core::type::Type* el,
uint32_t num_columns,
uint32_t num_rows) {
if (TINT_UNLIKELY(!el)) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.Matrix(el, ident->source))) {
return nullptr;
}
auto* column = Vec(ident, el, num_rows);
if (!column) {
return nullptr;
}
return b.create<core::type::Matrix>(column, num_columns);
}
core::type::Type* Resolver::MatT(const ast::Identifier* ident,
core::BuiltinType builtin,
uint32_t num_columns,
uint32_t num_rows) {
auto* tmpl_ident = ident->As<ast::TemplatedIdentifier>();
if (!tmpl_ident) {
// 'vecN' has no template arguments, so return an incomplete type.
return b.create<IncompleteType>(builtin);
}
if (TINT_UNLIKELY(!CheckTemplatedIdentifierArgs(tmpl_ident, 1))) {
return nullptr;
}
auto* el_ty = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!el_ty)) {
return nullptr;
}
return Mat(ident, const_cast<core::type::Type*>(el_ty), num_columns, num_rows);
}
core::type::Type* Resolver::Array(const ast::Identifier* ident) {
auto* tmpl_ident = ident->As<ast::TemplatedIdentifier>();
if (!tmpl_ident) {
// 'array' has no template arguments, so return an incomplete type.
return b.create<IncompleteType>(core::BuiltinType::kArray);
}
if (TINT_UNLIKELY(!CheckTemplatedIdentifierArgs(tmpl_ident, 1, 2))) {
return nullptr;
}
auto* ast_el_ty = tmpl_ident->arguments[0];
auto* ast_count = (tmpl_ident->arguments.Length() > 1) ? tmpl_ident->arguments[1] : nullptr;
auto* el_ty = sem_.GetType(ast_el_ty);
if (!el_ty) {
return nullptr;
}
const core::type::ArrayCount* el_count =
ast_count ? ArrayCount(ast_count) : b.create<core::type::RuntimeArrayCount>();
if (!el_count) {
return nullptr;
}
// Look for explicit stride via @stride(n) attribute
uint32_t explicit_stride = 0;
if (!ArrayAttributes(tmpl_ident->attributes, el_ty, explicit_stride)) {
return nullptr;
}
auto* out = Array(tmpl_ident->source, //
ast_el_ty->source, //
ast_count ? ast_count->source : ident->source, //
el_ty, el_count, explicit_stride);
if (!out) {
return nullptr;
}
if (el_ty->Is<core::type::Atomic>()) {
atomic_composite_info_.Add(out, &ast_el_ty->source);
} else {
if (auto found = atomic_composite_info_.Get(el_ty)) {
atomic_composite_info_.Add(out, *found);
}
}
return out;
}
core::type::Atomic* Resolver::Atomic(const ast::Identifier* ident) {
auto* tmpl_ident = TemplatedIdentifier(ident, 1); // atomic<type>
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto* el_ty = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!el_ty)) {
return nullptr;
}
auto* out = b.create<core::type::Atomic>(el_ty);
if (TINT_UNLIKELY(!validator_.Atomic(tmpl_ident, out))) {
return nullptr;
}
return out;
}
core::type::Pointer* Resolver::Ptr(const ast::Identifier* ident) {
auto* tmpl_ident = TemplatedIdentifier(ident, 2, 3); // ptr<address, type [, access]>
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto address_space = sem_.GetAddressSpace(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(address_space == core::AddressSpace::kUndefined)) {
return nullptr;
}
auto* store_ty = const_cast<core::type::Type*>(sem_.GetType(tmpl_ident->arguments[1]));
if (TINT_UNLIKELY(!store_ty)) {
return nullptr;
}
core::Access access = core::Access::kUndefined;
if (tmpl_ident->arguments.Length() > 2) {
access = sem_.GetAccess(tmpl_ident->arguments[2]);
if (TINT_UNLIKELY(access == core::Access::kUndefined)) {
return nullptr;
}
} else {
access = DefaultAccessForAddressSpace(address_space);
}
auto* out = b.create<core::type::Pointer>(address_space, store_ty, access);
if (TINT_UNLIKELY(!validator_.Pointer(tmpl_ident, out))) {
return nullptr;
}
if (!ApplyAddressSpaceUsageToType(address_space, store_ty, tmpl_ident->arguments[1]->source)) {
AddNote(ident->source) << "while instantiating " << out->FriendlyName();
return nullptr;
}
return out;
}
core::type::SampledTexture* Resolver::SampledTexture(const ast::Identifier* ident,
core::type::TextureDimension dim) {
auto* tmpl_ident = TemplatedIdentifier(ident, 1);
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto* ty_expr = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!ty_expr)) {
return nullptr;
}
auto* out = b.create<core::type::SampledTexture>(dim, ty_expr);
return validator_.SampledTexture(out, ident->source) ? out : nullptr;
}
core::type::MultisampledTexture* Resolver::MultisampledTexture(const ast::Identifier* ident,
core::type::TextureDimension dim) {
auto* tmpl_ident = TemplatedIdentifier(ident, 1);
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto* ty_expr = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!ty_expr)) {
return nullptr;
}
auto* out = b.create<core::type::MultisampledTexture>(dim, ty_expr);
return validator_.MultisampledTexture(out, ident->source) ? out : nullptr;
}
core::type::StorageTexture* Resolver::StorageTexture(const ast::Identifier* ident,
core::type::TextureDimension dim) {
auto* tmpl_ident = TemplatedIdentifier(ident, 2);
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto format = sem_.GetTexelFormat(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(format == core::TexelFormat::kUndefined)) {
return nullptr;
}
auto access = sem_.GetAccess(tmpl_ident->arguments[1]);
if (TINT_UNLIKELY(access == core::Access::kUndefined)) {
return nullptr;
}
auto* subtype = core::type::StorageTexture::SubtypeFor(format, b.Types());
auto* tex = b.create<core::type::StorageTexture>(dim, format, access, subtype);
if (!validator_.StorageTexture(tex, ident->source)) {
return nullptr;
}
return tex;
}
core::type::InputAttachment* Resolver::InputAttachment(const ast::Identifier* ident) {
auto* tmpl_ident = TemplatedIdentifier(ident, 1);
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto* ty_expr = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!ty_expr)) {
return nullptr;
}
auto* out = b.create<core::type::InputAttachment>(ty_expr);
return validator_.InputAttachment(out, ident->source) ? out : nullptr;
}
core::type::Vector* Resolver::PackedVec3T(const ast::Identifier* ident) {
auto* tmpl_ident = TemplatedIdentifier(ident, 1);
if (TINT_UNLIKELY(!tmpl_ident)) {
return nullptr;
}
auto* el_ty = sem_.GetType(tmpl_ident->arguments[0]);
if (TINT_UNLIKELY(!el_ty)) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.Vector(el_ty, ident->source))) {
return nullptr;
}
return b.create<core::type::Vector>(el_ty, 3u, true);
}
const ast::TemplatedIdentifier* Resolver::TemplatedIdentifier(const ast::Identifier* ident,
size_t min_args,
size_t max_args /* = use min 0 */) {
auto* tmpl_ident = ident->As<ast::TemplatedIdentifier>();
if (!tmpl_ident) {
if (TINT_UNLIKELY(min_args != 0)) {
AddError(Source{ident->source.range.end}) << "expected " << style::Code("<") << " for "
<< style::Code(ident->symbol.NameView());
}
return nullptr;
}
return CheckTemplatedIdentifierArgs(tmpl_ident, min_args, max_args) ? tmpl_ident : nullptr;
}
bool Resolver::CheckTemplatedIdentifierArgs(const ast::TemplatedIdentifier* ident,
size_t min_args,
size_t max_args /* = use min 0 */) {
if (max_args == 0) {
max_args = min_args;
}
if (min_args == max_args) {
if (TINT_UNLIKELY(ident->arguments.Length() != min_args)) {
AddError(ident->source) << style::Code(ident->symbol.NameView()) << " requires "
<< min_args << " template arguments";
return false;
}
} else {
if (TINT_UNLIKELY(ident->arguments.Length() < min_args)) {
AddError(ident->source) << style::Code(ident->symbol.NameView())
<< " requires at least " << min_args << " template arguments";
return false;
}
if (TINT_UNLIKELY(ident->arguments.Length() > max_args)) {
AddError(ident->source) << style::Code(ident->symbol.NameView()) << " requires at most "
<< max_args << " template arguments";
return false;
}
}
return ident;
}
size_t Resolver::NestDepth(const core::type::Type* ty) const {
return Switch(
ty, //
[](const core::type::Vector*) { return size_t{1}; },
[](const core::type::Matrix*) { return size_t{2}; },
[&](Default) {
if (auto d = nest_depth_.Get(ty)) {
return *d;
}
return size_t{0};
});
}
void Resolver::CollectTextureSamplerPairs(const sem::BuiltinFn* builtin,
VectorRef<const sem::ValueExpression*> args) const {
if (builtin->Fn() == wgsl::BuiltinFn::kInputAttachmentLoad) {
// inputAttachmentLoad() is considered a texture function, however it doesn't need sampler,
// and its parameter has ParameterUsage::kInputAttachment, so return early.
