diff --git a/3rdparty/glslang/SPIRV/GlslangToSpv.cpp b/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
index 8eb06b7cd..6d9c7cfec 100644
--- a/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
+++ b/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
@@ -2784,6 +2784,7 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
         break;
 
     case glslang::EOpAtomicAdd:
+    case glslang::EOpAtomicSubtract:
     case glslang::EOpAtomicMin:
     case glslang::EOpAtomicMax:
     case glslang::EOpAtomicAnd:
@@ -2955,6 +2956,7 @@ bool TGlslangToSpvTraverser::visitAggregate(glslang::TVisit visit, glslang::TInt
             break;
 
         case glslang::EOpAtomicAdd:
+        case glslang::EOpAtomicSubtract:
         case glslang::EOpAtomicMin:
         case glslang::EOpAtomicMax:
         case glslang::EOpAtomicAnd:
@@ -6894,6 +6896,7 @@ spv::Id TGlslangToSpvTraverser::createAtomicOperation(glslang::TOperator op, spv
                 builder.addCapability(spv::CapabilityAtomicFloat64AddEXT);
         }
         break;
+    case glslang::EOpAtomicSubtract:
     case glslang::EOpAtomicCounterSubtract:
         opCode = spv::OpAtomicISub;
         break;
diff --git a/3rdparty/glslang/StandAlone/StandAlone.cpp b/3rdparty/glslang/StandAlone/StandAlone.cpp
index fdbf02799..923ded305 100644
--- a/3rdparty/glslang/StandAlone/StandAlone.cpp
+++ b/3rdparty/glslang/StandAlone/StandAlone.cpp
@@ -110,6 +110,7 @@ bool SpvToolsValidate = false;
 bool NaNClamp = false;
 bool stripDebugInfo = false;
 bool beQuiet = false;
+bool VulkanRulesRelaxed = false;
 
 //
 // Return codes from main/exit().
@@ -195,6 +196,17 @@ std::array<std::array<unsigned int, EShLangCount>, glslang::EResCount> baseBindi
 std::array<std::array<TPerSetBaseBinding, EShLangCount>, glslang::EResCount> baseBindingForSet;
 std::array<std::vector<std::string>, EShLangCount> baseResourceSetBinding;
 
+std::vector<std::pair<std::string, glslang::TBlockStorageClass>> blockStorageOverrides;
+
+bool setGlobalUniformBlock = false;
+std::string globalUniformName;
+unsigned int globalUniformBinding;
+unsigned int globalUniformSet;
+
+bool setGlobalBufferBlock = false;
+std::string atomicCounterBlockName;
+unsigned int atomicCounterBlockSet;
+
 // Add things like "#define ..." to a preamble to use in the beginning of the shader.
 class TPreamble {
 public:
@@ -396,6 +408,115 @@ void ProcessResourceSetBindingBase(int& argc, char**& argv, std::array<std::vect
     }
 }
 
+//
+// Process an optional binding base of one the forms:
+//   --argname name {uniform|buffer|push_constant}
+void ProcessBlockStorage(int& argc, char**& argv, std::vector<std::pair<std::string, glslang::TBlockStorageClass>>& storage)
+{
+    if (argc < 3)
+        usage();
+
+    glslang::TBlockStorageClass blockStorage = glslang::EbsNone;
+
+    std::string strBacking(argv[2]);
+    if (strBacking == "uniform")
+        blockStorage = glslang::EbsUniform;
+    else if (strBacking == "buffer")
+        blockStorage = glslang::EbsStorageBuffer;
+    else if (strBacking == "push_constant")
+        blockStorage = glslang::EbsPushConstant;
+    else {
+        printf("%s: invalid block storage\n", strBacking.c_str());
+        usage();
+    }
+
+    storage.push_back(std::make_pair(std::string(argv[1]), blockStorage));
+
+    argc -= 2;
+    argv += 2;
+}
+
+inline bool isNonDigit(char c) {
+    // a non-digit character valid in a glsl identifier
+    return (c == '_') || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z');
+}
+
+// whether string isa  valid identifier to be used in glsl
+bool isValidIdentifier(const char* str) {
+    std::string idn(str);
+
+    if (idn.length() == 0) {
+        return false;
+    }
+
+    if (idn.length() >= 3 && idn.substr(0, 3) == "gl_") {
+        // identifiers startin with "gl_" are reserved
+        return false;
+    }
+
+    if (!isNonDigit(idn[0])) {
+        return false;
+    }
+
+    for (unsigned int i = 1; i < idn.length(); ++i) {
+        if (!(isdigit(idn[i]) || isNonDigit(idn[i]))) {
+            return false;
+        }
+    }
+
+    return true;
+}
+
+// Process settings for either the global buffer block or global unfirom block
+// of the form:
+//      --argname name set binding
+void ProcessGlobalBlockSettings(int& argc, char**& argv, std::string* name, unsigned int* set, unsigned int* binding)
+{
+    if (argc < 4)
+        usage();
+
+    unsigned int curArg = 1;
+
+    assert(name || set || binding);
+
+    if (name) {
+        if (!isValidIdentifier(argv[curArg])) {
+            printf("%s: invalid identifier\n", argv[curArg]);
+            usage();
+        }
+        *name = argv[curArg];
+
+        curArg++;
+    }
+
+    if (set) {
+        errno = 0;
+        int setVal = ::strtol(argv[curArg], NULL, 10);
+        if (errno || setVal < 0) {
+            printf("%s: invalid set\n", argv[curArg]);
+            usage();
+        }
+        *set = setVal;
+
+        curArg++;
+    }
+
+    if (binding) {
+        errno = 0;
+        int bindingVal = ::strtol(argv[curArg], NULL, 10);
+        if (errno || bindingVal < 0) {
+            printf("%s: invalid binding\n", argv[curArg]);
+            usage();
+        }
+        *binding = bindingVal;
+
+        curArg++;
+    }
+
+    argc -= (curArg - 1);
+    argv += (curArg - 1);
+}
+
 //
 // Do all command-line argument parsing.  This includes building up the work-items
 // to be processed later, and saving all the command-line options.
@@ -569,6 +690,17 @@ void ProcessArguments(std::vector<std::unique_ptr<glslang::TWorkItem>>& workItem
                                lowerword == "resource-set-binding"  ||
                                lowerword == "rsb") {
                         ProcessResourceSetBindingBase(argc, argv, baseResourceSetBinding);
+                    } else if (lowerword == "set-block-storage" ||
+                               lowerword == "sbs") {
+                        ProcessBlockStorage(argc, argv, blockStorageOverrides);
+                    } else if (lowerword == "set-atomic-counter-block" ||
+                               lowerword == "sacb") {
+                        ProcessGlobalBlockSettings(argc, argv, &atomicCounterBlockName, &atomicCounterBlockSet, nullptr);
+                        setGlobalBufferBlock = true;
+                    } else if (lowerword == "set-default-uniform-block" ||
+                               lowerword == "sdub") {
+                        ProcessGlobalBlockSettings(argc, argv, &globalUniformName, &globalUniformSet, &globalUniformBinding);
+                        setGlobalUniformBlock = true;
                     } else if (lowerword == "shift-image-bindings" ||  // synonyms
                                lowerword == "shift-image-binding"  ||
                                lowerword == "sib") {
@@ -721,6 +853,9 @@ void ProcessArguments(std::vector<std::unique_ptr<glslang::TWorkItem>>& workItem
                 else
                     Error("unknown -O option");
                 break;
+            case 'R':
+                VulkanRulesRelaxed = true;
+                break;
             case 'S':
                 if (argc <= 1)
                     Error("no <stage> specified for -S");
@@ -1068,6 +1203,24 @@ void CompileAndLinkShaderUnits(std::vector<ShaderCompUnit> compUnits)
         shader->setUniformLocationBase(uniformBase);
 #endif
 
+        if (VulkanRulesRelaxed) {
+            for (auto& storageOverride : blockStorageOverrides) {
+                shader->addBlockStorageOverride(storageOverride.first.c_str(),
+                    storageOverride.second);
+            }
+
+            if (setGlobalBufferBlock) {
+                shader->setAtomicCounterBlockName(atomicCounterBlockName.c_str());
+                shader->setAtomicCounterBlockSet(atomicCounterBlockSet);
+            }
+
+            if (setGlobalUniformBlock) {
+                shader->setGlobalUniformBlockName(globalUniformName.c_str());
+                shader->setGlobalUniformSet(globalUniformSet);
+                shader->setGlobalUniformBinding(globalUniformBinding);
+            }
+        }
+
         shader->setNanMinMaxClamp(NaNClamp);
 
 #ifdef ENABLE_HLSL
@@ -1091,6 +1244,8 @@ void CompileAndLinkShaderUnits(std::vector<ShaderCompUnit> compUnits)
             if (targetHlslFunctionality1)
                 shader->setEnvTargetHlslFunctionality1();
 #endif
+            if (VulkanRulesRelaxed)
+                shader->setEnvInputVulkanRulesRelaxed();
         }
 
         shaders.push_back(shader);
@@ -1572,6 +1727,9 @@ void usage()
            "              is searched first, followed by left-to-right order of -I\n"
            "  -Od         disables optimization; may cause illegal SPIR-V for HLSL\n"
            "  -Os         optimizes SPIR-V to minimize size\n"
+           "  -R          use relaxed verification rules for generating Vulkan SPIR-V,\n"
+           "              allowing the use of default uniforms, atomic_uints, and\n"
+           "              gl_VertexID and gl_InstanceID keywords.\n"
            "  -S <stage>  uses specified stage rather than parsing the file extension\n"
            "              choices for <stage> are vert, tesc, tese, geom, frag, or comp\n"
            "  -U<name> | --undef-macro <name> | --U <name>\n"
@@ -1649,6 +1807,22 @@ void usage()
            "  --resource-set-binding [stage] set\n"
            "                                    set descriptor set for all resources\n"
            "  --rsb                             synonym for --resource-set-binding\n"
+           "  --set-block-backing name {uniform|buffer|push_constant}\n"
+           "                                    changes the backing type of a uniform, buffer,\n"
+           "                                    or push_constant block declared in\n"
+           "                                    in the program, when using -R option.\n"
+           "                                    This can be used to change the backing\n"
+           "                                    for existing blocks as well as implicit ones\n"
+           "                                    such as 'gl_DefaultUniformBlock'.\n"
+           "  --sbs                             synonym for set-block-storage\n"
+           "  --set-atomic-counter-block name set\n"
+           "                                    set name, and descriptor set for\n"
+           "                                    atomic counter blocks, with -R opt\n"
+           "  --sacb                            synonym for set-atomic-counter-block\n"
+           "  --set-default-uniform-block name set binding\n"
+           "                                    set name, descriptor set, and binding for\n"
+           "                                    global default-uniform-block, with -R opt\n"
+           "  --sdub                            synonym for set-default-uniform-block\n"
            "  --shift-image-binding [stage] num\n"
            "                                    base binding number for images (uav)\n"
            "  --shift-image-binding [stage] [num set]...\n"
diff --git a/3rdparty/glslang/glslang/Include/Types.h b/3rdparty/glslang/glslang/Include/Types.h
index 603203db9..04250843a 100644
--- a/3rdparty/glslang/glslang/Include/Types.h
+++ b/3rdparty/glslang/glslang/Include/Types.h
@@ -501,6 +501,7 @@ public:
         noContraction = false;
         nullInit = false;
 #endif
+        defaultBlock = false;
     }
 
     // drop qualifiers that don't belong in a temporary variable
@@ -514,6 +515,7 @@ public:
         specConstant = false;
         nonUniform = false;
         nullInit = false;
+        defaultBlock = false;
         clearLayout();
     }
 
@@ -572,6 +574,7 @@ public:
     bool specConstant : 1;
     bool nonUniform   : 1;
     bool explicitOffset   : 1;
+    bool defaultBlock : 1; // default blocks with matching names have structures merged when linking
 
 #ifdef GLSLANG_WEB
     bool isWriteOnly() const { return false; }
@@ -756,6 +759,46 @@ public:
         }
     }
 
+    TBlockStorageClass getBlockStorage() const {
+        if (storage == EvqUniform && !isPushConstant()) {
+            return EbsUniform;
+        }
+        else if (storage == EvqUniform) {
+            return EbsPushConstant;
+        }
+        else if (storage == EvqBuffer) {
+            return EbsStorageBuffer;
+        }
+        return EbsNone;
+    }
+
+    void setBlockStorage(TBlockStorageClass newBacking) {
+#ifndef GLSLANG_WEB
+        layoutPushConstant = (newBacking == EbsPushConstant);
+#endif
+        switch (newBacking) {
+        case EbsUniform :
+            if (layoutPacking == ElpStd430) {
+                // std430 would not be valid
+                layoutPacking = ElpStd140;
+            }
+            storage = EvqUniform;
+            break;
+        case EbsStorageBuffer : 
+            storage = EvqBuffer;
+            break;
+#ifndef GLSLANG_WEB
+        case EbsPushConstant :
+            storage = EvqUniform;
+            layoutSet = TQualifier::layoutSetEnd;
+            layoutBinding = TQualifier::layoutBindingEnd;
+            break;
+#endif
+        default:
+            break;
+        }
+    }
+
 #ifdef GLSLANG_WEB
     bool isPerView() const { return false; }
     bool isTaskMemory() const { return false; }
@@ -852,6 +895,7 @@ public:
         return hasNonXfbLayout() ||
                hasXfb();
     }
+
     TLayoutMatrix  layoutMatrix  : 3;
     TLayoutPacking layoutPacking : 4;
     int layoutOffset;
@@ -2292,6 +2336,17 @@ public:
         name += ';' ;
     }
 
+    // These variables are inconsistently declared inside and outside of gl_PerVertex in glslang right now.
+    // They are declared inside of 'in gl_PerVertex', but sitting as standalone when they are 'out'puts.
+    bool isInconsistentGLPerVertexMember(const TString& name) const
+    {
+        if (name == "gl_SecondaryPositionNV" ||
+            name == "gl_PositionPerViewNV")
+            return true;
+        return false;
+    }
+
+
     // Do two structure types match?  They could be declared independently,
     // in different places, but still might satisfy the definition of matching.
     // From the spec:
@@ -2307,22 +2362,48 @@ public:
             (isStruct() && right.isStruct() && structure == right.structure))
             return true;
 
-        // Both being nullptr was caught above, now they both have to be structures of the same number of elements
-        if (!isStruct() || !right.isStruct() ||
-            structure->size() != right.structure->size())
-            return false;
-
         // Structure names have to match
         if (*typeName != *right.typeName)
             return false;
 
-        // Compare the names and types of all the members, which have to match
-        for (unsigned int i = 0; i < structure->size(); ++i) {
-            if ((*structure)[i].type->getFieldName() != (*right.structure)[i].type->getFieldName())
-                return false;
+        // There are inconsistencies with how gl_PerVertex is setup. For now ignore those as errors if they
+        // are known inconsistencies.
+        bool isGLPerVertex = *typeName == "gl_PerVertex";
 
-            if (*(*structure)[i].type != *(*right.structure)[i].type)
-                return false;
+        // Both being nullptr was caught above, now they both have to be structures of the same number of elements
+        if (!isStruct() || !right.isStruct() ||
+            (structure->size() != right.structure->size() && !isGLPerVertex))
+            return false;
+
+        // Compare the names and types of all the members, which have to match
+        for (int li = 0, ri = 0; li < structure->size() || ri < right.structure->size(); ++li, ++ri) {
+            if (li < structure->size() && ri < right.structure->size()) {
+                if ((*structure)[li].type->getFieldName() == (*right.structure)[ri].type->getFieldName()) {
+                    if (*(*structure)[li].type != *(*right.structure)[ri].type)
+                        return false;
+                } else {
+                    // If one of the members is something that's inconsistently declared, skip over it
+                    // for now.
+                    if (isGLPerVertex) {
+                        if (isInconsistentGLPerVertexMember((*structure)[li].type->getFieldName())) {
+                            ri--;
+                            continue;
+                        } else if (isInconsistentGLPerVertexMember((*right.structure)[ri].type->getFieldName())) {
+                            li--;
+                            continue;
+                        }
+                    } else {
+                        return false;
+                    }
+                }
+            // If we get here, then there should only be inconsistently declared members left
+            } else if (li < structure->size()) {
+                if (!isInconsistentGLPerVertexMember((*structure)[li].type->getFieldName()))
+                    return false;
+            } else {
+                if (!isInconsistentGLPerVertexMember((*right.structure)[ri].type->getFieldName()))
+                    return false;
+            }
         }
 
         return true;
diff --git a/3rdparty/glslang/glslang/Include/intermediate.h b/3rdparty/glslang/glslang/Include/intermediate.h
index 85edd6358..1440c205e 100644
--- a/3rdparty/glslang/glslang/Include/intermediate.h
+++ b/3rdparty/glslang/glslang/Include/intermediate.h
@@ -593,6 +593,7 @@ enum TOperator {
     EOpTime,
 
     EOpAtomicAdd,
+    EOpAtomicSubtract,
     EOpAtomicMin,
     EOpAtomicMax,
     EOpAtomicAnd,
@@ -1135,6 +1136,8 @@ public:
     virtual TBasicType getBasicType() const { return type.getBasicType(); }
     virtual TQualifier& getQualifier() { return type.getQualifier(); }
     virtual const TQualifier& getQualifier() const { return type.getQualifier(); }
+    virtual TArraySizes* getArraySizes() { return type.getArraySizes(); }
+    virtual const TArraySizes* getArraySizes() const { return type.getArraySizes(); }
     virtual void propagatePrecision(TPrecisionQualifier);
     virtual int getVectorSize() const { return type.getVectorSize(); }
     virtual int getMatrixCols() const { return type.getMatrixCols(); }
diff --git a/3rdparty/glslang/glslang/MachineIndependent/Initialize.cpp b/3rdparty/glslang/glslang/MachineIndependent/Initialize.cpp
index c94a8a4b5..a9e5af456 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/Initialize.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/Initialize.cpp
@@ -1627,6 +1627,36 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
                 "uint atomicCounterExchange(atomic_uint, uint);"
                 "uint atomicCounterCompSwap(atomic_uint, uint, uint);"
 
+                "\n");
+        }
+    }
+    else if (spvVersion.vulkanRelaxed) {
+        //
+        // Atomic counter functions act as aliases to normal atomic functions.
+        // replace definitions to take 'volatile coherent uint' instead of 'atomic_uint'
+        // and map to equivalent non-counter atomic op
+        //
+        if ((profile != EEsProfile && version >= 300) ||
+            (profile == EEsProfile && version >= 310)) {
+            commonBuiltins.append(
+                "uint atomicCounterIncrement(volatile coherent uint);"
+                "uint atomicCounterDecrement(volatile coherent uint);"
+                "uint atomicCounter(volatile coherent uint);"
+
+                "\n");
+        }
+        if (profile != EEsProfile && version >= 460) {
+            commonBuiltins.append(
+                "uint atomicCounterAdd(volatile coherent uint, uint);"
+                "uint atomicCounterSubtract(volatile coherent uint, uint);"
+                "uint atomicCounterMin(volatile coherent uint, uint);"
+                "uint atomicCounterMax(volatile coherent uint, uint);"
+                "uint atomicCounterAnd(volatile coherent uint, uint);"
+                "uint atomicCounterOr(volatile coherent uint, uint);"
+                "uint atomicCounterXor(volatile coherent uint, uint);"
+                "uint atomicCounterExchange(volatile coherent uint, uint);"
+                "uint atomicCounterCompSwap(volatile coherent uint, uint, uint);"
+
                 "\n");
         }
     }
@@ -4124,7 +4154,7 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
     }
 #ifndef GLSLANG_WEB
     if ((profile != EEsProfile && version >= 420) || esBarrier) {
-        if (spvVersion.vulkan == 0) {
+        if (spvVersion.vulkan == 0 || spvVersion.vulkanRelaxed) {
             commonBuiltins.append("void memoryBarrierAtomicCounter();");
         }
         commonBuiltins.append("void memoryBarrierImage();");
@@ -4848,6 +4878,13 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
                 "in int gl_VertexIndex;"
                 "in int gl_InstanceIndex;"
                 );
+
+        if (spvVersion.vulkan > 0 && version >= 140 && spvVersion.vulkanRelaxed)
+            stageBuiltins[EShLangVertex].append(
+                "in int gl_VertexID;"         // declare with 'in' qualifier
+                "in int gl_InstanceID;"
+                );
+
         if (version >= 440) {
             stageBuiltins[EShLangVertex].append(
                 "in int gl_BaseVertexARB;"
@@ -4885,7 +4922,7 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
                 "mediump float gl_PointSize;" // needs qualifier fixed later
                 );
         } else {
-            if (spvVersion.vulkan == 0)
+            if (spvVersion.vulkan == 0 || spvVersion.vulkanRelaxed)
                 stageBuiltins[EShLangVertex].append(
                     "in highp int gl_VertexID;"      // needs qualifier fixed later
                     "in highp int gl_InstanceID;"    // needs qualifier fixed later
@@ -7436,6 +7473,12 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
             SpecialQualifier("gl_InstanceID", EvqInstanceId, EbvInstanceId, symbolTable);
         }
 
+        if (spvVersion.vulkan > 0 && spvVersion.vulkanRelaxed) {
+            // treat these built-ins as aliases of VertexIndex and InstanceIndex
+            BuiltInVariable("gl_VertexID", EbvVertexIndex, symbolTable);
+            BuiltInVariable("gl_InstanceID", EbvInstanceIndex, symbolTable);
+        }
+
         if (profile != EEsProfile) {
             if (version >= 440) {
                 symbolTable.setVariableExtensions("gl_BaseVertexARB",   1, &E_GL_ARB_shader_draw_parameters);
@@ -8912,6 +8955,14 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
     symbolTable.relateToOperator("memoryBarrierAtomicCounter", EOpMemoryBarrierAtomicCounter);
     symbolTable.relateToOperator("memoryBarrierImage",         EOpMemoryBarrierImage);
 
+    if (spvVersion.vulkanRelaxed) {
+        //
+        // functions signature have been replaced to take uint operations on buffer variables
+        // remap atomic counter functions to atomic operations
+        //
+        symbolTable.relateToOperator("memoryBarrierAtomicCounter", EOpMemoryBarrierBuffer);
+    }
+
     symbolTable.relateToOperator("atomicLoad",     EOpAtomicLoad);
     symbolTable.relateToOperator("atomicStore",    EOpAtomicStore);
 
@@ -8919,6 +8970,20 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
     symbolTable.relateToOperator("atomicCounterDecrement", EOpAtomicCounterDecrement);
     symbolTable.relateToOperator("atomicCounter",          EOpAtomicCounter);
 
+    if (spvVersion.vulkanRelaxed) {
+        //
+        // functions signature have been replaced to take uint operations
+        // remap atomic counter functions to atomic operations
+        //
+        // these atomic counter functions do not match signatures of glsl
+        // atomic functions, so they will be remapped to semantically
+        // equivalent functions in the parser
+        //
+        symbolTable.relateToOperator("atomicCounterIncrement", EOpNull);
+        symbolTable.relateToOperator("atomicCounterDecrement", EOpNull);
+        symbolTable.relateToOperator("atomicCounter", EOpNull);
+    }
+
     symbolTable.relateToOperator("clockARB",     EOpReadClockSubgroupKHR);
     symbolTable.relateToOperator("clock2x32ARB", EOpReadClockSubgroupKHR);
 
@@ -8937,6 +9002,23 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
         symbolTable.relateToOperator("atomicCounterCompSwap", EOpAtomicCounterCompSwap);
     }
 
+    if (spvVersion.vulkanRelaxed) {
+        //
+        // functions signature have been replaced to take 'uint' instead of 'atomic_uint'
+        // remap atomic counter functions to non-counter atomic ops so
+        // functions act as aliases to non-counter atomic ops
+        //
+        symbolTable.relateToOperator("atomicCounterAdd", EOpAtomicAdd);
+        symbolTable.relateToOperator("atomicCounterSubtract", EOpAtomicSubtract);
+        symbolTable.relateToOperator("atomicCounterMin", EOpAtomicMin);
+        symbolTable.relateToOperator("atomicCounterMax", EOpAtomicMax);
+        symbolTable.relateToOperator("atomicCounterAnd", EOpAtomicAnd);
+        symbolTable.relateToOperator("atomicCounterOr", EOpAtomicOr);
+        symbolTable.relateToOperator("atomicCounterXor", EOpAtomicXor);
+        symbolTable.relateToOperator("atomicCounterExchange", EOpAtomicExchange);
+        symbolTable.relateToOperator("atomicCounterCompSwap", EOpAtomicCompSwap);
+    }
+
     symbolTable.relateToOperator("fma",               EOpFma);
     symbolTable.relateToOperator("frexp",             EOpFrexp);
     symbolTable.relateToOperator("ldexp",             EOpLdexp);
diff --git a/3rdparty/glslang/glslang/MachineIndependent/ParseContextBase.cpp b/3rdparty/glslang/glslang/MachineIndependent/ParseContextBase.cpp
index 3efa27aca..02cca409e 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/ParseContextBase.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/ParseContextBase.cpp
@@ -601,7 +601,6 @@ void TParseContextBase::parseSwizzleSelector(const TSourceLoc& loc, const TStrin
         selector.push_back(0);
 }
 
-#ifdef ENABLE_HLSL
 //
 // Make the passed-in variable information become a member of the
 // global uniform block.  If this doesn't exist yet, make it.
@@ -646,7 +645,67 @@ void TParseContextBase::growGlobalUniformBlock(const TSourceLoc& loc, TType& mem
 
     ++firstNewMember;
 }
-#endif
+
+void TParseContextBase::growAtomicCounterBlock(int binding, const TSourceLoc& loc, TType& memberType, const TString& memberName, TTypeList* typeList) {
+    // Make the atomic counter block, if not yet made.
+    const auto &at  = atomicCounterBuffers.find(binding);
+    if (at == atomicCounterBuffers.end()) {
+        atomicCounterBuffers.insert({binding, (TVariable*)nullptr });
+        atomicCounterBlockFirstNewMember.insert({binding, 0});
+    }
+
+    TVariable*& atomicCounterBuffer = atomicCounterBuffers[binding];
+    int& bufferNewMember = atomicCounterBlockFirstNewMember[binding];
+
+    if (atomicCounterBuffer == nullptr) {
+        TQualifier blockQualifier;
+        blockQualifier.clear();
+        blockQualifier.storage = EvqBuffer;
+        
+        char charBuffer[512];
+        if (binding != TQualifier::layoutBindingEnd) {
+            snprintf(charBuffer, 512, "%s_%d", getAtomicCounterBlockName(), binding);
+        } else {
+            snprintf(charBuffer, 512, "%s_0", getAtomicCounterBlockName());
+        }
+        
+        TType blockType(new TTypeList, *NewPoolTString(charBuffer), blockQualifier);
+        setUniformBlockDefaults(blockType);
+        blockType.getQualifier().layoutPacking = ElpStd430;
+        atomicCounterBuffer = new TVariable(NewPoolTString(""), blockType, true);
+        // If we arn't auto mapping bindings then set the block to use the same
+        // binding as what the atomic was set to use
+        if (!intermediate.getAutoMapBindings()) {
+            atomicCounterBuffer->getWritableType().getQualifier().layoutBinding = binding;
+        }
+        bufferNewMember = 0;
+
+        atomicCounterBuffer->getWritableType().getQualifier().layoutSet = atomicCounterBlockSet;
+    }
+
+    // Add the requested member as a member to the global block.
+    TType* type = new TType;
+    type->shallowCopy(memberType);
+    type->setFieldName(memberName);
+    if (typeList)
+        type->setStruct(typeList);
+    TTypeLoc typeLoc = {type, loc};
+    atomicCounterBuffer->getType().getWritableStruct()->push_back(typeLoc);
+
+    // Insert into the symbol table.
+    if (bufferNewMember == 0) {
+        // This is the first request; we need a normal symbol table insert
+        if (symbolTable.insert(*atomicCounterBuffer))
+            trackLinkage(*atomicCounterBuffer);
+        else
+            error(loc, "failed to insert the global constant buffer", "buffer", "");
+    } else {
+        // This is a follow-on request; we need to amend the first insert
+        symbolTable.amend(*atomicCounterBuffer, bufferNewMember);
+    }
+
+    ++bufferNewMember;
+}
 
 void TParseContextBase::finish()
 {
diff --git a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
index 12d68a3e6..550fce679 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
@@ -225,6 +225,108 @@ void TParseContext::parserError(const char* s)
         error(getCurrentLoc(), "compilation terminated", "", "");
 }
 
+void TParseContext::growGlobalUniformBlock(const TSourceLoc& loc, TType& memberType, const TString& memberName, TTypeList* typeList)
+{
+    bool createBlock = globalUniformBlock == nullptr;
+
+    if (createBlock) {
+        globalUniformBinding = intermediate.getGlobalUniformBinding();
+        globalUniformSet = intermediate.getGlobalUniformSet();
+    }
+
+    // use base class function to create/expand block
+    TParseContextBase::growGlobalUniformBlock(loc, memberType, memberName, typeList);
+
+    if (spvVersion.vulkan > 0 && spvVersion.vulkanRelaxed) {
+        // check for a block storage override
+        TBlockStorageClass storageOverride = intermediate.getBlockStorageOverride(getGlobalUniformBlockName());
+        TQualifier& qualifier = globalUniformBlock->getWritableType().getQualifier();
+        qualifier.defaultBlock = true;
+
+        if (storageOverride != EbsNone) {
+            if (createBlock) {
+                // Remap block storage
+                qualifier.setBlockStorage(storageOverride);
+
+                // check that the change didn't create errors
+                blockQualifierCheck(loc, qualifier, false);
+            }
+
+            // remap meber storage as well
+            memberType.getQualifier().setBlockStorage(storageOverride);
+        }
+    }
+}
+
+void TParseContext::growAtomicCounterBlock(int binding, const TSourceLoc& loc, TType& memberType, const TString& memberName, TTypeList* typeList)
+{
+    bool createBlock = atomicCounterBuffers.find(binding) == atomicCounterBuffers.end();
+
+    if (createBlock) {
+        atomicCounterBlockSet = intermediate.getAtomicCounterBlockSet();
+    }
+
+    // use base class function to create/expand block
+    TParseContextBase::growAtomicCounterBlock(binding, loc, memberType, memberName, typeList);
+    TQualifier& qualifier = atomicCounterBuffers[binding]->getWritableType().getQualifier();
+    qualifier.defaultBlock = true;
+
+    if (spvVersion.vulkan > 0 && spvVersion.vulkanRelaxed) {
+        // check for a Block storage override
+        TBlockStorageClass storageOverride = intermediate.getBlockStorageOverride(getAtomicCounterBlockName());
+
+        if (storageOverride != EbsNone) {
+            if (createBlock) {
+                // Remap block storage
+
+                qualifier.setBlockStorage(storageOverride);
+
+                // check that the change didn't create errors
+                blockQualifierCheck(loc, qualifier, false);
+            }
+
+            // remap meber storage as well
+            memberType.getQualifier().setBlockStorage(storageOverride);
+        }
+    }
+}
+
+const char* TParseContext::getGlobalUniformBlockName() const
+{
+    const char* name = intermediate.getGlobalUniformBlockName();
+    if (std::string(name) == "")
+        return "gl_DefaultUniformBlock";
+    else
+        return name;
+}
+void TParseContext::finalizeGlobalUniformBlockLayout(TVariable&)
+{
+}
+void TParseContext::setUniformBlockDefaults(TType& block) const
+{
+    block.getQualifier().layoutPacking = ElpStd140;
+    block.getQualifier().layoutMatrix = ElmColumnMajor;
+}
+
+
+const char* TParseContext::getAtomicCounterBlockName() const
+{
+    const char* name = intermediate.getAtomicCounterBlockName();
+    if (std::string(name) == "")
+        return "gl_AtomicCounterBlock";
+    else
+        return name;
+}
+void TParseContext::finalizeAtomicCounterBlockLayout(TVariable&)
+{
+}
+
+void TParseContext::setAtomicCounterBlockDefaults(TType& block) const
+{
+    block.getQualifier().layoutPacking = ElpStd430;
+    block.getQualifier().layoutMatrix = ElmRowMajor;
+}
+
 void TParseContext::handlePragma(const TSourceLoc& loc, const TVector<TString>& tokens)
 {
 #ifndef GLSLANG_WEB
@@ -1135,6 +1237,14 @@ TIntermTyped* TParseContext::handleFunctionCall(const TSourceLoc& loc, TFunction
 {
     TIntermTyped* result = nullptr;
 
+    if (spvVersion.vulkan != 0 && spvVersion.vulkanRelaxed) {
+        // allow calls that are invalid in Vulkan Semantics to be invisibily
+        // remapped to equivalent valid functions
+        result = vkRelaxedRemapFunctionCall(loc, function, arguments);
+        if (result)
+            return result;
+    }
+
     if (function->getBuiltInOp() == EOpArrayLength)
         result = handleLengthMethod(loc, function, arguments);
     else if (function->getBuiltInOp() != EOpNull) {
@@ -1727,6 +1837,7 @@ void TParseContext::memorySemanticsCheck(const TSourceLoc& loc, const TFunction&
     // Grab the semantics and storage class semantics from the operands, based on opcode
     switch (callNode.getOp()) {
     case EOpAtomicAdd:
+    case EOpAtomicSubtract:
     case EOpAtomicMin:
     case EOpAtomicMax:
     case EOpAtomicAnd:
@@ -2176,6 +2287,7 @@ void TParseContext::builtInOpCheck(const TSourceLoc& loc, const TFunction& fnCan
     }
 
     case EOpAtomicAdd:
+    case EOpAtomicSubtract:
     case EOpAtomicMin:
     case EOpAtomicMax:
     case EOpAtomicAnd:
@@ -3388,7 +3500,7 @@ void TParseContext::transparentOpaqueCheck(const TSourceLoc& loc, const TType& t
 
     if (type.containsNonOpaque()) {
         // Vulkan doesn't allow transparent uniforms outside of blocks
-        if (spvVersion.vulkan > 0)
+        if (spvVersion.vulkan > 0 && !spvVersion.vulkanRelaxed)
             vulkanRemoved(loc, "non-opaque uniforms outside a block");
         // OpenGL wants locations on these (unless they are getting automapped)
         if (spvVersion.openGl > 0 && !type.getQualifier().hasLocation() && !intermediate.getAutoMapLocations())
@@ -5019,14 +5131,22 @@ void TParseContext::setLayoutQualifier(const TSourceLoc& loc, TPublicType& publi
         return;
     }
     if (id == TQualifier::getLayoutPackingString(ElpPacked)) {
-        if (spvVersion.spv != 0)
-            spvRemoved(loc, "packed");
+        if (spvVersion.spv != 0) {
+            if (spvVersion.vulkanRelaxed)
+                return; // silently ignore qualifier
+            else
+                spvRemoved(loc, "packed");
+        }
         publicType.qualifier.layoutPacking = ElpPacked;
         return;
     }
     if (id == TQualifier::getLayoutPackingString(ElpShared)) {
-        if (spvVersion.spv != 0)
-            spvRemoved(loc, "shared");
+        if (spvVersion.spv != 0) {
+            if (spvVersion.vulkanRelaxed)
+                return; // silently ignore qualifier
+            else
+                spvRemoved(loc, "shared");
+        }
         publicType.qualifier.layoutPacking = ElpShared;
         return;
     }
@@ -5928,7 +6048,7 @@ void TParseContext::layoutTypeCheck(const TSourceLoc& loc, const TType& type)
                 error(loc, "sampler binding not less than gl_MaxCombinedTextureImageUnits", "binding", type.isArray() ? "(using array)" : "");
 #endif
         }
-        if (type.isAtomic()) {
+        if (type.isAtomic() && !spvVersion.vulkanRelaxed) {
             if (qualifier.layoutBinding >= (unsigned int)resources.maxAtomicCounterBindings) {
                 error(loc, "atomic_uint binding is too large; see gl_MaxAtomicCounterBindings", "binding", "");
                 return;
@@ -6598,6 +6718,68 @@ const TFunction* TParseContext::findFunctionExplicitTypes(const TSourceLoc& loc,
     return bestMatch;
 }
 
+//
+// Adjust function calls that aren't declared in Vulkan to a
+// calls with equivalent effects
+//
+TIntermTyped* TParseContext::vkRelaxedRemapFunctionCall(const TSourceLoc& loc, TFunction* function, TIntermNode* arguments)
+{
+    TIntermTyped* result = nullptr;
+
+#ifndef GLSLANG_WEB
+    if (function->getBuiltInOp() != EOpNull) {
+        return nullptr;
+    }
+
+    if (function->getName() == "atomicCounterIncrement") {
+        // change atomicCounterIncrement into an atomicAdd of 1
+        TString name("atomicAdd");
+        TType uintType(EbtUint);
+
+        TFunction realFunc(&name, function->getType());
+
+        for (int i = 0; i < function->getParamCount(); ++i) {
+            realFunc.addParameter((*function)[i]);
+        }
+
+        TParameter tmpP = { 0, &uintType };
+        realFunc.addParameter(tmpP);
+        arguments = intermediate.growAggregate(arguments, intermediate.addConstantUnion(1, loc, true));
+
+        result = handleFunctionCall(loc, &realFunc, arguments);
+    } else if (function->getName() == "atomicCounterDecrement") {
+        // change atomicCounterDecrement into an atomicAdd with -1
+        // and subtract 1 from result, to return post-decrement value
+        TString name("atomicAdd");
+        TType uintType(EbtUint);
+
+        TFunction realFunc(&name, function->getType());
+
+        for (int i = 0; i < function->getParamCount(); ++i) {
+            realFunc.addParameter((*function)[i]);
+        }
+
+        TParameter tmpP = { 0, &uintType };
+        realFunc.addParameter(tmpP);
+        arguments = intermediate.growAggregate(arguments, intermediate.addConstantUnion(-1, loc, true));
+
+        result = handleFunctionCall(loc, &realFunc, arguments);
+
+        // post decrement, so that it matches AtomicCounterDecrement semantics
+        if (result) {
+            result = handleBinaryMath(loc, "-", EOpSub, result, intermediate.addConstantUnion(1, loc, true));
+        }
+    } else if (function->getName() == "atomicCounter") {
+        // change atomicCounter into a direct read of the variable
+        if (arguments->getAsTyped()) {
+            result = arguments->getAsTyped();
+        }
+    }
+#endif
+
+    return result;
+}
+
 // When a declaration includes a type, but not a variable name, it can be used
 // to establish defaults.
 void TParseContext::declareTypeDefaults(const TSourceLoc& loc, const TPublicType& publicType)
@@ -6622,6 +6804,91 @@ void TParseContext::declareTypeDefaults(const TSourceLoc& loc, const TPublicType
 #endif
 }
 
+bool TParseContext::vkRelaxedRemapUniformVariable(const TSourceLoc& loc, TString& identifier, const TPublicType&,
+    TArraySizes*, TIntermTyped* initializer, TType& type)
+{
+    if (parsingBuiltins || symbolTable.atBuiltInLevel() || !symbolTable.atGlobalLevel() ||
+        type.getQualifier().storage != EvqUniform ||
+        !(type.containsNonOpaque()
+#ifndef GLSLANG_WEB
+            || type.getBasicType() == EbtAtomicUint
+#endif
+        )) {
+        return false;
+    }
+
+    if (type.getQualifier().hasLocation()) {
+        warn(loc, "ignoring layout qualifier for uniform", identifier.c_str(), "location");
+        type.getQualifier().layoutLocation = TQualifier::layoutLocationEnd;
+    }
+
+    if (initializer) {
+        warn(loc, "Ignoring initializer for uniform", identifier.c_str(), "");
+        initializer = nullptr;
+    }
+
+    if (type.isArray()) {
+        // do array size checks here
+        arraySizesCheck(loc, type.getQualifier(), type.getArraySizes(), initializer, false);
+
+        if (arrayQualifierError(loc, type.getQualifier()) || arrayError(loc, type)) {
+            error(loc, "array param error", identifier.c_str(), "");
+        }
+    }
+
+    // do some checking on the type as it was declared
+    layoutTypeCheck(loc, type);
+
+    int bufferBinding = TQualifier::layoutBindingEnd;
+    TVariable* updatedBlock = nullptr;
+
+#ifndef GLSLANG_WEB
+    // Convert atomic_uint into members of a buffer block
+    if (type.isAtomic()) {
+        type.setBasicType(EbtUint);
+        type.getQualifier().storage = EvqBuffer;
+
+        type.getQualifier().volatil = true;
+        type.getQualifier().coherent = true;
+
+        // xxTODO: use logic from fixOffset() to apply explicit member offset
+        bufferBinding = type.getQualifier().layoutBinding;
+        type.getQualifier().layoutBinding = TQualifier::layoutBindingEnd;
+        type.getQualifier().explicitOffset = false;
+        growAtomicCounterBlock(bufferBinding, loc, type, identifier, nullptr);
+        updatedBlock = atomicCounterBuffers[bufferBinding];
+    }
+#endif
+
+    if (!updatedBlock) {
+        growGlobalUniformBlock(loc, type, identifier, nullptr);
+        updatedBlock = globalUniformBlock;
+    }
+
+    //
+    //      don't assign explicit member offsets here
+    //      if any are assigned, need to be updated here and in the merge/link step
+    // fixBlockUniformOffsets(updatedBlock->getWritableType().getQualifier(), *updatedBlock->getWritableType().getWritableStruct());
+
+    // checks on update buffer object
+    layoutObjectCheck(loc, *updatedBlock);
+
+    TSymbol* symbol = symbolTable.find(identifier);
+
+    if (!symbol) {
+        if (updatedBlock == globalUniformBlock)
+            error(loc, "error adding uniform to default uniform block", identifier.c_str(), "");
+        else
+            error(loc, "error adding atomic counter to atomic counter block", identifier.c_str(), "");
+        return false;
+    }
+
+    // merge qualifiers
+    mergeObjectLayoutQualifiers(updatedBlock->getWritableType().getQualifier(), type.getQualifier(), true);
+
+    return true;
+}
+
 //
 // Do everything necessary to handle a variable (non-block) declaration.
 // Either redeclaring a variable, or making a new one, updating the symbol
@@ -6733,6 +7000,14 @@ TIntermNode* TParseContext::declareVariable(const TSourceLoc& loc, TString& iden
     if (symbol == nullptr)
         reservedErrorCheck(loc, identifier);
 
+    if (symbol == nullptr && spvVersion.vulkan > 0 && spvVersion.vulkanRelaxed) {
+        bool remapped = vkRelaxedRemapUniformVariable(loc, identifier, publicType, arraySizes, initializer, type);
+
+        if (remapped) {
+            return nullptr;
+        }
+    }
+
     inheritGlobalDefaults(type.getQualifier());
 
     // Declare the variable
@@ -7625,6 +7900,8 @@ void TParseContext::inheritMemoryQualifiers(const TQualifier& from, TQualifier&
 void TParseContext::declareBlock(const TSourceLoc& loc, TTypeList& typeList, const TString* instanceName,
     TArraySizes* arraySizes)
 {
+    if (spvVersion.vulkan > 0 && spvVersion.vulkanRelaxed)
+        blockStorageRemap(loc, blockName, currentBlockQualifier);
     blockStageIoCheck(loc, currentBlockQualifier);
     blockQualifierCheck(loc, currentBlockQualifier, instanceName != nullptr);
     if (arraySizes != nullptr) {
@@ -7914,6 +8191,17 @@ void TParseContext::declareBlock(const TSourceLoc& loc, TTypeList& typeList, con
     trackLinkage(variable);
 }
 
+//
+// allow storage type of block to be remapped at compile time
+//
+void TParseContext::blockStorageRemap(const TSourceLoc&, const TString* instanceName, TQualifier& qualifier)
+{
+    TBlockStorageClass type = intermediate.getBlockStorageOverride(instanceName->c_str());
+    if (type != EbsNone) {
+        qualifier.setBlockStorage(type);
+    }
+}
+
 // Do all block-declaration checking regarding the combination of in/out/uniform/buffer
 // with a particular stage.
 void TParseContext::blockStageIoCheck(const TSourceLoc& loc, const TQualifier& qualifier)
diff --git a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.h b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.h
index ad9b8d6ac..6f00621af 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.h
+++ b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.h
@@ -92,7 +92,8 @@ public:
             limits(resources.limits),
             globalUniformBlock(nullptr),
             globalUniformBinding(TQualifier::layoutBindingEnd),
-            globalUniformSet(TQualifier::layoutSetEnd)
+            globalUniformSet(TQualifier::layoutSetEnd),
+            atomicCounterBlockSet(TQualifier::layoutSetEnd)
     {
         if (entryPoint != nullptr)
             sourceEntryPointName = *entryPoint;
@@ -154,10 +155,11 @@ public:
             extensionCallback(line, extension, behavior);
     }
 
-#ifdef ENABLE_HLSL
     // Manage the global uniform block (default uniforms in GLSL, $Global in HLSL)
     virtual void growGlobalUniformBlock(const TSourceLoc&, TType&, const TString& memberName, TTypeList* typeList = nullptr);
-#endif
+
+    // Manage global buffer (used for backing atomic counters in GLSL when using relaxed Vulkan semantics)
+    virtual void growAtomicCounterBlock(int binding, const TSourceLoc&, TType&, const TString& memberName, TTypeList* typeList = nullptr);
 
     // Potentially rename shader entry point function
     void renameShaderFunction(TString*& name) const
@@ -230,7 +232,24 @@ protected:
     // override this to set the language-specific name
     virtual const char* getGlobalUniformBlockName() const { return ""; }
     virtual void setUniformBlockDefaults(TType&) const { }
-    virtual void finalizeGlobalUniformBlockLayout(TVariable&) { }
+    virtual void finalizeGlobalUniformBlockLayout(TVariable&) {}
+
+    // Manage the atomic counter block (used for atomic_uints with Vulkan-Relaxed)
+    TMap<int, TVariable*> atomicCounterBuffers;
+    unsigned int atomicCounterBlockSet;
+    TMap<int, int> atomicCounterBlockFirstNewMember;
+    // override this to set the language-specific name
+    virtual const char* getAtomicCounterBlockName() const { return ""; }
+    virtual void setAtomicCounterBlockDefaults(TType&) const {}
+    virtual void finalizeAtomicCounterBlockLayout(TVariable&) {}
+    bool isAtomicCounterBlock(const TSymbol& symbol) {
+        const TVariable* var = symbol.getAsVariable();
+        if (!var)
+            return false;
+        const auto& at = atomicCounterBuffers.find(var->getType().getQualifier().layoutBinding);
+        return (at != atomicCounterBuffers.end() && (*at).second->getType() == var->getType());
+    }
+
     virtual void outputMessage(const TSourceLoc&, const char* szReason, const char* szToken,
                                const char* szExtraInfoFormat, TPrefixType prefix,
                                va_list args);
@@ -293,6 +312,9 @@ public:
     bool parseShaderStrings(TPpContext&, TInputScanner& input, bool versionWillBeError = false) override;
     void parserError(const char* s);     // for bison's yyerror
 
+    virtual void growGlobalUniformBlock(const TSourceLoc&, TType&, const TString& memberName, TTypeList* typeList = nullptr) override;
+    virtual void growAtomicCounterBlock(int binding, const TSourceLoc&, TType&, const TString& memberName, TTypeList* typeList = nullptr) override;
+
     void reservedErrorCheck(const TSourceLoc&, const TString&);
     void reservedPpErrorCheck(const TSourceLoc&, const char* name, const char* op) override;
     bool lineContinuationCheck(const TSourceLoc&, bool endOfComment) override;
@@ -340,6 +362,10 @@ public:
     void checkPrecisionQualifier(const TSourceLoc&, TPrecisionQualifier);
     void memorySemanticsCheck(const TSourceLoc&, const TFunction&, const TIntermOperator& callNode);
 
+    TIntermTyped* vkRelaxedRemapFunctionCall(const TSourceLoc&, TFunction*, TIntermNode*);
+    // returns true if the variable was remapped to something else
+    bool vkRelaxedRemapUniformVariable(const TSourceLoc&, TString&, const TPublicType&, TArraySizes*, TIntermTyped*, TType&);
+
     void assignError(const TSourceLoc&, const char* op, TString left, TString right);
     void unaryOpError(const TSourceLoc&, const char* op, TString operand);
     void binaryOpError(const TSourceLoc&, const char* op, TString left, TString right);
@@ -417,6 +443,7 @@ public:
     TIntermTyped* constructBuiltIn(const TType&, TOperator, TIntermTyped*, const TSourceLoc&, bool subset);
     void inheritMemoryQualifiers(const TQualifier& from, TQualifier& to);
     void declareBlock(const TSourceLoc&, TTypeList& typeList, const TString* instanceName = 0, TArraySizes* arraySizes = 0);
+    void blockStorageRemap(const TSourceLoc&, const TString*, TQualifier&);
     void blockStageIoCheck(const TSourceLoc&, const TQualifier&);
     void blockQualifierCheck(const TSourceLoc&, const TQualifier&, bool instanceName);
     void fixBlockLocations(const TSourceLoc&, TQualifier&, TTypeList&, bool memberWithLocation, bool memberWithoutLocation);
@@ -461,6 +488,14 @@ protected:
     void finish() override;
 #endif
 
+    virtual const char* getGlobalUniformBlockName() const override;
+    virtual void finalizeGlobalUniformBlockLayout(TVariable&) override;
+    virtual void setUniformBlockDefaults(TType& block) const override;
+
+    virtual const char* getAtomicCounterBlockName() const override;
+    virtual void finalizeAtomicCounterBlockLayout(TVariable&) override;
+    virtual void setAtomicCounterBlockDefaults(TType& block) const override;
+
 public:
     //
     // Generally, bison productions, the scanner, and the PP need read/write access to these; just give them direct access
diff --git a/3rdparty/glslang/glslang/MachineIndependent/ShaderLang.cpp b/3rdparty/glslang/glslang/MachineIndependent/ShaderLang.cpp
index 3f793a743..d02eae6fc 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/ShaderLang.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/ShaderLang.cpp
@@ -159,7 +159,7 @@ int MapVersionToIndex(int version)
     return index;
 }
 
-const int SpvVersionCount = 3;  // index range in MapSpvVersionToIndex
+const int SpvVersionCount = 4;  // index range in MapSpvVersionToIndex
 
 int MapSpvVersionToIndex(const SpvVersion& spvVersion)
 {
@@ -167,8 +167,12 @@ int MapSpvVersionToIndex(const SpvVersion& spvVersion)
 
     if (spvVersion.openGl > 0)
         index = 1;
-    else if (spvVersion.vulkan > 0)
-        index = 2;
+    else if (spvVersion.vulkan > 0) {
+        if (!spvVersion.vulkanRelaxed)
+            index = 2;
+        else
+            index = 3;
+    }
 
     assert(index < SpvVersionCount);
 
@@ -723,6 +727,7 @@ void TranslateEnvironment(const TEnvironment* environment, EShMessages& messages
                 break;
             case EShClientVulkan:
                 spvVersion.vulkanGlsl = environment->input.dialectVersion;
+                spvVersion.vulkanRelaxed = environment->input.vulkanRulesRelaxed;
                 break;
             case EShClientOpenGL:
                 spvVersion.openGl = environment->input.dialectVersion;
@@ -1762,6 +1767,7 @@ TShader::TShader(EShLanguage s)
     // clear environment (avoid constructors in them for use in a C interface)
     environment.input.languageFamily = EShSourceNone;
     environment.input.dialect = EShClientNone;
+    environment.input.vulkanRulesRelaxed = false;
     environment.client.client = EShClientNone;
     environment.target.language = EShTargetNone;
     environment.target.hlslFunctionality1 = false;
@@ -1868,6 +1874,15 @@ void TShader::setResourceSetBinding(const std::vector<std::string>& base)   { in
 void TShader::setTextureSamplerTransformMode(EShTextureSamplerTransformMode mode) { intermediate->setTextureSamplerTransformMode(mode); }
 #endif
 
+void TShader::addBlockStorageOverride(const char* nameStr, TBlockStorageClass backing) { intermediate->addBlockStorageOverride(nameStr, backing); }
+
+void TShader::setGlobalUniformBlockName(const char* name) { intermediate->setGlobalUniformBlockName(name); }
+void TShader::setGlobalUniformSet(unsigned int set) { intermediate->setGlobalUniformSet(set); }
+void TShader::setGlobalUniformBinding(unsigned int binding) { intermediate->setGlobalUniformBinding(binding); }
+
+void TShader::setAtomicCounterBlockName(const char* name) { intermediate->setAtomicCounterBlockName(name); }
+void TShader::setAtomicCounterBlockSet(unsigned int set) { intermediate->setAtomicCounterBlockSet(set); }
+
 #ifdef ENABLE_HLSL
 // See comment above TDefaultHlslIoMapper in iomapper.cpp:
 void TShader::setHlslIoMapping(bool hlslIoMap)          { intermediate->setHlslIoMapping(hlslIoMap); }
@@ -1983,7 +1998,10 @@ bool TProgram::link(EShMessages messages)
             error = true;
     }
 
-    // TODO: Link: cross-stage error checking
+    if (!error) {
+        if (! crossStageCheck(messages))
+            error = true;
+    }
 
     return ! error;
 }
@@ -2060,6 +2078,69 @@ bool TProgram::linkStage(EShLanguage stage, EShMessages messages)
     return intermediate[stage]->getNumErrors() == 0;
 }
 
+//
+// Check that there are no errors in linker objects accross stages
+//
+// Return true if no errors.
+//
+bool TProgram::crossStageCheck(EShMessages) {
+
+    // make temporary intermediates to hold the linkage symbols for each linking interface
+    // while we do the checks
+    // Independent interfaces are:
+    //                  all uniform variables and blocks
+    //                  all buffer blocks
+    //                  all in/out on a stage boundary
+
+    TVector<TIntermediate*> activeStages;
+    for (int s = 0; s < EShLangCount; ++s) {
+        if (intermediate[s])
+            activeStages.push_back(intermediate[s]);
+    }
+
+    // no extra linking if there is only one stage
+    if (! (activeStages.size() > 1))
+        return true;
+
+    // setup temporary tree to hold unfirom objects from different stages
+    TIntermediate* firstIntermediate = activeStages.front();
+    TIntermediate uniforms(EShLangCount,
+                           firstIntermediate->getVersion(),
+                           firstIntermediate->getProfile());
+    uniforms.setSpv(firstIntermediate->getSpv());
+
+    TIntermAggregate uniformObjects(EOpLinkerObjects);
+    TIntermAggregate root(EOpSequence);
+    root.getSequence().push_back(&uniformObjects);
+    uniforms.setTreeRoot(&root);
+
+    bool error = false;
+
+    // merge uniforms from all stages into a single intermediate
+    for (unsigned int i = 0; i < activeStages.size(); ++i) {
+        uniforms.mergeUniformObjects(*infoSink, *activeStages[i]);
+    }
+    error |= uniforms.getNumErrors() != 0;
+
+    // copy final definition of global block back into each stage
+    for (unsigned int i = 0; i < activeStages.size(); ++i) {
+        // We only want to merge into already existing global uniform blocks.
+        // A stage that doesn't already know about the global doesn't care about it's content.
+        // Otherwise we end up pointing to the same object between different stages
+        // and that will break binding/set remappings
+        bool mergeExistingOnly = true;
+        activeStages[i]->mergeGlobalUniformBlocks(*infoSink, uniforms, mergeExistingOnly);
+    }
+
+    // compare cross stage symbols for each stage boundary
+    for (unsigned int i = 1; i < activeStages.size(); ++i) {
+        activeStages[i - 1]->checkStageIO(*infoSink, *activeStages[i]);
+        error |= (activeStages[i - 1]->getNumErrors() != 0);
+    }
+
+    return !error;
+}
+
 const char* TProgram::getInfoLog()
 {
     return infoSink->info.c_str();
diff --git a/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp b/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
index 488c98c5c..7554bfd76 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
@@ -1273,7 +1273,7 @@ void TParseVersions::spvRemoved(const TSourceLoc& loc, const char* op)
 // Call for any operation removed because Vulkan SPIR-V is being generated.
 void TParseVersions::vulkanRemoved(const TSourceLoc& loc, const char* op)
 {
-    if (spvVersion.vulkan > 0)
+    if (spvVersion.vulkan > 0 && !spvVersion.vulkanRelaxed)
         error(loc, "not allowed when using GLSL for Vulkan", op, "");
 }
 
diff --git a/3rdparty/glslang/glslang/MachineIndependent/Versions.h b/3rdparty/glslang/glslang/MachineIndependent/Versions.h
index f377973a8..25feb0b7b 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/Versions.h
+++ b/3rdparty/glslang/glslang/MachineIndependent/Versions.h
@@ -87,11 +87,12 @@ inline const char* ProfileName(EProfile profile)
 // The union of all requested rule sets will be applied.
 //
 struct SpvVersion {
-    SpvVersion() : spv(0), vulkanGlsl(0), vulkan(0), openGl(0) {}
+    SpvVersion() : spv(0), vulkanGlsl(0), vulkan(0), openGl(0), vulkanRelaxed(false) {}
     unsigned int spv; // the version of SPIR-V to target, as defined by "word 1" of the SPIR-V binary header
     int vulkanGlsl;   // the version of GLSL semantics for Vulkan, from GL_KHR_vulkan_glsl, for "#define VULKAN XXX"
     int vulkan;       // the version of Vulkan, for which SPIR-V execution environment rules to use
     int openGl;       // the version of GLSL semantics for OpenGL, from GL_ARB_gl_spirv, for "#define GL_SPIRV XXX"
+    bool vulkanRelaxed; // relax changes to GLSL for Vulkan, allowing some GL-specific to be compiled to Vulkan SPIR-V target
 };
 
 //
diff --git a/3rdparty/glslang/glslang/MachineIndependent/gl_types.h b/3rdparty/glslang/glslang/MachineIndependent/gl_types.h
index b9372d4bb..d6c939374 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/gl_types.h
+++ b/3rdparty/glslang/glslang/MachineIndependent/gl_types.h
@@ -49,9 +49,9 @@
 #define GL_INT64_VEC4_ARB                 0x8FEB
 
 #define GL_UNSIGNED_INT64_ARB             0x140F
-#define GL_UNSIGNED_INT64_VEC2_ARB        0x8FE5
-#define GL_UNSIGNED_INT64_VEC3_ARB        0x8FE6
-#define GL_UNSIGNED_INT64_VEC4_ARB        0x8FE7
+#define GL_UNSIGNED_INT64_VEC2_ARB        0x8FF5
+#define GL_UNSIGNED_INT64_VEC3_ARB        0x8FF6
+#define GL_UNSIGNED_INT64_VEC4_ARB        0x8FF7
 #define GL_UNSIGNED_INT16_VEC2_NV         0x8FF1
 #define GL_UNSIGNED_INT16_VEC3_NV         0x8FF2
 #define GL_UNSIGNED_INT16_VEC4_NV         0x8FF3
diff --git a/3rdparty/glslang/glslang/MachineIndependent/intermOut.cpp b/3rdparty/glslang/glslang/MachineIndependent/intermOut.cpp
index 5ce3e4728..c718f949a 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/intermOut.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/intermOut.cpp
@@ -886,6 +886,7 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
     case EOpTime:                       out.debug << "time";                  break;
 
     case EOpAtomicAdd:                  out.debug << "AtomicAdd";             break;
+    case EOpAtomicSubtract:             out.debug << "AtomicSubtract";        break;
     case EOpAtomicMin:                  out.debug << "AtomicMin";             break;
     case EOpAtomicMax:                  out.debug << "AtomicMax";             break;
     case EOpAtomicAnd:                  out.debug << "AtomicAnd";             break;
diff --git a/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp b/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
index c42e74fa5..7e12864f3 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
@@ -210,8 +210,8 @@ struct TResolverUniformAdaptor {
         ent.newIndex = -1;
         const bool isValid = resolver.validateBinding(stage, ent);
         if (isValid) {
-            resolver.resolveBinding(ent.stage, ent);
             resolver.resolveSet(ent.stage, ent);
+            resolver.resolveBinding(ent.stage, ent);
             resolver.resolveUniformLocation(ent.stage, ent);
 
             if (ent.newBinding != -1) {
@@ -317,15 +317,13 @@ private:
 };
 
 // The class is used for reserving explicit uniform locations and ubo/ssbo/opaque bindings
+// xxTODO: maybe this logic should be moved into the resolver's "validateInOut" and "validateUniform"
 
 struct TSymbolValidater
 {
     TSymbolValidater(TIoMapResolver& r, TInfoSink& i, TVarLiveMap* in[EShLangCount], TVarLiveMap* out[EShLangCount],
                      TVarLiveMap* uniform[EShLangCount], bool& hadError, EProfile profile, int version)
-        : preStage(EShLangCount)
-        , currentStage(EShLangCount)
-        , nextStage(EShLangCount)
-        , resolver(r)
+        : resolver(r)
         , infoSink(i)
         , hadError(hadError)
         , profile(profile)
@@ -438,17 +436,23 @@ struct TSymbolValidater
         TIntermSymbol* base = ent1.symbol;
         const TType& type = ent1.symbol->getType();
         const TString& name = entKey.first;
-        EShLanguage stage = ent1.stage;
         TString mangleName1, mangleName2;
-        if (currentStage != stage) {
-            preStage = currentStage;
-            currentStage = stage;
-            nextStage = EShLangCount;
-            for (int i = currentStage + 1; i < EShLangCount; i++) {
-                if (inVarMaps[i] != nullptr) {
-                    nextStage = static_cast<EShLanguage>(i);
-                    break;
-                }
+        EShLanguage stage = ent1.stage;
+        EShLanguage preStage, currentStage, nextStage;
+
+        preStage = EShLangCount;
+        for (int i = stage - 1; i >= 0; i--) {
+            if (inVarMaps[i] != nullptr) {
+                preStage = static_cast<EShLanguage>(i);
+                break;
+            }
+        }
+        currentStage = stage;
+        nextStage = EShLangCount;
+        for (int i = stage + 1; i < EShLangCount; i++) {
+            if (inVarMaps[i] != nullptr) {
+                nextStage = static_cast<EShLanguage>(i);
+                break;
             }
         }
 
@@ -459,6 +463,9 @@ struct TSymbolValidater
             type.appendMangledName(mangleName1);
         }
 
+
+        // basic checking that symbols match
+        // more extensive checking in the link stage
         if (base->getQualifier().storage == EvqVaryingIn) {
             // validate stage in;
             if (preStage == EShLangCount)
@@ -484,8 +491,7 @@ struct TSymbolValidater
                     if (ent2->second.symbol->getType().getQualifier().isArrayedIo(preStage)) {
                         TType subType(ent2->second.symbol->getType(), 0);
                         subType.appendMangledName(mangleName2);
-                    }
-                    else {
+                    } else {
                         ent2->second.symbol->getType().appendMangledName(mangleName2);
                     }
 
@@ -536,8 +542,7 @@ struct TSymbolValidater
                     if (ent2->second.symbol->getType().getQualifier().isArrayedIo(nextStage)) {
                         TType subType(ent2->second.symbol->getType(), 0);
                         subType.appendMangledName(mangleName2);
-                    }
-                    else {
+                    } else {
                         ent2->second.symbol->getType().appendMangledName(mangleName2);
                     }
                     if (mangleName1 == mangleName2)
@@ -550,7 +555,7 @@ struct TSymbolValidater
                 }
                 return;
             }
-        } else if (base->getQualifier().isUniformOrBuffer() && ! base->getQualifier().isPushConstant()) {
+        } else if (base->getQualifier().isUniformOrBuffer() && !base->getQualifier().isPushConstant()) {
             // validate uniform type;
             for (int i = 0; i < EShLangCount; i++) {
                 if (i != currentStage && outVarMaps[i] != nullptr) {
@@ -558,6 +563,7 @@ struct TSymbolValidater
                     if (ent2 != uniformVarMap[i]->end()) {
                         ent2->second.symbol->getType().appendMangledName(mangleName2);
                         if (mangleName1 != mangleName2) {
+                            ent2->second.symbol->getType().sameElementType(type);
                             TString err = "Invalid Uniform variable type : " + entKey.first;
                             infoSink.info.message(EPrefixInternalError, err.c_str());
                             hadError = true;
@@ -608,8 +614,7 @@ struct TSymbolValidater
     }
 
     TVarLiveMap *inVarMaps[EShLangCount], *outVarMaps[EShLangCount], *uniformVarMap[EShLangCount];
-    // Use for mark pre stage, to get more interface symbol information.
-    EShLanguage preStage, currentStage, nextStage;
+
     // Use for mark current shader stage for resolver
     TIoMapResolver& resolver;
     TInfoSink& infoSink;
@@ -749,14 +754,18 @@ TDefaultIoResolverBase::TDefaultIoResolverBase(const TIntermediate& intermediate
     , nextOutputLocation(0)
 {
     memset(stageMask, false, sizeof(bool) * (EShLangCount + 1));
+    memset(stageIntermediates, 0, sizeof(TIntermediate*) * (EShLangCount));
+    stageIntermediates[intermediate.getStage()] = &intermediate;
 }
 
-int TDefaultIoResolverBase::getBaseBinding(TResourceType res, unsigned int set) const {
-    return selectBaseBinding(intermediate.getShiftBinding(res), intermediate.getShiftBindingForSet(res, set));
+int TDefaultIoResolverBase::getBaseBinding(EShLanguage stage, TResourceType res, unsigned int set) const {
+    return stageIntermediates[stage] ? selectBaseBinding(stageIntermediates[stage]->getShiftBinding(res), stageIntermediates[stage]->getShiftBindingForSet(res, set))
+                                     : selectBaseBinding(intermediate.getShiftBinding(res), intermediate.getShiftBindingForSet(res, set));
 }
 
-const std::vector<std::string>& TDefaultIoResolverBase::getResourceSetBinding() const {
-    return intermediate.getResourceSetBinding();
+const std::vector<std::string>& TDefaultIoResolverBase::getResourceSetBinding(EShLanguage stage) const {
+    return stageIntermediates[stage] ? stageIntermediates[stage]->getResourceSetBinding()
+                                     : intermediate.getResourceSetBinding();
 }
 
 bool TDefaultIoResolverBase::doAutoBindingMapping() const { return intermediate.getAutoMapBindings(); }
@@ -797,14 +806,14 @@ int TDefaultIoResolverBase::getFreeSlot(int set, int base, int size) {
     return reserveSlot(set, base, size);
 }
 
-int TDefaultIoResolverBase::resolveSet(EShLanguage /*stage*/, TVarEntryInfo& ent) {
+int TDefaultIoResolverBase::resolveSet(EShLanguage stage, TVarEntryInfo& ent) {
     const TType& type = ent.symbol->getType();
     if (type.getQualifier().hasSet()) {
         return ent.newSet = type.getQualifier().layoutSet;
     }
     // If a command line or API option requested a single descriptor set, use that (if not overrided by spaceN)
-    if (getResourceSetBinding().size() == 1) {
-        return ent.newSet = atoi(getResourceSetBinding()[0].c_str());
+    if (getResourceSetBinding(stage).size() == 1) {
+        return ent.newSet = atoi(getResourceSetBinding(stage)[0].c_str());
     }
     return ent.newSet = 0;
 }
@@ -925,7 +934,7 @@ int TDefaultGlslIoResolver::resolveInOutLocation(EShLanguage stage, TVarEntryInf
         preStage = currentStage;
         currentStage = stage;
     }
-    // kick out of not doing this
+    // kick out if not doing this
     if (! doAutoLocationMapping()) {
         return ent.newLocation = -1;
     }
@@ -1073,7 +1082,7 @@ int TDefaultGlslIoResolver::resolveUniformLocation(EShLanguage /*stage*/, TVarEn
     return ent.newLocation = location;
 }
 
-int TDefaultGlslIoResolver::resolveBinding(EShLanguage /*stage*/, TVarEntryInfo& ent) {
+int TDefaultGlslIoResolver::resolveBinding(EShLanguage stage, TVarEntryInfo& ent) {
     const TType& type = ent.symbol->getType();
     const TString& name = ent.symbol->getAccessName();
     // On OpenGL arrays of opaque types take a separate binding for each element
@@ -1086,30 +1095,32 @@ int TDefaultGlslIoResolver::resolveBinding(EShLanguage /*stage*/, TVarEntryInfo&
     // There is no 'set' qualifier in OpenGL shading language, each resource has its own
     // binding name space, so remap the 'set' to resource type which make each resource
     // binding is valid from 0 to MAX_XXRESOURCE_BINDINGS
-    int set = resource;
+    int set = intermediate.getSpv().openGl != 0 ? resource : ent.newSet;
+    int resourceKey = set;
     if (resource < EResCount) {
         if (type.getQualifier().hasBinding()) {
-            ent.newBinding = reserveSlot(set, getBaseBinding(resource, set) + type.getQualifier().layoutBinding, numBindings);
-            return ent.newBinding;
-        } else if (ent.live && doAutoBindingMapping()) {
+            int newBinding = reserveSlot(resourceKey, getBaseBinding(stage, resource, set) + type.getQualifier().layoutBinding, numBindings);
+            return ent.newBinding = newBinding;
+
+        } else {
             // The resource in current stage is not declared with binding, but it is possible declared
             // with explicit binding in other stages, find the resourceSlotMap firstly to check whether
             // the resource has binding, don't need to allocate if it already has a binding
             bool hasBinding = false;
-            if (! resourceSlotMap[resource].empty()) {
-                TVarSlotMap::iterator iter = resourceSlotMap[resource].find(name);
-                if (iter != resourceSlotMap[resource].end()) {
+            ent.newBinding = -1; // leave as -1 if it isn't set below
+
+            if (! resourceSlotMap[resourceKey].empty()) {
+                TVarSlotMap::iterator iter = resourceSlotMap[resourceKey].find(name);
+                if (iter != resourceSlotMap[resourceKey].end()) {
                     hasBinding = true;
                     ent.newBinding = iter->second;
                 }
             }
-            if (! hasBinding) {
-                TVarSlotMap varSlotMap;
+            if (!hasBinding && (ent.live && doAutoBindingMapping())) {
                 // find free slot, the caller did make sure it passes all vars with binding
                 // first and now all are passed that do not have a binding and needs one
-                int binding = getFreeSlot(resource, getBaseBinding(resource, set), numBindings);
-                varSlotMap[name] = binding;
-                resourceSlotMap[resource] = varSlotMap;
+                int binding = getFreeSlot(resourceKey, getBaseBinding(stage, resource, set), numBindings);
+                resourceSlotMap[resourceKey][name] = binding;
                 ent.newBinding = binding;
             }
             return ent.newBinding;
@@ -1211,16 +1222,20 @@ void TDefaultGlslIoResolver::reserverStorageSlot(TVarEntryInfo& ent, TInfoSink&
 void TDefaultGlslIoResolver::reserverResourceSlot(TVarEntryInfo& ent, TInfoSink& infoSink) {
     const TType& type = ent.symbol->getType();
     const TString& name = ent.symbol->getAccessName();
-    int resource = getResourceType(type);
+    TResourceType resource = getResourceType(type);
+    int set = intermediate.getSpv().openGl != 0 ? resource : resolveSet(ent.stage, ent);
+    int resourceKey = set;
+
     if (type.getQualifier().hasBinding()) {
-        TVarSlotMap& varSlotMap = resourceSlotMap[resource];
+        TVarSlotMap& varSlotMap = resourceSlotMap[resourceKey];
         TVarSlotMap::iterator iter = varSlotMap.find(name);
-        int binding = type.getQualifier().layoutBinding;
+        int binding = type.getQualifier().layoutBinding + getBaseBinding(ent.stage, resource, set);
+
         if (iter == varSlotMap.end()) {
             // Reserve the slots for the ubo, ssbo and opaques who has explicit binding
-            int numBindings = type.isSizedArray() ? type.getCumulativeArraySize() : 1;
+            int numBindings = intermediate.getSpv().openGl != 0 && type.isSizedArray() ? type.getCumulativeArraySize() : 1;
             varSlotMap[name] = binding;
-            reserveSlot(resource, binding, numBindings);
+            reserveSlot(resourceKey, binding, numBindings);
         } else {
             // Allocate binding by name for OpenGL driver, so the resource in different
             // stages should be declared with the same binding
@@ -1269,7 +1284,7 @@ struct TDefaultIoResolver : public TDefaultIoResolverBase {
         return EResCount;
     }
 
-    int resolveBinding(EShLanguage /*stage*/, TVarEntryInfo& ent) override {
+    int resolveBinding(EShLanguage stage, TVarEntryInfo& ent) override {
         const TType& type = ent.symbol->getType();
         const int set = getLayoutSet(type);
         // On OpenGL arrays of opaque types take a seperate binding for each element
@@ -1278,11 +1293,11 @@ struct TDefaultIoResolver : public TDefaultIoResolverBase {
         if (resource < EResCount) {
             if (type.getQualifier().hasBinding()) {
                 return ent.newBinding = reserveSlot(
-                           set, getBaseBinding(resource, set) + type.getQualifier().layoutBinding, numBindings);
+                           set, getBaseBinding(stage, resource, set) + type.getQualifier().layoutBinding, numBindings);
             } else if (ent.live && doAutoBindingMapping()) {
                 // find free slot, the caller did make sure it passes all vars with binding
                 // first and now all are passed that do not have a binding and needs one
-                return ent.newBinding = getFreeSlot(set, getBaseBinding(resource, set), numBindings);
+                return ent.newBinding = getFreeSlot(set, getBaseBinding(stage, resource, set), numBindings);
             }
         }
         return ent.newBinding = -1;
@@ -1354,17 +1369,17 @@ struct TDefaultHlslIoResolver : public TDefaultIoResolverBase {
         return EResCount;
     }
 
-    int resolveBinding(EShLanguage /*stage*/, TVarEntryInfo& ent) override {
+    int resolveBinding(EShLanguage stage, TVarEntryInfo& ent) override {
         const TType& type = ent.symbol->getType();
         const int set = getLayoutSet(type);
         TResourceType resource = getResourceType(type);
         if (resource < EResCount) {
             if (type.getQualifier().hasBinding()) {
-                return ent.newBinding = reserveSlot(set, getBaseBinding(resource, set) + type.getQualifier().layoutBinding);
+                return ent.newBinding = reserveSlot(set, getBaseBinding(stage, resource, set) + type.getQualifier().layoutBinding);
             } else if (ent.live && doAutoBindingMapping()) {
                 // find free slot, the caller did make sure it passes all vars with binding
                 // first and now all are passed that do not have a binding and needs one
-                return ent.newBinding = getFreeSlot(set, getBaseBinding(resource, set));
+                return ent.newBinding = getFreeSlot(set, getBaseBinding(stage, resource, set));
             }
         }
         return ent.newBinding = -1;
@@ -1403,10 +1418,10 @@ bool TIoMapper::addStage(EShLanguage stage, TIntermediate& intermediate, TInfoSi
         else
             resolver = &defaultResolver;
     }
-    resolver->addStage(stage);
 #else
     resolver = &defaultResolver;
 #endif
+    resolver->addStage(stage, intermediate);
 
     TVarLiveMap inVarMap, outVarMap, uniformVarMap;
     TVarLiveVector inVector, outVector, uniformVector;
@@ -1502,10 +1517,21 @@ bool TGlslIoMapper::addStage(EShLanguage stage, TIntermediate& intermediate, TIn
     }
     // if no resolver is provided, use the default resolver with the given shifts and auto map settings
     TDefaultGlslIoResolver defaultResolver(intermediate);
+#ifdef ENABLE_HLSL
+    TDefaultHlslIoResolver defaultHlslResolver(intermediate);
+    if (resolver == nullptr) {
+        // TODO: use a passed in IO mapper for this
+        if (intermediate.usingHlslIoMapping())
+            resolver = &defaultHlslResolver;
+        else
+            resolver = &defaultResolver;
+    }
+#else
     if (resolver == nullptr) {
         resolver = &defaultResolver;
     }
-    resolver->addStage(stage);
+#endif
+    resolver->addStage(stage, intermediate);
     inVarMaps[stage] = new TVarLiveMap(); outVarMaps[stage] = new TVarLiveMap(); uniformVarMap[stage] = new TVarLiveMap();
     TVarGatherTraverser iter_binding_all(intermediate, true, *inVarMaps[stage], *outVarMaps[stage],
                                          *uniformVarMap[stage]);
@@ -1547,15 +1573,51 @@ bool TGlslIoMapper::doMap(TIoMapResolver* resolver, TInfoSink& infoSink) {
         TResolverInOutAdaptor inOutResolve(EShLangCount, *resolver, infoSink, hadError);
         TSymbolValidater symbolValidater(*resolver, infoSink, inVarMaps,
                                          outVarMaps, uniformVarMap, hadError, profile, version);
+
+        TVarLiveVector inVectors[EShLangCount];
+        TVarLiveVector outVectors[EShLangCount];
         TVarLiveVector uniformVector;
+
         resolver->beginResolve(EShLangCount);
         for (int stage = EShLangVertex; stage < EShLangCount; stage++) {
             if (inVarMaps[stage] != nullptr) {
                 inOutResolve.setStage(EShLanguage(stage));
-                for (auto& var : *(inVarMaps[stage])) { symbolValidater(var); }
-                for (auto& var : *(inVarMaps[stage])) { inOutResolve(var); }
-                for (auto& var : *(outVarMaps[stage])) { symbolValidater(var); }
-                for (auto& var : *(outVarMaps[stage])) { inOutResolve(var); }
+
+                // copy vars into a sorted list
+                std::for_each(inVarMaps[stage]->begin(), inVarMaps[stage]->end(),
+                        [&inVectors, stage](TVarLivePair p) { inVectors[stage].push_back(p); });
+                std::sort(inVectors[stage].begin(), inVectors[stage].end(),
+                        [](const TVarLivePair& p1, const TVarLivePair& p2) -> bool {
+                            return TVarEntryInfo::TOrderByPriority()(p1.second, p2.second);
+                });
+
+                std::for_each(outVarMaps[stage]->begin(), outVarMaps[stage]->end(),
+                        [&outVectors, stage](TVarLivePair p) { outVectors[stage].push_back(p); });
+                std::sort(outVectors[stage].begin(), outVectors[stage].end(),
+                        [](const TVarLivePair& p1, const TVarLivePair& p2) -> bool {
+                            return TVarEntryInfo::TOrderByPriority()(p1.second, p2.second);
+                });
+
+                for (auto& var : inVectors[stage]) { symbolValidater(var); }
+                for (auto& var : inVectors[stage]) { inOutResolve(var); }
+                for (auto& var : outVectors[stage]) { symbolValidater(var); }
+                for (auto& var : outVectors[stage]) { inOutResolve(var); }
+
+                // copy results back into maps
+                std::for_each(inVectors[stage].begin(), inVectors[stage].end(),
+                    [this, stage](TVarLivePair p) {
+                        auto at = inVarMaps[stage]->find(p.first);
+                        if (at != inVarMaps[stage]->end())
+                            at->second = p.second;
+                });
+
+                std::for_each(outVectors[stage].begin(), outVectors[stage].end(),
+                    [this, stage](TVarLivePair p) {
+                        auto at = outVarMaps[stage]->find(p.first);
+                        if (at != outVarMaps[stage]->end())
+                            at->second = p.second;
+                });
+
             }
             if (uniformVarMap[stage] != nullptr) {
                 uniformResolve.setStage(EShLanguage(stage));
@@ -1563,7 +1625,7 @@ bool TGlslIoMapper::doMap(TIoMapResolver* resolver, TInfoSink& infoSink) {
             }
         }
         std::sort(uniformVector.begin(), uniformVector.end(), [](const TVarLivePair& p1, const TVarLivePair& p2) -> bool {
-            return TVarEntryInfo::TOrderByPriority()(p1.second, p2.second);
+            return TVarEntryInfo::TOrderByPriorityAndLive()(p1.second, p2.second);
         });
         for (auto& var : uniformVector) { symbolValidater(var); }
         for (auto& var : uniformVector) { uniformResolve(var); }
diff --git a/3rdparty/glslang/glslang/MachineIndependent/iomapper.h b/3rdparty/glslang/glslang/MachineIndependent/iomapper.h
index 1dce8ff54..07357c2ef 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/iomapper.h
+++ b/3rdparty/glslang/glslang/MachineIndependent/iomapper.h
@@ -87,6 +87,35 @@ struct TVarEntryInfo {
             return lPoints > rPoints;
         }
     };
+
+    struct TOrderByPriorityAndLive {
+        // ordering:
+        // 1) do live variables first
+        // 2) has both binding and set
+        // 3) has binding but no set
+        // 4) has no binding but set
+        // 5) has no binding and no set
+        inline bool operator()(const TVarEntryInfo& l, const TVarEntryInfo& r) {
+
+            const TQualifier& lq = l.symbol->getQualifier();
+            const TQualifier& rq = r.symbol->getQualifier();
+
+            // simple rules:
+            // has binding gives 2 points
+            // has set gives 1 point
+            // who has the most points is more important.
+            int lPoints = (lq.hasBinding() ? 2 : 0) + (lq.hasSet() ? 1 : 0);
+            int rPoints = (rq.hasBinding() ? 2 : 0) + (rq.hasSet() ? 1 : 0);
+
+            if (l.live != r.live)
+                return l.live > r.live;
+
+            if (lPoints != rPoints)
+                return lPoints > rPoints;
+
+            return l.id < r.id;
+        }
+    };
 };
 
 // Base class for shared TIoMapResolver services, used by several derivations.
@@ -107,8 +136,8 @@ public:
     void endCollect(EShLanguage) override {}
     void reserverResourceSlot(TVarEntryInfo& /*ent*/, TInfoSink& /*infoSink*/) override {}
     void reserverStorageSlot(TVarEntryInfo& /*ent*/, TInfoSink& /*infoSink*/) override {}
-    int getBaseBinding(TResourceType res, unsigned int set) const;
-    const std::vector<std::string>& getResourceSetBinding() const;
+    int getBaseBinding(EShLanguage stage, TResourceType res, unsigned int set) const;
+    const std::vector<std::string>& getResourceSetBinding(EShLanguage stage) const;
     virtual TResourceType getResourceType(const glslang::TType& type) = 0;
     bool doAutoBindingMapping() const;
     bool doAutoLocationMapping() const;
@@ -122,9 +151,11 @@ public:
     int resolveInOutLocation(EShLanguage stage, TVarEntryInfo& ent) override;
     int resolveInOutComponent(EShLanguage /*stage*/, TVarEntryInfo& ent) override;
     int resolveInOutIndex(EShLanguage /*stage*/, TVarEntryInfo& ent) override;
-    void addStage(EShLanguage stage) override {
-        if (stage < EShLangCount)
+    void addStage(EShLanguage stage, TIntermediate& stageIntermediate) override {
+        if (stage < EShLangCount) {
             stageMask[stage] = true;
+            stageIntermediates[stage] = &stageIntermediate;
+        }
     }
     uint32_t computeTypeLocationSize(const TType& type, EShLanguage stage);
 
@@ -139,6 +170,8 @@ protected:
     int nextInputLocation;
     int nextOutputLocation;
     bool stageMask[EShLangCount + 1];
+    const TIntermediate* stageIntermediates[EShLangCount];
+
     // Return descriptor set specific base if there is one, and the generic base otherwise.
     int selectBaseBinding(int base, int descriptorSetBase) const {
         return descriptorSetBase != -1 ? descriptorSetBase : base;
diff --git a/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp b/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
index 3f3a3127c..33af7bbe0 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
@@ -90,6 +90,56 @@ void TIntermediate::merge(TInfoSink& infoSink, TIntermediate& unit)
 #endif
 }
 
+//
+// check that link objects between stages
+//
+void TIntermediate::mergeUniformObjects(TInfoSink& infoSink, TIntermediate& unit) {
+    if (unit.treeRoot == nullptr || treeRoot == nullptr)
+        return;
+
+    // Get the linker-object lists
+    TIntermSequence& linkerObjects = findLinkerObjects()->getSequence();
+    TIntermSequence unitLinkerObjects = unit.findLinkerObjects()->getSequence();
+
+    // filter unitLinkerObjects to only contain uniforms
+    auto end = std::remove_if(unitLinkerObjects.begin(), unitLinkerObjects.end(),
+        [](TIntermNode* node) {return node->getAsSymbolNode()->getQualifier().storage != EvqUniform &&
+                                      node->getAsSymbolNode()->getQualifier().storage != EvqBuffer; });
+    unitLinkerObjects.resize(end - unitLinkerObjects.begin());
+
+    // merge uniforms and do error checking
+    bool mergeExistingOnly = false;
+    mergeGlobalUniformBlocks(infoSink, unit, mergeExistingOnly);
+    mergeLinkerObjects(infoSink, linkerObjects, unitLinkerObjects, unit.getStage());
+}
+
+//
+// do error checking on the shader boundary in / out vars 
+//
+void TIntermediate::checkStageIO(TInfoSink& infoSink, TIntermediate& unit) {
+    if (unit.treeRoot == nullptr || treeRoot == nullptr)
+        return;
+
+    // Get copies of the linker-object lists
+    TIntermSequence linkerObjects = findLinkerObjects()->getSequence();
+    TIntermSequence unitLinkerObjects = unit.findLinkerObjects()->getSequence();
+
+    // filter linkerObjects to only contain out variables
+    auto end = std::remove_if(linkerObjects.begin(), linkerObjects.end(),
+        [](TIntermNode* node) {return node->getAsSymbolNode()->getQualifier().storage != EvqVaryingOut; });
+    linkerObjects.resize(end - linkerObjects.begin());
+
+    // filter unitLinkerObjects to only contain in variables
+    auto unitEnd = std::remove_if(unitLinkerObjects.begin(), unitLinkerObjects.end(),
+        [](TIntermNode* node) {return node->getAsSymbolNode()->getQualifier().storage != EvqVaryingIn; });
+    unitLinkerObjects.resize(unitEnd - unitLinkerObjects.begin());
+
+    // do matching and error checking
+    mergeLinkerObjects(infoSink, linkerObjects, unitLinkerObjects, unit.getStage());
+
+    // TODO: final check; make sure that any statically used `in` have matching `out` written to
+}
+
 void TIntermediate::mergeCallGraphs(TInfoSink& infoSink, TIntermediate& unit)
 {
     if (unit.getNumEntryPoints() > 0) {
@@ -137,6 +187,7 @@ void TIntermediate::mergeModes(TInfoSink& infoSink, TIntermediate& unit)
     MERGE_MAX(spvVersion.vulkanGlsl);
     MERGE_MAX(spvVersion.vulkan);
     MERGE_MAX(spvVersion.openGl);
+    MERGE_TRUE(spvVersion.vulkanRelaxed);
 
     numErrors += unit.getNumErrors();
     // Only one push_constant is allowed, mergeLinkerObjects() will ensure the push_constant
@@ -312,7 +363,9 @@ void TIntermediate::mergeTrees(TInfoSink& infoSink, TIntermediate& unit)
     remapIds(idMaps, idShift + 1, unit);
 
     mergeBodies(infoSink, globals, unitGlobals);
-    mergeLinkerObjects(infoSink, linkerObjects, unitLinkerObjects);
+    bool mergeExistingOnly = false;
+    mergeGlobalUniformBlocks(infoSink, unit, mergeExistingOnly);
+    mergeLinkerObjects(infoSink, linkerObjects, unitLinkerObjects, unit.getStage());
     ioAccessed.insert(unit.ioAccessed.begin(), unit.ioAccessed.end());
 }
 
@@ -456,11 +509,193 @@ void TIntermediate::mergeBodies(TInfoSink& infoSink, TIntermSequence& globals, c
     globals.insert(globals.end() - 1, unitGlobals.begin(), unitGlobals.end() - 1);
 }
 
+static inline bool isSameInterface(TIntermSymbol* symbol, EShLanguage stage, TIntermSymbol* unitSymbol, EShLanguage unitStage) {
+    return // 1) same stage and same shader interface
+        (stage == unitStage && symbol->getType().getShaderInterface() == unitSymbol->getType().getShaderInterface()) ||
+        // 2) accross stages and both are uniform or buffer
+        (symbol->getQualifier().storage == EvqUniform  && unitSymbol->getQualifier().storage == EvqUniform) ||
+        (symbol->getQualifier().storage == EvqBuffer   && unitSymbol->getQualifier().storage == EvqBuffer) ||
+        // 3) in/out matched across stage boundary
+        (stage < unitStage && symbol->getQualifier().storage == EvqVaryingOut  && unitSymbol->getQualifier().storage == EvqVaryingIn) ||
+        (unitStage < stage && symbol->getQualifier().storage == EvqVaryingIn && unitSymbol->getQualifier().storage == EvqVaryingOut);
+}
+
+//
+// Global Unfiform block stores any default uniforms (i.e. uniforms without a block)
+// If two linked stages declare the same member, they are meant to be the same uniform
+// and need to be in the same block
+// merge the members of different stages to allow them to be linked properly
+// as a single block
+//
+void TIntermediate::mergeGlobalUniformBlocks(TInfoSink& infoSink, TIntermediate& unit, bool mergeExistingOnly)
+{
+    TIntermSequence& linkerObjects = findLinkerObjects()->getSequence();
+    TIntermSequence& unitLinkerObjects = unit.findLinkerObjects()->getSequence();
+
+    // build lists of default blocks from the intermediates
+    TIntermSequence defaultBlocks;
+    TIntermSequence unitDefaultBlocks;
+
+    auto filter = [](TIntermSequence& list, TIntermNode* node) {
+        if (node->getAsSymbolNode()->getQualifier().defaultBlock) {
+            list.push_back(node);
+        }
+    };
+
+    std::for_each(linkerObjects.begin(), linkerObjects.end(),
+        [&defaultBlocks, &filter](TIntermNode* node) {
+            filter(defaultBlocks, node);
+        });
+    std::for_each(unitLinkerObjects.begin(), unitLinkerObjects.end(),
+        [&unitDefaultBlocks, &filter](TIntermNode* node) {
+            filter(unitDefaultBlocks, node);
+    });
+
+    auto itUnitBlock = unitDefaultBlocks.begin();
+    for (; itUnitBlock != unitDefaultBlocks.end(); itUnitBlock++) {
+
+        bool add = !mergeExistingOnly;
+        auto itBlock = defaultBlocks.begin();
+
+        for (; itBlock != defaultBlocks.end(); itBlock++) {
+            TIntermSymbol* block = (*itBlock)->getAsSymbolNode();
+            TIntermSymbol* unitBlock = (*itUnitBlock)->getAsSymbolNode();
+
+            assert(block && unitBlock);
+
+            // if the two default blocks match, then merge their definitions
+            if (block->getType().getTypeName() == unitBlock->getType().getTypeName() &&
+                block->getQualifier().storage == unitBlock->getQualifier().storage) {
+                add = false;
+                mergeBlockDefinitions(infoSink, block, unitBlock, &unit);
+            }
+        }
+        if (add) {
+            // push back on original list; won't change the size of the list we're iterating over
+            linkerObjects.push_back(*itUnitBlock);
+        }
+    }
+}
+
+void TIntermediate::mergeBlockDefinitions(TInfoSink& infoSink, TIntermSymbol* block, TIntermSymbol* unitBlock, TIntermediate* unit) {
+    if (block->getType() == unitBlock->getType()) {
+        return;
+    }
+
+    if (block->getType().getTypeName() != unitBlock->getType().getTypeName() ||
+        block->getType().getBasicType() != unitBlock->getType().getBasicType() ||
+        block->getQualifier().storage != unitBlock->getQualifier().storage ||
+        block->getQualifier().layoutSet != unitBlock->getQualifier().layoutSet) {
+        // different block names likely means different blocks
+        return;
+    }
+
+    // merge the struct
+    // order of declarations doesn't matter and they matched based on member name
+    TTypeList* memberList = block->getType().getWritableStruct();
+    TTypeList* unitMemberList = unitBlock->getType().getWritableStruct();
+
+    // keep track of which members have changed position
+    // so we don't have to search the array again
+    std::map<unsigned int, unsigned int> memberIndexUpdates;
+
+    size_t memberListStartSize = memberList->size();
+    for (unsigned int i = 0; i < unitMemberList->size(); ++i) {
+        bool merge = true;
+        for (unsigned int j = 0; j < memberListStartSize; ++j) {
+            if ((*memberList)[j].type->getFieldName() == (*unitMemberList)[i].type->getFieldName()) {
+                merge = false;
+                const TType* memberType = (*memberList)[j].type;
+                const TType* unitMemberType = (*unitMemberList)[i].type;
+
+                // compare types
+                // don't need as many checks as when merging symbols, since
+                // initializers and most qualifiers are stripped when the member is moved into the block
+                if ((*memberType) != (*unitMemberType)) {
+                    error(infoSink, "Types must match:");
+                    infoSink.info << "    " << memberType->getFieldName() << ": ";
+                    infoSink.info << "\"" << memberType->getCompleteString() << "\" versus ";
+                    infoSink.info << "\"" << unitMemberType->getCompleteString() << "\"\n";
+                }
+
+                memberIndexUpdates[i] = j;
+            }
+        }
+        if (merge) {
+            memberList->push_back((*unitMemberList)[i]);
+            memberIndexUpdates[i] = (unsigned int)memberList->size() - 1;
+        }
+    }
+
+    TType unitType;
+    unitType.shallowCopy(unitBlock->getType());
+
+    // update symbol node in unit tree,
+    // and other nodes that may reference it
+    class TMergeBlockTraverser : public TIntermTraverser {
+    public:
+        TMergeBlockTraverser(const glslang::TType &type, const glslang::TType& unitType,
+                             glslang::TIntermediate& unit,
+                             const std::map<unsigned int, unsigned int>& memberIdxUpdates) :
+            newType(type), unitType(unitType), unit(unit), memberIndexUpdates(memberIdxUpdates)
+        { }
+        virtual ~TMergeBlockTraverser() { }
+
+        const glslang::TType& newType;          // type with modifications
+        const glslang::TType& unitType;         // copy of original type
+        glslang::TIntermediate& unit;           // intermediate that is being updated
+        const std::map<unsigned int, unsigned int>& memberIndexUpdates;
+
+        virtual void visitSymbol(TIntermSymbol* symbol)
+        {
+            glslang::TType& symType = symbol->getWritableType();
+
+            if (symType == unitType) {
+                // each symbol node has a local copy of the unitType
+                //  if merging involves changing properties that aren't shared objects
+                //  they should be updated in all instances
+
+                // e.g. the struct list is a ptr to an object, so it can be updated
+                // once, outside the traverser
+                //*symType.getWritableStruct() = *newType.getStruct();
+            }
+
+        }
+
+        virtual bool visitBinary(TVisit, glslang::TIntermBinary* node)
+        {
+            if (node->getOp() == EOpIndexDirectStruct && node->getLeft()->getType() == unitType) {
+                // this is a dereference to a member of the block since the
+                // member list changed, need to update this to point to the
+                // right index
+                assert(node->getRight()->getAsConstantUnion());
+
+                glslang::TIntermConstantUnion* constNode = node->getRight()->getAsConstantUnion();
+                unsigned int memberIdx = constNode->getConstArray()[0].getUConst();
+                unsigned int newIdx = memberIndexUpdates.at(memberIdx);
+                TIntermTyped* newConstNode = unit.addConstantUnion(newIdx, node->getRight()->getLoc());
+
+                node->setRight(newConstNode);
+                delete constNode;
+
+                return true;
+            }
+            return true;
+        }
+    } finalLinkTraverser(block->getType(), unitType, *unit, memberIndexUpdates);
+
+    // update the tree to use the new type
+    unit->getTreeRoot()->traverse(&finalLinkTraverser);
+
+    // update the member list
+    (*unitMemberList) = (*memberList);
+}
+
 //
 // Merge the linker objects from unitLinkerObjects into linkerObjects.
 // Duplication is expected and filtered out, but contradictions are an error.
 //
-void TIntermediate::mergeLinkerObjects(TInfoSink& infoSink, TIntermSequence& linkerObjects, const TIntermSequence& unitLinkerObjects)
+void TIntermediate::mergeLinkerObjects(TInfoSink& infoSink, TIntermSequence& linkerObjects, const TIntermSequence& unitLinkerObjects, EShLanguage unitStage)
 {
     // Error check and merge the linker objects (duplicates should not be created)
     std::size_t initialNumLinkerObjects = linkerObjects.size();
@@ -475,7 +710,7 @@ void TIntermediate::mergeLinkerObjects(TInfoSink& infoSink, TIntermSequence& lin
             // If they are both blocks in the same shader interface,
             // match by the block-name, not the identifier name.
             if (symbol->getType().getBasicType() == EbtBlock && unitSymbol->getType().getBasicType() == EbtBlock) {
-                if (symbol->getType().getShaderInterface() == unitSymbol->getType().getShaderInterface()) {
+                if (isSameInterface(symbol, getStage(), unitSymbol, unitStage)) {
                     isSameSymbol = symbol->getType().getTypeName() == unitSymbol->getType().getTypeName();
                 }
             }
@@ -495,18 +730,54 @@ void TIntermediate::mergeLinkerObjects(TInfoSink& infoSink, TIntermSequence& lin
                 if (! symbol->getQualifier().hasBinding() && unitSymbol->getQualifier().hasBinding())
                     symbol->getQualifier().layoutBinding = unitSymbol->getQualifier().layoutBinding;
 
+                // Similarly for location
+                if (!symbol->getQualifier().hasLocation() && unitSymbol->getQualifier().hasLocation()) {
+                    symbol->getQualifier().layoutLocation = unitSymbol->getQualifier().layoutLocation;
+                }
+
                 // Update implicit array sizes
                 mergeImplicitArraySizes(symbol->getWritableType(), unitSymbol->getType());
 
                 // Check for consistent types/qualification/initializers etc.
-                mergeErrorCheck(infoSink, *symbol, *unitSymbol, false);
+                mergeErrorCheck(infoSink, *symbol, *unitSymbol, unitStage);
             }
             // If different symbols, verify they arn't push_constant since there can only be one per stage
-            else if (symbol->getQualifier().isPushConstant() && unitSymbol->getQualifier().isPushConstant())
+            else if (symbol->getQualifier().isPushConstant() && unitSymbol->getQualifier().isPushConstant() && getStage() == unitStage)
                 error(infoSink, "Only one push_constant block is allowed per stage");
         }
-        if (merge)
+        if (merge) {
             linkerObjects.push_back(unitLinkerObjects[unitLinkObj]);
+
+            // for anonymous blocks, check that their members don't conflict with other names
+            if (unitLinkerObjects[unitLinkObj]->getAsSymbolNode()->getBasicType() == EbtBlock &&
+                IsAnonymous(unitLinkerObjects[unitLinkObj]->getAsSymbolNode()->getName())) {
+                for (std::size_t linkObj = 0; linkObj < initialNumLinkerObjects; ++linkObj) {
+                    TIntermSymbol* symbol = linkerObjects[linkObj]->getAsSymbolNode();
+                    TIntermSymbol* unitSymbol = unitLinkerObjects[unitLinkObj]->getAsSymbolNode();
+                    assert(symbol && unitSymbol);
+
+                    auto checkName = [this, unitSymbol, &infoSink](const TString& name) {
+                        for (unsigned int i = 0; i < unitSymbol->getType().getStruct()->size(); ++i) {
+                            if (name == (*unitSymbol->getType().getStruct())[i].type->getFieldName()) {
+                                error(infoSink, "Anonymous member name used for global variable or other anonymous member: ");
+                                infoSink.info << (*unitSymbol->getType().getStruct())[i].type->getCompleteString() << "\n";
+                            }
+                        }
+                    };
+
+                    if (isSameInterface(symbol, getStage(), unitSymbol, unitStage)) {
+                        checkName(symbol->getName());
+
+                        // check members of other anonymous blocks
+                        if (symbol->getBasicType() == EbtBlock && IsAnonymous(symbol->getName())) {
+                            for (unsigned int i = 0; i < symbol->getType().getStruct()->size(); ++i) {
+                                checkName((*symbol->getType().getStruct())[i].type->getFieldName());
+                            }
+                        }
+                    }
+                }
+            }
+        }
     }
 }
 
@@ -538,26 +809,74 @@ void TIntermediate::mergeImplicitArraySizes(TType& type, const TType& unitType)
 //
 // This function only does one of intra- or cross-stage matching per call.
 //
-void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& symbol, const TIntermSymbol& unitSymbol, bool crossStage)
+void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& symbol, const TIntermSymbol& unitSymbol, EShLanguage unitStage)
 {
 #if !defined(GLSLANG_WEB) && !defined(GLSLANG_ANGLE)
+    bool crossStage = getStage() != unitStage;
     bool writeTypeComparison = false;
 
     // Types have to match
-    if (symbol.getType() != unitSymbol.getType()) {
+    {
         // but, we make an exception if one is an implicit array and the other is sized
-        if (! (symbol.getType().isArray() && unitSymbol.getType().isArray() &&
-                symbol.getType().sameElementType(unitSymbol.getType()) &&
-                (symbol.getType().isUnsizedArray() || unitSymbol.getType().isUnsizedArray()))) {
-            error(infoSink, "Types must match:");
+        // or if the array sizes differ because of the extra array dimension on some in/out boundaries
+        bool arraysMatch = false;
+        if (isIoResizeArray(symbol.getType(), getStage()) || isIoResizeArray(unitSymbol.getType(), unitStage)) {
+            // if the arrays have an extra dimension because of the stage.
+            // compare dimensions while ignoring the outer dimension
+            unsigned int firstDim = isIoResizeArray(symbol.getType(), getStage()) ? 1 : 0;
+            unsigned int numDim = symbol.getArraySizes()
+                ? symbol.getArraySizes()->getNumDims() : 0;
+            unsigned int unitFirstDim = isIoResizeArray(unitSymbol.getType(), unitStage) ? 1 : 0;
+            unsigned int unitNumDim = unitSymbol.getArraySizes()
+                ? unitSymbol.getArraySizes()->getNumDims() : 0;
+            arraysMatch = (numDim - firstDim) == (unitNumDim - unitFirstDim);
+            // check that array sizes match as well
+            for (unsigned int i = 0; i < (numDim - firstDim) && arraysMatch; i++) {
+                if (symbol.getArraySizes()->getDimSize(firstDim + i) !=
+                    unitSymbol.getArraySizes()->getDimSize(unitFirstDim + i)) {
+                    arraysMatch = false;
+                    break;
+                }
+            }
+        }
+        else {
+            arraysMatch = symbol.getType().sameArrayness(unitSymbol.getType()) ||
+                (symbol.getType().isArray() && unitSymbol.getType().isArray() &&
+                (symbol.getType().isUnsizedArray() || unitSymbol.getType().isUnsizedArray()));
+        }
+
+        if (!symbol.getType().sameElementType(unitSymbol.getType()) ||
+            !symbol.getType().sameTypeParameters(unitSymbol.getType()) ||
+            !arraysMatch ) {
             writeTypeComparison = true;
+            error(infoSink, "Types must match:");
+        }
+    }
+
+    // Interface block  member-wise layout qualifiers have to match
+    if (symbol.getType().getBasicType() == EbtBlock && unitSymbol.getType().getBasicType() == EbtBlock &&
+        symbol.getType().getStruct() && unitSymbol.getType().getStruct() &&
+        symbol.getType().sameStructType(unitSymbol.getType())) {
+        for (unsigned int i = 0; i < symbol.getType().getStruct()->size(); ++i) {
+            const TQualifier& qualifier = (*symbol.getType().getStruct())[i].type->getQualifier();
+            const TQualifier& unitQualifier = (*unitSymbol.getType().getStruct())[i].type->getQualifier();
+            if (qualifier.layoutMatrix     != unitQualifier.layoutMatrix ||
+                qualifier.layoutOffset     != unitQualifier.layoutOffset ||
+                qualifier.layoutAlign      != unitQualifier.layoutAlign ||
+                qualifier.layoutLocation   != unitQualifier.layoutLocation ||
+                qualifier.layoutComponent  != unitQualifier.layoutComponent) {
+                error(infoSink, "Interface block member layout qualifiers must match:");
+                writeTypeComparison = true;
+            }
         }
     }
 
     // Qualifiers have to (almost) match
 
     // Storage...
-    if (symbol.getQualifier().storage != unitSymbol.getQualifier().storage) {
+    if (symbol.getQualifier().storage != unitSymbol.getQualifier().storage &&
+        !((crossStage && symbol.getQualifier().storage == EvqVaryingIn && unitSymbol.getQualifier().storage == EvqVaryingOut) ||
+          (crossStage && symbol.getQualifier().storage == EvqVaryingOut && unitSymbol.getQualifier().storage == EvqVaryingIn))) {
         error(infoSink, "Storage qualifiers must match:");
         writeTypeComparison = true;
     }
@@ -597,12 +916,16 @@ void TIntermediate::mergeErrorCheck(TInfoSink& infoSink, const TIntermSymbol& sy
     }
 
     // Auxiliary and interpolation...
-    if (symbol.getQualifier().centroid  != unitSymbol.getQualifier().centroid ||
+    // "interpolation qualification (e.g., flat) and auxiliary qualification (e.g. centroid) may differ.  
+    //  These mismatches are allowed between any pair of stages ...
+    //  those provided in the fragment shader supersede those provided in previous stages."
+    if (!crossStage &&
+        (symbol.getQualifier().centroid  != unitSymbol.getQualifier().centroid ||
         symbol.getQualifier().smooth    != unitSymbol.getQualifier().smooth ||
         symbol.getQualifier().flat      != unitSymbol.getQualifier().flat ||
         symbol.getQualifier().isSample()!= unitSymbol.getQualifier().isSample() ||
         symbol.getQualifier().isPatch() != unitSymbol.getQualifier().isPatch() ||
-        symbol.getQualifier().isNonPerspective() != unitSymbol.getQualifier().isNonPerspective()) {
+        symbol.getQualifier().isNonPerspective() != unitSymbol.getQualifier().isNonPerspective())) {
         error(infoSink, "Interpolation and auxiliary storage qualifiers must match:");
         writeTypeComparison = true;
     }
@@ -1830,4 +2153,17 @@ int TIntermediate::computeBufferReferenceTypeSize(const TType& type)
     return size;
 }
 
+#ifndef GLSLANG_WEB
+bool TIntermediate::isIoResizeArray(const TType& type, EShLanguage language) {
+    return type.isArray() &&
+            ((language == EShLangGeometry    && type.getQualifier().storage == EvqVaryingIn) ||
+            (language == EShLangTessControl && type.getQualifier().storage == EvqVaryingOut &&
+                ! type.getQualifier().patch) ||
+            (language == EShLangFragment && type.getQualifier().storage == EvqVaryingIn &&
+                type.getQualifier().pervertexNV) ||
+            (language == EShLangMeshNV && type.getQualifier().storage == EvqVaryingOut &&
+                !type.getQualifier().perTaskNV));
+}
+#endif // not GLSLANG_WEB
+
 } // end namespace glslang
diff --git a/3rdparty/glslang/glslang/MachineIndependent/localintermediate.h b/3rdparty/glslang/glslang/MachineIndependent/localintermediate.h
index 9fe684a33..c9a1d8114 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/localintermediate.h
+++ b/3rdparty/glslang/glslang/MachineIndependent/localintermediate.h
@@ -293,7 +293,12 @@ public:
         useStorageBuffer(false),
         nanMinMaxClamp(false),
         depthReplacing(false),
-        uniqueId(0)
+        uniqueId(0),
+        globalUniformBlockName(""),
+        atomicCounterBlockName(""),
+        globalUniformBlockSet(TQualifier::layoutSetEnd),
+        globalUniformBlockBinding(TQualifier::layoutBindingEnd),
+        atomicCounterBlockSet(TQualifier::layoutSetEnd)
 #ifndef GLSLANG_WEB
         ,
         implicitThisName("@this"), implicitCounterName("@count"),
@@ -537,6 +542,19 @@ public:
     void addSymbolLinkageNode(TIntermAggregate*& linkage, const TSymbol&);
     TIntermAggregate* findLinkerObjects() const;
 
+    void setGlobalUniformBlockName(const char* name) { globalUniformBlockName = std::string(name); }
+    const char* getGlobalUniformBlockName() const { return globalUniformBlockName.c_str(); }
+    void setGlobalUniformSet(unsigned int set) { globalUniformBlockSet = set; }
+    unsigned int getGlobalUniformSet() const { return globalUniformBlockSet; }
+    void setGlobalUniformBinding(unsigned int binding) { globalUniformBlockBinding = binding; }
+    unsigned int getGlobalUniformBinding() const { return globalUniformBlockBinding; }
+
+    void setAtomicCounterBlockName(const char* name) { atomicCounterBlockName = std::string(name); }
+    const char* getAtomicCounterBlockName() const { return atomicCounterBlockName.c_str(); }
+    void setAtomicCounterBlockSet(unsigned int set) { atomicCounterBlockSet = set; }
+    unsigned int getAtomicCounterBlockSet() const { return atomicCounterBlockSet; }
+
+
     void setUseStorageBuffer() { useStorageBuffer = true; }
     bool usingStorageBuffer() const { return useStorageBuffer; }
     void setDepthReplacing() { depthReplacing = true; }
@@ -848,6 +866,20 @@ public:
     bool getBinaryDoubleOutput() { return binaryDoubleOutput; }
 #endif // GLSLANG_WEB
 
+    void addBlockStorageOverride(const char* nameStr, TBlockStorageClass backing)
+    {
+        std::string name(nameStr);
+        blockBackingOverrides[name] = backing;
+    }
+    TBlockStorageClass getBlockStorageOverride(const char* nameStr) const
+    {
+        std::string name = nameStr;
+        auto pos = blockBackingOverrides.find(name);
+        if (pos == blockBackingOverrides.end())
+            return EbsNone;
+        else
+            return pos->second;
+    }
 #ifdef ENABLE_HLSL
     void setHlslFunctionality1() { hlslFunctionality1 = true; }
     bool getHlslFunctionality1() const { return hlslFunctionality1; }
@@ -883,6 +915,10 @@ public:
     void merge(TInfoSink&, TIntermediate&);
     void finalCheck(TInfoSink&, bool keepUncalled);
 
+    void mergeGlobalUniformBlocks(TInfoSink& infoSink, TIntermediate& unit, bool mergeExistingOnly);
+    void mergeUniformObjects(TInfoSink& infoSink, TIntermediate& unit);
+    void checkStageIO(TInfoSink&, TIntermediate&);
+
     bool buildConvertOp(TBasicType dst, TBasicType src, TOperator& convertOp) const;
     TIntermTyped* createConversion(TBasicType convertTo, TIntermTyped* node) const;
 
@@ -906,6 +942,8 @@ public:
     static int getOffset(const TType& type, int index);
     static int getBlockSize(const TType& blockType);
     static int computeBufferReferenceTypeSize(const TType&);
+    static bool isIoResizeArray(const TType& type, EShLanguage language);
+
     bool promote(TIntermOperator*);
     void setNanMinMaxClamp(bool setting) { nanMinMaxClamp = setting; }
     bool getNanMinMaxClamp() const { return nanMinMaxClamp; }
@@ -963,9 +1001,10 @@ protected:
     void seedIdMap(TIdMaps& idMaps, long long& IdShift);
     void remapIds(const TIdMaps& idMaps, long long idShift, TIntermediate&);
     void mergeBodies(TInfoSink&, TIntermSequence& globals, const TIntermSequence& unitGlobals);
-    void mergeLinkerObjects(TInfoSink&, TIntermSequence& linkerObjects, const TIntermSequence& unitLinkerObjects);
+    void mergeLinkerObjects(TInfoSink&, TIntermSequence& linkerObjects, const TIntermSequence& unitLinkerObjects, EShLanguage);
+    void mergeBlockDefinitions(TInfoSink&, TIntermSymbol* block, TIntermSymbol* unitBlock, TIntermediate* unitRoot);
     void mergeImplicitArraySizes(TType&, const TType&);
-    void mergeErrorCheck(TInfoSink&, const TIntermSymbol&, const TIntermSymbol&, bool crossStage);
+    void mergeErrorCheck(TInfoSink&, const TIntermSymbol&, const TIntermSymbol&, EShLanguage);
     void checkCallGraphCycles(TInfoSink&);
     void checkCallGraphBodies(TInfoSink&, bool keepUncalled);
     void inOutLocationCheck(TInfoSink&);
@@ -1015,6 +1054,13 @@ protected:
     bool localSizeNotDefault[3];
     int localSizeSpecId[3];
     unsigned long long uniqueId;
+
+    std::string globalUniformBlockName;
+    std::string atomicCounterBlockName;
+    unsigned int globalUniformBlockSet;
+    unsigned int globalUniformBlockBinding;
+    unsigned int atomicCounterBlockSet;
+
 #ifndef GLSLANG_WEB
 public:
     const char* const implicitThisName;
@@ -1075,6 +1121,7 @@ protected:
     int uniformLocationBase;
     TNumericFeatures numericFeatures;
 #endif
+    std::unordered_map<std::string, TBlockStorageClass> blockBackingOverrides;
 
     std::unordered_set<int> usedConstantId; // specialization constant ids used
     std::vector<TOffsetRange> usedAtomics;  // sets of bindings used by atomic counters
diff --git a/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp b/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
index 9870a400b..9ea48c452 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
@@ -907,8 +907,8 @@ public:
             case EbtFloat16:    return GL_FLOAT16_VEC2_NV             + offset;
             case EbtInt:        return GL_INT_VEC2                    + offset;
             case EbtUint:       return GL_UNSIGNED_INT_VEC2           + offset;
-            case EbtInt64:      return GL_INT64_ARB                   + offset;
-            case EbtUint64:     return GL_UNSIGNED_INT64_ARB          + offset;
+            case EbtInt64:      return GL_INT64_VEC2_ARB              + offset;
+            case EbtUint64:     return GL_UNSIGNED_INT64_VEC2_ARB     + offset;
             case EbtBool:       return GL_BOOL_VEC2                   + offset;
             case EbtAtomicUint: return GL_UNSIGNED_INT_ATOMIC_COUNTER + offset;
             default:            return 0;
diff --git a/3rdparty/glslang/glslang/Public/ShaderLang.h b/3rdparty/glslang/glslang/Public/ShaderLang.h
index e147b0dc8..14683daf1 100644
--- a/3rdparty/glslang/glslang/Public/ShaderLang.h
+++ b/3rdparty/glslang/glslang/Public/ShaderLang.h
@@ -187,6 +187,7 @@ struct TInputLanguage {
     EShLanguage stage;        // redundant information with other input, this one overrides when not EShSourceNone
     EShClient dialect;
     int dialectVersion;       // version of client's language definition, not the client (when not EShClientNone)
+    bool vulkanRulesRelaxed;
 };
 
 struct TClient {
@@ -427,6 +428,14 @@ enum TResourceType {
     EResCount
 };
 
+enum TBlockStorageClass
+{
+    EbsUniform = 0,
+    EbsStorageBuffer,
+    EbsPushConstant,
+    EbsNone,    // not a uniform or buffer variable
+    EbsCount,
+};
 
 // Make one TShader per shader that you will link into a program. Then
 //  - provide the shader through setStrings() or setStringsWithLengths()
@@ -483,6 +492,14 @@ public:
     GLSLANG_EXPORT void setNoStorageFormat(bool useUnknownFormat);
     GLSLANG_EXPORT void setNanMinMaxClamp(bool nanMinMaxClamp);
     GLSLANG_EXPORT void setTextureSamplerTransformMode(EShTextureSamplerTransformMode mode);
+    GLSLANG_EXPORT void addBlockStorageOverride(const char* nameStr, glslang::TBlockStorageClass backing);
+
+    GLSLANG_EXPORT void setGlobalUniformBlockName(const char* name);
+    GLSLANG_EXPORT void setAtomicCounterBlockName(const char* name);
+    GLSLANG_EXPORT void setGlobalUniformSet(unsigned int set);
+    GLSLANG_EXPORT void setGlobalUniformBinding(unsigned int binding);
+    GLSLANG_EXPORT void setAtomicCounterBlockSet(unsigned int set);
+    GLSLANG_EXPORT void setAtomicCounterBlockBinding(unsigned int binding);
 
     // For setting up the environment (cleared to nothingness in the constructor).
     // These must be called so that parsing is done for the right source language and
@@ -539,6 +556,9 @@ public:
     bool getEnvTargetHlslFunctionality1() const { return false; }
 #endif
 
+    void setEnvInputVulkanRulesRelaxed() { environment.input.vulkanRulesRelaxed = true; }
+    bool getEnvInputVulkanRulesRelaxed() const { return environment.input.vulkanRulesRelaxed; }
+
     // Interface to #include handlers.
     //
     // To support #include, a client of Glslang does the following:
@@ -806,7 +826,7 @@ public:
     // Called by TSlotCollector to resolve resource locations or bindings
     virtual void reserverResourceSlot(TVarEntryInfo& ent, TInfoSink& infoSink) = 0;
     // Called by mapIO.addStage to set shader stage mask to mark a stage be added to this pipeline
-    virtual void addStage(EShLanguage stage) = 0;
+    virtual void addStage(EShLanguage stage, TIntermediate& stageIntermediate) = 0;
 };
 
 #endif // !GLSLANG_WEB && !GLSLANG_ANGLE
@@ -928,6 +948,7 @@ public:
 
 protected:
     GLSLANG_EXPORT bool linkStage(EShLanguage, EShMessages);
+    GLSLANG_EXPORT bool crossStageCheck(EShMessages);
 
     TPoolAllocator* pool;
     std::list<TShader*> stages[EShLangCount];