Commit 2c7f34c8 by Olli Etuaho Committed by Commit Bot

Initialize uninitialized GLSL arrays in a for loop

Previously, creating nodes for initializing each single array element could result in memory bloat during translation when dealing with large arrays. The resulting shader could also end up very long. Initialize most arrays using a simple for loop instead. The loop is compatible with ESSL 1.00 Appendix A limitations. An exception is made for fragment outputs, so that they are not indexed by non-constant values. On some platforms using the a loop to initialize variables can cause problems, so we also have a compiler flag for turning this behavior off. The flag was already added earlier for a staggered rollout of this functionality. BUG=chromium:735497 TEST=angle_unittests, angle_end2end_tests, WebGL conformance tests Change-Id: Iec727821d8137db56b440ddbe007879b1b55f61f Reviewed-on: https://chromium-review.googlesource.com/707195 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: 's avatarCorentin Wallez <cwallez@chromium.org> Reviewed-by: 's avatarJamie Madill <jmadill@chromium.org>
parent bb27c3a1
...@@ -595,7 +595,9 @@ bool TCompiler::checkAndSimplifyAST(TIntermBlock *root, ...@@ -595,7 +595,9 @@ bool TCompiler::checkAndSimplifyAST(TIntermBlock *root,
// on ESSL >= 3.00, and the initializers that need to be deferred can only exist in ESSL < 3.00. // on ESSL >= 3.00, and the initializers that need to be deferred can only exist in ESSL < 3.00.
bool initializeLocalsAndGlobals = bool initializeLocalsAndGlobals =
(compileOptions & SH_INITIALIZE_UNINITIALIZED_LOCALS) && !IsOutputHLSL(getOutputType()); (compileOptions & SH_INITIALIZE_UNINITIALIZED_LOCALS) && !IsOutputHLSL(getOutputType());
DeferGlobalInitializers(root, initializeLocalsAndGlobals, &symbolTable); bool canUseLoopsToInitialize = !(compileOptions & SH_DONT_USE_LOOPS_TO_INITIALIZE_VARIABLES);
DeferGlobalInitializers(root, initializeLocalsAndGlobals, canUseLoopsToInitialize, &symbolTable);
if (initializeLocalsAndGlobals) if (initializeLocalsAndGlobals)
{ {
...@@ -615,7 +617,8 @@ bool TCompiler::checkAndSimplifyAST(TIntermBlock *root, ...@@ -615,7 +617,8 @@ bool TCompiler::checkAndSimplifyAST(TIntermBlock *root,
&getSymbolTable(), getShaderVersion()); &getSymbolTable(), getShaderVersion());
} }
InitializeUninitializedLocals(root, getShaderVersion()); InitializeUninitializedLocals(root, getShaderVersion(), canUseLoopsToInitialize,
&getSymbolTable());
} }
if (getShaderType() == GL_VERTEX_SHADER && (compileOptions & SH_CLAMP_POINT_SIZE)) if (getShaderType() == GL_VERTEX_SHADER && (compileOptions & SH_CLAMP_POINT_SIZE))
...@@ -1055,7 +1058,7 @@ void TCompiler::initializeGLPosition(TIntermBlock *root) ...@@ -1055,7 +1058,7 @@ void TCompiler::initializeGLPosition(TIntermBlock *root)
sh::ShaderVariable var(GL_FLOAT_VEC4); sh::ShaderVariable var(GL_FLOAT_VEC4);
var.name = "gl_Position"; var.name = "gl_Position";
list.push_back(var); list.push_back(var);
InitializeVariables(root, list, symbolTable, shaderVersion, extensionBehavior); InitializeVariables(root, list, &symbolTable, shaderVersion, extensionBehavior, false);
} }
void TCompiler::useAllMembersInUnusedStandardAndSharedBlocks(TIntermBlock *root) void TCompiler::useAllMembersInUnusedStandardAndSharedBlocks(TIntermBlock *root)
...@@ -1097,7 +1100,7 @@ void TCompiler::initializeOutputVariables(TIntermBlock *root) ...@@ -1097,7 +1100,7 @@ void TCompiler::initializeOutputVariables(TIntermBlock *root)
list.push_back(var); list.push_back(var);
} }
} }
InitializeVariables(root, list, symbolTable, shaderVersion, extensionBehavior); InitializeVariables(root, list, &symbolTable, shaderVersion, extensionBehavior, false);
} }
const TExtensionBehavior &TCompiler::getExtensionBehavior() const const TExtensionBehavior &TCompiler::getExtensionBehavior() const
......