return;
}
// Collect a texture/sampler pair for this builtin.
const auto& signature = builtin->Signature();
int texture_index = signature.IndexOf(core::ParameterUsage::kTexture);
if (TINT_UNLIKELY(texture_index == -1)) {
TINT_ICE() << "texture builtin without texture parameter";
}
if (auto* user =
args[static_cast<size_t>(texture_index)]->UnwrapLoad()->As<sem::VariableUser>()) {
auto* texture = user->Variable();
if (!texture->Type()->UnwrapRef()->Is<core::type::StorageTexture>()) {
int sampler_index = signature.IndexOf(core::ParameterUsage::kSampler);
const sem::Variable* sampler = sampler_index != -1
? args[static_cast<size_t>(sampler_index)]
->UnwrapLoad()
->As<sem::VariableUser>()
->Variable()
: nullptr;
current_function_->AddTextureSamplerPair(texture, sampler);
}
}
}
sem::Call* Resolver::FunctionCall(const ast::CallExpression* expr,
sem::Function* target,
VectorRef<const sem::ValueExpression*> args_in,
sem::Behaviors arg_behaviors) {
Vector<const sem::ValueExpression*, 8> args = std::move(args_in);
if (!MaybeMaterializeAndLoadArguments(args, target)) {
return nullptr;
}
auto stage = not_evaluated_.Contains(expr) ? core::EvaluationStage::kNotEvaluated
: core::EvaluationStage::kRuntime;
// TODO(crbug.com/tint/1420): For now, assume all function calls have side effects.
bool has_side_effects = true;
auto* call = b.create<sem::Call>(expr, target, stage, std::move(args), current_statement_,
/* constant_value */ nullptr, has_side_effects);
target->AddCallSite(call);
call->Behaviors() = arg_behaviors + target->Behaviors();
if (!validator_.FunctionCall(call, current_statement_)) {
return nullptr;
}
if (current_function_) {
// Note: Requires called functions to be resolved first.
// This is currently guaranteed as functions must be declared before
// use.
current_function_->AddTransitivelyCalledFunction(target);
current_function_->AddDirectCall(call);
for (auto* transitive_call : target->TransitivelyCalledFunctions()) {
current_function_->AddTransitivelyCalledFunction(transitive_call);
}
// We inherit any referenced variables from the callee.
for (auto* var : target->TransitivelyReferencedGlobals()) {
current_function_->AddTransitivelyReferencedGlobal(var);
}
if (!AliasAnalysis(call)) {
return nullptr;
}
// Note: Validation *must* be performed before calling this method.
CollectTextureSamplerPairs(target, call->Arguments());
}
return call;
}
void Resolver::CollectTextureSamplerPairs(sem::Function* func,
VectorRef<const sem::ValueExpression*> args) const {
// Map all texture/sampler pairs from the target function to the
// current function. These can only be global or parameter
// variables. Resolve any parameter variables to the corresponding
// argument passed to the current function. Leave global variables
// as-is. Then add the mapped pair to the current function's list of
// texture/sampler pairs.
Hashset<const sem::Variable*, 4> texture_sampler_set;
for (sem::VariablePair pair : func->TextureSamplerPairs()) {
const sem::Variable* texture = pair.first;
const sem::Variable* sampler = pair.second;
if (auto* param = As<sem::Parameter>(texture)) {
texture = args[param->Index()]->UnwrapLoad()->As<sem::VariableUser>()->Variable();
texture_sampler_set.Add(texture);
}
if (auto* param = As<sem::Parameter>(sampler)) {
sampler = args[param->Index()]->UnwrapLoad()->As<sem::VariableUser>()->Variable();
texture_sampler_set.Add(sampler);
}
current_function_->AddTextureSamplerPair(texture, sampler);
}
// Add any possible texture/sampler not essentially passed to builtins from the function param.
// This could be unused texture/sampler or texture/sampler passed to builtins that are emulated.
const auto& signature = func->Signature();
for (size_t i = 0; i < signature.parameters.Length(); i++) {
auto* param = signature.parameters[i];
if (param->Type()->Is<core::type::Texture>()) {
auto* user = args[i]->UnwrapLoad()->As<sem::VariableUser>();
auto* texture = user->Variable();
if (!texture_sampler_set.Contains(texture)) {
current_function_->AddTextureSamplerPair(texture, nullptr);
func->AddTextureSamplerPair(param, nullptr);
texture_sampler_set.Add(texture);
}
} else if (param->Type()->Is<core::type::Sampler>()) {
auto* user = args[i]->UnwrapLoad()->As<sem::VariableUser>();
auto* sampler = user->Variable();
if (!texture_sampler_set.Contains(sampler)) {
current_function_->AddTextureSamplerPair(nullptr, sampler);
func->AddTextureSamplerPair(nullptr, param);
texture_sampler_set.Add(sampler);
}
}
}
}
sem::ValueExpression* Resolver::Literal(const ast::LiteralExpression* literal) {
auto* ty = Switch(
literal,
[&](const ast::IntLiteralExpression* i) -> core::type::Type* {
switch (i->suffix) {
case ast::IntLiteralExpression::Suffix::kNone:
return b.create<core::type::AbstractInt>();
case ast::IntLiteralExpression::Suffix::kI:
return b.create<core::type::I32>();
case ast::IntLiteralExpression::Suffix::kU:
return b.create<core::type::U32>();
}
TINT_UNREACHABLE() << "Unhandled integer literal suffix: " << i->suffix;
},
[&](const ast::FloatLiteralExpression* f) -> core::type::Type* {
switch (f->suffix) {
case ast::FloatLiteralExpression::Suffix::kNone:
return b.create<core::type::AbstractFloat>();
case ast::FloatLiteralExpression::Suffix::kF:
return b.create<core::type::F32>();
case ast::FloatLiteralExpression::Suffix::kH:
return validator_.CheckF16Enabled(literal->source) ? b.create<core::type::F16>()
: nullptr;
}
TINT_UNREACHABLE() << "Unhandled float literal suffix: " << f->suffix;
},
[&](const ast::BoolLiteralExpression*) { return b.create<core::type::Bool>(); }, //
TINT_ICE_ON_NO_MATCH);
if (ty == nullptr) {
return nullptr;
}
const core::constant::Value* val = nullptr;
auto stage = core::EvaluationStage::kConstant;
if (not_evaluated_.Contains(literal)) {
stage = core::EvaluationStage::kNotEvaluated;
}
if (stage == core::EvaluationStage::kConstant) {
val = Switch(
literal,
[&](const ast::BoolLiteralExpression* lit) { return b.constants.Get(lit->value); },
[&](const ast::IntLiteralExpression* lit) -> const core::constant::Value* {
switch (lit->suffix) {
case ast::IntLiteralExpression::Suffix::kNone:
return b.constants.Get(AInt(lit->value));
case ast::IntLiteralExpression::Suffix::kI:
return b.constants.Get(i32(lit->value));
case ast::IntLiteralExpression::Suffix::kU:
return b.constants.Get(u32(lit->value));
}
return nullptr;
},
[&](const ast::FloatLiteralExpression* lit) -> const core::constant::Value* {
switch (lit->suffix) {
case ast::FloatLiteralExpression::Suffix::kNone:
return b.constants.Get(AFloat(lit->value));
case ast::FloatLiteralExpression::Suffix::kF:
return b.constants.Get(f32(lit->value));
case ast::FloatLiteralExpression::Suffix::kH:
return b.constants.Get(f16(lit->value));
}
return nullptr;
});
}
return b.create<sem::ValueExpression>(literal, ty, stage, current_statement_, std::move(val),
/* has_side_effects */ false);
}
sem::Expression* Resolver::Identifier(const ast::IdentifierExpression* expr) {
auto* ident = expr->identifier;
Mark(ident);
auto resolved = dependencies_.resolved_identifiers.Get(ident);
if (!resolved) {
ICE(expr->source) << "identifier '" << ident->symbol.NameView() << "' was not resolved";
}
if (auto* ast_node = resolved->Node()) {
auto* resolved_node = sem_.Get(ast_node);
return Switch(
resolved_node, //
[&](sem::Variable* variable) -> sem::VariableUser* {
if (!TINT_LIKELY(CheckNotTemplated("variable", ident))) {
return nullptr;
}
auto stage = variable->Stage();
const core::constant::Value* value = variable->ConstantValue();
if (not_evaluated_.Contains(expr)) {
// This expression is short-circuited by an ancestor expression.
