diff --git a/3rdparty/glslang/SPIRV/CMakeLists.txt b/3rdparty/glslang/SPIRV/CMakeLists.txt
index 90406099a..8451446db 100644
--- a/3rdparty/glslang/SPIRV/CMakeLists.txt
+++ b/3rdparty/glslang/SPIRV/CMakeLists.txt
@@ -98,6 +98,5 @@ if(ENABLE_GLSLANG_INSTALL)
 
     install(EXPORT SPIRVTargets DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake)
 
-    install(FILES ${HEADERS} ${SPVREMAP_HEADERS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/SPIRV/)
     install(FILES ${HEADERS} ${SPVREMAP_HEADERS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/glslang/SPIRV/)
 endif(ENABLE_GLSLANG_INSTALL)
diff --git a/3rdparty/glslang/SPIRV/GlslangToSpv.cpp b/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
index e41bb5de6..24203a16e 100644
--- a/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
+++ b/3rdparty/glslang/SPIRV/GlslangToSpv.cpp
@@ -1292,7 +1292,8 @@ bool IsDescriptorResource(const glslang::TType& type)
     // basically samplerXXX/subpass/sampler/texture are all included
     // if they are the global-scope-class, not the function parameter
     // (or local, if they ever exist) class.
-    if (type.getBasicType() == glslang::EbtSampler)
+    if (type.getBasicType() == glslang::EbtSampler ||
+        type.getBasicType() == glslang::EbtAccStruct)
         return type.getQualifier().isUniformOrBuffer();
 
     // None of the above.
@@ -1710,16 +1711,19 @@ void TGlslangToSpvTraverser::visitSymbol(glslang::TIntermSymbol* symbol)
     spv::Id id = getSymbolId(symbol);
 
     if (builder.isPointer(id)) {
-        // Include all "static use" and "linkage only" interface variables on the OpEntryPoint instruction
-        // Consider adding to the OpEntryPoint interface list.
-        // Only looking at structures if they have at least one member.
-        if (!symbol->getType().isStruct() || symbol->getType().getStruct()->size() > 0) {
-            spv::StorageClass sc = builder.getStorageClass(id);
-            // Before SPIR-V 1.4, we only want to include Input and Output.
-            // Starting with SPIR-V 1.4, we want all globals.
-            if ((glslangIntermediate->getSpv().spv >= glslang::EShTargetSpv_1_4 && sc != spv::StorageClassFunction) ||
-                (sc == spv::StorageClassInput || sc == spv::StorageClassOutput)) {
-                iOSet.insert(id);
+        if (!symbol->getType().getQualifier().isParamInput() &&
+            !symbol->getType().getQualifier().isParamOutput()) {
+            // Include all "static use" and "linkage only" interface variables on the OpEntryPoint instruction
+            // Consider adding to the OpEntryPoint interface list.
+            // Only looking at structures if they have at least one member.
+            if (!symbol->getType().isStruct() || symbol->getType().getStruct()->size() > 0) {
+                spv::StorageClass sc = builder.getStorageClass(id);
+                // Before SPIR-V 1.4, we only want to include Input and Output.
+                // Starting with SPIR-V 1.4, we want all globals.
+                if ((glslangIntermediate->getSpv().spv >= glslang::EShTargetSpv_1_4 && sc != spv::StorageClassFunction) ||
+                    (sc == spv::StorageClassInput || sc == spv::StorageClassOutput)) {
+                    iOSet.insert(id);
+                }
             }
         }
 
@@ -3524,7 +3528,18 @@ spv::Id TGlslangToSpvTraverser::createSpvVariable(const glslang::TIntermSymbol*
     if (glslang::IsAnonymous(name))
         name = "";
 
-    return builder.createVariable(storageClass, spvType, name);
+    spv::Id initializer = spv::NoResult;
+
+    if (node->getType().getQualifier().storage == glslang::EvqUniform &&
+        !node->getConstArray().empty()) {
+            int nextConst = 0;
+            initializer = createSpvConstantFromConstUnionArray(node->getType(),
+                                                               node->getConstArray(),
+                                                               nextConst,
+                                                               false /* specConst */);
+    }
+
+    return builder.createVariable(storageClass, spvType, name, initializer);
 }
 
 // Return type Id of the sampled type.
@@ -4388,8 +4403,10 @@ bool TGlslangToSpvTraverser::writableParam(glslang::TStorageQualifier qualifier)
     assert(qualifier == glslang::EvqIn ||
            qualifier == glslang::EvqOut ||
            qualifier == glslang::EvqInOut ||
+           qualifier == glslang::EvqUniform ||
            qualifier == glslang::EvqConstReadOnly);
-    return qualifier != glslang::EvqConstReadOnly;
+    return qualifier != glslang::EvqConstReadOnly &&
+           qualifier != glslang::EvqUniform;
 }
 
 // Is parameter pass-by-original?
@@ -8270,6 +8287,18 @@ spv::Id TGlslangToSpvTraverser::createSpvConstant(const glslang::TIntermTyped& n
 
