Performance: Simple upgrade/cleanup of stl c++11 containers (10% perf. increase).

SPIRV/GlslangToSpv.cpp

@@ -117,11 +117,11 @@ protected:
     const glslang::TIntermediate* glslangIntermediate;
     spv::Id stdBuiltins;
 
-    std::map<int, spv::Id> symbolValues;
-    std::set<int> constReadOnlyParameters;  // set of formal function parameters that have glslang qualifier constReadOnly, so we know they are not local function "const" that are write-once
-    std::map<std::string, spv::Function*> functionMap;
-    std::map<const glslang::TTypeList*, spv::Id> structMap;
-    std::map<const glslang::TTypeList*, std::vector<int> > memberRemapper;  // for mapping glslang block indices to spv indices (e.g., due to hidden members)
+    std::unordered_map<int, spv::Id> symbolValues;
+    std::unordered_set<int> constReadOnlyParameters;  // set of formal function parameters that have glslang qualifier constReadOnly, so we know they are not local function "const" that are write-once
+    std::unordered_map<std::string, spv::Function*> functionMap;
+    std::unordered_map<const glslang::TTypeList*, spv::Id> structMap;
+    std::unordered_map<const glslang::TTypeList*, std::vector<int> > memberRemapper;  // for mapping glslang block indices to spv indices (e.g., due to hidden members)
     std::stack<bool> breakForLoop;  // false means break for switch
     std::stack<glslang::TIntermTyped*> loopTerminal;  // code from the last part of a for loop: for(...; ...; terminal), needed for e.g., continue };
 };
@@ -361,8 +361,8 @@ TGlslangToSpvTraverser::TGlslangToSpvTraverser(const glslang::TIntermediate* gls
     builder.addEntryPoint(executionModel, shaderEntry);
 
     // Add the source extensions
-    const std::set<std::string>& sourceExtensions = glslangIntermediate->getRequestedExtensions();
-    for (std::set<std::string>::const_iterator it = sourceExtensions.begin(); it != sourceExtensions.end(); ++it)
+    const auto& sourceExtensions = glslangIntermediate->getRequestedExtensions();
+    for (auto it = sourceExtensions.begin(); it != sourceExtensions.end(); ++it)
         builder.addSourceExtension(it->c_str());
 
     // Add the top-level modes for this shader.
@@ -2385,8 +2385,7 @@ spv::Id TGlslangToSpvTraverser::createNoArgOperation(glslang::TOperator op)
 
 spv::Id TGlslangToSpvTraverser::getSymbolId(const glslang::TIntermSymbol* symbol)
 {
-    std::map<int, spv::Id>::iterator iter;
-    iter = symbolValues.find(symbol->getId());
+    auto iter = symbolValues.find(symbol->getId());
     spv::Id id;
     if (symbolValues.end() != iter) {
         id = iter->second;

glslang/Include/Common.h

@@ -59,8 +59,10 @@
 #endif
 
 #include <set>
+#include <unordered_set>
 #include <vector>
 #include <map>
+#include <unordered_map>
 #include <list>
 #include <algorithm>
 #include <string>
@@ -82,10 +84,6 @@
     void operator delete[](void*) { }                                 \
     void operator delete[](void *, void *) { }
 
-#define TBaseMap std::map
-#define TBaseList std::list
-#define TBaseSet std::set
-
 namespace glslang {
 
 //
@@ -93,10 +91,15 @@ namespace glslang {
 //
 typedef pool_allocator<char> TStringAllocator;
 typedef std::basic_string <char, std::char_traits<char>, TStringAllocator> TString;
+
+struct TStringHash {
+    size_t operator()(const TString& string) const { return std::hash<TString>()(string); }
+};
+
 inline TString* NewPoolTString(const char* s)
 {
-	void* memory = GetThreadPoolAllocator().allocate(sizeof(TString));
-	return new(memory) TString(s);
+    void* memory = GetThreadPoolAllocator().allocate(sizeof(TString));
+    return new(memory) TString(s);
 }
 
 template<class T> inline T* NewPoolObject(T)
@@ -123,28 +126,32 @@ public:
     TVector(size_type i, const T& val) : std::vector<T, pool_allocator<T> >(i, val) {}
 };
 