// Do not const-eval.
stage = core::EvaluationStage::kNotEvaluated;
value = nullptr;
}
auto* user =
b.create<sem::VariableUser>(expr, stage, current_statement_, value, variable);
if (current_statement_) {
// If identifier is part of a loop continuing block, make sure it
// doesn't refer to a variable that is bypassed by a continue statement
// in the loop's body block.
if (auto* continuing_block =
current_statement_
->FindFirstParent<sem::LoopContinuingBlockStatement>()) {
auto* loop_block =
continuing_block->FindFirstParent<sem::LoopBlockStatement>();
if (loop_block->FirstContinue()) {
// If our identifier is in loop_block->decls, make sure its index is
// less than first_continue
auto symbol = ident->symbol;
if (auto decl = loop_block->Decls().Get(symbol)) {
if (decl->order >= loop_block->NumDeclsAtFirstContinue()) {
AddError(loop_block->FirstContinue()->source)
<< "continue statement bypasses declaration of '"
<< symbol.NameView() << "'";
AddNote(decl->variable->Declaration()->source)
<< "identifier '" << symbol.NameView() << "' declared here";
AddNote(expr->source)
<< "identifier '" << symbol.NameView()
<< "' referenced in continuing block here";
return nullptr;
}
}
}
}
}
if (auto* global = variable->As<sem::GlobalVariable>()) {
for (auto& fn : on_transitively_reference_global_) {
fn(global);
}
if (!current_function_ && variable->Declaration()->Is<ast::Var>()) {
// Use of a module-scope 'var' outside of a function.
AddError(expr->source)
<< style::Keyword("var ") << style::Variable(ident->symbol.NameView())
<< " cannot be referenced at module-scope";
sem_.NoteDeclarationSource(variable->Declaration());
return nullptr;
}
}
variable->AddUser(user);
return user;
},
[&](const core::type::Type* ty) -> sem::TypeExpression* {
// User declared types cannot be templated.
if (!TINT_LIKELY(CheckNotTemplated("type", ident))) {
return nullptr;
}
// Notify callers of all transitively referenced globals.
if (auto* arr = ty->As<sem::Array>()) {
for (auto& fn : on_transitively_reference_global_) {
for (auto* ref : arr->TransitivelyReferencedOverrides()) {
fn(ref);
}
}
}
return b.create<sem::TypeExpression>(expr, current_statement_, ty);
},
[&](const sem::Function* fn) -> sem::FunctionExpression* {
if (!TINT_LIKELY(CheckNotTemplated("function", ident))) {
return nullptr;
}
return b.create<sem::FunctionExpression>(expr, current_statement_, fn);
});
}
if (auto builtin_ty = resolved->BuiltinType(); builtin_ty != core::BuiltinType::kUndefined) {
auto* ty = BuiltinType(builtin_ty, ident);
if (!ty) {
return nullptr;
}
return b.create<sem::TypeExpression>(expr, current_statement_, ty);
}
if (auto fn = resolved->BuiltinFn(); fn != wgsl::BuiltinFn::kNone) {
return b.create<sem::BuiltinEnumExpression<wgsl::BuiltinFn>>(expr, current_statement_, fn);
}
if (auto access = resolved->Access(); access != core::Access::kUndefined) {
return CheckNotTemplated("access", ident)
? b.create<sem::BuiltinEnumExpression<core::Access>>(expr, current_statement_,
access)
: nullptr;
}
if (auto addr = resolved->AddressSpace(); addr != core::AddressSpace::kUndefined) {
return CheckNotTemplated("address space", ident)
? b.create<sem::BuiltinEnumExpression<core::AddressSpace>>(
expr, current_statement_, addr)
: nullptr;
}
if (auto i_smpl = resolved->InterpolationSampling();
i_smpl != core::InterpolationSampling::kUndefined) {
return CheckNotTemplated("interpolation sampling", ident)
? b.create<sem::BuiltinEnumExpression<core::InterpolationSampling>>(
expr, current_statement_, i_smpl)
: nullptr;
}
if (auto i_type = resolved->InterpolationType();
i_type != core::InterpolationType::kUndefined) {
return CheckNotTemplated("interpolation type", ident)
? b.create<sem::BuiltinEnumExpression<core::InterpolationType>>(
expr, current_statement_, i_type)
: nullptr;
}
if (auto fmt = resolved->TexelFormat(); fmt != core::TexelFormat::kUndefined) {
return CheckNotTemplated("texel format", ident)
? b.create<sem::BuiltinEnumExpression<core::TexelFormat>>(
expr, current_statement_, fmt)
: nullptr;
}
if (resolved->Unresolved()) {
return b.create<UnresolvedIdentifier>(expr, current_statement_);
}
TINT_UNREACHABLE() << "unhandled resolved identifier: " << resolved->String();
}
sem::ValueExpression* Resolver::MemberAccessor(const ast::MemberAccessorExpression* expr) {
auto* object = sem_.GetVal(expr->object);
if (!object) {
return nullptr;
}
auto* object_ty = object->Type();
auto* const memory_view = object_ty->As<core::type::MemoryView>();
const core::type::Type* storage_ty = object_ty->UnwrapRef();
if (memory_view) {
if (memory_view->Is<core::type::Pointer>() &&
!allowed_features_.features.count(wgsl::LanguageFeature::kPointerCompositeAccess)) {
AddError(expr->source)
<< "pointer composite access requires the pointer_composite_access language "
"feature, which is not allowed in the current environment";
return nullptr;
}
storage_ty = memory_view->StoreType();
}
auto* root_ident = object->RootIdentifier();
const core::type::Type* ty = nullptr;
// Object may be a side-effecting expression (e.g. function call).
bool has_side_effects = object->HasSideEffects();
Mark(expr->member);
return Switch(
storage_ty, //
[&](const core::type::Struct* str) -> sem::ValueExpression* {
auto symbol = expr->member->symbol;
const core::type::StructMember* member = nullptr;
for (auto* m : str->Members()) {
if (m->Name() == symbol) {
member = m;
break;
}
}
if (member == nullptr) {
AddError(expr->source) << "struct member " << symbol.NameView() << " not found";
return nullptr;
}
ty = member->Type();
// If we're extracting from a memory view, we return a reference.
if (memory_view) {
ty = b.create<core::type::Reference>(memory_view->AddressSpace(), ty,
memory_view->Access());
}
const core::constant::Value* val = nullptr;
if (auto* obj_val = object->ConstantValue()) {
val = obj_val->Index(static_cast<size_t>(member->Index()));
}
return b.create<sem::StructMemberAccess>(expr, ty, current_statement_, val, object,
member, has_side_effects, root_ident);
},
[&](const core::type::Vector* vec) -> sem::ValueExpression* {
std::string_view s = expr->member->symbol.NameView();
auto size = s.size();
Vector<uint32_t, 4> swizzle;
swizzle.Reserve(s.size());
for (auto c : s) {
switch (c) {
case 'x':
case 'r':
swizzle.Push(0u);
break;
case 'y':
case 'g':
swizzle.Push(1u);
break;
case 'z':
case 'b':
swizzle.Push(2u);
break;
case 'w':
case 'a':
swizzle.Push(3u);
break;
default:
AddError(expr->member->source.Begin() +
static_cast<uint32_t>(swizzle.Length()))
<< "invalid vector swizzle character";
return nullptr;
}
if (swizzle.Back() >= vec->Width()) {
AddError(expr->member->source) << "invalid vector swizzle member";
return nullptr;
}
}
if (size < 1 || size > 4) {
AddError(expr->member->source) << "invalid vector swizzle size";
return nullptr;
}
// All characters are valid, check if they're being mixed
auto is_rgba = [](char c) { return c == 'r' || c == 'g' || c == 'b' || c == 'a'; };
auto is_xyzw = [](char c) { return c == 'x' || c == 'y' || c == 'z' || c == 'w'; };
if (!std::all_of(s.begin(), s.end(), is_rgba) &&
!std::all_of(s.begin(), s.end(), is_xyzw)) {
AddError(expr->member->source)
<< "invalid mixing of vector swizzle characters rgba with xyzw";
return nullptr;
}
const sem::ValueExpression* obj_expr = object;
if (size == 1) {
// A single element swizzle is just the type of the vector.
ty = vec->type();
// If we're extracting from a memory view, we return a reference.
if (memory_view) {
ty = b.create<core::type::Reference>(memory_view->AddressSpace(), ty,
memory_view->Access());
}
} else {
// The vector will have a number of components equal to the length of
// the swizzle.
ty = b.create<core::type::Vector>(vec->type(), static_cast<uint32_t>(size));
// The load rule is invoked before the swizzle, if necessary.
obj_expr = Load(object);
}
const core::constant::Value* val = nullptr;
if (auto* obj_val = object->ConstantValue()) {
auto res = const_eval_.Swizzle(ty, obj_val, swizzle);
if (res != Success) {
return nullptr;
}
val = res.Get();
}
return b.create<sem::Swizzle>(expr, ty, current_statement_, val, obj_expr,
std::move(swizzle), has_side_effects, root_ident);
},
[&](Default) {
AddError(expr->object->source)
<< "cannot index into expression of type '" << sem_.TypeNameOf(storage_ty) << "'";
return nullptr;
});
}
sem::ValueExpression* Resolver::Binary(const ast::BinaryExpression* expr) {
const auto* lhs = sem_.GetVal(expr->lhs);
const auto* rhs = sem_.GetVal(expr->rhs);
if (!lhs || !rhs) {
return nullptr;
}
// Load arguments if they are references
lhs = Load(lhs);
if (!lhs) {
return nullptr;
}
rhs = Load(rhs);
if (!rhs) {
return nullptr;
}
auto stage = core::EarliestStage(lhs->Stage(), rhs->Stage());
auto overload = intrinsic_table_.Lookup(expr->op, lhs->Type()->UnwrapRef(),
rhs->Type()->UnwrapRef(), stage, false);
if (overload != Success) {
AddError(expr->source) << overload.Failure();
return nullptr;
}
auto* res_ty = overload->return_type;
// Parameter types
auto* lhs_ty = overload->parameters[0].type;
auto* rhs_ty = overload->parameters[1].type;
if (ShouldMaterializeArgument(lhs_ty)) {
lhs = Materialize(lhs, lhs_ty);
if (!lhs) {
return nullptr;
}
}
if (ShouldMaterializeArgument(rhs_ty)) {
rhs = Materialize(rhs, rhs_ty);
if (!rhs) {
return nullptr;
}
}
const core::constant::Value* value = nullptr;
if (not_evaluated_.Contains(expr)) {
// This expression is short-circuited by an ancestor expression.
// Do not const-eval.
stage = core::EvaluationStage::kNotEvaluated;
} else if (lhs->Stage() == core::EvaluationStage::kConstant &&
rhs->Stage() == core::EvaluationStage::kNotEvaluated) {
// Short-circuiting binary expression. Use the LHS value and stage.
value = lhs->ConstantValue();
stage = core::EvaluationStage::kConstant;
} else if (stage == core::EvaluationStage::kConstant) {
// Both LHS and RHS have expressions that are constant evaluation stage.
auto const_eval_fn = overload->const_eval_fn;
if (const_eval_fn) { // Do we have a @const operator?
// Yes. Perform any required abstract argument values implicit conversions to the
// overload parameter types, and const-eval.
Vector const_args{lhs->ConstantValue(), rhs->ConstantValue()};
// Implicit conversion (e.g. AInt -> AFloat)
if (!Convert(const_args[0], lhs_ty, lhs->Declaration()->source)) {
return nullptr;
}
if (!Convert(const_args[1], rhs_ty, rhs->Declaration()->source)) {
return nullptr;
}
auto r = (const_eval_.*const_eval_fn)(res_ty, const_args, expr->source);
if (r != Success) {
return nullptr;
}
value = r.Get();
} else {
// The arguments have constant values, but the operator cannot be const-evaluated.
// This can only be evaluated at runtime.
stage = core::EvaluationStage::kRuntime;
}
}
bool has_side_effects = lhs->HasSideEffects() || rhs->HasSideEffects();
auto* sem = b.create<sem::ValueExpression>(expr, res_ty, stage, current_statement_, value,
has_side_effects);
sem->Behaviors() = lhs->Behaviors() + rhs->Behaviors();
return sem;
}
sem::ValueExpression* Resolver::UnaryOp(const ast::UnaryOpExpression* unary) {
const auto* expr = sem_.GetVal(unary->expr);
if (!expr) {
return nullptr;
}
auto* expr_ty = expr->Type();
const core::type::Type* ty = nullptr;
const sem::Variable* root_ident = nullptr;
const core::constant::Value* value = nullptr;
auto stage = core::EvaluationStage::kRuntime;
switch (unary->op) {
case core::UnaryOp::kAddressOf:
if (auto* ref = expr_ty->As<core::type::Reference>()) {
if (ref->StoreType()->UnwrapRef()->is_handle()) {
AddError(unary->expr->source)
<< "cannot take the address of " << sem_.Describe(expr)
<< " in handle address space";
return nullptr;
}
auto* array = unary->expr->As<ast::IndexAccessorExpression>();
auto* member = unary->expr->As<ast::MemberAccessorExpression>();
if ((array && sem_.TypeOf(array->object)->UnwrapRef()->Is<core::type::Vector>()) ||
(member &&
sem_.TypeOf(member->object)->UnwrapRef()->Is<core::type::Vector>())) {
AddError(unary->expr->source)
<< "cannot take the address of a vector component";
return nullptr;
}
ty = b.create<core::type::Pointer>(ref->AddressSpace(), ref->StoreType(),
ref->Access());
root_ident = expr->RootIdentifier();
} else {
AddError(unary->expr->source)
<< "cannot take the address of " << sem_.Describe(expr);
return nullptr;
}
break;
case core::UnaryOp::kIndirection:
if (auto* ptr = expr_ty->As<core::type::Pointer>()) {
ty = b.create<core::type::Reference>(ptr->AddressSpace(), ptr->StoreType(),
ptr->Access());
root_ident = expr->RootIdentifier();
} else {
AddError(unary->expr->source) << "cannot dereference expression of type "
<< style::Type(sem_.TypeNameOf(expr_ty));
return nullptr;
}
break;
default: {
stage = expr->Stage();
auto overload = intrinsic_table_.Lookup(unary->op, expr_ty->UnwrapRef(), stage);
if (overload != Success) {
AddError(unary->source) << overload.Failure();
return nullptr;
}
ty = overload->return_type;
auto* param_ty = overload->parameters[0].type;
if (ShouldMaterializeArgument(param_ty)) {
expr = Materialize(expr, param_ty);
if (!expr) {
return nullptr;
}
}
// Load expr if it is a reference
expr = Load(expr);
if (!expr) {
return nullptr;
}
stage = expr->Stage();
if (stage == core::EvaluationStage::kConstant) {
if (auto const_eval_fn = overload->const_eval_fn) {
auto r = (const_eval_.*const_eval_fn)(ty, Vector{expr->ConstantValue()},
expr->Declaration()->source);
if (r != Success) {
return nullptr;
}
value = r.Get();
} else {
stage = core::EvaluationStage::kRuntime;
}
}
break;
}
}
auto* sem = b.create<sem::ValueExpression>(unary, ty, stage, current_statement_, value,
expr->HasSideEffects(), root_ident);
sem->Behaviors() = expr->Behaviors();
return sem;
}
tint::Result<uint32_t> Resolver::LocationAttribute(const ast::LocationAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@location value"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@location") << " must be an "
<< style::Type("i32") << " or " << style::Type("u32") << " value";
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source) << style::Attribute("@location") << " value must be non-negative";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<uint32_t> Resolver::ColorAttribute(const ast::ColorAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@color value"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@color") << " must be an " << style::Type("i32")
<< " or " << style::Type("u32") << " value";
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source) << style::Attribute("@color") << " value must be non-negative";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<uint32_t> Resolver::BlendSrcAttribute(const ast::BlendSrcAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@blend_src value"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@blend_src") << " value must be "
<< style::Type("i32") << " or " << style::Type("u32");
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value != 0 && value != 1) {
AddError(attr->source) << style::Attribute("@blend_src") << " value must be zero or one";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<uint32_t> Resolver::BindingAttribute(const ast::BindingAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@binding"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@binding") << " must be an "
<< style::Type("i32") << " or " << style::Type("u32") << " value";
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source) << style::Attribute("@binding") << " value must be non-negative";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<uint32_t> Resolver::GroupAttribute(const ast::GroupAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@group"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@group") << " must be an " << style::Type("i32")
<< " or " << style::Type("u32") << " value";
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source) << style::Attribute("@group") << " value must be non-negative";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<uint32_t> Resolver::InputAttachmentIndexAttribute(
const ast::InputAttachmentIndexAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@input_attachment_index"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return Failure{};
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source) << style::Attribute("@input_attachment_index") << " must be an "
<< style::Type("i32") << " or " << style::Type("u32") << " value";
return Failure{};
}
auto const_value = materialized->ConstantValue();
auto value = const_value->ValueAs<AInt>();
if (value < 0) {
AddError(attr->source) << style::Attribute("@input_attachment_index")
<< " value must be non-negative";
return Failure{};
}
return static_cast<uint32_t>(value);
}
tint::Result<sem::WorkgroupSize> Resolver::WorkgroupAttribute(const ast::WorkgroupAttribute* attr) {
// Set work-group size defaults.