     // We now know we have a specialization constant to build
 
+    // Extra capabilities may be needed.
+    if (node.getType().contains8BitInt())
+        builder.addCapability(spv::CapabilityInt8);
+    if (node.getType().contains16BitFloat())
+        builder.addCapability(spv::CapabilityFloat16);
+    if (node.getType().contains16BitInt())
+        builder.addCapability(spv::CapabilityInt16);
+    if (node.getType().contains64BitInt())
+        builder.addCapability(spv::CapabilityInt64);
+    if (node.getType().containsDouble())
+        builder.addCapability(spv::CapabilityFloat64);
+
     // gl_WorkGroupSize is a special case until the front-end handles hierarchical specialization constants,
     // even then, it's specialization ids are handled by special case syntax in GLSL: layout(local_size_x = ...
     if (node.getType().getQualifier().builtIn == glslang::EbvWorkGroupSize) {
@@ -8625,7 +8654,8 @@ int GetSpirvGeneratorVersion()
     // return 6; // revert version 5 change, which makes a different (new) kind of incorrect code,
                  // versions 4 and 6 each generate OpArrayLength as it has long been done
     // return 7; // GLSL volatile keyword maps to both SPIR-V decorations Volatile and Coherent
-    return 8; // switch to new dead block eliminator; use OpUnreachable
+    // return 8; // switch to new dead block eliminator; use OpUnreachable
+    return 9; // don't include opaque function parameters in OpEntryPoint global's operand list
 }
 
 // Write SPIR-V out to a binary file
diff --git a/3rdparty/glslang/StandAlone/StandAlone.cpp b/3rdparty/glslang/StandAlone/StandAlone.cpp
index a7ce53d4f..ab8d904d0 100644
--- a/3rdparty/glslang/StandAlone/StandAlone.cpp
+++ b/3rdparty/glslang/StandAlone/StandAlone.cpp
@@ -555,6 +555,12 @@ void ProcessArguments(std::vector<std::unique_ptr<glslang::TWorkItem>>& workItem
                         ReflectOptions |= EShReflectionAllBlockVariables;
                     } else if (lowerword == "reflect-unwrap-io-blocks") {
                         ReflectOptions |= EShReflectionUnwrapIOBlocks;
+                    } else if (lowerword == "reflect-all-io-variables") {
+                        ReflectOptions |= EShReflectionAllIOVariables;
+                    } else if (lowerword == "reflect-shared-std140-ubo") {
+                        ReflectOptions |= EShReflectionSharedStd140UBO;
+                    } else if (lowerword == "reflect-shared-std140-ssbo") {
+                        ReflectOptions |= EShReflectionSharedStd140SSBO;
                     } else if (lowerword == "resource-set-bindings" ||  // synonyms
                                lowerword == "resource-set-binding"  ||
                                lowerword == "rsb") {
@@ -1593,7 +1599,8 @@ void usage()
            "  --hlsl-iomap                      perform IO mapping in HLSL register space\n"
            "  --hlsl-enable-16bit-types         allow 16-bit types in SPIR-V for HLSL\n"
            "  --hlsl-dx9-compatible             interprets sampler declarations as a\n"
-           "                                    texture/sampler combo like DirectX9 would.\n"
+           "                                    texture/sampler combo like DirectX9 would,\n"
+           "                                    and recognizes DirectX9-specific semantics\n"
            "  --invert-y | --iy                 invert position.Y output in vertex shader\n"
            "  --keep-uncalled | --ku            don't eliminate uncalled functions\n"
            "  --nan-clamp                       favor non-NaN operand in min, max, and clamp\n"
diff --git a/3rdparty/glslang/glslang/Include/glslang_c_shader_types.h b/3rdparty/glslang/glslang/Include/glslang_c_shader_types.h
index 15bf1e14f..d01a115f6 100644
--- a/3rdparty/glslang/glslang/Include/glslang_c_shader_types.h
+++ b/3rdparty/glslang/glslang/Include/glslang_c_shader_types.h
@@ -164,7 +164,9 @@ typedef enum {
     GLSLANG_REFLECTION_SEPARATE_BUFFERS_BIT = (1 << 3),
     GLSLANG_REFLECTION_ALL_BLOCK_VARIABLES_BIT = (1 << 4),
     GLSLANG_REFLECTION_UNWRAP_IO_BLOCKS_BIT = (1 << 5),
-    GLSLANG_REFLECTION_SHARED_STD140_BLOCKS_BIT = (1 << 6),
+    GLSLANG_REFLECTION_ALL_IO_VARIABLES_BIT = (1 << 6),
+    GLSLANG_REFLECTION_SHARED_STD140_SSBO_BIT = (1 << 7),
+    GLSLANG_REFLECTION_SHARED_STD140_UBO_BIT = (1 << 8),
     LAST_ELEMENT_MARKER(GLSLANG_REFLECTION_COUNT),
 } glslang_reflection_options_t;
 
diff --git a/3rdparty/glslang/glslang/Include/revision.h b/3rdparty/glslang/glslang/Include/revision.h
index 693557bec..94c2dd5f7 100644
--- a/3rdparty/glslang/glslang/Include/revision.h
+++ b/3rdparty/glslang/glslang/Include/revision.h
@@ -1,3 +1,3 @@
 // This header is generated by the make-revision script.
 