...@@ -29,7 +29,9 @@ namespace ...@@ -29,7 +29,9 @@ namespace
void GetDeferredInitializers(TIntermDeclaration *declaration, void GetDeferredInitializers(TIntermDeclaration *declaration,
bool initializeUninitializedGlobals, bool initializeUninitializedGlobals,
TIntermSequence *deferredInitializersOut) bool canUseLoopsToInitialize,
TIntermSequence *deferredInitializersOut,
TSymbolTable *symbolTable)
{ {
// SeparateDeclarations should have already been run. // SeparateDeclarations should have already been run.
ASSERT(declaration->getSequence()->size() == 1); ASSERT(declaration->getSequence()->size() == 1);
...@@ -81,7 +83,8 @@ void GetDeferredInitializers(TIntermDeclaration *declaration, ...@@ -81,7 +83,8 @@ void GetDeferredInitializers(TIntermDeclaration *declaration,
if (symbolNode->getQualifier() == EvqGlobal && symbolNode->getSymbol() != "") if (symbolNode->getQualifier() == EvqGlobal && symbolNode->getSymbol() != "")
{ {
TIntermSequence *initCode = CreateInitCode(symbolNode); TIntermSequence *initCode =
CreateInitCode(symbolNode, canUseLoopsToInitialize, symbolTable);
deferredInitializersOut->insert(deferredInitializersOut->end(), initCode->begin(), deferredInitializersOut->insert(deferredInitializersOut->end(), initCode->begin(),
initCode->end()); initCode->end());
} }
...@@ -117,6 +120,7 @@ void InsertInitCallToMain(TIntermBlock *root, ...@@ -117,6 +120,7 @@ void InsertInitCallToMain(TIntermBlock *root,
void DeferGlobalInitializers(TIntermBlock *root, void DeferGlobalInitializers(TIntermBlock *root,
bool initializeUninitializedGlobals, bool initializeUninitializedGlobals,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable) TSymbolTable *symbolTable)
{ {
TIntermSequence *deferredInitializers = new TIntermSequence(); TIntermSequence *deferredInitializers = new TIntermSequence();
...@@ -129,7 +133,7 @@ void DeferGlobalInitializers(TIntermBlock *root, ...@@ -129,7 +133,7 @@ void DeferGlobalInitializers(TIntermBlock *root,
if (declaration) if (declaration)
{ {
GetDeferredInitializers(declaration, initializeUninitializedGlobals, GetDeferredInitializers(declaration, initializeUninitializedGlobals,
deferredInitializers); canUseLoopsToInitialize, deferredInitializers, symbolTable);
} }
} }
......
...@@ -24,6 +24,7 @@ class TSymbolTable; ...@@ -24,6 +24,7 @@ class TSymbolTable;
void DeferGlobalInitializers(TIntermBlock *root, void DeferGlobalInitializers(TIntermBlock *root,
bool initializeUninitializedGlobals, bool initializeUninitializedGlobals,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable); TSymbolTable *symbolTable);
} // namespace sh } // namespace sh
......