sem::WorkgroupSize ws;
for (size_t i = 0; i < 3; i++) {
ws[i] = 1;
}
auto values = attr->Values();
Vector<const sem::ValueExpression*, 3> args;
Vector<const core::type::Type*, 3> arg_tys;
auto err_bad_expr = [&](const ast::Expression* value) {
AddError(value->source) << style::Attribute("@workgroup_size")
<< " argument must be a constant or override-expression of type "
<< style::Type("abstract-integer") << ", " << style::Type("i32")
<< " or " << style::Type("u32");
};
for (size_t i = 0; i < 3; i++) {
// Each argument to this attribute can either be a literal, an identifier for a
// module-scope constants, a const-expression, or nullptr if not specified.
auto* value = values[i];
if (!value) {
break;
}
const auto* expr = ValueExpression(value);
if (!expr) {
return Failure{};
}
auto* ty = expr->Type();
if (!ty->IsAnyOf<core::type::I32, core::type::U32, core::type::AbstractInt>()) {
err_bad_expr(value);
return Failure{};
}
if (expr->Stage() != core::EvaluationStage::kConstant &&
expr->Stage() != core::EvaluationStage::kOverride) {
err_bad_expr(value);
return Failure{};
}
args.Push(expr);
arg_tys.Push(ty);
}
auto* common_ty = core::type::Type::Common(arg_tys);
if (!common_ty) {
AddError(attr->source) << style::Attribute("@workgroup_size")
<< " arguments must be of the same type, either "
<< style::Type("i32") << " or " << style::Type("u32");
return Failure{};
}
// If all arguments are abstract-integers, then materialize to i32.
if (common_ty->Is<core::type::AbstractInt>()) {
common_ty = b.create<core::type::I32>();
}
for (size_t i = 0; i < args.Length(); i++) {
auto* materialized = Materialize(args[i], common_ty);
if (!materialized) {
return Failure{};
}
if (auto* value = materialized->ConstantValue()) {
if (value->ValueAs<AInt>() < 1) {
AddError(values[i]->source)
<< style::Attribute("@workgroup_size") << " argument must be at least 1";
return Failure{};
}
ws[i] = value->ValueAs<u32>();
} else {
ws[i] = std::nullopt;
}
}
uint64_t total_size = static_cast<uint64_t>(ws[0].value_or(1));
for (size_t i = 1; i < 3; i++) {
total_size *= static_cast<uint64_t>(ws[i].value_or(1));
if (total_size > 0xffffffff) {
AddError(values[i]->source) << "total workgroup grid size cannot exceed 0xffffffff";
return Failure{};
}
}
return ws;
}
tint::Result<tint::core::BuiltinValue> Resolver::BuiltinAttribute(
const ast::BuiltinAttribute* attr) {
const ast::BuiltinValueName* builtin_val = attr->builtin;
Mark(builtin_val);
const ast::Identifier* ident = builtin_val->name;
if (!TINT_LIKELY(CheckNotTemplated("builtin value", ident))) {
return Failure{};
}
Mark(ident);
core::BuiltinValue builtin = core::ParseBuiltinValue(ident->symbol.NameView());
if (builtin == core::BuiltinValue::kUndefined) {
sem_.ErrorUnexpectedIdent(ident, "builtin value", core::kBuiltinValueStrings);
return Failure{};
}
auto* sem = b.create<sem::BuiltinAttribute>(attr, builtin);
// Apply the resolved tint::sem::BuiltinAttribute to the attribute.
b.Sem().Add(attr, sem);
return builtin;
}
bool Resolver::DiagnosticAttribute(const ast::DiagnosticAttribute* attr) {
return DiagnosticControl(attr->control);
}
bool Resolver::StageAttribute(const ast::StageAttribute*) {
return true;
}
bool Resolver::MustUseAttribute(const ast::MustUseAttribute*) {
return true;
}
bool Resolver::InvariantAttribute(const ast::InvariantAttribute*) {
return true;
}
bool Resolver::StrideAttribute(const ast::StrideAttribute*) {
return true;
}
tint::Result<core::Interpolation> Resolver::InterpolateAttribute(
const ast::InterpolateAttribute* attr) {
core::Interpolation out;
auto* type = InterpolationType(attr->type);
if (!type) {
return Failure{};
}
out.type = type->Value();
if (attr->sampling) {
auto* sampling = InterpolationSampling(attr->sampling);
if (!sampling) {
return Failure{};
}
out.sampling = sampling->Value();
}
return out;
}
bool Resolver::InternalAttribute(const ast::InternalAttribute* attr) {
for (auto* dep : attr->dependencies) {
if (!Expression(dep)) {
return false;
}
}
return true;
}
bool Resolver::DiagnosticControl(const ast::DiagnosticControl& control) {
Mark(control.rule_name);
Mark(control.rule_name->name);
auto name = control.rule_name->name->symbol.NameView();
if (control.rule_name->category) {
Mark(control.rule_name->category);
if (control.rule_name->category->symbol.NameView() == "chromium") {
auto rule = wgsl::ParseChromiumDiagnosticRule(name);
if (rule != wgsl::ChromiumDiagnosticRule::kUndefined) {
validator_.DiagnosticFilters().Set(rule, control.severity);
} else {
auto& warning = AddWarning(control.rule_name->source)
<< "unrecognized diagnostic rule " << style::Code("chromium.", name)
<< "\n";
tint::SuggestAlternativeOptions opts;
opts.prefix = "chromium.";
tint::SuggestAlternatives(name, wgsl::kChromiumDiagnosticRuleStrings,
warning.message, opts);
}
}
return true;
}
auto rule = wgsl::ParseCoreDiagnosticRule(name);
if (rule != wgsl::CoreDiagnosticRule::kUndefined) {
validator_.DiagnosticFilters().Set(rule, control.severity);
} else {
auto& warning = AddWarning(control.rule_name->source)
<< "unrecognized diagnostic rule " << style::Code(name) << "\n";
tint::SuggestAlternatives(name, wgsl::kCoreDiagnosticRuleStrings, warning.message);
}
return true;
}
bool Resolver::Enable(const ast::Enable* enable) {
for (auto* ext : enable->extensions) {
Mark(ext);
enabled_extensions_.Add(ext->name);
if (!allowed_features_.extensions.count(ext->name)) {
AddError(ext->source) << "extension " << style::Code(ext->name)
<< " is not allowed in the current environment";
return false;
}
}
return true;
}
bool Resolver::Requires(const ast::Requires* req) {
for (auto feature : req->features) {
if (!allowed_features_.features.count(feature)) {
AddError(req->source) << "language feature " << style::Code(wgsl::ToString(feature))
<< " is not allowed in the current environment";
return false;
}
}
return true;
}
core::type::Type* Resolver::TypeDecl(const ast::TypeDecl* named_type) {
Mark(named_type->name);
core::type::Type* result = nullptr;
if (auto* alias = named_type->As<ast::Alias>()) {
result = Alias(alias);
} else if (auto* str = named_type->As<ast::Struct>()) {
result = Structure(str);
} else {
TINT_UNREACHABLE() << "Unhandled TypeDecl";
}
if (!result) {
return nullptr;
}
b.Sem().Add(named_type, result);
return result;
}
const core::type::ArrayCount* Resolver::ArrayCount(const ast::Expression* count_expr) {
// Evaluate the constant array count expression.
const auto* count_sem = Materialize(sem_.GetVal(count_expr));
if (!count_sem) {
return nullptr;
}
switch (count_sem->Stage()) {
case core::EvaluationStage::kNotEvaluated:
// Happens in expressions like:
// false && array<T, N>()[i]
// The end result will not be used, so just make N=1.
return b.create<core::type::ConstantArrayCount>(static_cast<uint32_t>(1));
case core::EvaluationStage::kOverride: {
// array count is an override expression.
// Is the count a named 'override'?
if (auto* user = count_sem->UnwrapMaterialize()->As<sem::VariableUser>()) {
if (auto* global = user->Variable()->As<sem::GlobalVariable>()) {
return b.create<sem::NamedOverrideArrayCount>(global);
}
}
return b.create<sem::UnnamedOverrideArrayCount>(count_sem);
}
case core::EvaluationStage::kConstant: {
auto* count_val = count_sem->ConstantValue();
if (auto* ty = count_val->Type(); !ty->is_integer_scalar()) {
AddError(count_expr->source)
<< "array count must evaluate to a constant integer expression, but is type "
<< style::Type(ty->FriendlyName());
return nullptr;
}
int64_t count = count_val->ValueAs<AInt>();
if (count < 1) {
AddError(count_expr->source)
<< "array count (" << count << ") must be greater than 0";
return nullptr;
}
return b.create<core::type::ConstantArrayCount>(static_cast<uint32_t>(count));
}
default: {
AddError(count_expr->source)
<< "array count must evaluate to a constant integer expression "
"or override variable";
return nullptr;
}
}
}
bool Resolver::ArrayAttributes(VectorRef<const ast::Attribute*> attributes,
const core::type::Type* el_ty,
uint32_t& explicit_stride) {
if (!validator_.NoDuplicateAttributes(attributes)) {
return false;
}
for (auto* attribute : attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::StrideAttribute* attr) {
// If the element type is not plain, then el_ty->Align() may be 0, in which case we
// could get a DBZ in ArrayStrideAttribute(). In this case, validation will error
// about the invalid array element type (which is tested later), so this is just a
// seatbelt.
if (IsPlain(el_ty)) {
explicit_stride = attr->stride;
if (!validator_.ArrayStrideAttribute(attr, el_ty->Size(), el_ty->Align())) {
return false;
}
}
return true;
},
[&](Default) {
ErrorInvalidAttribute(attribute, StyledText{} << style::Type("array") << " types");
return false;
});
if (!ok) {
return false;
}
}
return true;
}
sem::Array* Resolver::Array(const Source& array_source,
const Source& el_source,
const Source& count_source,
const core::type::Type* el_ty,
const core::type::ArrayCount* el_count,
uint32_t explicit_stride) {
uint32_t el_align = el_ty->Align();
uint32_t el_size = el_ty->Size();
uint64_t implicit_stride = el_size ? tint::RoundUp<uint64_t>(el_align, el_size) : 0;
uint64_t stride = explicit_stride ? explicit_stride : implicit_stride;
uint64_t size = 0;
if (auto const_count = el_count->As<core::type::ConstantArrayCount>()) {
size = const_count->value * stride;
if (size > std::numeric_limits<uint32_t>::max()) {
AddError(count_source) << "array byte size (0x" << std::hex << size
<< ") must not exceed 0xffffffff bytes";
return nullptr;
}
} else if (el_count->Is<core::type::RuntimeArrayCount>()) {
size = stride;
}
auto* out =
b.create<sem::Array>(el_ty, el_count, el_align, static_cast<uint32_t>(size),
static_cast<uint32_t>(stride), static_cast<uint32_t>(implicit_stride));
// Maximum nesting depth of composite types
// https://gpuweb.github.io/gpuweb/wgsl/#limits
const size_t nest_depth = 1 + NestDepth(el_ty);
if (nest_depth > kMaxNestDepthOfCompositeType) {
AddError(array_source) << "array has nesting depth of " << nest_depth << ", maximum is "
<< kMaxNestDepthOfCompositeType;
return nullptr;
}
nest_depth_.Add(out, nest_depth);
if (!validator_.Array(out, el_source)) {
return nullptr;
}
return out;
}
core::type::Type* Resolver::Alias(const ast::Alias* alias) {
auto* ty = Type(alias->type);
if (TINT_UNLIKELY(!ty)) {
return nullptr;
}
if (TINT_UNLIKELY(!validator_.Alias(alias))) {
return nullptr;
}
return ty;
}
sem::Struct* Resolver::Structure(const ast::Struct* str) {
auto struct_name = [&] { //
return str->name->symbol.NameView();
};
if (validator_.IsValidationEnabled(str->attributes,
ast::DisabledValidation::kIgnoreStructMemberLimit)) {
// Maximum number of members in a structure type
// https://gpuweb.github.io/gpuweb/wgsl/#limits
const size_t kMaxNumStructMembers = 16383;
if (str->members.Length() > kMaxNumStructMembers) {
AddError(str->source) << style::Keyword("struct ") << style::Type(struct_name())
<< " has " << str->members.Length() << " members, maximum is "
<< kMaxNumStructMembers;
return nullptr;
}
}
if (!validator_.NoDuplicateAttributes(str->attributes)) {
return nullptr;
}
for (auto* attribute : str->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, [&](const ast::InternalAttribute* attr) { return InternalAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute,
StyledText{} << style::Keyword("struct") << " declarations");
return false;
});
if (!ok) {
return nullptr;
}
}
Vector<const sem::StructMember*, 8> sem_members;
sem_members.Reserve(str->members.Length());
// Calculate the effective size and alignment of each field, and the overall size of the
// structure. For size, use the size attribute if provided, otherwise use the default size
// for the type. For alignment, use the alignment attribute if provided, otherwise use the
// default alignment for the member type. Diagnostic errors are raised if a basic rule is
// violated. Validation of storage-class rules requires analyzing the actual variable usage
// of the structure, and so is performed as part of the variable validation.
uint64_t struct_size = 0;
uint64_t struct_align = 1;
Hashmap<Symbol, const ast::StructMember*, 8> member_map;
size_t members_nest_depth = 0;
for (auto* member : str->members) {
Mark(member);
Mark(member->name);
if (auto added = member_map.Add(member->name->symbol, member); !added) {
AddError(member->source)
<< "redefinition of " << style::Code(member->name->symbol.NameView());
AddNote(added.value->source) << "previous definition is here";
return nullptr;
}
// Resolve member type
auto type = Type(member->type);
if (!type) {
return nullptr;
}
members_nest_depth = std::max(members_nest_depth, NestDepth(type));
// validator_.Validate member type
if (!validator_.IsPlain(type)) {
AddError(member->source)
<< sem_.TypeNameOf(type) << " cannot be used as the type of a structure member";
return nullptr;
}
uint64_t offset = struct_size;
uint64_t align = type->Align();
uint64_t size = type->Size();
if (!validator_.NoDuplicateAttributes(member->attributes)) {
return nullptr;
}
bool has_offset_attr = false;
bool has_align_attr = false;
bool has_size_attr = false;
core::IOAttributes attributes;
for (auto* attribute : member->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::StructMemberOffsetAttribute* attr) {
// Offset attributes are not part of the WGSL spec, but are emitted by the
// SPIR-V reader.
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant,
"@offset value"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return false;
}
auto const_value = materialized->ConstantValue();
if (!const_value) {
AddError(attr->expr->source) << "@offset must be constant expression";
return false;
}
offset = const_value->ValueAs<uint64_t>();
if (offset < struct_size) {
AddError(attr->source) << "offsets must be in ascending order";
return false;
}
has_offset_attr = true;
return true;
},
[&](const ast::StructMemberAlignAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@align"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return false;
}
if (!materialized->Type()->IsAnyOf<core::type::I32, core::type::U32>()) {
AddError(attr->source)
<< style::Attribute("@align") << " value must be an "
<< style::Type("i32") << " or " << style::Type("u32");
return false;
}
auto const_value = materialized->ConstantValue();
if (!const_value) {
AddError(attr->source)
<< style::Attribute("@align") << " value must be constant expression";
return false;
}
auto value = const_value->ValueAs<AInt>();
if (value <= 0 || !tint::IsPowerOfTwo(value)) {
AddError(attr->source) << style::Attribute("@align")
<< " value must be a positive, power-of-two integer";
return false;
}
align = u32(value);
has_align_attr = true;
return true;
},
[&](const ast::StructMemberSizeAttribute* attr) {
ExprEvalStageConstraint constraint{core::EvaluationStage::kConstant, "@size"};
TINT_SCOPED_ASSIGNMENT(expr_eval_stage_constraint_, constraint);
auto* materialized = Materialize(ValueExpression(attr->expr));
if (!materialized) {
return false;
}
if (!materialized->Type()->IsAnyOf<core::type::U32, core::type::I32>()) {
AddError(attr->source)
<< style::Attribute("@size") << " value must be an "
<< style::Type("i32") << " or " << style::Type("u32");
return false;
}
auto const_value = materialized->ConstantValue();
if (!const_value) {
AddError(attr->expr->source)
<< style::Attribute("@size") << " value must be constant expression";
return false;
}
{
auto value = const_value->ValueAs<AInt>();
if (value <= 0) {
AddError(attr->source)
<< style::Attribute("@size") << " value must be a positive integer";
return false;
}
}
auto value = const_value->ValueAs<uint64_t>();
if (value < size) {
AddError(attr->source)
<< style::Attribute("@size")
<< " must be at least as big as the type's size (" << size << ")";
return false;
}
size = u32(value);
has_size_attr = true;
return true;
},
[&](const ast::LocationAttribute* attr) {
auto value = LocationAttribute(attr);
if (value != Success) {
return false;
}
attributes.location = value.Get();
return true;
},
[&](const ast::BlendSrcAttribute* attr) {
auto value = BlendSrcAttribute(attr);
if (value != Success) {
return false;
}
attributes.blend_src = value.Get();
return true;
},
[&](const ast::ColorAttribute* attr) {
auto value = ColorAttribute(attr);
if (value != Success) {
return false;
}
attributes.color = value.Get();
return true;
},
[&](const ast::BuiltinAttribute* attr) {
auto value = BuiltinAttribute(attr);
if (value != Success) {
return false;
}
attributes.builtin = value.Get();
return true;
},
[&](const ast::InterpolateAttribute* attr) {
auto value = InterpolateAttribute(attr);
if (value != Success) {
return false;
}
attributes.interpolation = value.Get();
return true;
},
[&](const ast::InvariantAttribute* attr) {
if (!InvariantAttribute(attr)) {
return false;
}
attributes.invariant = true;
return true;
},
[&](const ast::StrideAttribute* attr) {
if (validator_.IsValidationEnabled(
member->attributes, ast::DisabledValidation::kIgnoreStrideAttribute)) {
ErrorInvalidAttribute(
attribute, StyledText{} << style::Keyword("struct") << " members");
return false;
}
return StrideAttribute(attr);
},
[&](const ast::InternalAttribute* attr) { return InternalAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute,
StyledText{} << style::Keyword("struct") << " members");
return false;
});
if (!ok) {
return nullptr;
}
}
if (has_offset_attr && (has_align_attr || has_size_attr)) {
AddError(member->source)
<< style::Attribute("@offset") << " cannot be used with "
<< style::Attribute("@align") << " or " << style::Attribute("@size");
return nullptr;
}
offset = tint::RoundUp(align, offset);
if (offset > std::numeric_limits<uint32_t>::max()) {
AddError(member->source)
<< "struct member offset (0x" << std::hex << offset << ") must not exceed 0x"
<< std::hex << std::numeric_limits<uint32_t>::max() << " bytes";
return nullptr;
}
auto* sem_member = b.create<sem::StructMember>(
member, member->name->symbol, type, static_cast<uint32_t>(sem_members.Length()),
static_cast<uint32_t>(offset), static_cast<uint32_t>(align),
static_cast<uint32_t>(size), attributes);
b.Sem().Add(member, sem_member);
sem_members.Push(sem_member);
struct_size = offset + size;
struct_align = std::max(struct_align, align);
}
uint64_t size_no_padding = struct_size;
struct_size = tint::RoundUp(struct_align, struct_size);
if (struct_size > std::numeric_limits<uint32_t>::max()) {
AddError(str->source) << "struct size (0x" << std::hex << struct_size
<< ") must not exceed 0xffffffff bytes";
return nullptr;
}
if (TINT_UNLIKELY(struct_align > std::numeric_limits<uint32_t>::max())) {
ICE(str->source) << "calculated struct stride exceeds uint32";
}
auto* out = b.create<sem::Struct>(
str, str->name->symbol, std::move(sem_members), static_cast<uint32_t>(struct_align),
static_cast<uint32_t>(struct_size), static_cast<uint32_t>(size_no_padding));
for (size_t i = 0; i < sem_members.Length(); i++) {
auto* mem_type = sem_members[i]->Type();
if (mem_type->Is<core::type::Atomic>()) {
atomic_composite_info_.Add(out, &sem_members[i]->Declaration()->source);
break;
} else {
if (auto found = atomic_composite_info_.Get(mem_type)) {
atomic_composite_info_.Add(out, *found);
break;
}
}
const_cast<sem::StructMember*>(sem_members[i])->SetStruct(out);
}
auto stage = current_function_ ? current_function_->Declaration()->PipelineStage()
: ast::PipelineStage::kNone;
if (!validator_.Structure(out, stage)) {
return nullptr;
}
// Maximum nesting depth of composite types
// https://gpuweb.github.io/gpuweb/wgsl/#limits
const size_t nest_depth = 1 + members_nest_depth;
if (nest_depth > kMaxNestDepthOfCompositeType) {
AddError(str->source) << style::Keyword("struct ") << style::Type(struct_name())
<< " has nesting depth of " << nest_depth << ", maximum is "
<< kMaxNestDepthOfCompositeType;
return nullptr;
}
nest_depth_.Add(out, nest_depth);
return out;
}
sem::Statement* Resolver::ReturnStatement(const ast::ReturnStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto& behaviors = current_statement_->Behaviors();
behaviors = sem::Behavior::kReturn;
const core::type::Type* value_ty = nullptr;
if (auto* value = stmt->value) {
const auto* expr = Load(ValueExpression(value));
if (!expr) {
return false;
}
if (auto* ret_ty = current_function_->ReturnType(); !ret_ty->Is<core::type::Void>()) {
expr = Materialize(expr, ret_ty);
if (!expr) {
return false;
}
}
behaviors.Add(expr->Behaviors() - sem::Behavior::kNext);
value_ty = expr->Type();
} else {
value_ty = b.create<core::type::Void>();
}
// Validate after processing the return value expression so that its type
// is available for validation.
return validator_.Return(stmt, current_function_->ReturnType(), value_ty,
current_statement_);
});
}
sem::SwitchStatement* Resolver::SwitchStatement(const ast::SwitchStatement* stmt) {
auto* sem =
b.create<sem::SwitchStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto& behaviors = sem->Behaviors();
const auto* cond = Load(ValueExpression(stmt->condition));
if (!cond) {
return false;
}
behaviors = cond->Behaviors() - sem::Behavior::kNext;
auto* cond_ty = cond->Type();
// Determine the common type across all selectors and the switch expression
// This must materialize to an integer scalar (non-abstract).
Vector<const core::type::Type*, 8> types;
types.Push(cond_ty);
for (auto* case_stmt : stmt->body) {
for (auto* sel : case_stmt->selectors) {
if (sel->IsDefault()) {
continue;
}
auto* sem_expr = ValueExpression(sel->expr);
if (!sem_expr) {
return false;
}
types.Push(sem_expr->Type()->UnwrapRef());
}
}
auto* common_ty = core::type::Type::Common(types);
if (!common_ty || !common_ty->is_integer_scalar()) {
// No common type found or the common type was abstract.
// Pick i32 and let validation deal with any mismatches.
common_ty = b.create<core::type::I32>();
}
cond = Materialize(cond, common_ty);
if (!cond) {
return false;
}
// Handle switch body attributes.
for (auto* attribute : stmt->body_attributes) {
Mark(attribute);
bool ok = Switch(
attribute,
[&](const ast::DiagnosticAttribute* attr) { return DiagnosticAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute,
StyledText{} << style::Keyword("switch") << " body");
return false;
});
if (!ok) {
return false;
}
}
if (!validator_.NoDuplicateAttributes(stmt->body_attributes)) {
return false;
}
Vector<sem::CaseStatement*, 4> cases;
cases.Reserve(stmt->body.Length());
for (auto* case_stmt : stmt->body) {
Mark(case_stmt);
auto* c = CaseStatement(case_stmt, common_ty);
if (!c) {
return false;
}
cases.Push(c);
behaviors.Add(c->Behaviors());
sem->Cases().emplace_back(c);
ApplyDiagnosticSeverities(c);
}
if (behaviors.Contains(sem::Behavior::kBreak)) {
behaviors.Add(sem::Behavior::kNext);
}
behaviors.Remove(sem::Behavior::kBreak);
return validator_.SwitchStatement(stmt);
});
}
sem::Statement* Resolver::VariableDeclStatement(const ast::VariableDeclStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
Mark(stmt->variable);
auto* variable = Variable(stmt->variable, /* is_global */ false);
if (!variable) {
return false;
}
current_compound_statement_->AddDecl(variable->As<sem::LocalVariable>());
if (auto* ctor = variable->Initializer()) {
sem->Behaviors() = ctor->Behaviors();
}
return validator_.LocalVariable(variable);
});
}
sem::Statement* Resolver::AssignmentStatement(const ast::AssignmentStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto* lhs = ValueExpression(stmt->lhs);
if (!lhs) {
return false;
}
const bool is_phony_assignment = stmt->lhs->Is<ast::PhonyExpression>();
const auto* rhs = ValueExpression(stmt->rhs);
if (!rhs) {
return false;
}
if (!is_phony_assignment) {
rhs = Materialize(rhs, lhs->Type()->UnwrapRef());
if (!rhs) {
return false;
}
}
rhs = Load(rhs);
if (!rhs) {
return false;
}
auto& behaviors = sem->Behaviors();
behaviors = rhs->Behaviors();
if (!is_phony_assignment) {
behaviors.Add(lhs->Behaviors());
}
if (!is_phony_assignment) {
RegisterStore(lhs);
}
return validator_.Assignment(stmt, sem_.TypeOf(stmt->rhs));
});
}
sem::Statement* Resolver::BreakStatement(const ast::BreakStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
sem->Behaviors() = sem::Behavior::kBreak;
return validator_.BreakStatement(sem, current_statement_);
});
}
sem::Statement* Resolver::BreakIfStatement(const ast::BreakIfStatement* stmt) {
auto* sem =
b.create<sem::BreakIfStatement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto* cond = Load(ValueExpression(stmt->condition));
if (!cond) {
return false;
}
sem->SetCondition(cond);
sem->Behaviors() = cond->Behaviors();
sem->Behaviors().Add(sem::Behavior::kBreak);
return validator_.BreakIfStatement(sem, current_statement_);
});
}
sem::Statement* Resolver::CallStatement(const ast::CallStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
if (auto* expr = ValueExpression(stmt->expr)) {
sem->Behaviors() = expr->Behaviors();
return true;
}
return false;
});
}
sem::Statement* Resolver::CompoundAssignmentStatement(
const ast::CompoundAssignmentStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto* lhs = ValueExpression(stmt->lhs);
if (!lhs) {
return false;
}
const auto* rhs = ValueExpression(stmt->rhs);
if (!rhs) {
return false;
}
RegisterStore(lhs);
sem->Behaviors() = rhs->Behaviors() + lhs->Behaviors();
auto stage = core::EarliestStage(lhs->Stage(), rhs->Stage());
auto overload = intrinsic_table_.Lookup(stmt->op, lhs->Type()->UnwrapRef(),
rhs->Type()->UnwrapRef(), stage, true);
if (overload != Success) {
AddError(stmt->source) << overload.Failure();
return false;
}
// Load or materialize the RHS if necessary.