-template <class T> class TList   : public TBaseList  <T, pool_allocator<T> > {
+template <class T> class TList  : public std::list<T, pool_allocator<T> > {
 public:
-    typedef typename TBaseList<T, pool_allocator<T> >::size_type size_type;
-    TList() : TBaseList<T, pool_allocator<T> >() {}
-    TList(const pool_allocator<T>& a) : TBaseList<T, pool_allocator<T> >(a) {}
-    TList(size_type i): TBaseList<T, pool_allocator<T> >(i) {}
+    typedef typename std::list<T, pool_allocator<T> >::size_type size_type;
+    TList() : std::list<T, pool_allocator<T> >() {}
+    TList(const pool_allocator<T>& a) : std::list<T, pool_allocator<T> >(a) {}
+    TList(size_type i): std::list<T, pool_allocator<T> >(i) {}
 };
 
-// This is called TStlSet, because TSet is taken by an existing compiler class.
-template <class T, class CMP> class TStlSet : public std::set<T, CMP, pool_allocator<T> > {
-    // No pool allocator versions of constructors in std::set.
-};
+template <class K, class D, class CMP = std::less<K> > 
+class TMap : public std::map<K, D, CMP, pool_allocator<std::pair<K, D> > > {
+public:
+    typedef pool_allocator<std::pair <K, D> > tAllocator;
 
+    TMap() : std::map<K, D, CMP, tAllocator >() {}
+    // use correct two-stage name lookup supported in gcc 3.4 and above
+    TMap(const tAllocator& a) : std::map<K, D, CMP, tAllocator>(TBaseMap<K, D, CMP, tAllocator >::key_compare(), a) {}
+};
 
-template <class K, class D, class CMP = std::less<K> > 
-class TMap : public TBaseMap<K, D, CMP, pool_allocator<std::pair<K, D> > > {
+template <class K, class D, class HASH = std::hash<K>, class PRED = std::equal_to<K> > 
+class TUnorderedMap : public std::unordered_map<K, D, HASH, PRED, pool_allocator<std::pair<K, D> > > {
 public:
     typedef pool_allocator<std::pair <K, D> > tAllocator;
 
-    TMap() : TBaseMap<K, D, CMP, tAllocator >() {}
+    TUnorderedMap() : std::unordered_map<K, D, HASH, PRED, tAllocator >() {}
     // use correct two-stage name lookup supported in gcc 3.4 and above
-    TMap(const tAllocator& a) : TBaseMap<K, D, CMP, tAllocator>(TBaseMap<K, D, CMP, tAllocator >::key_compare(), a) {}
+    TUnorderedMap(const tAllocator& a) : std::unordered_map<K, D, HASH, PRED, tAllocator>(TBaseMap<K, D, HASH, PRED, tAllocator >::key_compare(), a) {}
 };
 
 //
@@ -184,7 +191,6 @@ struct TSourceLoc {
 };
 
 typedef TMap<TString, TString> TPragmaTable;
-typedef TMap<TString, TString>::tAllocator TPragmaTableAllocator;
 
 const int GlslangMaxTokenLength = 1024;
 

glslang/Include/Types.h

@@ -849,9 +849,6 @@ public:
     }
 };
 
-typedef std::map<TTypeList*, TTypeList*> TStructureMap;
-typedef std::map<TTypeList*, TTypeList*>::const_iterator TStructureMapIterator;
-
 //
 // Base class for things that have a type.
 //
@@ -957,7 +954,6 @@ public:
 
         if (copyOf.structure) {
             structure = new TTypeList;
-            TStructureMapIterator iter;
             for (unsigned int i = 0; i < copyOf.structure->size(); ++i) {
                 TTypeLoc typeLoc;
                 typeLoc.loc = (*copyOf.structure)[i].loc;

glslang/MachineIndependent/ParseHelper.h

@@ -48,10 +48,10 @@
 namespace glslang {
 
 struct TPragma {
-	TPragma(bool o, bool d) : optimize(o), debug(d) { }
-	bool optimize;
-	bool debug;
-	TPragmaTable pragmaTable;
+    TPragma(bool o, bool d) : optimize(o), debug(d) { }
+    bool optimize;
+    bool debug;
+    TPragmaTable pragmaTable;
 };
 
 class TScanContext;
@@ -306,7 +306,7 @@ protected:
     TInputScanner* currentScanner;
     int numErrors;               // number of compile-time errors encountered
     bool parsingBuiltins;        // true if parsing built-in symbols/functions
-    TMap<TString, TExtensionBehavior> extensionBehavior;    // for each extension string, what its current behavior is set to
+    std::unordered_map<TString, TExtensionBehavior> extensionBehavior;    // for each extension string, what its current behavior is set to
     static const int maxSamplerIndex = EsdNumDims * (EbtNumTypes * (2 * 2 * 2)); // see computeSamplerTypeIndex()
     TPrecisionQualifier defaultSamplerPrecision[maxSamplerIndex];
     bool afterEOF;

glslang/MachineIndependent/Scan.cpp

@@ -39,6 +39,8 @@
 //
 
 #include <string.h>
+#include <unordered_map>
+#include <unordered_set>
 
 #include "../Include/Types.h"
 #include "SymbolTable.h"
@@ -293,8 +295,8 @@ namespace {
 
 // A single global usable by all threads, by all versions, by all languages.
 // After a single process-level initialization, this is read only and thread safe
-std::map<std::string, int>* KeywordMap = 0;
-std::set<std::string>* ReservedSet = 0;
+std::unordered_map<std::string, int>* KeywordMap = 0;
+std::unordered_set<std::string>* ReservedSet = 0;
 
 };
 
@@ -307,7 +309,7 @@ void TScanContext::fillInKeywordMap()
         // but, the only risk is if two threads called simultaneously
         return;
     }
-    KeywordMap = new std::map<std::string, int>;
+    KeywordMap = new std::unordered_map<std::string, int>;
 
     (*KeywordMap)["const"] =                   CONST;
     (*KeywordMap)["uniform"] =                 UNIFORM;
@@ -476,7 +478,7 @@ void TScanContext::fillInKeywordMap()
     (*KeywordMap)["resource"] =                RESOURCE;
     (*KeywordMap)["superp"] =                  SUPERP;
 
-    ReservedSet = new std::set<std::string>;
+    ReservedSet = new std::unordered_set<std::string>;
     
     ReservedSet->insert("common");
     ReservedSet->insert("partition");
@@ -610,7 +612,7 @@ int TScanContext::tokenizeIdentifier()
     if (ReservedSet->find(tokenText) != ReservedSet->end())
         return reservedWord();
 
-    std::map<std::string, int>::const_iterator it = KeywordMap->find(tokenText);
+    auto it = KeywordMap->find(tokenText);
     if (it == KeywordMap->end()) {
         // Should have an identifier of some sort
         return identifierOrType();

glslang/MachineIndependent/Versions.cpp

@@ -441,7 +441,7 @@ void TParseContext::requireExtensions(TSourceLoc loc, int numExtensions, const c
 
 TExtensionBehavior TParseContext::getExtensionBehavior(const char* extension)
 {
-    TMap<TString, TExtensionBehavior>::iterator iter = extensionBehavior.find(TString(extension));
+    auto iter = extensionBehavior.find(TString(extension));
     if (iter == extensionBehavior.end())
         return EBhMissing;
     else
@@ -527,19 +527,18 @@ void TParseContext::updateExtensionBehavior(int line, const char* extension, con
 void TParseContext::updateExtensionBehavior(const char* extension, TExtensionBehavior behavior)
 {
     // Update the current behavior
-    TMap<TString, TExtensionBehavior>::iterator iter;
     if (strcmp(extension, "all") == 0) {
         // special case for the 'all' extension; apply it to every extension present
         if (behavior == EBhRequire || behavior == EBhEnable) {
             error(getCurrentLoc(), "extension 'all' cannot have 'require' or 'enable' behavior", "#extension", "");
             return;
         } else {
-            for (iter = extensionBehavior.begin(); iter != extensionBehavior.end(); ++iter)
+            for (auto iter = extensionBehavior.begin(); iter != extensionBehavior.end(); ++iter)
                 iter->second = behavior;
         }
     } else {
         // Do the update for this single extension
-        iter = extensionBehavior.find(TString(extension));
+        auto iter = extensionBehavior.find(TString(extension));
         if (iter == extensionBehavior.end()) {
             switch (behavior) {
             case EBhRequire:

glslang/MachineIndependent/intermOut.cpp

@@ -649,7 +649,7 @@ void TIntermediate::output(TInfoSink& infoSink, bool tree)
 {
     infoSink.debug << "Shader version: " << version << "\n";
     if (requestedExtensions.size() > 0) {
-        for (std::set<std::string>::const_iterator extIt = requestedExtensions.begin(); extIt != requestedExtensions.end(); ++extIt)
+        for (auto extIt = requestedExtensions.begin(); extIt != requestedExtensions.end(); ++extIt)
             infoSink.debug << "Requested " << *extIt << "\n";
     }
 

glslang/MachineIndependent/preprocessor/PpAtom.cpp

@@ -130,7 +130,7 @@ namespace glslang {
 //
 int TPpContext::LookUpAddString(const char* s)
 {
-    TAtomMap::const_iterator it = atomMap.find(s);
+    auto it = atomMap.find(s);
     if (it == atomMap.end())
         return AddAtomFixed(s, nextAtom++);
     else
@@ -161,7 +161,7 @@ const char* TPpContext::GetAtomString(int atom)
 //
 int TPpContext::AddAtomFixed(const char* s, int atom)
 {
-    TAtomMap::const_iterator it = atomMap.insert(std::pair<TString, int>(s, atom)).first;
+    auto it = atomMap.insert(std::pair<TString, int>(s, atom)).first;
     if (stringMap.size() < (size_t)atom + 1)
         stringMap.resize(atom + 100, 0);
     stringMap[atom] = &it->first;

glslang/MachineIndependent/preprocessor/PpContext.h

@@ -78,6 +78,8 @@ NVIDIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #ifndef PPCONTEXT_H
 #define PPCONTEXT_H
 
+#include <unordered_map>
+
 #include "../ParseHelper.h"
 
 #pragma warning(disable : 4127)
@@ -190,7 +192,7 @@ public:
     };
 
     MemoryPool *pool;
-    typedef std::map<int, Symbol*> TSymbolMap;
+    typedef TUnorderedMap<int, Symbol*> TSymbolMap;
     TSymbolMap symbols; // this has light use... just defined macros
 
 protected:
@@ -459,7 +461,7 @@ protected:
     //
     // From PpAtom.cpp
     //
-    typedef std::map<const TString, int> TAtomMap;
+    typedef TUnorderedMap<const TString, int, TStringHash> TAtomMap;
     typedef TVector<const TString*> TStringMap;
     TAtomMap atomMap;
     TStringMap stringMap;

glslang/MachineIndependent/reflection.h

@@ -108,7 +108,7 @@ public:
 protected:
     friend class glslang::TLiveTraverser;
 
-    typedef std::map<TString, int> TNameToIndex;
+    typedef std::unordered_map<TString, int> TNameToIndex;
     typedef std::vector<TObjectReflection> TMapIndexToReflection;
 
     TObjectReflection badReflection; // return for queries of -1 or generally out of range; has expected descriptions with in it for this