...@@ -21,7 +21,14 @@ namespace ...@@ -21,7 +21,14 @@ namespace
{ {
void AddArrayZeroInitSequence(const TIntermTyped *initializedNode, void AddArrayZeroInitSequence(const TIntermTyped *initializedNode,
TIntermSequence *initSequenceOut); bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable);
void AddStructZeroInitSequence(const TIntermTyped *initializedNode,
bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable);
TIntermBinary *CreateZeroInitAssignment(const TIntermTyped *initializedNode) TIntermBinary *CreateZeroInitAssignment(const TIntermTyped *initializedNode)
{ {
...@@ -29,8 +36,32 @@ TIntermBinary *CreateZeroInitAssignment(const TIntermTyped *initializedNode) ...@@ -29,8 +36,32 @@ TIntermBinary *CreateZeroInitAssignment(const TIntermTyped *initializedNode)
return new TIntermBinary(EOpAssign, initializedNode->deepCopy(), zero); return new TIntermBinary(EOpAssign, initializedNode->deepCopy(), zero);
} }
void AddZeroInitSequence(const TIntermTyped *initializedNode,
bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable)
{
if (initializedNode->isArray())
{
AddArrayZeroInitSequence(initializedNode, canUseLoopsToInitialize, initSequenceOut,
symbolTable);
}
else if (initializedNode->getType().isStructureContainingArrays() ||
initializedNode->getType().isNamelessStruct())
{
AddStructZeroInitSequence(initializedNode, canUseLoopsToInitialize, initSequenceOut,
symbolTable);
}
else
{
initSequenceOut->push_back(CreateZeroInitAssignment(initializedNode));
}
}
void AddStructZeroInitSequence(const TIntermTyped *initializedNode, void AddStructZeroInitSequence(const TIntermTyped *initializedNode,
TIntermSequence *initSequenceOut) bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable)
{ {
ASSERT(initializedNode->getBasicType() == EbtStruct); ASSERT(initializedNode->getBasicType() == EbtStruct);
const TStructure *structType = initializedNode->getType().getStruct(); const TStructure *structType = initializedNode->getType().getStruct();
...@@ -38,55 +69,87 @@ void AddStructZeroInitSequence(const TIntermTyped *initializedNode, ...@@ -38,55 +69,87 @@ void AddStructZeroInitSequence(const TIntermTyped *initializedNode,
{ {
TIntermBinary *element = new TIntermBinary(EOpIndexDirectStruct, TIntermBinary *element = new TIntermBinary(EOpIndexDirectStruct,
initializedNode->deepCopy(), CreateIndexNode(i)); initializedNode->deepCopy(), CreateIndexNode(i));
if (element->isArray()) // Structs can't be defined inside structs, so the type of a struct field can't be a
{ // nameless struct.
AddArrayZeroInitSequence(element, initSequenceOut); ASSERT(!element->getType().isNamelessStruct());
} AddZeroInitSequence(element, canUseLoopsToInitialize, initSequenceOut, symbolTable);
else if (element->getType().isStructureContainingArrays())
{
AddStructZeroInitSequence(element, initSequenceOut);
}
else
{
// Structs can't be defined inside structs, so the type of a struct field can't be a
// nameless struct.
ASSERT(!element->getType().isNamelessStruct());
initSequenceOut->push_back(CreateZeroInitAssignment(element));
}
} }
} }
void AddArrayZeroInitSequence(const TIntermTyped *initializedNode, TIntermSequence *initSequenceOut) void AddArrayZeroInitStatementList(const TIntermTyped *initializedNode,
bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable)
{ {
ASSERT(initializedNode->isArray());
// Assign the array elements one by one to keep the AST compatible with ESSL 1.00 which
// doesn't have array assignment.