rhs = Load(Materialize(rhs, overload->parameters[1].type));
if (!rhs) {
return false;
}
return validator_.Assignment(stmt, overload->return_type);
});
}
sem::Statement* Resolver::ContinueStatement(const ast::ContinueStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
sem->Behaviors() = sem::Behavior::kContinue;
// Set if we've hit the first continue statement in our parent loop
if (auto* block = sem->FindFirstParent<sem::LoopBlockStatement>()) {
if (!block->FirstContinue()) {
const_cast<sem::LoopBlockStatement*>(block)->SetFirstContinue(
stmt, block->Decls().Count());
}
}
return validator_.ContinueStatement(sem, current_statement_);
});
}
sem::Statement* Resolver::DiscardStatement(const ast::DiscardStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
current_function_->SetDiscardStatement(sem);
return true;
});
}
sem::Statement* Resolver::IncrementDecrementStatement(
const ast::IncrementDecrementStatement* stmt) {
auto* sem = b.create<sem::Statement>(stmt, current_compound_statement_, current_function_);
return StatementScope(stmt, sem, [&] {
auto* lhs = ValueExpression(stmt->lhs);
if (!lhs) {
return false;
}
sem->Behaviors() = lhs->Behaviors();
RegisterStore(lhs);
return validator_.IncrementDecrementStatement(stmt);
});
}
bool Resolver::ApplyAddressSpaceUsageToType(core::AddressSpace address_space,
core::type::Type* ty,
const Source& usage) {
ty = const_cast<core::type::Type*>(ty->UnwrapRef());
if (auto* str = ty->As<sem::Struct>()) {
if (str->AddressSpaceUsage().Contains(address_space)) {
return true; // Already applied
}
str->AddUsage(address_space);
for (auto* member : str->Members()) {
auto decl = member->Declaration();
if (decl && !ApplyAddressSpaceUsageToType(address_space,
const_cast<core::type::Type*>(member->Type()),
decl->type->source)) {
AddNote(member->Declaration()->source)
<< "while analyzing structure member " << sem_.TypeNameOf(str) << "."
<< member->Name().Name();
return false;
}
}
return true;
}
if (auto* arr = ty->As<sem::Array>()) {
if (address_space != core::AddressSpace::kStorage) {
if (arr->Count()->Is<core::type::RuntimeArrayCount>()) {
AddError(usage)
<< "runtime-sized arrays can only be used in the <storage> address space";
return false;
}
auto count = arr->ConstantCount();
if (count.has_value() && count.value() >= kMaxArrayElementCount) {
AddError(usage) << "array count (" << count.value() << ") must be less than "
<< kMaxArrayElementCount;
return false;
}
}
return ApplyAddressSpaceUsageToType(address_space,
const_cast<core::type::Type*>(arr->ElemType()), usage);
}
if (core::IsHostShareable(address_space) && !validator_.IsHostShareable(ty)) {
AddError(usage) << "type " << style::Type(sem_.TypeNameOf(ty))
<< " cannot be used in address space " << style::Enum(address_space)
<< " as it is non-host-shareable";
return false;
}
return true;
}
template <typename SEM, typename F>
SEM* Resolver::StatementScope(const ast::Statement* ast, SEM* sem, F&& callback) {
b.Sem().Add(ast, sem);
auto* as_compound =
As<sem::CompoundStatement, tint::CastFlags::kDontErrorOnImpossibleCast>(sem);
// Helper to handle attributes that are supported on certain types of statement.
auto handle_attributes = [&](auto* stmt, sem::Statement* sem_stmt, const char* use) {
for (auto* attribute : stmt->attributes) {
Mark(attribute);
bool ok = Switch(
attribute, //
[&](const ast::DiagnosticAttribute* attr) { return DiagnosticAttribute(attr); },
[&](Default) {
ErrorInvalidAttribute(attribute, StyledText{} << use);
return false;
});
if (!ok) {
return false;
}
}
if (!validator_.NoDuplicateAttributes(stmt->attributes)) {
return false;
}
ApplyDiagnosticSeverities(sem_stmt);
return true;
};
// Handle attributes, if necessary.
// Some statements can take diagnostic filtering attributes, so push a new diagnostic filter
// scope to capture them.
validator_.DiagnosticFilters().Push();
TINT_DEFER(validator_.DiagnosticFilters().Pop());
if (!Switch(
ast, //
[&](const ast::BlockStatement* block) {
return handle_attributes(block, sem, "block statements");
},
[&](const ast::ForLoopStatement* f) {
return handle_attributes(f, sem, "for statements");
},
[&](const ast::IfStatement* i) { return handle_attributes(i, sem, "if statements"); },
[&](const ast::LoopStatement* l) {
return handle_attributes(l, sem, "loop statements");
},
[&](const ast::SwitchStatement* s) {
return handle_attributes(s, sem, "switch statements");
},
[&](const ast::WhileStatement* w) {
return handle_attributes(w, sem, "while statements");
},
[&](Default) { return true; })) {
return nullptr;
}
TINT_SCOPED_ASSIGNMENT(current_statement_, sem);
TINT_SCOPED_ASSIGNMENT(current_compound_statement_,
as_compound ? as_compound : current_compound_statement_);
TINT_SCOPED_ASSIGNMENT(current_scoping_depth_, current_scoping_depth_ + 1);
if (current_scoping_depth_ > kMaxStatementDepth) {
AddError(ast->source) << "statement nesting depth / chaining length exceeds limit of "
<< kMaxStatementDepth;
return nullptr;
}
if (!callback()) {
return nullptr;
}
return sem;
}
bool Resolver::Mark(const ast::Node* node) {
if (TINT_UNLIKELY(node == nullptr)) {
TINT_ICE() << "Resolver::Mark() called with nullptr";
}
auto marked_bit_ref = marked_[node->node_id.value];
if (TINT_LIKELY(!marked_bit_ref)) {
marked_bit_ref = true;
return true;
}
ICE(node->source) << "AST node '" << node->TypeInfo().name
<< "' was encountered twice in the same AST of a Program\n"
<< "Pointer: " << node;
}
template <typename NODE>
void Resolver::ApplyDiagnosticSeverities(NODE* node) {
for (auto itr : validator_.DiagnosticFilters().Top()) {
node->SetDiagnosticSeverity(itr.key, itr.value);
}
}
bool Resolver::CheckNotTemplated(const char* use, const ast::Identifier* ident) {
if (TINT_UNLIKELY(ident->Is<ast::TemplatedIdentifier>())) {
AddError(ident->source) << use << " " << style::Code(ident->symbol.NameView())
<< " does not take template arguments";
if (auto resolved = dependencies_.resolved_identifiers.Get(ident)) {
if (auto* ast_node = resolved->Node()) {
sem_.NoteDeclarationSource(ast_node);
}
}
return false;
}
return true;
}
void Resolver::ErrorInvalidAttribute(const ast::Attribute* attr, StyledText use) {
AddError(attr->source) << style::Attribute("@", attr->Name()) << " is not valid for " << use;
}
diag::Diagnostic& Resolver::AddError(const Source& source) const {
return diagnostics_.AddError(source);
}
diag::Diagnostic& Resolver::AddWarning(const Source& source) const {
return diagnostics_.AddWarning(source);
}
diag::Diagnostic& Resolver::AddNote(const Source& source) const {
return diagnostics_.AddNote(source);
}
} // namespace tint::resolver