-#define GLSLANG_PATCH_LEVEL 3766
+#define GLSLANG_PATCH_LEVEL 3795
diff --git a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
index a35d41a05..698006581 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/ParseHelper.cpp
@@ -5776,6 +5776,8 @@ void TParseContext::layoutTypeCheck(const TSourceLoc& loc, const TType& type)
         int repeated = intermediate.addXfbBufferOffset(type);
         if (repeated >= 0)
             error(loc, "overlapping offsets at", "xfb_offset", "offset %d in buffer %d", repeated, qualifier.layoutXfbBuffer);
+        if (type.isUnsizedArray())
+            error(loc, "unsized array", "xfb_offset", "in buffer %d", qualifier.layoutXfbBuffer);
 
         // "The offset must be a multiple of the size of the first component of the first
         // qualified variable or block member, or a compile-time error results. Further, if applied to an aggregate
@@ -7461,8 +7463,8 @@ void TParseContext::declareBlock(const TSourceLoc& loc, TTypeList& typeList, con
             arraySizesCheck(memberLoc, currentBlockQualifier, memberType.getArraySizes(), nullptr, member == typeList.size() - 1);
         if (memberQualifier.hasOffset()) {
             if (spvVersion.spv == 0) {
-                requireProfile(memberLoc, ~EEsProfile, "offset on block member");
-                profileRequires(memberLoc, ~EEsProfile, 440, E_GL_ARB_enhanced_layouts, "offset on block member");
+                profileRequires(memberLoc, ~EEsProfile, 440, E_GL_ARB_enhanced_layouts, "\"offset\" on block member");
+                profileRequires(memberLoc, EEsProfile, 300, E_GL_ARB_enhanced_layouts, "\"offset\" on block member");
             }
         }
 
diff --git a/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp b/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
index d4dad8ac3..3bf140cce 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/Versions.cpp
@@ -167,7 +167,7 @@ void TParseVersions::initializeExtensionBehavior()
 
     const extensionData exts[] = { {E_GL_EXT_ray_tracing, EShTargetSpv_1_4} };
 
-    for (int ii = 0; ii < sizeof(exts) / sizeof(exts[0]); ii++) {
+    for (size_t ii = 0; ii < sizeof(exts) / sizeof(exts[0]); ii++) {
         // Add only extensions which require > spv1.0 to save space in map
         if (exts[ii].minSpvVersion > EShTargetSpv_1_0) {
             extensionMinSpv[E_GL_EXT_ray_tracing] = exts[ii].minSpvVersion;
diff --git a/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp b/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
index 4e409e06a..fadcb1bf9 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/iomapper.cpp
@@ -309,26 +309,43 @@ struct TSymbolValidater
         TIntermSymbol* base = ent1.symbol;
         const TType& type = ent1.symbol->getType();
         const TString& name = entKey.first;
-        TString mangleName1, mangleName2;
-        type.appendMangledName(mangleName1);
         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)
+                if (inVarMaps[i] != nullptr) {
                     nextStage = static_cast<EShLanguage>(i);
+                    break;
+                }
             }
         }
+
+        if (type.getQualifier().isArrayedIo(stage)) {
+            TType subType(type, 0);
+            subType.appendMangledName(mangleName1);
+        } else {
+            type.appendMangledName(mangleName1);
+        }
+
         if (base->getQualifier().storage == EvqVaryingIn) {
             // validate stage in;
             if (preStage == EShLangCount)
                 return;
+            if (name == "gl_PerVertex")
+                return;
             if (outVarMaps[preStage] != nullptr) {
                 auto ent2 = outVarMaps[preStage]->find(name);
                 if (ent2 != outVarMaps[preStage]->end()) {
-                    ent2->second.symbol->getType().appendMangledName(mangleName2);
+                    if (ent2->second.symbol->getType().getQualifier().isArrayedIo(preStage)) {
+                        TType subType(ent2->second.symbol->getType(), 0);
+                        subType.appendMangledName(mangleName2);
+                    }
+                    else {
+                        ent2->second.symbol->getType().appendMangledName(mangleName2);
+                    }
                     if (mangleName1 == mangleName2)
                         return;
                     else {
@@ -343,10 +360,18 @@ struct TSymbolValidater
             // validate stage out;
             if (nextStage == EShLangCount)
                 return;
+            if (name == "gl_PerVertex")
+                return;
             if (outVarMaps[nextStage] != nullptr) {
                 auto ent2 = inVarMaps[nextStage]->find(name);
                 if (ent2 != inVarMaps[nextStage]->end()) {
-                    ent2->second.symbol->getType().appendMangledName(mangleName2);
+                    if (ent2->second.symbol->getType().getQualifier().isArrayedIo(nextStage)) {
+                        TType subType(ent2->second.symbol->getType(), 0);
+                        subType.appendMangledName(mangleName2);
+                    }
+                    else {
+                        ent2->second.symbol->getType().appendMangledName(mangleName2);
+                    }
                     if (mangleName1 == mangleName2)
                         return;
                     else {
diff --git a/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp b/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
index 045d45e3c..f0bede53a 100755
--- a/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/linkValidate.cpp
@@ -1362,9 +1362,9 @@ unsigned int TIntermediate::computeTypeXfbSize(const TType& type, bool& contains
     // that component's size.  Aggregate types are flattened down to the component
     // level to get this sequence of components."
 
-    if (type.isArray()) {
+    if (type.isSizedArray()) {
         // TODO: perf: this can be flattened by using getCumulativeArraySize(), and a deref that discards all arrayness
-        assert(type.isSizedArray());
+        // Unsized array use to xfb should be a compile error.
         TType elementType(type, 0);
         return type.getOuterArraySize() * computeTypeXfbSize(elementType, contains64BitType, contains16BitType, contains16BitType);
     }
@@ -1550,7 +1550,9 @@ int TIntermediate::getBaseAlignment(const TType& type, int& size, int& stride, T
         RoundToPow2(size, alignment);
         stride = size;  // uses full matrix size for stride of an array of matrices (not quite what rule 6/8, but what's expected)
                         // uses the assumption for rule 10 in the comment above
-        size = stride * type.getOuterArraySize();
+        // use one element to represent the last member of SSBO which is unsized array
+        int arraySize = (type.isUnsizedArray() && (type.getOuterArraySize() == 0)) ? 1 : type.getOuterArraySize();
+        size = stride * arraySize;
         return alignment;
     }
 
diff --git a/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp b/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
index 55caa1f44..03d6fe145 100644
--- a/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
+++ b/3rdparty/glslang/glslang/MachineIndependent/reflection.cpp
@@ -107,22 +107,13 @@ public:
                 else
                     baseName = "";
 
-                if (base.getType().isArray()) {
-                    TType derefType(base.getType(), 0);
-
-                    assert(!anonymous);
-                    for (int e = 0; e < base.getType().getCumulativeArraySize(); ++e)
-                        blockIndex = addBlockName(blockName + "[" + String(e) + "]", derefType,
-                            intermediate.getBlockSize(base.getType()));
-                }
-                else
-                    blockIndex = addBlockName(blockName, base.getType(), intermediate.getBlockSize(base.getType()));
+                blockIndex = addBlockName(blockName, base.getType(), intermediate.getBlockSize(base.getType()));
             }
 
             // Use a degenerate (empty) set of dereferences to immediately put as at the end of
             // the dereference change expected by blowUpActiveAggregate.
-            blowUpActiveAggregate(base.getType(), baseName, derefs, derefs.end(), offset, blockIndex, 0, 0,
-                                  base.getQualifier().storage, updateStageMasks);
+            blowUpActiveAggregate(base.getType(), baseName, derefs, derefs.end(), offset, blockIndex, 0, -1, 0,
+                                    base.getQualifier().storage, updateStageMasks);
         }
     }
 
@@ -259,7 +250,7 @@ public:
     // A value of 0 for arraySize will mean to use the full array's size.
     void blowUpActiveAggregate(const TType& baseType, const TString& baseName, const TList<TIntermBinary*>& derefs,
                                TList<TIntermBinary*>::const_iterator deref, int offset, int blockIndex, int arraySize,
-                               int topLevelArrayStride, TStorageQualifier baseStorage, bool active)
+                               int topLevelArraySize, int topLevelArrayStride, TStorageQualifier baseStorage, bool active)
     {
         // when strictArraySuffix is enabled, we closely follow the rules from ARB_program_interface_query.
         // Broadly:
@@ -288,14 +279,15 @@ public:
                 // Visit all the indices of this array, and for each one add on the remaining dereferencing
                 for (int i = 0; i < std::max(visitNode->getLeft()->getType().getOuterArraySize(), 1); ++i) {
                     TString newBaseName = name;
-                    if (strictArraySuffix && blockParent)
+                    if (terminalType->getBasicType() == EbtBlock) {}
+                    else if (strictArraySuffix && blockParent)
                         newBaseName.append(TString("[0]"));
                     else if (strictArraySuffix || baseType.getBasicType() != EbtBlock)
                         newBaseName.append(TString("[") + String(i) + "]");
                     TList<TIntermBinary*>::const_iterator nextDeref = deref;
                     ++nextDeref;
                     blowUpActiveAggregate(*terminalType, newBaseName, derefs, nextDeref, offset, blockIndex, arraySize,
-                                          topLevelArrayStride, baseStorage, active);
+                                          topLevelArraySize, topLevelArrayStride, baseStorage, active);
 
                     if (offset >= 0)
                         offset += stride;
@@ -308,9 +300,10 @@ public:
                 int stride = getArrayStride(baseType, visitNode->getLeft()->getType());
 
                 index = visitNode->getRight()->getAsConstantUnion()->getConstArray()[0].getIConst();
-                if (strictArraySuffix && blockParent) {
+                if (terminalType->getBasicType() == EbtBlock) {}
+                else if (strictArraySuffix && blockParent)
                     name.append(TString("[0]"));
-                } else if (strictArraySuffix || baseType.getBasicType() != EbtBlock) {
+                else if (strictArraySuffix || baseType.getBasicType() != EbtBlock) {
                     name.append(TString("[") + String(index) + "]");
 
                     if (offset >= 0)
@@ -320,7 +313,10 @@ public:
                 if (topLevelArrayStride == 0)
                     topLevelArrayStride = stride;
 
-                blockParent = false;
+                // expand top-level arrays in blocks with [0] suffix
+                if (topLevelArrayStride != 0 && visitNode->getLeft()->getType().isArray()) {
+                    blockParent = false;
+                }
                 break;
             }
             case EOpIndexDirectStruct:
@@ -330,6 +326,12 @@ public:
                 if (name.size() > 0)
                     name.append(".");
                 name.append((*visitNode->getLeft()->getType().getStruct())[index].type->getFieldName());
+
+                // expand non top-level arrays with [x] suffix
+                if (visitNode->getLeft()->getType().getBasicType() != EbtBlock && terminalType->isArray())
+                {
+                    blockParent = false;
+                }
                 break;
             default:
                 break;
@@ -349,14 +351,16 @@ public:
                 if (offset >= 0)
                     stride = getArrayStride(baseType, *terminalType);
 
-                if (topLevelArrayStride == 0)
-                    topLevelArrayStride = stride;
-
                 int arrayIterateSize = std::max(terminalType->getOuterArraySize(), 1);
 
                 // for top-level arrays in blocks, only expand [0] to avoid explosion of items
-                if (strictArraySuffix && blockParent)
+                if ((strictArraySuffix && blockParent) ||
+                    ((topLevelArraySize == arrayIterateSize) && (topLevelArrayStride == 0))) {
                     arrayIterateSize = 1;
+                }
+
+                if (topLevelArrayStride == 0)
+                    topLevelArrayStride = stride;
 
                 for (int i = 0; i < arrayIterateSize; ++i) {
                     TString newBaseName = name;
@@ -367,7 +371,7 @@ public:
                         offset = baseOffset + stride * i;
 
                     blowUpActiveAggregate(derefType, newBaseName, derefs, derefs.end(), offset, blockIndex, 0,
-                                          topLevelArrayStride, baseStorage, active);
+                                          topLevelArraySize, topLevelArrayStride, baseStorage, active);
                 }
             } else {
                 // Visit all members of this aggregate, and for each one,
@@ -396,8 +400,31 @@ public:
                         arrayStride = getArrayStride(baseType, derefType);
                     }
 
-                    blowUpActiveAggregate(derefType, newBaseName, derefs, derefs.end(), offset, blockIndex, 0,
-                                          arrayStride, baseStorage, active);
+                    if (topLevelArraySize == -1 && arrayStride == 0 && blockParent)
+                        topLevelArraySize = 1;
+
+                    if (strictArraySuffix && blockParent) {
+                        // if this member is an array, store the top-level array stride but start the explosion from
+                        // the inner struct type.
+                        if (derefType.isArray() && derefType.isStruct()) {
+                            newBaseName.append("[0]");
+                            auto dimSize = derefType.isUnsizedArray() ? 0 : derefType.getArraySizes()->getDimSize(0);
+                            blowUpActiveAggregate(TType(derefType, 0), newBaseName, derefs, derefs.end(), memberOffsets[i],
+                                blockIndex, 0, dimSize, arrayStride, terminalType->getQualifier().storage, false);
+                        }
+                        else if (derefType.isArray()) {
+                            auto dimSize = derefType.isUnsizedArray() ? 0 : derefType.getArraySizes()->getDimSize(0);
+                            blowUpActiveAggregate(derefType, newBaseName, derefs, derefs.end(), memberOffsets[i], blockIndex,
+                                0, dimSize, 0, terminalType->getQualifier().storage, false);
+                        }
+                        else {
+                            blowUpActiveAggregate(derefType, newBaseName, derefs, derefs.end(), memberOffsets[i], blockIndex,
+                                0, 1, 0, terminalType->getQualifier().storage, false);
+                        }
+                    } else {
+                        blowUpActiveAggregate(derefType, newBaseName, derefs, derefs.end(), offset, blockIndex, 0,
+                                              topLevelArraySize, arrayStride, baseStorage, active);
+                    }
                 }
             }
 
@@ -433,6 +460,7 @@ public:
             if ((reflection.options & EShReflectionSeparateBuffers) && terminalType->isAtomic())
                 reflection.atomicCounterUniformIndices.push_back(uniformIndex);
 
+            variables.back().topLevelArraySize = topLevelArraySize;
             variables.back().topLevelArrayStride = topLevelArrayStride;
             
             if ((reflection.options & EShReflectionAllBlockVariables) && active) {
@@ -564,65 +592,17 @@ public:
             if (! anonymous)
                 baseName = blockName;
 
-            if (base->getType().isArray()) {
-                TType derefType(base->getType(), 0);
-
-                assert(! anonymous);
-                for (int e = 0; e < base->getType().getCumulativeArraySize(); ++e)
-                    blockIndex = addBlockName(blockName + "[" + String(e) + "]", derefType,
-                                              intermediate.getBlockSize(base->getType()));
-                baseName.append(TString("[0]"));
-            } else
-                blockIndex = addBlockName(blockName, base->getType(), intermediate.getBlockSize(base->getType()));
+            blockIndex = addBlockName(blockName, base->getType(), intermediate.getBlockSize(base->getType()));
 
             if (reflection.options & EShReflectionAllBlockVariables) {
                 // Use a degenerate (empty) set of dereferences to immediately put as at the end of
                 // the dereference change expected by blowUpActiveAggregate.
                 TList<TIntermBinary*> derefs;
 
-                // because we don't have any derefs, the first thing blowUpActiveAggregate will do is iterate over each
-                // member in the struct definition. This will lose any information about whether the parent was a buffer
-                // block. So if we're using strict array rules which don't expand the first child of a buffer block we
-                // instead iterate over the children here.
-                const bool strictArraySuffix = (reflection.options & EShReflectionStrictArraySuffix);
-                bool blockParent = (base->getType().getBasicType() == EbtBlock && base->getQualifier().storage == EvqBuffer);
-
-                if (strictArraySuffix && blockParent) {
-                    TType structDerefType(base->getType(), 0);
-
-                    const TType &structType = base->getType().isArray() ? structDerefType : base->getType();
-                    const TTypeList& typeList = *structType.getStruct();
-
-                    TVector<int> memberOffsets;
-
-                    memberOffsets.resize(typeList.size());
-                    getOffsets(structType, memberOffsets);
-
-                    for (int i = 0; i < (int)typeList.size(); ++i) {
-                        TType derefType(structType, i);
-                        TString name = baseName;
-                        if (name.size() > 0)
-                            name.append(".");
-                        name.append(typeList[i].type->getFieldName());
-
-                        // if this member is an array, store the top-level array stride but start the explosion from
-                        // the inner struct type.
-                        if (derefType.isArray() && derefType.isStruct()) {
-                            name.append("[0]");
-                            blowUpActiveAggregate(TType(derefType, 0), name, derefs, derefs.end(), memberOffsets[i],
-                                                  blockIndex, 0, getArrayStride(structType, derefType),
-                                                  base->getQualifier().storage, false);
-                        } else {
-                            blowUpActiveAggregate(derefType, name, derefs, derefs.end(), memberOffsets[i], blockIndex,
-                                                  0, 0, base->getQualifier().storage, false);
-                        }
-                    }
-                } else {
-                    // otherwise - if we're not using strict array suffix rules, or this isn't a block so we are
-                    // expanding root arrays anyway, just start the iteration from the base block type.
-                    blowUpActiveAggregate(base->getType(), baseName, derefs, derefs.end(), 0, blockIndex, 0, 0,
+                // otherwise - if we're not using strict array suffix rules, or this isn't a block so we are
+                // expanding root arrays anyway, just start the iteration from the base block type.
+                blowUpActiveAggregate(base->getType(), baseName, derefs, derefs.end(), 0, blockIndex, 0, -1, 0,
                                           base->getQualifier().storage, false);
-                }
             }
         }
 
@@ -653,31 +633,37 @@ public:
             else
                 baseName = base->getName();
         }
-        blowUpActiveAggregate(base->getType(), baseName, derefs, derefs.begin(), offset, blockIndex, arraySize, 0,
+        blowUpActiveAggregate(base->getType(), baseName, derefs, derefs.begin(), offset, blockIndex, arraySize, -1, 0,
                               base->getQualifier().storage, true);
     }
 
     int addBlockName(const TString& name, const TType& type, int size)
     {
-        TReflection::TMapIndexToReflection& blocks = reflection.GetBlockMapForStorage(type.getQualifier().storage);
-
         int blockIndex;
-        TReflection::TNameToIndex::const_iterator it = reflection.nameToIndex.find(name.c_str());
-        if (reflection.nameToIndex.find(name.c_str()) == reflection.nameToIndex.end()) {
-            blockIndex = (int)blocks.size();
-            reflection.nameToIndex[name.c_str()] = blockIndex;
-            blocks.push_back(TObjectReflection(name.c_str(), type, -1, -1, size, -1));
+        if (type.isArray()) {
+            TType derefType(type, 0);
+            for (int e = 0; e < type.getOuterArraySize(); ++e) {
+                int memberBlockIndex = addBlockName(name + "[" + String(e) + "]", derefType, size);
+                if (e == 0)
+                    blockIndex = memberBlockIndex;
+            }
+        } else {
+            TReflection::TMapIndexToReflection& blocks = reflection.GetBlockMapForStorage(type.getQualifier().storage);
 
-            blocks.back().numMembers = countAggregateMembers(type);
+            TReflection::TNameToIndex::const_iterator it = reflection.nameToIndex.find(name.c_str());
+            if (reflection.nameToIndex.find(name.c_str()) == reflection.nameToIndex.end()) {
+                blockIndex = (int)blocks.size();
+                reflection.nameToIndex[name.c_str()] = blockIndex;
+                blocks.push_back(TObjectReflection(name.c_str(), type, -1, -1, size, blockIndex));
+
+                blocks.back().numMembers = countAggregateMembers(type);
 
-            if (updateStageMasks) {
                 EShLanguageMask& stages = blocks.back().stages;
                 stages = static_cast<EShLanguageMask>(stages | 1 << intermediate.getStage());
             }
-        } else {
-            blockIndex = it->second;
+            else {
+                blockIndex = it->second;
 
-            if (updateStageMasks) {
                 EShLanguageMask& stages = blocks[blockIndex].stages;
                 stages = static_cast<EShLanguageMask>(stages | 1 << intermediate.getStage());
             }
@@ -1064,7 +1050,7 @@ void TReflectionTraverser::visitSymbol(TIntermSymbol* base)
 {
     if (base->getQualifier().storage == EvqUniform) {
         if (base->getBasicType() == EbtBlock) {
-            if (reflection.options & EShReflectionSharedStd140Blocks) {
+            if (reflection.options & EShReflectionSharedStd140UBO) {
                 addUniform(*base);
             }
         } else {
@@ -1072,6 +1058,13 @@ void TReflectionTraverser::visitSymbol(TIntermSymbol* base)
         }
     }
 
+    // #TODO add std140/layout active rules for ssbo, same with ubo.
+    // Storage buffer blocks will be collected and expanding in this part.
+    if((reflection.options & EShReflectionSharedStd140SSBO) &&
+       (base->getQualifier().storage == EvqBuffer && base->getBasicType() == EbtBlock &&
+        (base->getQualifier().layoutPacking == ElpStd140 || base->getQualifier().layoutPacking == ElpShared)))
+        addUniform(*base);
+
     if ((intermediate.getStage() == reflection.firstStage && base->getQualifier().isPipeInput()) ||
         (intermediate.getStage() == reflection.lastStage && base->getQualifier().isPipeOutput()))
         addPipeIOVariable(*base);
@@ -1182,15 +1175,23 @@ bool TReflection::addStage(EShLanguage stage, const TIntermediate& intermediate)
                 TIntermAggregate* linkerObjects = sequnence->getAsAggregate();
                 for (auto& sequnence : linkerObjects->getSequence()) {
                     auto pNode = sequnence->getAsSymbolNode();
-                    if (pNode != nullptr && pNode->getQualifier().storage == EvqUniform &&
-                        (options & EShReflectionSharedStd140Blocks)) {
-                        if (pNode->getBasicType() == EbtBlock) {
+                    if (pNode != nullptr) {
+                        if ((pNode->getQualifier().storage == EvqUniform &&
+                            (options & EShReflectionSharedStd140UBO)) ||
+                           (pNode->getQualifier().storage == EvqBuffer &&
+                            (options & EShReflectionSharedStd140SSBO))) {
                             // collect std140 and shared uniform block form AST
-                            if (pNode->getQualifier().layoutPacking == ElpStd140 ||
-                                pNode->getQualifier().layoutPacking == ElpShared) {
-                                pNode->traverse(&it);
+                            if ((pNode->getBasicType() == EbtBlock) &&
+                                ((pNode->getQualifier().layoutPacking == ElpStd140) ||
+                                 (pNode->getQualifier().layoutPacking == ElpShared))) {
+                                   pNode->traverse(&it);
                             }
                         }
+                        else if ((options & EShReflectionAllIOVariables) &&
+                            (pNode->getQualifier().isPipeInput() || pNode->getQualifier().isPipeOutput()))
+                        {
+                            pNode->traverse(&it);
+                        }
                     }
                 }
             } else {
diff --git a/3rdparty/glslang/glslang/Public/ShaderLang.h b/3rdparty/glslang/glslang/Public/ShaderLang.h
index 5b63ff7b6..7905b3d16 100644
--- a/3rdparty/glslang/glslang/Public/ShaderLang.h
+++ b/3rdparty/glslang/glslang/Public/ShaderLang.h
@@ -255,7 +255,7 @@ enum EShMessages : unsigned {
     EShMsgDebugInfo        = (1 << 10), // save debug information
     EShMsgHlslEnable16BitTypes  = (1 << 11), // enable use of 16-bit types in SPIR-V for HLSL
     EShMsgHlslLegalization  = (1 << 12), // enable HLSL Legalization messages
-    EShMsgHlslDX9Compatible = (1 << 13), // enable HLSL DX9 compatible mode (right now only for samplers)
+    EShMsgHlslDX9Compatible = (1 << 13), // enable HLSL DX9 compatible mode (for samplers and semantics)
     EShMsgBuiltinSymbolTable = (1 << 14), // print the builtin symbol table
     LAST_ELEMENT_MARKER(EShMsgCount),
 };
@@ -271,7 +271,9 @@ typedef enum {
     EShReflectionSeparateBuffers    = (1 << 3), // buffer variables and buffer blocks are reflected separately
     EShReflectionAllBlockVariables  = (1 << 4), // reflect all variables in blocks, even if they are inactive
     EShReflectionUnwrapIOBlocks     = (1 << 5), // unwrap input/output blocks the same as with uniform blocks
-    EShReflectionSharedStd140Blocks = (1 << 6), // Apply std140/shared rules for ubo to ssbo
+    EShReflectionAllIOVariables     = (1 << 6), // reflect all input/output variables, even if they are inactive
+    EShReflectionSharedStd140SSBO   = (1 << 7), // Apply std140/shared rules for ubo to ssbo
+    EShReflectionSharedStd140UBO    = (1 << 8), // Apply std140/shared rules for ubo to ssbo
     LAST_ELEMENT_MARKER(EShReflectionCount),
 } EShReflectionOptions;
 
@@ -696,6 +698,7 @@ public:
     int counterIndex;
     int numMembers;
     int arrayStride;            // stride of an array variable
+    int topLevelArraySize;      // size of the top-level variable in a storage buffer member
     int topLevelArrayStride;    // stride of the top-level variable in a storage buffer member
     EShLanguageMask stages;
 
diff --git a/3rdparty/glslang/hlsl/hlslGrammar.cpp b/3rdparty/glslang/hlsl/hlslGrammar.cpp
index 47ced29c0..5bfc53fe5 100644
--- a/3rdparty/glslang/hlsl/hlslGrammar.cpp
+++ b/3rdparty/glslang/hlsl/hlslGrammar.cpp
@@ -382,6 +382,16 @@ bool HlslGrammar::acceptDeclaration(TIntermNode*& nodeList)
     if (forbidDeclarators)
         return true;
 
+    // Check if there are invalid in/out qualifiers
+    switch (declaredType.getQualifier().storage) {
+    case EvqIn:
+    case EvqOut:
+    case EvqInOut:
+        parseContext.error(token.loc, "in/out qualifiers are only valid on parameters", token.string->c_str(), "");
+    default:
+        break;
+    }
+
     // declarator_list
     //    : declarator
     //         : identifier
@@ -697,7 +707,9 @@ bool HlslGrammar::acceptQualifier(TQualifier& qualifier)
             qualifier.noContraction = true;
             break;
         case EHTokIn:
-            qualifier.storage = (qualifier.storage == EvqOut) ? EvqInOut : EvqIn;
+            if (qualifier.storage != EvqUniform) {
+                qualifier.storage = (qualifier.storage == EvqOut) ? EvqInOut : EvqIn;
+            }
             break;
         case EHTokOut:
             qualifier.storage = (qualifier.storage == EvqIn) ? EvqInOut : EvqOut;
diff --git a/3rdparty/glslang/hlsl/hlslParseHelper.cpp b/3rdparty/glslang/hlsl/hlslParseHelper.cpp
index 87ca4c8bf..18ee4c460 100755
--- a/3rdparty/glslang/hlsl/hlslParseHelper.cpp
+++ b/3rdparty/glslang/hlsl/hlslParseHelper.cpp
@@ -2111,6 +2111,23 @@ TIntermNode* HlslParseContext::transformEntryPoint(const TSourceLoc& loc, TFunct
                 makeVariableInOut(*(*it));
     }
 
+    // Add uniform parameters to the $Global uniform block.
+    TVector<TVariable*> opaque_uniforms;
+    for (int i = 0; i < userFunction.getParamCount(); i++) {
+        TType& paramType = *userFunction[i].type;
+        TString& paramName = *userFunction[i].name;
+        if (paramType.getQualifier().storage == EvqUniform) {
+            if (!paramType.containsOpaque()) {
+                // Add it to the global uniform block.
+                growGlobalUniformBlock(loc, paramType, paramName);
+            } else {
+                // Declare it as a separate variable.
+                TVariable *var = makeInternalVariable(paramName.c_str(), paramType);
+                opaque_uniforms.push_back(var);
+            }
+        }
+    }
+
     // Synthesize the call
 
     pushScope(); // matches the one in handleFunctionBody()
@@ -2131,6 +2148,7 @@ TIntermNode* HlslParseContext::transformEntryPoint(const TSourceLoc& loc, TFunct
     TVector<TVariable*> argVars;
     TIntermAggregate* synthBody = new TIntermAggregate();
     auto inputIt = inputs.begin();
+    auto opaqueUniformIt = opaque_uniforms.begin();
     TIntermTyped* callingArgs = nullptr;
 
     for (int i = 0; i < userFunction.getParamCount(); i++) {
@@ -2149,6 +2167,17 @@ TIntermNode* HlslParseContext::transformEntryPoint(const TSourceLoc& loc, TFunct
                                                                intermediate.addSymbol(**inputIt)));
             inputIt++;
         }
+        if (param.type->getQualifier().storage == EvqUniform) {
+            if (!param.type->containsOpaque()) {
+                // Look it up in the $Global uniform block.
+                intermediate.growAggregate(synthBody, handleAssign(loc, EOpAssign, arg,
+                                                                   handleVariable(loc, param.name)));
+            } else {
+                intermediate.growAggregate(synthBody, handleAssign(loc, EOpAssign, arg,
+                                                                   intermediate.addSymbol(**opaqueUniformIt)));
+                ++opaqueUniformIt;
+            }
+        }
     }
 
     // Call
@@ -6100,6 +6129,32 @@ void HlslParseContext::handleSemantic(TSourceLoc loc, TQualifier& qualifier, TBu
         return semanticNum;
     };
 
+    if (builtIn == EbvNone && hlslDX9Compatible()) {
+        if (language == EShLangVertex) {
+            if (qualifier.isParamOutput()) {
+                if (upperCase == "POSITION") {
+                    builtIn = EbvPosition;
+                }
+                if (upperCase == "PSIZE") {
+                    builtIn = EbvPointSize;
+                }
+            }
+        } else if (language == EShLangFragment) {
+            if (qualifier.isParamInput() && upperCase == "VPOS") {
+                builtIn = EbvFragCoord;
+            }
+            if (qualifier.isParamOutput()) {
+                if (upperCase.compare(0, 5, "COLOR") == 0) {
+                    qualifier.layoutLocation = getSemanticNumber(upperCase, 0, nullptr);
+                    nextOutLocation = std::max(nextOutLocation, qualifier.layoutLocation + 1u);
+                }
+                if (upperCase == "DEPTH") {
+                    builtIn = EbvFragDepth;
+                }
+            }
+        }
+    }
+
     switch(builtIn) {
     case EbvNone:
         // Get location numbers from fragment outputs, instead of
@@ -6914,7 +6969,6 @@ void HlslParseContext::paramFix(TType& type)
         type.getQualifier().storage = EvqConstReadOnly;
         break;
     case EvqGlobal:
-    case EvqUniform:
     case EvqTemporary:
         type.getQualifier().storage = EvqIn;
         break;