// Note that it is important to have the array init in the right order to workaround
// http://crbug.com/709317
for (unsigned int i = 0; i < initializedNode->getOutermostArraySize(); ++i) for (unsigned int i = 0; i < initializedNode->getOutermostArraySize(); ++i)
{ {
TIntermBinary *element = TIntermBinary *element =
new TIntermBinary(EOpIndexDirect, initializedNode->deepCopy(), CreateIndexNode(i)); new TIntermBinary(EOpIndexDirect, initializedNode->deepCopy(), CreateIndexNode(i));
if (element->isArray()) AddZeroInitSequence(element, canUseLoopsToInitialize, initSequenceOut, symbolTable);
{ }
AddArrayZeroInitSequence(element, initSequenceOut); }
}
else if (element->getType().isStructureContainingArrays()) void AddArrayZeroInitForLoop(const TIntermTyped *initializedNode,
{ TIntermSequence *initSequenceOut,
AddStructZeroInitSequence(element, initSequenceOut); TSymbolTable *symbolTable)
} {
else ASSERT(initializedNode->isArray());
{ TSymbolUniqueId indexSymbol(symbolTable);
initSequenceOut->push_back(CreateZeroInitAssignment(element));
} TIntermSymbol *indexSymbolNode = CreateTempSymbolNode(indexSymbol, TType(EbtInt), EvqTemporary);
TIntermDeclaration *indexInit =
CreateTempInitDeclarationNode(indexSymbol, CreateZeroNode(TType(EbtInt)), EvqTemporary);
TIntermConstantUnion *arraySizeNode = CreateIndexNode(initializedNode->getOutermostArraySize());
TIntermBinary *indexSmallerThanSize =
new TIntermBinary(EOpLessThan, indexSymbolNode->deepCopy(), arraySizeNode);
TIntermUnary *indexIncrement = new TIntermUnary(EOpPreIncrement, indexSymbolNode->deepCopy());
TIntermBlock *forLoopBody = new TIntermBlock();
TIntermSequence *forLoopBodySeq = forLoopBody->getSequence();
TIntermBinary *element = new TIntermBinary(EOpIndexIndirect, initializedNode->deepCopy(),
indexSymbolNode->deepCopy());
AddZeroInitSequence(element, true, forLoopBodySeq, symbolTable);
TIntermLoop *forLoop =
new TIntermLoop(ELoopFor, indexInit, indexSmallerThanSize, indexIncrement, forLoopBody);
initSequenceOut->push_back(forLoop);
}
void AddArrayZeroInitSequence(const TIntermTyped *initializedNode,
bool canUseLoopsToInitialize,
TIntermSequence *initSequenceOut,
TSymbolTable *symbolTable)
{
// The array elements are assigned one by one to keep the AST compatible with ESSL 1.00 which
// doesn't have array assignment. We'll do this either with a for loop or just a list of
// statements assigning to each array index. Note that it is important to have the array init in
// the right order to workaround http://crbug.com/709317
bool isSmallArray = initializedNode->getOutermostArraySize() <= 1u ||
(initializedNode->getBasicType() != EbtStruct &&
!initializedNode->getType().isArrayOfArrays() &&
initializedNode->getOutermostArraySize() <= 3u);
if (initializedNode->getQualifier() == EvqFragData ||
initializedNode->getQualifier() == EvqFragmentOut || isSmallArray ||
!canUseLoopsToInitialize)
{
// Fragment outputs should not be indexed by non-constant indices.
// Also it doesn't make sense to use loops to initialize very small arrays.
AddArrayZeroInitStatementList(initializedNode, canUseLoopsToInitialize, initSequenceOut,
symbolTable);
}
else
{
AddArrayZeroInitForLoop(initializedNode, initSequenceOut, symbolTable);
} }
} }
void InsertInitCode(TIntermSequence *mainBody, void InsertInitCode(TIntermSequence *mainBody,
const InitVariableList &variables, const InitVariableList &variables,
const TSymbolTable &symbolTable, TSymbolTable *symbolTable,
int shaderVersion, int shaderVersion,
const TExtensionBehavior &extensionBehavior) const TExtensionBehavior &extensionBehavior,
bool canUseLoopsToInitialize)
{ {
for (const auto &var : variables) for (const auto &var : variables)
{ {
...@@ -100,7 +163,7 @@ void InsertInitCode(TIntermSequence *mainBody, ...@@ -100,7 +163,7 @@ void InsertInitCode(TIntermSequence *mainBody,
TIntermTyped *initializedSymbol = nullptr; TIntermTyped *initializedSymbol = nullptr;
if (var.isBuiltIn()) if (var.isBuiltIn())
{ {
initializedSymbol = ReferenceBuiltInVariable(name, symbolTable, shaderVersion); initializedSymbol = ReferenceBuiltInVariable(name, *symbolTable, shaderVersion);
if (initializedSymbol->getQualifier() == EvqFragData && if (initializedSymbol->getQualifier() == EvqFragData &&
!IsExtensionEnabled(extensionBehavior, TExtension::EXT_draw_buffers)) !IsExtensionEnabled(extensionBehavior, TExtension::EXT_draw_buffers))
{ {
...@@ -116,11 +179,12 @@ void InsertInitCode(TIntermSequence *mainBody, ...@@ -116,11 +179,12 @@ void InsertInitCode(TIntermSequence *mainBody,
} }
else else
{ {
initializedSymbol = ReferenceGlobalVariable(name, symbolTable); initializedSymbol = ReferenceGlobalVariable(name, *symbolTable);
} }
ASSERT(initializedSymbol != nullptr); ASSERT(initializedSymbol != nullptr);
TIntermSequence *initCode = CreateInitCode(initializedSymbol); TIntermSequence *initCode =
CreateInitCode(initializedSymbol, canUseLoopsToInitialize, symbolTable);
mainBody->insert(mainBody->begin(), initCode->begin(), initCode->end()); mainBody->insert(mainBody->begin(), initCode->begin(), initCode->end());
} }
} }
...@@ -128,8 +192,12 @@ void InsertInitCode(TIntermSequence *mainBody, ...@@ -128,8 +192,12 @@ void InsertInitCode(TIntermSequence *mainBody,
class InitializeLocalsTraverser : public TIntermTraverser class InitializeLocalsTraverser : public TIntermTraverser
{ {
public: public:
InitializeLocalsTraverser(int shaderVersion) InitializeLocalsTraverser(int shaderVersion,
: TIntermTraverser(true, false, false), mShaderVersion(shaderVersion) TSymbolTable *symbolTable,
bool canUseLoopsToInitialize)
: TIntermTraverser(true, false, false, symbolTable),
mShaderVersion(shaderVersion),
mCanUseLoopsToInitialize(canUseLoopsToInitialize)
{ {
} }
...@@ -154,6 +222,9 @@ class InitializeLocalsTraverser : public TIntermTraverser ...@@ -154,6 +222,9 @@ class InitializeLocalsTraverser : public TIntermTraverser
mShaderVersion == 100; mShaderVersion == 100;
// Nameless struct constructors can't be referred to, so they also need to be // Nameless struct constructors can't be referred to, so they also need to be
// initialized one element at a time. // initialized one element at a time.
// TODO(oetuaho): Check if it makes sense to initialize using a loop, even if we
// could use an initializer. It could at least reduce code size for very large
// arrays, but could hurt runtime performance.
if (arrayConstructorUnavailable || symbol->getType().isNamelessStruct()) if (arrayConstructorUnavailable || symbol->getType().isNamelessStruct())
{ {
// SimplifyLoopConditions should have been run so the parent node of this node // SimplifyLoopConditions should have been run so the parent node of this node
...@@ -163,7 +234,9 @@ class InitializeLocalsTraverser : public TIntermTraverser ...@@ -163,7 +234,9 @@ class InitializeLocalsTraverser : public TIntermTraverser
// about further declarators in this declaration depending on the effects of // about further declarators in this declaration depending on the effects of
// this declarator. // this declarator.
ASSERT(node->getSequence()->size() == 1); ASSERT(node->getSequence()->size() == 1);
insertStatementsInParentBlock(TIntermSequence(), *CreateInitCode(symbol)); insertStatementsInParentBlock(
TIntermSequence(),
*CreateInitCode(symbol, mCanUseLoopsToInitialize, mSymbolTable));
} }
else else
{ {
...@@ -178,44 +251,40 @@ class InitializeLocalsTraverser : public TIntermTraverser ...@@ -178,44 +251,40 @@ class InitializeLocalsTraverser : public TIntermTraverser
private: private:
int mShaderVersion; int mShaderVersion;
bool mCanUseLoopsToInitialize;
}; };
} // namespace anonymous } // namespace anonymous
TIntermSequence *CreateInitCode(const TIntermTyped *initializedSymbol) TIntermSequence *CreateInitCode(const TIntermTyped *initializedSymbol,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable)
{ {
TIntermSequence *initCode = new TIntermSequence(); TIntermSequence *initCode = new TIntermSequence();
if (initializedSymbol->isArray()) AddZeroInitSequence(initializedSymbol, canUseLoopsToInitialize, initCode, symbolTable);
{
AddArrayZeroInitSequence(initializedSymbol, initCode);
}
else if (initializedSymbol->getType().isStructureContainingArrays() ||
initializedSymbol->getType().isNamelessStruct())
{
AddStructZeroInitSequence(initializedSymbol, initCode);
}
else
{
initCode->push_back(CreateZeroInitAssignment(initializedSymbol));
}
return initCode; return initCode;
} }
void InitializeUninitializedLocals(TIntermBlock *root, int shaderVersion) void InitializeUninitializedLocals(TIntermBlock *root,
int shaderVersion,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable)
{ {
InitializeLocalsTraverser traverser(shaderVersion); InitializeLocalsTraverser traverser(shaderVersion, symbolTable, canUseLoopsToInitialize);
root->traverse(&traverser); root->traverse(&traverser);
traverser.updateTree(); traverser.updateTree();
} }
void InitializeVariables(TIntermBlock *root, void InitializeVariables(TIntermBlock *root,
const InitVariableList &vars, const InitVariableList &vars,
const TSymbolTable &symbolTable, TSymbolTable *symbolTable,
int shaderVersion, int shaderVersion,
const TExtensionBehavior &extensionBehavior) const TExtensionBehavior &extensionBehavior,
bool canUseLoopsToInitialize)
{ {
TIntermBlock *body = FindMainBody(root); TIntermBlock *body = FindMainBody(root);
InsertInitCode(body->getSequence(), vars, symbolTable, shaderVersion, extensionBehavior); InsertInitCode(body->getSequence(), vars, symbolTable, shaderVersion, extensionBehavior,
canUseLoopsToInitialize);
} }
} // namespace sh } // namespace sh
...@@ -18,12 +18,20 @@ class TSymbolTable; ...@@ -18,12 +18,20 @@ class TSymbolTable;
typedef std::vector<sh::ShaderVariable> InitVariableList; typedef std::vector<sh::ShaderVariable> InitVariableList;
// For all of the functions below: If canUseLoopsToInitialize is set, for loops are used instead of
// a large number of initializers where it can make sense, such as for initializing large arrays.
// Return a sequence of assignment operations to initialize "initializedSymbol". initializedSymbol // Return a sequence of assignment operations to initialize "initializedSymbol". initializedSymbol
// may be an array, struct or any combination of these, as long as it contains only basic types. // may be an array, struct or any combination of these, as long as it contains only basic types.
TIntermSequence *CreateInitCode(const TIntermTyped *initializedSymbol); TIntermSequence *CreateInitCode(const TIntermTyped *initializedSymbol,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable);
// Initialize all uninitialized local variables, so that undefined behavior is avoided. // Initialize all uninitialized local variables, so that undefined behavior is avoided.
void InitializeUninitializedLocals(TIntermBlock *root, int shaderVersion); void InitializeUninitializedLocals(TIntermBlock *root,
int shaderVersion,
bool canUseLoopsToInitialize,
TSymbolTable *symbolTable);
// This function can initialize all the types that CreateInitCode is able to initialize. All // This function can initialize all the types that CreateInitCode is able to initialize. All
// variables must be globals which can be found in the symbol table. For now it is used for the // variables must be globals which can be found in the symbol table. For now it is used for the
...@@ -35,9 +43,10 @@ void InitializeUninitializedLocals(TIntermBlock *root, int shaderVersion); ...@@ -35,9 +43,10 @@ void InitializeUninitializedLocals(TIntermBlock *root, int shaderVersion);
// enabled extensions. // enabled extensions.
void InitializeVariables(TIntermBlock *root, void InitializeVariables(TIntermBlock *root,
const InitVariableList &vars, const InitVariableList &vars,
const TSymbolTable &symbolTable, TSymbolTable *symbolTable,
int shaderVersion, int shaderVersion,
const TExtensionBehavior &extensionBehavior); const TExtensionBehavior &extensionBehavior,
bool canUseLoopsToInitialize);
} // namespace sh } // namespace sh
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment