Collection of updates from Unity internal repo at changeset 17a9acd02335496aff4e2eec1b2685085c9b5b0d

include/hlslcc.h

@@ -43,6 +43,8 @@ typedef struct GlExtensions {
 	uint32_t ARB_explicit_attrib_location : 1;
 	uint32_t ARB_explicit_uniform_location : 1;
 	uint32_t ARB_shading_language_420pack : 1;
+	uint32_t OVR_multiview : 1;
+	uint32_t EXT_shader_framebuffer_fetch : 1;
 } GlExtensions;
 
 #include "ShaderInfo.h"
@@ -197,13 +199,17 @@ private:
 		}
 	}
 
-
+	typedef std::map<std::string, uint32_t> SpecializationConstantMap;
+	SpecializationConstantMap m_SpecConstantMap;
+	uint32_t m_NextSpecID;
 
 public:
 	GLSLCrossDependencyData()
 		: eTessPartitioning(),
 		eTessOutPrim(),
-		ui32ProgramStages(0)
+		ui32ProgramStages(0),
+		m_ExtBlendModes(),
+		m_NextSpecID(0)
 	{ 
 		memset(nextAvailableVaryingLocation, 0, sizeof(nextAvailableVaryingLocation));
 		memset(m_NextAvailableVulkanResourceBinding, 0, sizeof(m_NextAvailableVulkanResourceBinding));
@@ -289,6 +295,8 @@ public:
 	// Needed so we can construct proper shader input and output names
 	uint32_t ui32ProgramStages;
 
+	std::vector<std::string> m_ExtBlendModes; // The blend modes (from KHR_blend_equation_advanced) requested for this shader. See ext spec for list.
+	
 	inline INTERPOLATION_MODE GetInterpolationMode(uint32_t regNo)
 	{
 		if (regNo >= pixelInterpolation.size())
@@ -313,9 +321,21 @@ public:
 			varyingLocationsMap[i].clear();
 			nextAvailableVaryingLocation[i] = 0;
 		}
+		m_NextSpecID = 0;
+		m_SpecConstantMap.clear();
 	}
 
+	// Retrieve or allocate a layout slot for Vulkan specialization constant
+	inline uint32_t GetSpecializationConstantSlot(const std::string &name)
+	{
+		SpecializationConstantMap::iterator itr = m_SpecConstantMap.find(name);
+		if (itr != m_SpecConstantMap.end())
+			return itr->second;
 
+		m_SpecConstantMap.insert(std::make_pair(std::string(name), m_NextSpecID));
+
+		return m_NextSpecID++;
+	}
 };
 
 struct GLSLShader
@@ -339,16 +359,17 @@ public:
 
 	virtual void OnInputBinding(const std::string &name, int bindIndex) {}
 
+	// Returns false if this constant buffer is not needed for this shader. This info can be used for pruning unused 
+	// constant buffers and vars from compute shaders where we need broader context than a single kernel to know 
+	// if something can be dropped, as the constant buffers are shared between all kernels in a .compute file.
 	virtual bool OnConstantBuffer(const std::string &name, size_t bufferSize, size_t memberCount) { return true; }
 
+	// Returns false if this constant var is not needed for this shader. See above.
 	virtual bool OnConstant(const std::string &name, int bindIndex, SHADER_VARIABLE_TYPE cType, int rows, int cols, bool isMatrix, int arraySize) { return true; }
 
 	virtual void OnConstantBufferBinding(const std::string &name, int bindIndex) {}
-
-	virtual void OnTextureBinding(const std::string &name, int bindIndex, HLSLCC_TEX_DIMENSION dim, bool isUAV) {}
-
+	virtual void OnTextureBinding(const std::string &name, int bindIndex, int samplerIndex, HLSLCC_TEX_DIMENSION dim, bool isUAV) {}
 	virtual void OnBufferBinding(const std::string &name, int bindIndex, bool isUAV) {}
-	
 	virtual void OnThreadGroupSize(unsigned int xSize, unsigned int ySize, unsigned int zSize) {}
 };
 
@@ -423,6 +444,16 @@ static const unsigned int HLSLCC_FLAG_VULKAN_BINDINGS = 0x40000;
 // If set, metal output will use linear sampler for shadow compares, otherwise point sampler.
 static const unsigned int HLSLCC_FLAG_METAL_SHADOW_SAMPLER_LINEAR = 0x80000;
 
+// If set, emits for NVN, the Nvidia-provided graphics API for Nintendo Switch.
+static const unsigned int HLSLCC_FLAG_NVN_TARGET = 0x100000; 
+
+// If set, and generating Vulkan shaders, attempts to detect static branching and transforms them into specialization constants
+static const unsigned int HLSLCC_FLAG_VULKAN_SPECIALIZATION_CONSTANTS = 0x200000;
+
+// If set, this shader uses the GLSL extension EXT_shader_framebuffer_fetch
+static const unsigned int HLSLCC_FLAG_SHADER_FRAMEBUFFER_FETCH = 0x400000;
+
+
 #ifdef __cplusplus
 extern "C" {
 #endif

src/DataTypeAnalysis.cpp

@@ -250,11 +250,10 @@ void HLSLcc::DataTypeAnalysis::SetDataTypes(HLSLCrossCompilerContext* psContext,
 		// Only ever to int->float promotion (or int->uint), never the other way around
 		for (i = 0; i < (uint32_t)instructions.size(); ++i, psInst++)
 		{
-			int k = 0;
 			if (psInst->ui32NumOperands == 0)
 				continue;
 #ifdef _DEBUG
-			for (k = 0; k < (int)psInst->ui32NumOperands; k++)
+			for (int k = 0; k < (int)psInst->ui32NumOperands; k++)
 			{
 				if (psInst->asOperands[k].eType == OPERAND_TYPE_TEMP)
 				{

src/HLSLCrossCompilerContext.cpp

@@ -95,6 +95,14 @@ void HLSLCrossCompilerContext::AddIndentation()
 	}
 }
 
+void HLSLCrossCompilerContext::RequireExtension(const std::string &extName)
+{
+	if (m_EnabledExtensions.find(extName) != m_EnabledExtensions.end())
+		return;
+
+	m_EnabledExtensions.insert(extName);
+	bformata(extensions, "#extension %s : require\n", extName.c_str());
+}
 
 std::string HLSLCrossCompilerContext::GetDeclaredInputName(const Operand* psOperand, int *piRebase, int iIgnoreRedirect, uint32_t *puiIgnoreSwizzle) const
 {
@@ -186,7 +194,13 @@ std::string HLSLCrossCompilerContext::GetDeclaredOutputName(const Operand* psOpe
 	std::string res = "";
 	if (psTranslator->TranslateSystemValue(psOperand, psOut, res, puiIgnoreSwizzle, psShader->aIndexedOutput[regSpace][psOperand->ui32RegisterNumber], false))
 	{
-		if (psShader->eTargetLanguage == LANG_METAL && (iIgnoreRedirect == 0))
+		// HACK: i couldnt find better way to handle it
+		// clip planes will always have interim variable, as HLSL operates on float4 but we need to size output accordingly with actual planes count
+		// for some reason TranslateSystemValue return *outSkipPrefix = true for ALL system vars and then we simply ignore it here
+		const bool isClipPlanes = psOut && psOut->eSystemValueType == NAME_CLIP_DISTANCE;
+
+		if (psShader->eTargetLanguage == LANG_METAL && (iIgnoreRedirect == 0) && !isClipPlanes)
+
 			return outputPrefix + res;
 		else
 			return res;
@@ -247,6 +261,13 @@ bool HLSLCrossCompilerContext::OutputNeedsDeclaring(const Operand* psOperand, co
 		{
 			psShader->acOutputDeclared[regSpace][startIndex + offset] |= compMask;
 		}
+
+		if (psSignature && (psSignature->semanticName == "PSIZE") && (psShader->eTargetLanguage != LANG_METAL))
+		{
+			// gl_PointSize, doesn't need declaring. TODO: Metal doesn't have pointsize at all?
+			return false;
+		}
+
 		return true;
 	}
 

src/HLSLccToolkit.cpp

@@ -402,11 +402,13 @@ namespace HLSLcc
 			case RESOURCE_DIMENSION_TEXTURE1D:
 				return 1;
 			case RESOURCE_DIMENSION_TEXTURE2D:
+			case RESOURCE_DIMENSION_TEXTURE2DMS:
 			case RESOURCE_DIMENSION_TEXTURE1DARRAY:
 			case RESOURCE_DIMENSION_TEXTURECUBE:
 				return 2;
 			case RESOURCE_DIMENSION_TEXTURE3D:
 			case RESOURCE_DIMENSION_TEXTURE2DARRAY:
+			case RESOURCE_DIMENSION_TEXTURE2DMSARRAY:
 			case RESOURCE_DIMENSION_TEXTURECUBEARRAY:
 				return 3;
 			default:

src/Instruction.cpp

@@ -123,10 +123,17 @@ static Operand *GetSrcSwizzleOperand(Instruction *psInst)
 	case OPCODE_LD_STRUCTURED:
 		return &psInst->asOperands[3];
 
+	case OPCODE_SAMPLE_INFO:
+		return &psInst->asOperands[1];
+
 	case OPCODE_ISHL:
 	case OPCODE_ISHR:
 	case OPCODE_USHR:
+		// sm4 variant has single component selection on src1 -> only src0 has swizzle
+		if (psInst->asOperands[2].eSelMode == OPERAND_4_COMPONENT_SELECT_1_MODE)
 			return &psInst->asOperands[1];
+		else // whereas sm5 variant has swizzle also on src1 
+			return NULL;
 
 	default:
 		ASSERT(0);

src/Operand.cpp

@@ -274,7 +274,7 @@ SHADER_VARIABLE_TYPE Operand::GetDataType(HLSLCrossCompilerContext* psContext, S
 	{
 	case OPERAND_TYPE_TEMP:
 	{
-		SHADER_VARIABLE_TYPE eCurrentType;
+		SHADER_VARIABLE_TYPE eCurrentType = SVT_FLOAT;
 		int i = 0;
 
 		if (eSelMode == OPERAND_4_COMPONENT_SELECT_1_MODE)

src/Shader.cpp

@@ -861,6 +861,7 @@ void ShaderPhase::ExpandSWAPCs()
 		swapItr->eOpcode = OPCODE_MOVC;
 		swapItr->ui32NumOperands = 4;
 		swapItr->ui32FirstSrc = 1;
+		swapItr->asOperands[0] = origSwapInst.asOperands[0];
 		swapItr->asOperands[0].eType = OPERAND_TYPE_TEMP;
 		swapItr->asOperands[0].ui32RegisterNumber = extraTemp;
 		// mask is already fine
@@ -923,12 +924,10 @@ void Shader::ForcePositionToHighp()
 	{
 		if (decl.eOpcode == OPCODE_DCL_OUTPUT_SIV)
 		{
-			if (decl.asOperands[0].eSpecialName == NAME_POSITION)
-				return true;
-			if (decl.asOperands[0].eSpecialName != NAME_UNDEFINED)
-				return false;
-
-			// This might be SV_Position (because d3dcompiler is weird). Get signature and check
+			const SPECIAL_NAME specialName = decl.asOperands[0].eSpecialName;
+			if (specialName == NAME_POSITION ||
+				specialName == NAME_UNDEFINED) // This might be SV_Position (because d3dcompiler is weird).
+			{
 				const ShaderInfo::InOutSignature *sig = NULL;
 				sInfo.GetOutputSignatureFromRegister(decl.asOperands[0].ui32RegisterNumber, decl.asOperands[0].GetAccessMask(), 0, &sig);
 				ASSERT(sig != NULL);
@@ -937,6 +936,7 @@ void Shader::ForcePositionToHighp()
 					((ShaderInfo::InOutSignature *)sig)->eMinPrec = MIN_PRECISION_DEFAULT;
 					return true;
 				}
+			}
 			return false;
 		}
 		else if (decl.eOpcode == OPCODE_DCL_OUTPUT)

src/UseDefineChains.cpp

@@ -145,10 +145,6 @@ static DefineUseChainEntry *GetOrCreateDefinition(const BasicBlock::Definition &
 // Do flow control analysis on the instructions and build the define-use and use-define chains
 void BuildUseDefineChains(std::vector<Instruction> &instructions, uint32_t ui32NumTemps, DefineUseChains &psDUChain, UseDefineChains &psUDChain, HLSLcc::ControlFlow::ControlFlowGraph &cfg)
 {
-
-	Instruction *psFirstInstruction = &instructions[0];
-	Instruction *psLastInstruction = &instructions[instructions.size() - 1];
-
 	ActiveDefinitions lastSeenDefinitions(ui32NumTemps * 4, NULL); // Array of pointers to the currently active definition for each temp
 
 	psDUChain.clear();

src/cbstring/bstraux.c

@@ -193,10 +193,6 @@ int i, l, c;
 }
 
 static size_t readNothing (void *buff, size_t elsize, size_t nelem, void *parm) {
-	buff = buff;
-	elsize = elsize;
-	nelem = nelem;
-	parm = parm;
 	return 0; /* Immediately indicate EOF. */
 }
 

src/decode.cpp

@@ -178,7 +178,7 @@ static void MarkTextureSamplerPair(ShaderInfo* psShaderInfo, std::vector<Declara
 				//  set::insert returns a pair of which .second tells whether a new element was actually added
 				if (psDecl->samplersUsed.insert(psSamplerOperand->ui32RegisterNumber).second)
                 {
-                    // Record the texturename_X_samplername string in the TextureSamplerPair array that we return to the client
+                    // Record the <texturename>TEX_with_SMP<samplername> string in the TextureSamplerPair array that we return to the client
                     std::string combinedname = TextureSamplerName(psShaderInfo, psTextureOperand->ui32RegisterNumber, psSamplerOperand->ui32RegisterNumber, psDecl->ui32IsShadowTex);
 					samplers.push_back(combinedname);
                 }
@@ -1112,7 +1112,7 @@ const uint32_t* DecodeInstruction(const uint32_t* pui32Token, Instruction* psIns
 		{
             psInst->ui32NumOperands = 4;
 
-			if(eOpcode == OPCODE_IMUL)
+			if(eOpcode == OPCODE_IMUL || eOpcode == OPCODE_UDIV)
 			{
 				psInst->ui32FirstSrc = 2;
 			}
@@ -1244,6 +1244,16 @@ const uint32_t* DecodeInstruction(const uint32_t* pui32Token, Instruction* psIns
             ui32OperandOffset += DecodeOperand(pui32Token+ui32OperandOffset, &psInst->asOperands[2]);
             break;
         }
+		case OPCODE_SAMPLE_INFO:
+		{
+			psInst->ui32NumOperands = 2;
+
+			psInst->eResInfoReturnType = DecodeResInfoReturnType(pui32Token[0]);
+
+			ui32OperandOffset += DecodeOperand(pui32Token + ui32OperandOffset, &psInst->asOperands[0]);
+			ui32OperandOffset += DecodeOperand(pui32Token + ui32OperandOffset, &psInst->asOperands[1]);
+			break;
+		}
         case OPCODE_MSAD:
         default:
         {
@@ -1254,8 +1264,8 @@ const uint32_t* DecodeInstruction(const uint32_t* pui32Token, Instruction* psIns
 
     // For opcodes that sample textures, mark which samplers are used by each texture
     {
-        uint32_t ui32TextureRegisterNumber;
-        uint32_t ui32SamplerRegisterNumber;
+        uint32_t ui32TextureRegisterNumber = 0;
+        uint32_t ui32SamplerRegisterNumber = 0;
         uint32_t bTextureSampleInstruction = 0;
         switch (eOpcode)
         {

src/internal_includes/HLSLCrossCompilerContext.h

@@ -2,6 +2,7 @@
 
 #include <stdint.h>
 #include <string>
+#include <set>
 #include "bstrlib.h"
 
 class Shader;
@@ -47,4 +48,8 @@ public:
 
 	bool OutputNeedsDeclaring(const Operand* psOperand, const int count);
 
+	void RequireExtension(const std::string &extName);
+
+private:
+	std::set<std::string> m_EnabledExtensions;
 };

src/internal_includes/HLSLccToolkit.h

@@ -3,6 +3,7 @@
 #include "bstrlib.h"
 #include <vector>
 #include <string>
+#include <algorithm>
 
 #include "internal_includes/Instruction.h"
 #include "internal_includes/Operand.h"

src/internal_includes/Instruction.h

@@ -30,6 +30,8 @@ struct Instruction
 	, m_SkipTranslation(false)
 	, m_InductorRegister(0)
 	, bSaturate(0)
+	, m_IsStaticBranch(false)
+	, m_StaticBranchCondition(NULL)
 	{
 		m_LoopInductors[0] = m_LoopInductors[1] = m_LoopInductors[2] = m_LoopInductors[3] = 0;
 	}
@@ -85,6 +87,20 @@ struct Instruction
 		asOperands[3].eSelMode = OPERAND_4_COMPONENT_MASK_MODE;
 	}
 
+	// Returns true if this instruction is a conditional branch
+	bool IsConditionalBranchInstruction() const
+	{
+		switch (eOpcode)
+		{
+		case OPCODE_IF:
+		case OPCODE_BREAKC:
+		case OPCODE_CONTINUEC:
+		case OPCODE_RETC:
+			return true;
+		default:
+			return false;
+		}
+	}
 
 	bool IsPartialPrecisionSamplerInstruction(const ShaderInfo &info, OPERAND_MIN_PRECISION *pType) const;
 
@@ -114,6 +130,10 @@ struct Instruction
 	RESOURCE_DIMENSION eResDim;
 	int8_t iCausedSplit; // Nonzero if has caused a temp split. Later used by sampler datatype tweaking
 
+	bool m_IsStaticBranch; // If true, this instruction is a static branch
+	const Instruction *m_StaticBranchCondition; // If this is a static branch, this instruction points to the condition instruction. Can also be NULL if the operand itself is the condition
+	std::string m_StaticBranchName; // The name of the static branch variable, with the condition encoded in it.
+
 	struct Use
 	{
 		Use() : m_Inst(0), m_Op(0) {}

src/internal_includes/Shader.h

@@ -104,6 +104,9 @@ public:
 	uint32_t m_NextFreeTempRegister; // A counter for creating new temporaries for for-loops.
 	uint32_t m_NextTexCoordTemp; // A counter for creating tex coord temps for driver issue workarounds
 
+	// Instructions that are static branches (branches based on constant buffer values only)
+	std::vector<Instruction *> m_StaticBranchInstructions;
+
 private:
 	bool m_CFGInitialized;
 	HLSLcc::ControlFlow::ControlFlowGraph m_CFG;
@@ -142,10 +145,7 @@ public:
 		ui32CurrentVertexOutputStream(0),
 		textureSamplers(),
 		aui32StructuredBufferBindingPoints(MAX_RESOURCE_BINDINGS, 0),
-		ui32CurrentStructuredBufferIndex(),
-		m_CubemapArrayExtensionDeclared(false),
-		m_TextureBufferExtensionDeclared(false),
-		m_ClipDistanceExtensionDeclared(false)
+		ui32CurrentStructuredBufferIndex()
 	{
 	}
 
@@ -246,10 +246,6 @@ public:
 	std::vector<uint32_t> aui32StructuredBufferBindingPoints;
 	uint32_t ui32CurrentStructuredBufferIndex;
 
-	bool m_CubemapArrayExtensionDeclared;
-	bool m_TextureBufferExtensionDeclared;
-	bool m_ClipDistanceExtensionDeclared;
-
 	std::vector<char> psIntTempSizes; // Array for whether this temp register needs declaration as int temp
 	std::vector<char> psInt16TempSizes; // min16ints
 	std::vector<char> psInt12TempSizes; // min12ints

src/internal_includes/debug.h

@@ -12,7 +12,9 @@ static void CustomAssert(int expression)
     }
 }
 #else
-#define ASSERT(expr)
+#define UNUSED(EXPR_) \
+	do { if (false) (void)(EXPR_); } while(0)
+#define ASSERT(expr) UNUSED(expr)
 #endif
 
 #endif

src/internal_includes/languages.h

@@ -2,6 +2,8 @@
 #define LANGUAGES_H
 
 #include "hlslcc.h"
+#include "HLSLCrossCompilerContext.h"
+#include "Shader.h"
 
 static int InOutSupported(const GLLang eLang)
 {
@@ -42,9 +44,13 @@ static int HaveOverloadedTextureFuncs(const GLLang eLang)
 }
 
 //Only enable for ES.
-//Not present in 120, ignored in other desktop languages.
-static int HavePrecisionQualifers(const GLLang eLang)
+//Not present in 120, ignored in other desktop languages. Specifically enabled on Vulkan.
+static int HavePrecisionQualifiers(const HLSLCrossCompilerContext *psContext)
 {
+	if ((psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) != 0)
+		return 1;
+
+	const GLLang eLang = psContext->psShader->eTargetLanguage;
 	if(eLang >= LANG_ES_100 && eLang <= LANG_ES_310)
 	{
 		return 1;

src/internal_includes/toGLSL.h

@@ -20,10 +20,17 @@ public:
 	virtual void SetIOPrefixes();
 
 private:
+	// Vulkan-only: detect which branches only depend on uniforms and immediate values and can be turned into specialization constants.
+	void IdentifyStaticBranches(ShaderPhase *psPhase);
+	void BuildStaticBranchNameForInstruction(Instruction &inst);
 
+	void DeclareSpecializationConstants(ShaderPhase &phase);
+
+	void TranslateOperand(bstring glsl, const Operand *psOp, uint32_t flags, uint32_t ui32ComponentMask = OPERAND_4_COMPONENT_MASK_ALL);
 	void TranslateOperand(const Operand *psOp, uint32_t flags, uint32_t ui32ComponentMask = OPERAND_4_COMPONENT_MASK_ALL);
 	void TranslateInstruction(Instruction* psInst, bool isEmbedded = false);
 
+	void TranslateVariableNameWithMask(bstring glsl, const Operand* psOperand, uint32_t ui32TOFlag, uint32_t* pui32IgnoreSwizzle, uint32_t ui32CompMask, int *piRebase);
 	void TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui32TOFlag, uint32_t* pui32IgnoreSwizzle, uint32_t ui32CompMask, int *piRebase);
 
 	void TranslateOperandIndex(const Operand* psOperand, int index);

src/internal_includes/toGLSLOperand.h

@@ -12,6 +12,7 @@ class HLSLCrossCompilerContext;
 //void TranslateOperandWithMask(HLSLCrossCompilerContext* psContext, const Operand* psOperand, uint32_t ui32TOFlag, uint32_t ui32ComponentMask);
 
 void TranslateOperandSwizzle(HLSLCrossCompilerContext* psContext, const Operand* psOperand, int iRebase);
+void TranslateOperandSwizzleWithMask(bstring glsl, HLSLCrossCompilerContext* psContext, const Operand* psOperand, uint32_t ui32ComponentMask, int iRebase);
 void TranslateOperandSwizzleWithMask(HLSLCrossCompilerContext* psContext, const Operand* psOperand, uint32_t ui32ComponentMask, int iRebase);
 
 void ResourceName(bstring targetStr, HLSLCrossCompilerContext* psContext, ResourceGroup group, const uint32_t ui32RegisterNumber, const int bZCompare);

src/internal_includes/toMetal.h

@@ -26,7 +26,11 @@ class BindingSlotAllocator
 	typedef std::map<uint32_t, uint32_t> SlotMap;
 	SlotMap m_Allocations;
 public:
-	BindingSlotAllocator() : m_Allocations(), m_NextFreeSlot(0) {}
+	BindingSlotAllocator() : m_Allocations()
+	{
+		for(int i = MAX_RESOURCE_BINDINGS-1; i >= 0; i --)
+			m_FreeSlots.push_back(i);
+	}
 
 	enum BindType
 	{
@@ -36,8 +40,6 @@ public:
 		UAV
 	};
 
-	// isUAV is only meaningful for texture slots
-
 	uint32_t GetBindingSlot(uint32_t regNo, BindType type)
 	{
 		// The key is regNumber with the bindtype stored to highest 16 bits
@@ -45,14 +47,47 @@ public:
 		SlotMap::iterator itr = m_Allocations.find(key);
 		if(itr == m_Allocations.end())
 		{
-			m_Allocations.insert(std::make_pair(key, m_NextFreeSlot));
-			return m_NextFreeSlot++;
+			uint32_t slot = m_FreeSlots.back();
+			m_FreeSlots.pop_back();
+			m_Allocations.insert(std::make_pair(key, slot));
+			return slot;
 		}
 		return itr->second;
 	}
 	
+	// Func for reserving binding slots with the original reg number.
+	// Used for fragment shader UAVs (SetRandomWriteTarget etc).
+	void ReserveBindingSlot(uint32_t regNo, BindType type)
+	{
+		uint32_t key = regNo | (uint32_t(type) << 16);
+		m_Allocations.insert(std::make_pair(key, regNo));
+		
+		// Remove regNo from free slots
+		for (int i = m_FreeSlots.size() - 1; i >= 0; i--)
+		{
+			if (m_FreeSlots[i] == regNo)
+			{
+				m_FreeSlots.erase(m_FreeSlots.begin() + i);
+				return;
+			}
+		}
+	}
+
 private:
-	uint32_t m_NextFreeSlot;
+	std::vector<uint32_t> m_FreeSlots;
+};
+
+struct SamplerDesc
+{
+	std::string name;
+	uint32_t reg, slot;
+};
+struct TextureSamplerDesc
+{
+	std::string name;
+	int textureBind, samplerBind;
+	HLSLCC_TEX_DIMENSION dim;
+	bool uav;
 };
 
 
@@ -61,7 +96,12 @@ class ToMetal : public Translator
 protected:
 	GLLang language;
 public:
-	explicit ToMetal(HLSLCrossCompilerContext *ctx) : Translator(ctx), m_ShadowSamplerDeclared(false) {}
+	explicit ToMetal(HLSLCrossCompilerContext *ctx)
+        : Translator(ctx)
+        , m_ShadowSamplerDeclared(false)
+        , m_NeedFBOutputRemapDecl(false)
+        , m_NeedFBInputRemapDecl(false)
+    {}
 
 	virtual bool Translate();
 	virtual void TranslateDeclaration(const Declaration *psDecl);
@@ -75,6 +115,8 @@ private:
 
 	void DeclareBuiltinInput(const Declaration *psDecl);
 	void DeclareBuiltinOutput(const Declaration *psDecl);
+	void DeclareClipPlanes(const Declaration* decl, unsigned declCount);
+	void GenerateTexturesReflection(HLSLccReflection* refl);
 
 	// Retrieve the name of the output struct for this shader
 	std::string GetOutputStructName() const;
@@ -88,7 +130,7 @@ private:
 	void DeclareStructType(const std::string &name, const std::vector<ShaderVarType> &contents, bool withinCB = false, uint32_t cumulativeOffset = 0);
 	void DeclareStructVariable(const std::string &parentName, const ShaderVar &var, bool withinCB = false, uint32_t cumulativeOffset = 0);
 	void DeclareStructVariable(const std::string &parentName, const ShaderVarType &var, bool withinCB = false, uint32_t cumulativeOffset = 0);
-	void DeclareBufferVariable(const Declaration *psDecl, const bool isRaw, const bool isUAV);
+	void DeclareBufferVariable(const Declaration *psDecl, bool isRaw, bool isUAV);
 
 	void DeclareResource(const Declaration *psDecl);
 	void TranslateResourceTexture(const Declaration* psDecl, uint32_t samplerCanDoShadowCmp, HLSLCC_TEX_DIMENSION texDim);
@@ -174,11 +216,18 @@ private:
 	// A <function name, body text> map of extra helper functions we'll need.
 	FunctionDefinitions m_FunctionDefinitions;
 
-	BindingSlotAllocator m_TextureSlots;
+	BindingSlotAllocator m_TextureSlots, m_SamplerSlots;
 	BindingSlotAllocator m_BufferSlots;
 
+	std::vector<SamplerDesc> m_Samplers;
+	std::vector<TextureSamplerDesc> m_Textures;
+
 	std::string m_ExtraGlobalDefinitions;
 
+    // Flags for whether we need to add the declaration for the FB IO remaps
+    bool m_NeedFBInputRemapDecl;
+    bool m_NeedFBOutputRemapDecl;
+    
 	bool m_ShadowSamplerDeclared;
 
 	void EnsureShadowSamplerDeclared();
@@ -188,6 +237,9 @@ private:
 
 	// Move all lowp -> mediump
 	void ClampPartialPrecisions();
+	
+	// Reseve UAV slots in advance to match the original HLSL bindings -> correct bindings in SetRandomWriteTarget()
+	void ReserveUAVBindingSlots(ShaderPhase *phase);
 };
 
 

src/reflect.cpp

@@ -306,7 +306,6 @@ static const uint32_t* ReadConstantBuffer(ShaderInfo* psShaderInfo,
         //D3D11_SHADER_VARIABLE_DESC
         ShaderVar * const psVar = &psBuffer->asVars[i];
 
-        uint32_t ui32Flags;
         uint32_t ui32TypeOffset;
         uint32_t ui32DefaultValueOffset;
 
@@ -317,7 +316,10 @@ static const uint32_t* ReadConstantBuffer(ShaderInfo* psShaderInfo,
 
         psVar->ui32StartOffset = *pui32VarToken++;
         psVar->ui32Size = *pui32VarToken++;
-        ui32Flags = *pui32VarToken++;
+        
+        //skip ui32Flags
+        pui32VarToken++;
+        
         ui32TypeOffset = *pui32VarToken++;
 
 		psVar->sType.name = psVar->name;
@@ -367,12 +369,12 @@ static const uint32_t* ReadConstantBuffer(ShaderInfo* psShaderInfo,
 
 
     {
-        uint32_t ui32Flags;
-        uint32_t ui32BufferType;
-
         psBuffer->ui32TotalSizeInBytes = *pui32Tokens++;
-        ui32Flags = *pui32Tokens++;
-        ui32BufferType = *pui32Tokens++;
+        
+        //skip ui32Flags
+        pui32Tokens++;
+        //skip ui32BufferType
+        pui32Tokens++;
     }
 
     return pui32Tokens;

src/toGLSL.cpp

@@ -95,6 +95,7 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 	bstring glsl = *psContext->currentGLSLString;
 	bstring extensions = psContext->extensions;
 	bool isES = (psContext->psShader->eTargetLanguage >= LANG_ES_100 && psContext->psShader->eTargetLanguage <= LANG_ES_310);
+	bool GL_ARB_shader_storage_buffer_object = false;
 	bool GL_ARB_shader_image_load_store = false;
 
 	if(psContext->psShader->ui32MajorVersion > 3 && psContext->psShader->eTargetLanguage != LANG_ES_300 && psContext->psShader->eTargetLanguage != LANG_ES_310 && !(psContext->psShader->eTargetLanguage >= LANG_330))
@@ -114,7 +115,7 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 			psContext->psShader->aiOpcodeUsed[OPCODE_DCL_RESOURCE_STRUCTURED] ||
 			psContext->psShader->aiOpcodeUsed[OPCODE_DCL_RESOURCE_RAW])
 		{
-			bcatcstr(extensions, "#extension GL_ARB_shader_storage_buffer_object : enable\n");
+			GL_ARB_shader_storage_buffer_object = true;
 		}
 	}
 
@@ -129,8 +130,6 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 		}
 	}
 
-	if (!HaveImageAtomics(psContext->psShader->eTargetLanguage))
-	{
 	if (psContext->psShader->aiOpcodeUsed[OPCODE_ATOMIC_CMP_STORE] ||
 		psContext->psShader->aiOpcodeUsed[OPCODE_IMM_ATOMIC_AND] ||
 		psContext->psShader->aiOpcodeUsed[OPCODE_ATOMIC_AND] ||
@@ -148,6 +147,11 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 		psContext->psShader->aiOpcodeUsed[OPCODE_IMM_ATOMIC_XOR] ||
 		psContext->psShader->aiOpcodeUsed[OPCODE_IMM_ATOMIC_EXCH] ||
 		psContext->psShader->aiOpcodeUsed[OPCODE_IMM_ATOMIC_CMP_EXCH])
+	{
+		if (!HaveAtomicMem(psContext->psShader->eTargetLanguage))
+			GL_ARB_shader_storage_buffer_object = true;
+
+		if (!HaveImageAtomics(psContext->psShader->eTargetLanguage))
 		{
 			if (isES)
 				bcatcstr(extensions, "#extension GL_OES_shader_image_atomic : enable\n");
@@ -189,9 +193,15 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 		if(psContext->psShader->aiOpcodeUsed[OPCODE_RESINFO])
 		{
 			bcatcstr(extensions,"#extension GL_ARB_texture_query_levels : enable\n");
+			bcatcstr(extensions, "#extension GL_ARB_shader_image_size : enable\n");
 		}
 	}
 
+	if (psContext->psShader->aiOpcodeUsed[OPCODE_SAMPLE_INFO ])
+	{
+		bcatcstr(extensions, "#extension GL_ARB_shader_texture_image_samples : enable\n");
+	}
+
 	if(!HaveImageLoadStore(psContext->psShader->eTargetLanguage))
 	{
 		if(psContext->psShader->aiOpcodeUsed[OPCODE_STORE_UAV_TYPED] ||
@@ -236,9 +246,24 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 		}
 	}
 
+	if (GL_ARB_shader_storage_buffer_object)
+		bcatcstr(extensions, "#extension GL_ARB_shader_storage_buffer_object : enable\n");
+
 	if (GL_ARB_shader_image_load_store)
 		bcatcstr(extensions, "#extension GL_ARB_shader_image_load_store : enable\n");
 
+	if(psContext->psShader->eShaderType == PIXEL_SHADER && psContext->psShader->eTargetLanguage >= LANG_120 && !HaveFragmentCoordConventions(psContext->psShader->eTargetLanguage))
+	{
+		bcatcstr(extensions,"#extension GL_ARB_fragment_coord_conventions : require\n");
+	}
+
+	if (psContext->psShader->extensions->EXT_shader_framebuffer_fetch && psContext->psShader->eShaderType == PIXEL_SHADER && psContext->flags & HLSLCC_FLAG_SHADER_FRAMEBUFFER_FETCH)
+	{
+		bcatcstr(extensions, "#ifdef GL_EXT_shader_framebuffer_fetch\n");
+		bcatcstr(extensions, "#extension GL_EXT_shader_framebuffer_fetch : enable\n");
+		bcatcstr(extensions, "#endif\n");
+	}
+
 	//Handle fragment shader default precision
 	if ((psContext->psShader->eShaderType == PIXEL_SHADER) &&
 		(psContext->psShader->eTargetLanguage == LANG_ES_100 || psContext->psShader->eTargetLanguage == LANG_ES_300 || psContext->psShader->eTargetLanguage == LANG_ES_310))
@@ -253,11 +278,6 @@ static void AddVersionDependentCode(HLSLCrossCompilerContext* psContext)
 		bcatcstr(glsl, "precision highp int;\n");
 	}
 
-	if(psContext->psShader->eShaderType == PIXEL_SHADER && psContext->psShader->eTargetLanguage >= LANG_120 && !HaveFragmentCoordConventions(psContext->psShader->eTargetLanguage))
-	{
-		bcatcstr(extensions,"#extension GL_ARB_fragment_coord_conventions : require\n");
-	}
-
 	if(psContext->psShader->eShaderType == PIXEL_SHADER && psContext->psShader->eTargetLanguage >= LANG_150)
 	{
 		if(psContext->flags & HLSLCC_FLAG_ORIGIN_UPPER_LEFT)
@@ -435,7 +455,7 @@ static void DoHullShaderPassthrough(HLSLCrossCompilerContext *psContext)
 			outputName = oss.str();
 		}
 
-		const char * prec = HavePrecisionQualifers(psContext->psShader->eTargetLanguage) ? "highp ": "";
+		const char * prec = HavePrecisionQualifiers(psContext) ? "highp ": "";
 
 		psContext->AddIndentation();
 		if (ui32NumComponents > 1) // TODO Precision
@@ -519,6 +539,8 @@ bool ToGLSL::Translate()
 
 	if (psShader->extensions)
 	{
+		if(psContext->flags & HLSLCC_FLAG_NVN_TARGET)
+			bcatcstr(extensions, "#extension GL_ARB_separate_shader_objects : enable\n");
 		if (psShader->extensions->ARB_explicit_attrib_location)
 			bcatcstr(extensions, "#extension GL_ARB_explicit_attrib_location : require\n");
 		if (psShader->extensions->ARB_explicit_uniform_location)
@@ -531,6 +553,25 @@ bool ToGLSL::Translate()
 
     AddVersionDependentCode(psContext);
 
+    if(psShader->eShaderType == VERTEX_SHADER &&
+        HaveLimitedInOutLocationQualifier(language, psShader->extensions) &&
+        psContext->flags & HLSLCC_FLAG_NVN_TARGET)
+    {
+        bcatcstr(glsl, "out gl_PerVertex { vec4 gl_Position; };\n");
+    }
+
+	if (!psContext->psDependencies->m_ExtBlendModes.empty() && psShader->eShaderType == PIXEL_SHADER)
+	{
+		bcatcstr(extensions, "#extension GL_KHR_blend_equation_advanced : enable\n");
+		bcatcstr(glsl, "#if GL_KHR_blend_equation_advanced\n");
+		for (i = 0; i < psContext->psDependencies->m_ExtBlendModes.size(); i++)
+		{
+			bformata(glsl, "layout(%s) out;\n", psContext->psDependencies->m_ExtBlendModes[i].c_str());
+		}
+		bcatcstr(glsl, "#endif\n");
+	}
+
+
 	psShader->PrepareStructuredBufferBindingSlots();
 
 	for (ui32Phase = 0; ui32Phase < psShader->asPhases.size(); ui32Phase++)
@@ -541,6 +582,7 @@ bool ToGLSL::Translate()
 		phase.ResolveUAVProperties();
 		psShader->ResolveStructuredBufferBindingSlots(&phase);
 		phase.PruneConstArrays();
+
 	}
 
 	psShader->PruneTempRegisters();
@@ -550,6 +592,12 @@ bool ToGLSL::Translate()
 		// Loop transform can only be done after the temps have been pruned
 		ShaderPhase &phase = psShader->asPhases[ui32Phase];
 		HLSLcc::DoLoopTransform(phase);
+
+		if ((psContext->flags & HLSLCC_FLAG_VULKAN_SPECIALIZATION_CONSTANTS) != 0)
+		{
+			IdentifyStaticBranches(&phase);
+		}
+
 	}
 
 	//Special case. Can have multiple phases.
@@ -593,6 +641,12 @@ bool ToGLSL::Translate()
 				TranslateDeclaration(&psPhase->psDecl[i]);
 			}
 
+			if ((psContext->flags & HLSLCC_FLAG_VULKAN_SPECIALIZATION_CONSTANTS) != 0)
+			{
+				DeclareSpecializationConstants(psShader->asPhases[i]);
+			}
+
+
 			bformata(glsl, "void %s%d(int phaseInstanceID)\n{\n", GetPhaseFuncName(psPhase->ePhase), ui32Phase);
 			psContext->indent++;
 
@@ -768,6 +822,11 @@ bool ToGLSL::Translate()
 		TranslateDeclaration(&psShader->asPhases[0].psDecl[i]);
 	}
 
+	if ((psContext->flags & HLSLCC_FLAG_VULKAN_SPECIALIZATION_CONSTANTS) != 0)
+	{
+		DeclareSpecializationConstants(psShader->asPhases[0]);
+	}
+
     bcatcstr(glsl, "void main()\n{\n");
 
     psContext->indent++;
@@ -803,4 +862,215 @@ bool ToGLSL::Translate()
     return true;
 }
 
+void ToGLSL::DeclareSpecializationConstants(ShaderPhase &phase)
+{
+	bstring glsl = psContext->glsl;
+	// There may be several uses for the same branch condition, so we'll need to keep track of what we've already declared.
+	std::set<uint32_t> alreadyDeclared;
+	for (std::vector<Instruction *>::iterator itr = phase.m_StaticBranchInstructions.begin(); itr != phase.m_StaticBranchInstructions.end(); itr++)
+	{
+		Instruction &i = **itr;
+		uint32_t slot = psContext->psDependencies->GetSpecializationConstantSlot(i.m_StaticBranchName);
+		if(alreadyDeclared.insert(slot).second) // Only declare if the insertion actually succeeded
+			bformata(glsl, "layout(constant_id = %d) const bool %s = false;\n", slot, i.m_StaticBranchName.c_str());
+	}
+}
 
+std::string to64(uint32_t in)
+{
+	const char to64[] =
+		"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+
+	char c_[2];
+	c_[0] = to64[in];
+	c_[1] = 0;
+	char c = c_[0];
+	if (c == 'X')
+		return "XX";
+	if (c == '+')
+		return "XA";
+	if (c == '/')
+		return "XB";
+	return std::string(c_);
+}
+
+// Slightly custom base64, espace non-identifier chars with 'X'
+static void Base64Encode(const std::string &in, std::string& result)
+{
+
+	size_t len = in.length();
+	size_t outputLen = (len + 2) / 3 * 4;
+	unsigned char *bytes = (unsigned char *)&in[0];
+
+	result.clear();
+	result.reserve(outputLen);
+
+	int i = 0;
+	unsigned char b1, b2, b3;
+	for (int chunk = 0; len > 0; len -= 3, chunk++) {
+		b1 = bytes[i++];
+		b2 = len > 1 ? bytes[i++] : '\0';
+		result += to64(b1 >> 2);
+		result += to64(((b1 & 3) << 4) | ((b2 & 0xf0) >> 4));
+		if (len > 2)
+		{
+			b3 = bytes[i++];
+			result += to64(((b2 & 0xF) << 2) | ((b3 & 0xC0) >> 6));
+			result += to64(b3 & 0x3F);
+		}
+		else if (len == 2)
+		{
+			result += to64((b2 & 0xF) << 2);
+			result += "XC";
+			break;
+		}
+		else /* len == 1 */
+		{
+			result += "XC";
+			break;
+		}
+	}
+}
+
+
+void ToGLSL::BuildStaticBranchNameForInstruction(Instruction &inst)
+{
+	std::ostringstream oss;
+	if (!inst.m_StaticBranchCondition)
+	{
+		// Simple case, just get the value, check if nonzero
+		bstring varname = bfromcstr("");
+		SHADER_VARIABLE_TYPE argType = inst.asOperands[0].GetDataType(psContext);
+		uint32_t flag = TO_FLAG_NONE;
+		switch (argType)
+		{
+		case SVT_BOOL:
+			flag = TO_FLAG_BOOL;
+			break;
+		case SVT_INT:
+		case SVT_INT12:
+		case SVT_INT16:
+			flag = TO_FLAG_INTEGER;
+			break;
+		case SVT_UINT:
+		case SVT_UINT16:
+		case SVT_UINT8:
+			flag = TO_FLAG_UNSIGNED_INTEGER;
+			break;
+		default:
+			break;
+		}
+		TranslateOperand(varname, &inst.asOperands[0], flag);
+		char *str = bstr2cstr(varname, '\0');
+		oss << str;
+		bcstrfree(str);
+		bdestroy(varname);
+		oss << "!=0";
+		Base64Encode(oss.str(), inst.m_StaticBranchName);
+	}
+	else
+	{
+		// Indirect, just store the whole previous instruction and then the condition
+		bstring res = bfromcstr("");
+
+		bstring *oldglsl = psContext->currentGLSLString;
+		psContext->currentGLSLString = &res;
+		TranslateInstruction((Instruction *)inst.m_StaticBranchCondition, true);
+		psContext->currentGLSLString = oldglsl;
+
+		SHADER_VARIABLE_TYPE argType = inst.asOperands[0].GetDataType(psContext);
+		uint32_t flag = TO_FLAG_NONE;
+		switch (argType)
+		{
+		case SVT_BOOL:
+			flag = TO_FLAG_BOOL;
+			break;
+		case SVT_INT:
+		case SVT_INT12:
+		case SVT_INT16:
+			flag = TO_FLAG_INTEGER;
+			break;
+		case SVT_UINT:
+		case SVT_UINT16:
+		case SVT_UINT8:
+			flag = TO_FLAG_UNSIGNED_INTEGER;
+			break;
+		default:
+			break;
+		}
+
+		if (argType == SVT_BOOL)
+		{
+			if (inst.eBooleanTestType == INSTRUCTION_TEST_ZERO)
+				bcatcstr(res, "!");
+		}
+
+		TranslateOperand(res, &inst.asOperands[0], flag);
+		char *str = bstr2cstr(res, '\0');
+		oss << str;
+		bcstrfree(str);
+		bdestroy(res);
+		if(argType != SVT_BOOL)
+			oss << "!=0";
+		Base64Encode(oss.str(), inst.m_StaticBranchName);
+
+	}
+
+}
+
+void ToGLSL::IdentifyStaticBranches(ShaderPhase *psPhase)
+{
+	for (std::vector<Instruction>::iterator itr = psPhase->psInst.begin(); itr != psPhase->psInst.end(); itr++)
+	{
+		Instruction &i = *itr;
+
+		if (!i.IsConditionalBranchInstruction())
+			continue;
+
+		// Simple case, direct conditional branch
+		if (i.asOperands[0].eType == OPERAND_TYPE_CONSTANT_BUFFER)
+		{
+			psPhase->m_StaticBranchInstructions.push_back(&i);
+			i.m_IsStaticBranch = true;
+			i.m_StaticBranchCondition = NULL;
+			BuildStaticBranchNameForInstruction(i);
+		}
+		// Indirect, comparison via another instruction
+		if (i.asOperands[0].eType == OPERAND_TYPE_TEMP)
+		{
+			// Check that the temp only has one visible definition
+			if (i.asOperands[0].m_Defines.size() == 1)
+			{
+				// ...and that it only uses constant buffers and immediates
+
+				Instruction &def = *i.asOperands[0].m_Defines[0].m_Inst;
+				bool isStatic = true;
+				for (uint32_t k = def.ui32FirstSrc; k < def.ui32NumOperands; k++)
+				{
+					Operand &o = def.asOperands[k];
+					if (!(o.eType == OPERAND_TYPE_CONSTANT_BUFFER || o.eType == OPERAND_TYPE_IMMEDIATE32))
+					{
+						isStatic = false;
+						break;
+					}
+					// Also check that the constant buffer access is "simple"
+					if (o.eType == OPERAND_TYPE_CONSTANT_BUFFER)
+					{
+						if (o.m_SubOperands[0].get() || o.m_SubOperands[1].get())
+						{
+							isStatic = false;
+							break;
+						}
+					}
+				}
+				if (isStatic)
+				{
+					psPhase->m_StaticBranchInstructions.push_back(&i);
+					i.m_IsStaticBranch = true;
+					i.m_StaticBranchCondition = &def;
+					BuildStaticBranchNameForInstruction(i);
+				}
+			}
+		}
+	}
+}

src/toGLSLDeclaration.cpp

@@ -237,7 +237,8 @@ static void DeclareInput(
 
 		std::string locationQualifier = "";
 
-		if (HaveInOutLocationQualifier(psContext->psShader->eTargetLanguage))
+		if (HaveInOutLocationQualifier(psContext->psShader->eTargetLanguage) ||
+			((psContext->flags & HLSLCC_FLAG_NVN_TARGET) && HaveLimitedInOutLocationQualifier(psContext->psShader->eTargetLanguage, psContext->psShader->extensions)))
 		{
 			bool addLocation = false;
 
@@ -365,10 +366,9 @@ void ToGLSL::AddBuiltinOutput(const Declaration* psDecl, int arrayElements, cons
 			{
 				int max = psDecl->asOperands[0].GetMaxComponent();
 
-				if (IsESLanguage(psShader->eTargetLanguage) && !psShader->m_ClipDistanceExtensionDeclared)
+				if (IsESLanguage(psShader->eTargetLanguage))
 				{
-					bcatcstr(psContext->extensions, "#extension GL_EXT_clip_cull_distance : require\n");
-					psShader->m_ClipDistanceExtensionDeclared = true;
+					psContext->RequireExtension("GL_EXT_clip_cull_distance");
 				}
 
 				int applySwizzle = psDecl->asOperands[0].GetNumSwizzleElements() > 1 ? 1 : 0;
@@ -551,7 +551,7 @@ void ToGLSL::AddUserOutput(const Declaration* psDecl)
 		const Operand* psOperand = &psDecl->asOperands[0];
 		const char* Precision = "";
 		int iNumComponents;
-		bstring type;
+		bstring type = NULL;
 		int regSpace = psDecl->asOperands[0].GetRegisterSpace(psContext);
 		uint32_t ui32Reg = psDecl->asOperands[0].ui32RegisterNumber;
 
@@ -604,7 +604,7 @@ void ToGLSL::AddUserOutput(const Declaration* psDecl)
                 break;
 		}
 
-		if(HavePrecisionQualifers(psShader->eTargetLanguage))
+		if(HavePrecisionQualifiers(psContext))
 		{
 			switch(psOperand->eMinPrecision)
 			{
@@ -664,8 +664,7 @@ void ToGLSL::AddUserOutput(const Declaration* psDecl)
 					{
 						bcatcstr(extensions, "#ifdef GL_ARB_conservative_depth\n");
 						bcatcstr(extensions, "#extension GL_ARB_conservative_depth : enable\n");
-						bcatcstr(extensions, "layout (depth_less) out float gl_FragDepth;\n");
-						bcatcstr(glsl, "#endif\n");
+						bcatcstr(extensions, "#endif\n");
 						bcatcstr(glsl, "#ifdef GL_ARB_conservative_depth\n");
 						bcatcstr(glsl, "layout (depth_less) out float gl_FragDepth;\n");
 						bcatcstr(glsl, "#endif\n");
@@ -685,6 +684,13 @@ void ToGLSL::AddUserOutput(const Declaration* psDecl)
 							strncpy(OutputName, (char *)oname->data, 512);
 							bdestroy(oname);
 
+							bstring layoutQualifier = bformat("");
+							bool haveFramebufferFetch = (psShader->extensions->EXT_shader_framebuffer_fetch &&
+														 psShader->eShaderType == PIXEL_SHADER &&
+														 psContext->flags & HLSLCC_FLAG_SHADER_FRAMEBUFFER_FETCH);
+							if (haveFramebufferFetch)
+								bcatcstr(glsl, "#ifdef GL_EXT_shader_framebuffer_fetch\n");
+
 							if (HaveInOutLocationQualifier(psContext->psShader->eTargetLanguage) ||
 								HaveLimitedInOutLocationQualifier(psContext->psShader->eTargetLanguage, psContext->psShader->extensions))
 							{
@@ -698,15 +704,23 @@ void ToGLSL::AddUserOutput(const Declaration* psDecl)
 										renderTarget = 0;
 										index = 1;
 									}
-									bformata(glsl, "layout(location = %d, index = %d) ", renderTarget, index);
+									layoutQualifier = bformat("layout(location = %d, index = %d) ", renderTarget, index);
 								}
 								else
 								{
-									bformata(glsl, "layout(location = %d) ", renderTarget);
+									layoutQualifier = bformat("layout(location = %d) ", renderTarget);
 								}
 							}
 
-							bformata(glsl, "out %s%s %s;\n", Precision, type->data, OutputName);
+							if (haveFramebufferFetch)
+							{
+								bformata(glsl, "%sinout %s%s %s;\n", bstr2cstr(layoutQualifier, '\0'), Precision, type->data, OutputName);
+								bcatcstr(glsl, "#else\n");
+								bformata(glsl, "%sout %s%s %s;\n", bstr2cstr(layoutQualifier, '\0'), Precision, type->data, OutputName);
+								bcatcstr(glsl, "#endif\n");
+							}
+							else
+								bformata(glsl, "%sout %s%s %s;\n", bstr2cstr(layoutQualifier, '\0'), Precision, type->data, OutputName);
 						}
 						break;
 					}
@@ -1005,6 +1019,8 @@ static const char* GetSamplerType(HLSLCrossCompilerContext* psContext,
 	{
 		case RESOURCE_DIMENSION_BUFFER:
 		{
+			if (IsESLanguage(psContext->psShader->eTargetLanguage))
+				psContext->RequireExtension("GL_EXT_texture_buffer");
 			switch(eType)
 			{
 				case RETURN_TYPE_SINT:
@@ -1151,9 +1167,9 @@ static const char* GetSamplerType(HLSLCrossCompilerContext* psContext,
 	return "sampler2D";
 }
 
-static const char *GetSamplerPrecision(GLLang langVersion, REFLECT_RESOURCE_PRECISION ePrec)
+static const char *GetSamplerPrecision(const HLSLCrossCompilerContext *psContext, REFLECT_RESOURCE_PRECISION ePrec)
 {
-	if (!HavePrecisionQualifers(langVersion))
+	if (!HavePrecisionQualifiers(psContext))
 		return " ";
 
 	switch (ePrec)
@@ -1181,23 +1197,20 @@ static void TranslateResourceTexture(HLSLCrossCompilerContext* psContext, const
 		psDecl->asOperands[0].ui32RegisterNumber);
 
 	if (psDecl->value.eResourceDimension == RESOURCE_DIMENSION_TEXTURECUBEARRAY
-		&& !HaveCubemapArray(psContext->psShader->eTargetLanguage)
-		&& !psContext->psShader->m_CubemapArrayExtensionDeclared)
+		&& !HaveCubemapArray(psContext->psShader->eTargetLanguage))
 	{
 		// Need to enable extension (either OES or ARB), but we only need to add it once
 		if (IsESLanguage(psContext->psShader->eTargetLanguage))
-			bformata(psContext->extensions, "#extension GL_OES_texture_cube_map_array : enable\n");
+			psContext->RequireExtension("GL_OES_texture_cube_map_array");
 		else
-			bformata(psContext->extensions, "#extension GL_ARB_texture_cube_map_array : enable\n");
-
-		psContext->psShader->m_CubemapArrayExtensionDeclared = true;
+			psContext->RequireExtension("GL_ARB_texture_cube_map_array");
 	}
 
 	const ResourceBinding *psBinding = NULL;
 	psShader->sInfo.GetResourceFromBindingPoint(RGROUP_TEXTURE, psDecl->asOperands[0].ui32RegisterNumber, &psBinding);
 	ASSERT(psBinding != NULL);
 
-	samplerPrecision = GetSamplerPrecision(psContext->psShader->eTargetLanguage, psBinding ? psBinding->ePrecision : REFLECT_RESOURCE_PRECISION_UNKNOWN);
+	samplerPrecision = GetSamplerPrecision(psContext, psBinding ? psBinding->ePrecision : REFLECT_RESOURCE_PRECISION_UNKNOWN);
 
 	if (psContext->flags & HLSLCC_FLAG_COMBINE_TEXTURE_SAMPLERS)
 	{
@@ -1669,7 +1682,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 					StorageQualifier = "in";
 			}
 
-			if(HavePrecisionQualifers(psShader->eTargetLanguage))
+			if(HavePrecisionQualifiers(psContext))
 			{
 				switch(psOperand->eMinPrecision)
 				{
@@ -1741,6 +1754,16 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 			const char* Interpolation = "";
 			int hasNoPerspective = psContext->psShader->eTargetLanguage <= LANG_ES_310 ? 0 : 1;
 			inputName = psContext->GetDeclaredInputName(psOperand, NULL, 1, NULL);
+
+			// If this is a SV_Target input and framebuffer fetch is enabled then skip the declaration
+			if (psShader->extensions->EXT_shader_framebuffer_fetch &&
+				psShader->eShaderType == PIXEL_SHADER &&
+				psContext->flags & HLSLCC_FLAG_SHADER_FRAMEBUFFER_FETCH)
+			{
+				if(inputName == "vs_SV_Target0")
+					break;
+			}
+
 			if (InOutSupported(psContext->psShader->eTargetLanguage))
 			{
 				StorageQualifier = "in";
@@ -1787,7 +1810,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
                     break;
 			}
 
-			if(HavePrecisionQualifers(psShader->eTargetLanguage))
+			if(HavePrecisionQualifiers(psContext))
 			{
 				switch(psOperand->eMinPrecision)
 				{
@@ -1830,7 +1853,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 		{
 			uint32_t i = 0;
 			const uint32_t ui32NumTemps = psDecl->value.ui32NumTemps;
-			bool usePrecision = (HavePrecisionQualifers(psShader->eTargetLanguage) != 0);
+			bool usePrecision = (HavePrecisionQualifiers(psContext) != 0);
 
 			for (i = 0; i < ui32NumTemps; i++)
 			{
@@ -1867,8 +1890,6 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 			const Operand* psOperand = &psDecl->asOperands[0];
 			const uint32_t ui32BindingPoint = psOperand->aui32ArraySizes[0];
 
-			const char* StageName = "VS";
-
 			const ConstantBuffer* psCBuf = NULL;
 			psContext->psShader->sInfo.GetConstantBufferFromBindingPoint(RGROUP_CBUFFER, ui32BindingPoint, &psCBuf);
 
@@ -1882,37 +1903,109 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 				break;
 			}
 			
-			switch(psContext->psShader->eShaderType)
+			if (psCBuf->name.substr(0, 20) == "hlslcc_SubpassInput_" && psCBuf->name.length() >= 23 && !psCBuf->asVars.empty())
 			{
-				case PIXEL_SHADER:
+				// Special case for vulkan subpass input.
+
+				// The multisample versions have multiple members in the cbuffer, but we must only declare once.
+				// We still need to loop through all the variables and adjust names
+
+				// Pick up the type and index
+				char ty = psCBuf->name[20];
+				int idx = psCBuf->name[22] - '0';
+				bool isMS = false;
+				std::pair<uint32_t, uint32_t> binding = psContext->psDependencies->GetVulkanResourceBinding((std::string &)psCBuf->name, false, 2);
+
+				bool declared = false;
+				for (std::vector<ShaderVar>::const_iterator itr = psCBuf->asVars.begin(); itr != psCBuf->asVars.end(); itr++)
 				{
-					StageName = "PS";
+					ShaderVar &sv = (ShaderVar &)*itr;
+					if (sv.name.substr(0, 15) == "hlslcc_fbinput_")
+					{
+						if (!declared)
+						{
+							switch (ty)
+							{
+							case 'f':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform highp subpassInput %s;\n", idx, binding.first, binding.second, sv.name.c_str());
 								break;
-				}
-				case HULL_SHADER:
-				{
-					StageName = "HS";
+							case 'h':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform mediump subpassInput %s;\n", idx, binding.first, binding.second, sv.name.c_str());
 								break;
-				}
-				case DOMAIN_SHADER:
-				{
-					StageName = "DS";
+							case 'i':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform isubpassInput %s;\n", idx, binding.first, binding.second, sv.name.c_str());
 								break;
-				}
-				case GEOMETRY_SHADER:
-				{
-					StageName = "GS";
+							case 'u':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform usubpassInput %s;\n", idx, binding.first, binding.second, sv.name.c_str());
 								break;
-				}
-				case COMPUTE_SHADER:
-				{
-					StageName = "CS";
+							case 'F':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform highp subpassInputMS %s;\n", idx, binding.first, binding.second, sv.name.substr(0, 16).c_str());
+								isMS = true;
+								break;
+							case 'H':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform mediump subpassInputMS %s;\n", idx, binding.first, binding.second, sv.name.substr(0, 16).c_str());
+								isMS = true;
+								break;
+							case 'I':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform isubpassInputMS %s;\n", idx, binding.first, binding.second, sv.name.substr(0, 16).c_str());
+								isMS = true;
+								break;
+							case 'U':
+								bformata(glsl, "layout(input_attachment_index = %d, set = %d, binding = %d) uniform usubpassInputMS %s;\n", idx, binding.first, binding.second, sv.name.substr(0, 16).c_str());
+								isMS = true;
 								break;
-				}
 							default:
-				{
 								break;
 							}
+							declared = true;
+						}
+						else
+						{
+							if (ty == 'F' || ty == 'I' || ty == 'U')
+								isMS = true;
+						}
+						// Munge the name so it'll get the correct function call in GLSL directly
+						sv.name.insert(0, "subpassLoad(");
+						if (isMS)
+							sv.name[sv.name.length() - 2] = ',';
+						sv.name.append(")");
+						// Also update the type name
+						sv.sType.name = sv.name;
+						sv.sType.fullName = sv.name;
+					}
+				}
+
+				// Break out so this doesn't get declared. 
+				break;
+			}
+
+			if(psCBuf->name == "OVR_multiview")
+			{
+				// Special case for piggy-backing multiview info out
+				// This is not really a cbuffer, but if we see this being accessed, we know we need viewID
+				
+				// Extract numViews
+				uint32_t numViews = 0;
+				for(std::vector<ShaderVar>::const_iterator itr = psCBuf->asVars.begin(); itr != psCBuf->asVars.end(); itr++)
+				{
+					if(strncmp(itr->name.c_str(), "numViews_", 9) == 0)
+					{
+						// I really don't think we'll ever have more than 9 multiviews
+						numViews = itr->name[9] - '0';
+						break;
+					}
+				}
+				if(numViews > 0 && numViews < 10)
+				{
+					bcatcstr(extensions, "#extension GL_OVR_multiview : require\n");
+					bcatcstr(extensions, "#extension GL_OVR_multiview2 : enable\n");
+
+					if(psShader->eShaderType == VERTEX_SHADER)
+						bformata(glsl, "layout(num_views = %d) in;\n", numViews);
+					
+					break; // Break out so we don't actually declare this cbuffer
+
+				}
 			}
 
 			if(psContext->flags & HLSLCC_FLAG_UNIFORM_BUFFER_OBJECT)
@@ -1934,7 +2027,9 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 		}
 		case OPCODE_DCL_RESOURCE:
 		{
-			if (HaveUniformBindingsAndLocations(psContext->psShader->eTargetLanguage, psContext->psShader->extensions, psContext->flags))
+			// Skip the location declaration on Vulkan
+			if (HaveUniformBindingsAndLocations(psContext->psShader->eTargetLanguage, psContext->psShader->extensions, psContext->flags)
+				&& ((psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) == 0))
 			{
 				// Explicit layout bindings are not currently compatible with combined texture samplers. The layout below assumes there is exactly one GLSL sampler
 				// for each HLSL texture declaration, but when combining textures+samplers, there can be multiple OGL samplers for each HLSL texture declaration.
@@ -1950,7 +2045,19 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 			{
 				case RESOURCE_DIMENSION_BUFFER:
 				{
-					bformata(glsl, "uniform %s ", GetSamplerType(psContext,
+					if ((psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) != 0)
+					{
+						const ResourceBinding *psBinding = NULL;
+						psShader->sInfo.GetResourceFromBindingPoint(RGROUP_TEXTURE, psDecl->asOperands[0].ui32RegisterNumber, &psBinding);
+						std::string tname = psBinding->name;
+						GLSLCrossDependencyData::VulkanResourceBinding binding = psContext->psDependencies->GetVulkanResourceBinding(tname);
+						bformata(glsl, "layout(set = %d, binding = %d) ", binding.first, binding.second);
+					}
+
+					bcatcstr(glsl, "uniform ");
+					if (IsESLanguage(psContext->psShader->eTargetLanguage))
+						bcatcstr(glsl, "highp ");
+					bformata(glsl, "%s ", GetSamplerType(psContext,
 						RESOURCE_DIMENSION_BUFFER,
 						psDecl->asOperands[0].ui32RegisterNumber));
 					TranslateOperand(&psDecl->asOperands[0], TO_FLAG_NONE);
@@ -2347,7 +2454,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
                         
 				}
 
-				if (HavePrecisionQualifers(psShader->eTargetLanguage))
+				if (HavePrecisionQualifiers(psContext))
 				{
 					switch (psSignature->eMinPrec) // TODO What if the inputs in the indexed range are of different precisions?
 					{
@@ -2560,7 +2667,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 			// non-float images need either 'i' or 'u' prefix.
 			char imageTypePrefix[2] = { 0, 0 };
 			uint32_t bindpoint = psDecl->asOperands[0].ui32RegisterNumber;
-			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS);
+			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) != 0;
 
 			if(psDecl->sUAV.ui32GloballyCoherentAccess & GLOBALLY_COHERENT_ACCESS)
 			{
@@ -2585,12 +2692,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 				if ((psDecl->sUAV.ui32AccessFlags & ACCESS_FLAG_WRITE) && IsESLanguage(psShader->eTargetLanguage))
 				{
 					// Need to require the extension
-					if (!psShader->m_TextureBufferExtensionDeclared)
-					{
-						bcatcstr(psContext->extensions, "#extension GL_EXT_texture_buffer : require\n");
-						psShader->m_TextureBufferExtensionDeclared = true;
-					}
-
+					psContext->RequireExtension("GL_EXT_texture_buffer");
 				}
 
 				if(isVulkan)
@@ -2642,6 +2744,9 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 			{
 			case RESOURCE_DIMENSION_BUFFER:
 			{
+				if(IsESLanguage(psShader->eTargetLanguage))
+					psContext->RequireExtension("GL_EXT_texture_buffer");
+
 				bformata(glsl, "uniform %simageBuffer ", imageTypePrefix);
 				break;
 			}
@@ -2701,7 +2806,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 		}
 		case OPCODE_DCL_UNORDERED_ACCESS_VIEW_STRUCTURED:
 		{
-			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS);
+			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) != 0;
 			if(psDecl->sUAV.bCounter)
 			{
 				if (isVulkan) 
@@ -2715,7 +2820,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 				{
 					bcatcstr(glsl, "layout (binding = 0) uniform ");
 
-					if (HavePrecisionQualifers(psShader->eTargetLanguage))
+					if (HavePrecisionQualifiers(psContext))
 						bcatcstr(glsl, "highp ");
 					bcatcstr(glsl, "atomic_uint ");
 					ResourceName(glsl, psContext, RGROUP_UAV, psDecl->asOperands[0].ui32RegisterNumber, 0);
@@ -2729,7 +2834,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 		}
 		case OPCODE_DCL_UNORDERED_ACCESS_VIEW_RAW:
 		{
-			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS);
+			const bool isVulkan = (psContext->flags & HLSLCC_FLAG_VULKAN_BINDINGS) != 0;
 			if(psDecl->sUAV.bCounter)
 			{
 				if (isVulkan)
@@ -2742,7 +2847,7 @@ void ToGLSL::TranslateDeclaration(const Declaration* psDecl)
 				else
 				{
 					bcatcstr(glsl, "layout (binding = 0) uniform ");
-					if (HavePrecisionQualifers(psShader->eTargetLanguage))
+					if (HavePrecisionQualifiers(psContext))
 						bcatcstr(glsl, "highp ");
 					bcatcstr(glsl, "atomic_uint ");
 					ResourceName(glsl, psContext, RGROUP_UAV, psDecl->asOperands[0].ui32RegisterNumber, 0);
@@ -2989,6 +3094,13 @@ bool ToGLSL::TranslateSystemValue(const Operand *psOperand, const ShaderInfo::In
 		return true;
 	}
 
-	return false;
+	if (sig->semanticName == "PSIZE")
+	{
+		result = "gl_PointSize";
+		if (pui32IgnoreSwizzle)
+			*pui32IgnoreSwizzle = 1;
+		return true;
 	}
 
+	return false;
+}

src/toGLSLInstruction.cpp

@@ -90,7 +90,7 @@ void ToGLSL::AddOpAssignToDestWithMask(const Operand* psDest,
 	case SVT_FLOAT:
 	case SVT_FLOAT10:
 	case SVT_FLOAT16:
-		ASSERT(eSrcType != SVT_INT12 || eSrcType != SVT_INT16 && eSrcType != SVT_UINT16);
+		ASSERT(eSrcType != SVT_INT12 || (eSrcType != SVT_INT16 && eSrcType != SVT_UINT16));
 		if (psContext->psShader->ui32MajorVersion > 3)
 		{
 			if (eSrcType == SVT_INT)
@@ -892,7 +892,8 @@ void ToGLSL::GetResInfoData(Instruction* psInst, int index, int destElem)
 	bstring glsl = *psContext->currentGLSLString;
 	int numParenthesis = 0;
 	const RESINFO_RETURN_TYPE eResInfoReturnType = psInst->eResInfoReturnType;
-	const int isUAV = (psInst->asOperands[2].eType == OPERAND_TYPE_UNORDERED_ACCESS_VIEW);
+	bool isUAV = (psInst->asOperands[2].eType == OPERAND_TYPE_UNORDERED_ACCESS_VIEW);
+	bool isMS = psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DMS || psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DMSARRAY;
 
 	psContext->AddIndentation();
 	AddOpAssignToDestWithMask(&psInst->asOperands[0], eResInfoReturnType == RESINFO_INSTRUCTION_RETURN_UINT ? SVT_UINT : SVT_FLOAT, 1, "=", &numParenthesis, 1 << destElem);
@@ -922,7 +923,7 @@ void ToGLSL::GetResInfoData(Instruction* psInst, int index, int destElem)
 
 			TranslateOperand(&psInst->asOperands[2], TO_FLAG_NONE);
 
-			if (!isUAV)
+			if (!isUAV && !isMS)
 			{
 				bcatcstr(glsl, ", ");
 				TranslateOperand(&psInst->asOperands[1], TO_FLAG_INTEGER);
@@ -1077,7 +1078,7 @@ void ToGLSL::TranslateTextureSample(Instruction* psInst,
 	if (ui32Flags & TEXSMP_FLAG_GATHER)
 		funcName = "textureGather";
 
-	uint32_t uniqueNameCounter;
+	uint32_t uniqueNameCounter = 0;
 
 	// In GLSL, for every texture sampling func overload, except for cubemap arrays, the 
 	// depth compare reference value is given as the last component of the texture coord vector.
@@ -1391,7 +1392,6 @@ void ToGLSL::TranslateShaderStorageLoad(Instruction* psInst)
 	}
 	for (component = 0; component < 4; component++)
 	{
-		const ShaderVarType *psVar = NULL;
 		int addedBitcast = 0;
 		if (!(destMask & (1 << component)))
 			continue;
@@ -1835,6 +1835,14 @@ void ToGLSL::TranslateConditional(
 		statement = "return";
 	}
 
+	if (psInst->m_IsStaticBranch)
+	{
+		// Instead of the actual condition, use the specialization constant instead
+
+		// But first we'll have to make sure the original values don't get dropped out (we rely on glslang not being very smart)
+		bcatcstr(glsl, "if(false)\n {\n");
+	}
+
 	SHADER_VARIABLE_TYPE argType = psInst->asOperands[0].GetDataType(psContext);
 	if (argType == SVT_BOOL)
 	{
@@ -1884,6 +1892,32 @@ void ToGLSL::TranslateConditional(
 			bcatcstr(glsl, " {\n");
 		}
 	}
+	if (psInst->m_IsStaticBranch)
+	{
+		if (psInst->eOpcode == OPCODE_IF)
+		{
+			bcatcstr(glsl, "}\n}\n");
+		}
+		else
+		{
+			bcatcstr(glsl, "}\n");
+		}
+		bcatcstr(glsl, "if(");
+		if (psInst->eBooleanTestType != INSTRUCTION_TEST_NONZERO)
+			bcatcstr(glsl, "!");
+		bcatcstr(glsl, psInst->m_StaticBranchName.c_str());
+		if (psInst->eOpcode != OPCODE_IF)
+		{
+			bformata(glsl, "){%s;}\n", statement);
+		}
+		else
+		{
+			bcatcstr(glsl, "){\n");
+		}
+		return;
+
+	}
+
 }
 
 void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = false */)
@@ -1896,8 +1930,9 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 	{
 
 #ifdef _DEBUG
-		psContext->AddIndentation();
-		bformata(glsl, "//Instruction %d\n", psInst->id);
+		// Uncomment to print instruction IDs
+		//psContext->AddIndentation();
+		//bformata(glsl, "//Instruction %d\n", psInst->id);
 #if 0
 		if (psInst->id == 73)
 		{
@@ -2751,10 +2786,7 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 								  uint32_t varFound;
 
 								  uint32_t funcPointer;
-								  uint32_t funcTableIndex;
-								  uint32_t funcTable;
 								  uint32_t funcBodyIndex;
-								  uint32_t funcBody;
 								  uint32_t ui32NumBodiesPerTable;
 
 #ifdef _DEBUG
@@ -2765,15 +2797,10 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 								  ASSERT(psInst->asOperands[0].eIndexRep[0] == OPERAND_INDEX_IMMEDIATE32);
 
 								  funcPointer = psInst->asOperands[0].aui32ArraySizes[0];
-								  funcTableIndex = psInst->asOperands[0].aui32ArraySizes[1];
 								  funcBodyIndex = psInst->ui32FuncIndexWithinInterface;
 
 								  ui32NumBodiesPerTable = psContext->psShader->funcPointer[funcPointer].ui32NumBodiesPerTable;
 
-								  funcTable = psContext->psShader->funcPointer[funcPointer].aui32FuncTables[funcTableIndex];
-
-								  funcBody = psContext->psShader->funcTable[funcTable].aui32FuncBodies[funcBodyIndex];
-
 								  varFound = psContext->psShader->sInfo.GetInterfaceVarFromOffset(funcPointer, &psVar);
 
 								  ASSERT(varFound);
@@ -3630,7 +3657,7 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 									break;
 								}
 
-								SHADER_VARIABLE_TYPE srcDataType;
+								SHADER_VARIABLE_TYPE srcDataType = SVT_FLOAT;
 								const ResourceBinding* psBinding = 0;
 								psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_UAV, psSrc->ui32RegisterNumber, &psBinding);
 								switch (psBinding->ui32ReturnType)
@@ -4005,13 +4032,13 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 		psContext->AddIndentation();
 		AddAssignToDest(&psInst->asOperands[0], SVT_UINT, 1, &numParenthesis);
 		if (isVulkan)
-			bcatcstr(glsl, "atomicAdd(");
+			bcatcstr(glsl, "(atomicAdd(");
 		else
 			bcatcstr(glsl, "atomicCounterDecrement(");
 		ResourceName(glsl, psContext, RGROUP_UAV, psInst->asOperands[1].ui32RegisterNumber, 0);
 		bformata(glsl, "_counter");
 		if (isVulkan)
-			bcatcstr(glsl, ", 0xffffffffu)");
+			bcatcstr(glsl, ", 0xffffffffu) + 0xffffffffu)");
 		else
 			bcatcstr(glsl, ")");
 		AddAssignPrologue(numParenthesis);
@@ -4071,7 +4098,21 @@ void ToGLSL::TranslateInstruction(Instruction* psInst, bool isEmbedded /* = fals
 		AddAssignPrologue(numParenthesis);
 		break;
 	}
-
+	case OPCODE_SAMPLE_INFO:
+	{
+#ifdef _DEBUG
+		psContext->AddIndentation();
+		bcatcstr(glsl, "//SAMPLE_INFO\n");
+#endif
+		const RESINFO_RETURN_TYPE eResInfoReturnType = psInst->eResInfoReturnType;
+		psContext->AddIndentation();
+		AddAssignToDest(&psInst->asOperands[0], eResInfoReturnType == RESINFO_INSTRUCTION_RETURN_UINT ? SVT_UINT : SVT_FLOAT, 1, &numParenthesis);
+		bcatcstr(glsl, "textureSamples(");
+		TranslateOperand(&psInst->asOperands[1], TO_FLAG_NAME_ONLY);
+		bcatcstr(glsl, ")");
+		AddAssignPrologue(numParenthesis);
+		break;
+	}
 	case OPCODE_DMAX:
 	case OPCODE_DMIN:
 	case OPCODE_DMUL:

src/toGLSLOperand.cpp

@@ -6,6 +6,7 @@
 #include "internal_includes/debug.h"
 #include "internal_includes/Shader.h"
 #include "internal_includes/toGLSL.h"
+#include "internal_includes/languages.h"
 #include <cmath>
 
 #include <sstream>
@@ -57,7 +58,11 @@ void TranslateOperandSwizzle(HLSLCrossCompilerContext* psContext, const Operand*
 
 void TranslateOperandSwizzleWithMask(HLSLCrossCompilerContext* psContext, const Operand* psOperand, uint32_t ui32ComponentMask, int iRebase)
 {
-	bstring glsl = *psContext->currentGLSLString;
+	TranslateOperandSwizzleWithMask(*psContext->currentGLSLString, psContext, psOperand, ui32ComponentMask, iRebase);
+}
+
+void TranslateOperandSwizzleWithMask(bstring glsl, HLSLCrossCompilerContext* psContext, const Operand* psOperand, uint32_t ui32ComponentMask, int iRebase)
+{
 	uint32_t accessMask = ui32ComponentMask & psOperand->GetAccessMask();
 	if(psOperand->eType == OPERAND_TYPE_INPUT)
 	{
@@ -400,11 +405,15 @@ static void printImmediate32(HLSLCrossCompilerContext *psContext, uint32_t value
 }
 
 void ToGLSL::TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui32TOFlag, uint32_t* pui32IgnoreSwizzle, uint32_t ui32CompMask, int *piRebase)
+{
+	TranslateVariableNameWithMask(*psContext->currentGLSLString, psOperand, ui32TOFlag, pui32IgnoreSwizzle, ui32CompMask, piRebase);
+}
+
+void ToGLSL::TranslateVariableNameWithMask(bstring glsl, const Operand* psOperand, uint32_t ui32TOFlag, uint32_t* pui32IgnoreSwizzle, uint32_t ui32CompMask, int *piRebase)
 {
 	int numParenthesis = 0;
 	int hasCtor = 0;
 	int needsBoolUpscale = 0; // If nonzero, bools need * 0xffffffff in them
-	bstring glsl = *psContext->currentGLSLString;
 	SHADER_VARIABLE_TYPE requestedType = TypeFlagsToSVTType(ui32TOFlag);
 	SHADER_VARIABLE_TYPE eType = psOperand->GetDataType(psContext, requestedType);
 	int numComponents = psOperand->GetNumSwizzleElements(ui32CompMask);
@@ -606,8 +615,22 @@ void ToGLSL::TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui
 						else
 						{
 							std::string name = psContext->GetDeclaredInputName(psOperand, piRebase, 0, pui32IgnoreSwizzle);
+
+							// Rewrite the variable name if we're using framebuffer fetch
+							if (psContext->psShader->extensions->EXT_shader_framebuffer_fetch &&
+								psContext->psShader->eShaderType == PIXEL_SHADER &&
+								psContext->flags & HLSLCC_FLAG_SHADER_FRAMEBUFFER_FETCH)
+							{
+								if(name == "vs_SV_Target0")
+									bcatcstr(glsl, "SV_Target0");
+								else
 									bcatcstr(glsl, name.c_str());
 							}
+							else
+							{
+								bcatcstr(glsl, name.c_str());
+							}
+						}
 					}
 					break;
 				}
@@ -781,7 +804,7 @@ void ToGLSL::TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui
 			const ShaderVarType* psVarType = NULL;
 			int32_t index = -1;
 			std::vector<uint32_t> arrayIndices;
-			bool isArray;
+			bool isArray = false;
 			psContext->psShader->sInfo.GetConstantBufferFromBindingPoint(RGROUP_CBUFFER, psOperand->aui32ArraySizes[0], &psCBuf);
 
 			switch(psContext->psShader->eShaderType)
@@ -817,6 +840,14 @@ void ToGLSL::TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui
 				}
 			}
 
+			if(psCBuf && psCBuf->name == "OVR_multiview")
+			{
+				pui32IgnoreSwizzle[0] = 1;
+				bformata(glsl, "gl_ViewID_OVR");
+				break;
+			}
+			
+			
 			if(ui32TOFlag & TO_FLAG_DECLARATION_NAME)
 			{
 				pui32IgnoreSwizzle[0] = 1;
@@ -1411,7 +1442,11 @@ void ToGLSL::TranslateVariableNameWithMask(const Operand* psOperand, uint32_t ui
 
 void ToGLSL::TranslateOperand(const Operand* psOperand, uint32_t ui32TOFlag, uint32_t ui32ComponentMask)
 {
-	bstring glsl = *psContext->currentGLSLString;
+	TranslateOperand(*psContext->currentGLSLString, psOperand, ui32TOFlag, ui32ComponentMask);
+}
+
+void ToGLSL::TranslateOperand(bstring glsl, const Operand* psOperand, uint32_t ui32TOFlag, uint32_t ui32ComponentMask)
+{
 	uint32_t ui32IgnoreSwizzle = 0;
 	int iRebase = 0;
 
@@ -1426,7 +1461,7 @@ void ToGLSL::TranslateOperand(const Operand* psOperand, uint32_t ui32TOFlag, uin
 
 	if(ui32TOFlag & TO_FLAG_NAME_ONLY)
 	{
-		TranslateVariableNameWithMask(psOperand, ui32TOFlag, &ui32IgnoreSwizzle, OPERAND_4_COMPONENT_MASK_ALL, &iRebase);
+		TranslateVariableNameWithMask(glsl, psOperand, ui32TOFlag, &ui32IgnoreSwizzle, OPERAND_4_COMPONENT_MASK_ALL, &iRebase);
 		return;
 	}
 
@@ -1453,7 +1488,7 @@ void ToGLSL::TranslateOperand(const Operand* psOperand, uint32_t ui32TOFlag, uin
 		}
 	}
 
-	TranslateVariableNameWithMask(psOperand, ui32TOFlag, &ui32IgnoreSwizzle, ui32ComponentMask, &iRebase);
+	TranslateVariableNameWithMask(glsl, psOperand, ui32TOFlag, &ui32IgnoreSwizzle, ui32ComponentMask, &iRebase);
 
 	if(psContext->psShader->eShaderType == HULL_SHADER && psOperand->eType == OPERAND_TYPE_OUTPUT &&
 		psOperand->ui32RegisterNumber != 0 && psOperand->iArrayElements != 0 && psOperand->eIndexRep[0] != OPERAND_INDEX_IMMEDIATE32_PLUS_RELATIVE
@@ -1464,7 +1499,7 @@ void ToGLSL::TranslateOperand(const Operand* psOperand, uint32_t ui32TOFlag, uin
 
 	if(!ui32IgnoreSwizzle)
 	{
-		TranslateOperandSwizzleWithMask(psContext, psOperand, ui32ComponentMask, iRebase);
+		TranslateOperandSwizzleWithMask(glsl, psContext, psOperand, ui32ComponentMask, iRebase);
 	}
 
 	switch(psOperand->eModifier)
@@ -1592,18 +1627,11 @@ std::string TextureSamplerName(ShaderInfo* psShaderInfo, const uint32_t ui32Text
 
 	if(ui32ArrayOffset)
 	{
-		oss << texName << ui32ArrayOffset << "_X_" << psSamplerBinding->name;
+		oss << texName << ui32ArrayOffset << "TEX_with_SMP" << psSamplerBinding->name;
 	}
 	else
 	{
-		if((i>0) && (texName[i-1] == '_'))//Prevent double underscore which is reserved
-		{
-			oss << texName << "X_" << psSamplerBinding->name;
-		}
-		else
-		{
-			oss << texName << "_X_" << psSamplerBinding->name;
-		}
+		oss << texName << "TEX_with_SMP" << psSamplerBinding->name;
 	}
 
 	return oss.str();

src/toMetal.cpp

@@ -81,13 +81,11 @@ bool ToMetal::Translate()
 
 	ClampPartialPrecisions();
 
-	psShader->PrepareStructuredBufferBindingSlots();
-
 	ShaderPhase &phase = psShader->asPhases[0];
 	phase.UnvectorizeImmMoves();
 	psContext->DoDataTypeAnalysis(&phase);
 	phase.ResolveUAVProperties();
-	psShader->ResolveStructuredBufferBindingSlots(&phase);
+	ReserveUAVBindingSlots(&phase); // TODO: unify slot allocation code between gl/metal/vulkan
 	phase.PruneConstArrays();
 	HLSLcc::DoLoopTransform(phase);
 
@@ -95,11 +93,17 @@ bool ToMetal::Translate()
 
 	bcatcstr(glsl, "#include <metal_stdlib>\n#include <metal_texture>\nusing namespace metal;\n");
 
-
 	for (i = 0; i < psShader->asPhases[0].psDecl.size(); ++i)
-	{
 		TranslateDeclaration(&psShader->asPhases[0].psDecl[i]);
-	}
+
+    // Output default implementations for framebuffer index remap if needed
+    if(m_NeedFBOutputRemapDecl)
+        bcatcstr(glsl, "#ifndef XLT_REMAP_O\n#define XLT_REMAP_O {0, 1, 2, 3, 4, 5, 6, 7}\n#endif\nconstexpr constant uint xlt_remap_o[] = XLT_REMAP_O;\n");
+    if(m_NeedFBInputRemapDecl)
+        bcatcstr(glsl, "#ifndef XLT_REMAP_I\n#define XLT_REMAP_I {0, 1, 2, 3, 4, 5, 6, 7}\n#endif\nconstexpr constant uint xlt_remap_i[] = XLT_REMAP_I;\n");
+    
+    DeclareClipPlanes(&psShader->asPhases[0].psDecl[0], psShader->asPhases[0].psDecl.size());
+	GenerateTexturesReflection(&psContext->m_Reflection);
 
 	if (m_StructDefinitions[GetInputStructName()].m_Members.size() > 0)
 	{
@@ -263,3 +267,25 @@ void ToMetal::ClampPartialPrecisions()
 			o->eMinPrecision = OPERAND_MIN_PRECISION_FLOAT_16;
 	});
 }
+
+void ToMetal::ReserveUAVBindingSlots(ShaderPhase *phase)
+{
+	for (uint32_t p = 0; p < phase->psDecl.size(); ++p)
+	{
+		uint32_t regNo = phase->psDecl[p].asOperands[0].ui32RegisterNumber;
+		
+		if (phase->psDecl[p].eOpcode == OPCODE_DCL_UNORDERED_ACCESS_VIEW_RAW ||
+			phase->psDecl[p].eOpcode == OPCODE_DCL_UNORDERED_ACCESS_VIEW_STRUCTURED)
+		{
+			m_BufferSlots.ReserveBindingSlot(regNo, BindingSlotAllocator::RWBuffer);
+		}
+		else if (phase->psDecl[p].eOpcode == OPCODE_DCL_UNORDERED_ACCESS_VIEW_TYPED)
+		{
+			// Typed buffers are atm faked using structured buffers -> bind in buffer space
+			if (phase->psDecl[p].value.eResourceDimension == RESOURCE_DIMENSION_BUFFER)
+				m_BufferSlots.ReserveBindingSlot(regNo, BindingSlotAllocator::RWBuffer);
+			else
+				m_TextureSlots.ReserveBindingSlot(regNo, BindingSlotAllocator::UAV);
+		}
+	}
+}

src/toMetalDeclaration.cpp

@@ -36,9 +36,13 @@ bool ToMetal::TranslateSystemValue(const Operand *psOperand, const ShaderInfo::I
 		if (outSkipPrefix != NULL) *outSkipPrefix = true;
 		return true;
 	case NAME_CLIP_DISTANCE:
-		result = "mtl_ClipDistance";
+	{
+		// this is temp variabe, declaration and redirecting to actual output is handled in DeclareClipPlanes
+		char tmpName[128]; sprintf(tmpName, "phase%d_ClipDistance%d", psContext->currentPhase, sig->ui32SemanticIndex);
+		result = tmpName;
 		if (outSkipPrefix != NULL) *outSkipPrefix = true;
 		return true;
+	}
 		/*	case NAME_VIEWPORT_ARRAY_INDEX:
 		result = "gl_ViewportIndex";
 		if (puiIgnoreSwizzle)
@@ -179,6 +183,80 @@ void ToMetal::DeclareBuiltinInput(const Declaration *psDecl)
 	}
 }
 
+void ToMetal::DeclareClipPlanes(const Declaration* decl, unsigned declCount)
+{
+	unsigned planeCount = 0;
+	for(unsigned i = 0, n = declCount ; i < n ; ++i)
+	{
+		const Operand* operand = &decl[i].asOperands[0];
+		if(operand->eSpecialName == NAME_CLIP_DISTANCE)
+			planeCount += operand->GetMaxComponent();
+	}
+	if(planeCount == 0) return;
+
+	std::ostringstream oss; oss << "float mtl_ClipDistance [[ clip_distance ]]";
+	if(planeCount > 1) oss << "[" << planeCount << "]";
+	m_StructDefinitions[GetOutputStructName()].m_Members.push_back(oss.str());
+
+	Shader* shader = psContext->psShader;
+
+	unsigned compCount = 1;
+	const ShaderInfo::InOutSignature* psFirstClipSignature;
+	if(shader->sInfo.GetOutputSignatureFromSystemValue(NAME_CLIP_DISTANCE, 0, &psFirstClipSignature))
+	{
+		if(psFirstClipSignature->ui32Mask & (1 << 3)) 		compCount = 4;
+		else if(psFirstClipSignature->ui32Mask & (1 << 2))	compCount = 3;
+		else if(psFirstClipSignature->ui32Mask & (1 << 1))	compCount = 2;
+	}
+
+	ShaderPhase* phase = &shader->asPhases[psContext->currentPhase];
+	for(unsigned i = 0, n = declCount ; i < n ; ++i)
+	{
+		const Operand* operand = &decl[i].asOperands[0];
+		if(operand->eSpecialName != NAME_CLIP_DISTANCE) continue;
+
+		const ShaderInfo::InOutSignature* signature = 0;
+		shader->sInfo.GetOutputSignatureFromRegister(operand->ui32RegisterNumber, operand->ui32CompMask, 0, &signature);
+		const int semanticIndex = signature->ui32SemanticIndex;
+
+		bformata(phase->earlyMain, "    float4 phase%d_ClipDistance%d;\n", psContext->currentPhase, signature->ui32SemanticIndex);
+
+		const char* swizzleStr[] = { "x", "y", "z", "w" };
+		phase->hasPostShaderCode = 1;
+		if(planeCount > 1)
+		{
+			for(int i = 0 ; i < compCount ; ++i)
+			{
+				bformata(phase->postShaderCode, "    %s.mtl_ClipDistance[%d] = phase%d_ClipDistance%d.%s;\n",
+					"output", semanticIndex*compCount + i, psContext->currentPhase, semanticIndex, swizzleStr[i]
+				);
+			}
+		}
+		else
+		{
+			bformata(phase->postShaderCode, "    %s.mtl_ClipDistance = phase%d_ClipDistance%d.x;\n", "output", psContext->currentPhase, semanticIndex);
+		}
+	}
+}
+void ToMetal::GenerateTexturesReflection(HLSLccReflection* refl)
+{
+	for(unsigned i = 0, n = m_Textures.size() ; i < n ; ++i)
+	{
+		const std::string samplerName1 = m_Textures[i].name, samplerName2 = "sampler"+m_Textures[i].name;
+		for(unsigned j = 0, m = m_Samplers.size() ; j < m ; ++j)
+		{
+			if(m_Samplers[j].name == samplerName1 || m_Samplers[j].name == samplerName2)
+			{
+				m_Textures[i].samplerBind = m_Samplers[j].slot;
+				break;
+			}
+		}
+	}
+
+	for(unsigned i = 0, n = m_Textures.size() ; i < n ; ++i)
+		refl->OnTextureBinding(m_Textures[i].name, m_Textures[i].textureBind, m_Textures[i].samplerBind, m_Textures[i].dim, m_Textures[i].uav);
+}
+
 void ToMetal::DeclareBuiltinOutput(const Declaration *psDecl)
 {
 	std::string out = GetOutputStructName();
@@ -198,7 +276,7 @@ void ToMetal::DeclareBuiltinOutput(const Declaration *psDecl)
 #endif
 			break;
 		case NAME_CLIP_DISTANCE:
-			m_StructDefinitions[out].m_Members.push_back("float4 mtl_ClipDistance [[ clip_distance ]]");
+			// it will be done separately in DeclareClipPlanes
 			break;
 
 		case NAME_VIEWPORT_ARRAY_INDEX:
@@ -329,7 +407,6 @@ void ToMetal::HandleOutputRedirect(const Declaration *psDecl, const std::string
 {
 	const Operand *psOperand = &psDecl->asOperands[0];
 	Shader *psShader = psContext->psShader;
-	bstring glsl = *psContext->currentGLSLString;
 	int needsRedirect = 0;
 	const ShaderInfo::InOutSignature *psSig = NULL;
 
@@ -425,7 +502,6 @@ void ToMetal::HandleInputRedirect(const Declaration *psDecl, const std::string &
 {
 	Operand *psOperand = (Operand *)&psDecl->asOperands[0];
 	Shader *psShader = psContext->psShader;
-	bstring glsl = *psContext->currentGLSLString;
 	int needsRedirect = 0;
 	const ShaderInfo::InOutSignature *psSig = NULL;
 
@@ -525,7 +601,7 @@ void ToMetal::HandleInputRedirect(const Declaration *psDecl, const std::string &
 }
 
 static std::string TranslateResourceDeclaration(HLSLCrossCompilerContext* psContext,
-	const Declaration *psDecl,
+	const Declaration *psDecl, const std::string& textureName,
 	bool isDepthSampler, bool isUAV)
 {
 	std::ostringstream oss;
@@ -591,7 +667,15 @@ static std::string TranslateResourceDeclaration(HLSLCrossCompilerContext* psCont
 			access = "read";
 	}
 
-	std::string typeName = HLSLcc::GetConstructorForTypeMetal(HLSLcc::ResourceReturnTypeToSVTType(eType, ePrec), 1);
+	SHADER_VARIABLE_TYPE svtType = HLSLcc::ResourceReturnTypeToSVTType(eType, ePrec);
+	std::string typeName = HLSLcc::GetConstructorForTypeMetal(svtType, 1);
+
+	if ((textureName == "_CameraDepthTexture" || textureName == "_LastCameraDepthTexture") && svtType != SVT_FLOAT)
+	{
+		std::string msg = textureName + " should be float on Metal (use sampler2D or sampler2D_float). Incorrect type "
+						  "can cause Metal validation failures or undefined results on some devices.";
+		psContext->m_Reflection.OnDiagnostics(msg, 0, false);
+	}
 
 	switch (eDimension)
 	{
@@ -870,19 +954,16 @@ void ToMetal::DeclareConstantBuffer(const ConstantBuffer *psCBuf, uint32_t ui32B
 
 }
 
-void ToMetal::DeclareBufferVariable(const Declaration *psDecl, const bool isRaw, const bool isUAV)
+void ToMetal::DeclareBufferVariable(const Declaration *psDecl, bool isRaw, bool isUAV)
 {
-	uint32_t ui32BindingPoint = psDecl->asOperands[0].ui32RegisterNumber;
-	std::string BufName, BufType;
+	uint32_t regNo = psDecl->asOperands[0].ui32RegisterNumber;
+	std::string BufName, BufType, BufConst;
 
 	BufName = "";
 	BufType = "";
+	BufConst = "";
 
-	// Use original HLSL bindings for UAVs only. For non-UAV buffers we have resolved new binding points from the same register space.
-	if (!isUAV)
-		ui32BindingPoint = psContext->psShader->aui32StructuredBufferBindingPoints[psContext->psShader->ui32CurrentStructuredBufferIndex++];
-
-	BufName = ResourceName(isUAV ? RGROUP_UAV : RGROUP_TEXTURE, psDecl->asOperands[0].ui32RegisterNumber);
+	BufName = ResourceName(isUAV ? RGROUP_UAV : RGROUP_TEXTURE, regNo);
 
 	if (!isRaw) // declare struct containing uint array when needed
 	{
@@ -898,7 +979,8 @@ void ToMetal::DeclareBufferVariable(const Declaration *psDecl, const bool isRaw,
 	
 	if (!isUAV || ((psDecl->sUAV.ui32AccessFlags & ACCESS_FLAG_WRITE) == 0))
 	{
-		oss << "const ";
+		BufConst =  "const ";
+		oss << BufConst;
 	}
 	else
 	{	
@@ -911,11 +993,138 @@ void ToMetal::DeclareBufferVariable(const Declaration *psDecl, const bool isRaw,
 	else
 		oss << "device " << BufType << " *" << BufName;
 
-	uint32_t loc = m_BufferSlots.GetBindingSlot(ui32BindingPoint, BindingSlotAllocator::RWBuffer);
+	uint32_t loc = m_BufferSlots.GetBindingSlot(regNo, isUAV ? BindingSlotAllocator::RWBuffer : BindingSlotAllocator::Texture);
 	oss << " [[ buffer(" << loc << ") ]]";
 
 	m_StructDefinitions[""].m_Members.push_back(oss.str());
 	psContext->m_Reflection.OnBufferBinding(BufName, loc, isUAV);
+	
+	
+	// In addition to the actual declaration, we need pointer modification and possible counter declaration
+	// in early main:
+	std::ostringstream earlymainoss;
+	
+	// Possible counter is always in the beginning of the buffer
+	if (isUAV && psDecl->sUAV.bCounter)
+	{
+		earlymainoss << "    device atomic_uint *" << BufName << "_counter = reinterpret_cast<device atomic_uint *> (" << BufName << ");\n";
+	}
+	
+	// Some GPUs don't allow memory access below buffer binding offset in the shader so always bind compute buffer
+	// at offset 0 instead of GetDataOffset().
+	// We can't tell at shader compile time if the buffer actually has counter or not. Therefore we'll always reserve
+	// space for the counter and bump the data pointer to beginning of the actual data here.
+	earlymainoss << "    " << BufName << " = reinterpret_cast<" << BufConst
+		<< "device " << (isRaw ? "uint" : BufType) << " *> (reinterpret_cast<device "
+		<< BufConst << "atomic_uint *> (" << BufName << ") + 1);\n";
+	
+	bformata(psContext->psShader->asPhases[psContext->currentPhase].earlyMain, earlymainoss.str().c_str());
+}
+
+
+static int ParseInlineSamplerWrapMode(const std::string& samplerName, const std::string& wrapName)
+{
+	int res = 0;
+	const bool hasWrap = (samplerName.find(wrapName) != std::string::npos);
+	if (!hasWrap)
+		return res;
+
+	const bool hasU = (samplerName.find(wrapName + 'u') != std::string::npos);
+	const bool hasV = (samplerName.find(wrapName + 'v') != std::string::npos);
+	const bool hasW = (samplerName.find(wrapName + 'w') != std::string::npos);
+
+	if (hasWrap) res |= 1;
+	if (hasU) res |= 2;
+	if (hasV) res |= 4;
+	if (hasW) res |= 8;
+	return res;
+}
+
+
+static bool EmitInlineSampler(HLSLCrossCompilerContext* ctx, const std::string& name)
+{
+	// See if it's a sampler that goes with the texture, or an "inline" sampler
+	// where sampler states are hardcoded in the shader directly.
+	//
+	// The logic for "inline" samplers below must match what is recognized
+	// by other shader platforms in Unity (ParseInlineSamplerName function
+	// in the shader compiler).
+	
+	std::string samplerName(name); std::transform(samplerName.begin(), samplerName.end(), samplerName.begin(), ::tolower);
+	
+	// filter modes
+	const bool hasPoint = (samplerName.find("point") != std::string::npos);
+	const bool hasTrilinear = (samplerName.find("trilinear") != std::string::npos);
+	const bool hasLinear = (samplerName.find("linear") != std::string::npos);
+	const bool hasAnyFilter = hasPoint || hasTrilinear || hasLinear;
+	
+	// wrap modes
+	const int bitsClamp = ParseInlineSamplerWrapMode(samplerName, "clamp");
+	const int bitsRepeat = ParseInlineSamplerWrapMode(samplerName, "repeat");
+	const int bitsMirror = ParseInlineSamplerWrapMode(samplerName, "mirror");
+	const int bitsMirrorOnce = ParseInlineSamplerWrapMode(samplerName, "mirroronce");
+	
+	const bool hasAnyWrap = bitsClamp != 0 || bitsRepeat != 0 || bitsMirror != 0 || bitsMirrorOnce != 0;
+	
+	// depth comparison
+	const bool hasCompare = (samplerName.find("compare") != std::string::npos);
+	
+	// name must contain a filter mode and a wrap mode at least
+	if (!hasAnyFilter || !hasAnyWrap)
+	{
+		return false;
+	}
+	
+	bstring str = ctx->psShader->asPhases[ctx->currentPhase].earlyMain;
+	bformata(str, "\tconstexpr sampler %s(", name.c_str());
+
+	if (hasCompare)
+		bformata(str, "compare_func::greater_equal,");
+
+	if (hasTrilinear)
+		bformata(str, "filter::linear,mip_filter::linear,");
+	else if (hasLinear)
+		bformata(str, "filter::linear,");
+	else
+		bformata(str, "filter::nearest,");
+
+	const char* kTexWrapClamp = "clamp_to_edge";
+	const char* kTexWrapRepeat = "repeat";
+	const char* kTexWrapMirror = "mirrored_repeat";
+	const char* kTexWrapMirrorOnce = "mirrored_repeat"; // currently Metal shading language does not have syntax for inline sampler state that would do "mirror clamp to edge"
+	const char* wrapU = kTexWrapRepeat;
+	const char* wrapV = kTexWrapRepeat;
+	const char* wrapW = kTexWrapRepeat;
+
+	if (bitsClamp == 1)				wrapU = wrapV = wrapW = kTexWrapClamp;
+	else if (bitsRepeat == 1)		wrapU = wrapV = wrapW = kTexWrapRepeat;
+	else if (bitsMirrorOnce == 1)	wrapU = wrapV = wrapW = kTexWrapMirrorOnce;
+	else if (bitsMirror == 1)		wrapU = wrapV = wrapW = kTexWrapMirror;
+
+	if ((bitsClamp & 2) != 0)	wrapU = kTexWrapClamp;
+	if ((bitsClamp & 4) != 0)	wrapV = kTexWrapClamp;
+	if ((bitsClamp & 8) != 0)	wrapW = kTexWrapClamp;
+
+	if ((bitsRepeat & 2) != 0)	wrapU = kTexWrapRepeat;
+	if ((bitsRepeat & 4) != 0)	wrapV = kTexWrapRepeat;
+	if ((bitsRepeat & 8) != 0)	wrapW = kTexWrapRepeat;
+
+	if ((bitsMirrorOnce & 2) != 0)	wrapU = kTexWrapMirrorOnce;
+	if ((bitsMirrorOnce & 4) != 0)	wrapV = kTexWrapMirrorOnce;
+	if ((bitsMirrorOnce & 8) != 0)	wrapW = kTexWrapMirrorOnce;
+
+	if ((bitsMirror & 2) != 0)	wrapU = kTexWrapMirror;
+	if ((bitsMirror & 4) != 0)	wrapV = kTexWrapMirror;
+	if ((bitsMirror & 8) != 0)	wrapW = kTexWrapMirror;
+
+	if (wrapU == wrapV && wrapU == wrapW)
+		bformata(str, "address::%s", wrapU);
+	else
+		bformata(str, "s_address::%s,t_address::%s,r_address::%s", wrapU, wrapV, wrapW);
+
+	bformata(str, ");\n");
+
+	return true;
 }
 
 
@@ -946,12 +1155,10 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 
 		std::string name = psContext->GetDeclaredInputName(psOperand, nullptr, 1, nullptr);
 
-		//Already declared as part of an array?
-		if (psShader->aIndexedInput[0][psDecl->asOperands[0].ui32RegisterNumber] == -1)
-		{
-			ASSERT(0); // Find out what's happening
-			break;
-		}
+		// NB: unlike GL we keep arrays of 2-component vectors as is (without collapsing into float4)
+		// if(psShader->aIndexedInput[0][psDecl->asOperands[0].ui32RegisterNumber] == -1)
+		//     break;
+
 		// Already declared?
 		if ((ui32CompMask != 0) && ((ui32CompMask & ~psShader->acInputDeclared[0][ui32Reg]) == 0))
 		{
@@ -1041,7 +1248,8 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 				//
 				// TODO: Improve later when GLES3 support arrives, it requires
 				// single declaration through inout
-				oss << "color(" << psSig->ui32SemanticIndex << ")";
+				oss << "color(xlt_remap_i[" << psSig->ui32SemanticIndex << "])";
+                m_NeedFBInputRemapDecl = true;
 			}
 			else
 			{
@@ -1115,6 +1323,49 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 		const ConstantBuffer* psCBuf = NULL;
 		psContext->psShader->sInfo.GetConstantBufferFromBindingPoint(RGROUP_CBUFFER, psDecl->asOperands[0].aui32ArraySizes[0], &psCBuf);
 		ASSERT(psCBuf != NULL);
+
+		if (psCBuf->name.substr(0, 20) == "hlslcc_SubpassInput_" && psCBuf->name.length() >= 23 && !psCBuf->asVars.empty())
+		{
+			// Special case for framebuffer fetch.
+			char ty = psCBuf->name[20];
+			int idx = psCBuf->name[22] - '0';
+
+			const ShaderVar &sv = psCBuf->asVars[0];
+			if (sv.name.substr(0, 15) == "hlslcc_fbinput_")
+			{
+				// Pick up the type and index
+				std::ostringstream oss;
+                m_NeedFBInputRemapDecl = true;
+				switch (ty)
+				{
+				case 'f':
+				case 'F':
+					oss << "float4 " << sv.name << " [[ color(xlt_remap_i["<< idx <<"]) ]]";
+					m_StructDefinitions[""].m_Members.push_back(oss.str());
+					break;
+				case 'h':
+				case 'H':
+					oss << "half4 " << sv.name << " [[ color(xlt_remap_i[" << idx << "]) ]]";
+					m_StructDefinitions[""].m_Members.push_back(oss.str());
+					break;
+				case 'i':
+				case 'I':
+					oss << "int4 " << sv.name << " [[ color(xlt_remap_i[" << idx << "]) ]]";
+					m_StructDefinitions[""].m_Members.push_back(oss.str());
+					break;
+				case 'u':
+				case 'U':
+					oss << "uint4 " << sv.name << " [[ color(xlt_remap_i[" << idx << "]) ]]";
+					m_StructDefinitions[""].m_Members.push_back(oss.str());
+					break;
+				default:
+					break;
+				}
+			}
+			// Break out so this doesn't get declared. 
+			break;
+		}
+
 		DeclareConstantBuffer(psCBuf, psDecl->asOperands[0].aui32ArraySizes[0]);
 		break;
 	}
@@ -1373,9 +1624,9 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 				uint32_t destMask = psDecl->asOperands[0].ui32CompMask;
 				uint32_t rebase = 0;
 				const ShaderInfo::InOutSignature *psSig = NULL;
-				uint32_t regSpace = psDecl->asOperands[0].GetRegisterSpace(psContext) == 0;
+				uint32_t regSpace = psDecl->asOperands[0].GetRegisterSpace(psContext);
 
-				if (regSpace)
+				if (regSpace == 0)
 					if (isInput)
 						psContext->psShader->sInfo.GetInputSignatureFromRegister(startReg + i, destMask, &psSig);
 					else
@@ -1414,8 +1665,9 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 							}
 						}
 					}
-					bcatcstr(glsl, " = ");
-					bcatcstr(glsl, realName.c_str());
+
+					// for some reason input struct is missed here from GetDeclaredInputName result, so add it manually
+					bformata(glsl, " = input.%s", realName.c_str());
 					if (destMask != OPERAND_4_COMPONENT_MASK_ALL && destMask != psSig->ui32Mask)
 					{
 						int k;
@@ -1537,25 +1789,20 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 	}
 	case OPCODE_DCL_SAMPLER:
 	{
-		// Find out if the sampler is good for a builtin
 		std::string name = TranslateOperand(&psDecl->asOperands[0], TO_FLAG_NAME_ONLY);
-		std::transform(name.begin(), name.end(), name.begin(), ::tolower);
-		bool linear = (name.find("linear") != std::string::npos);
-		bool point = (name.find("point") != std::string::npos);
-		bool clamp = (name.find("clamp") != std::string::npos);
-		bool repeat = (name.find("repeat") != std::string::npos);
 		
-		// Declare only builtin samplers here. Default samplers are declared together with the texture.
-		if ((linear != point) && (clamp != repeat))
+		if (!EmitInlineSampler(psContext, name))
 		{
-			std::ostringstream oss;
-			oss << "constexpr sampler " << TranslateOperand(&psDecl->asOperands[0], TO_FLAG_NAME_ONLY) << "(";
-			oss << (linear ? "filter::linear, " : "filter::nearest, ");
-			oss << (clamp ? "address::clamp_to_edge" : "address::repeat");
-			oss << ")";
+			// for some reason we have some samplers start with "sampler" and some not
+			const bool startsWithSampler = name.find("sampler") == 0;
+			const uint32_t slot = m_SamplerSlots.GetBindingSlot(psDecl->asOperands[0].ui32RegisterNumber, BindingSlotAllocator::Texture);
+			std::ostringstream oss; oss << "sampler " << (startsWithSampler ? "" : "sampler") << name << " [[ sampler (" << slot << ") ]]";
+			m_StructDefinitions[""].m_Members.push_back(oss.str());
 
-			bformata(psContext->psShader->asPhases[psContext->currentPhase].earlyMain, "\t%s;\n", oss.str().c_str());
+			SamplerDesc desc = { name, psDecl->asOperands[0].ui32RegisterNumber, slot };
+			m_Samplers.push_back(desc);
 		}
+
 		break;
 	}
 	case OPCODE_DCL_HS_MAX_TESSFACTOR:
@@ -1565,57 +1812,48 @@ void ToMetal::TranslateDeclaration(const Declaration* psDecl)
 	}
 	case OPCODE_DCL_UNORDERED_ACCESS_VIEW_TYPED:
 	{
-		std::string samplerTypeName = TranslateResourceDeclaration(psContext,
-			psDecl, false, true);
+		// A hack to support single component 32bit RWBuffers: Declare as raw buffer.
+		// TODO: Use textures for RWBuffers when the scripting API has actual format selection etc
+		// way to flag the created ComputeBuffer as typed. Even then might want to leave this 
+		// hack path for 32bit (u)int typed buffers to continue support atomic ops on those formats.
+		if (psDecl->value.eResourceDimension == RESOURCE_DIMENSION_BUFFER)
+		{
+			DeclareBufferVariable(psDecl, true, true);
+			break;
+		}
 		std::string texName = ResourceName(RGROUP_UAV, psDecl->asOperands[0].ui32RegisterNumber);
+		std::string samplerTypeName = TranslateResourceDeclaration(psContext, psDecl, texName, false, true);
 		uint32_t slot = m_TextureSlots.GetBindingSlot(psDecl->asOperands[0].ui32RegisterNumber, BindingSlotAllocator::UAV);		std::ostringstream oss;
 		oss << samplerTypeName << " " << texName
 			<< " [[ texture (" << slot << ") ]] ";
 
 		m_StructDefinitions[""].m_Members.push_back(oss.str());
-		psContext->m_Reflection.OnTextureBinding(texName, slot, TD_2D, true); // TODO: translate psDecl->value.eResourceDimension into HLSLCC_TEX_DIMENSION
+
+		// TODO: translate psDecl->value.eResourceDimension into HLSLCC_TEX_DIMENSION
+		TextureSamplerDesc desc = {texName, (int)slot, -1, TD_2D, true};
+		m_Textures.push_back(desc);
 
 		break;
 	}
 
 	case OPCODE_DCL_UNORDERED_ACCESS_VIEW_STRUCTURED:
 	{
-		if (psDecl->sUAV.bCounter)
-		{
-			std::ostringstream oss;
-			std::string bufName = ResourceName(RGROUP_UAV, psDecl->asOperands[0].ui32RegisterNumber);
-			
-			// Some GPUs don't allow memory access below buffer binding offset in the shader so always bind compute buffer
-			// at offset 0 instead of GetDataOffset() to access counter value and translate the buffer pointer in the shader.
-			oss << "device atomic_uint *" << bufName << "_counter = reinterpret_cast<device atomic_uint *> (" << bufName << ");";
-			oss << "\n    " << bufName << " = reinterpret_cast<device " << bufName << "_Type *> (reinterpret_cast<device atomic_uint *> (" << bufName << ") + 1)";
-			bformata(psContext->psShader->asPhases[psContext->currentPhase].earlyMain, "    %s;\n", oss.str().c_str());
-		}
-
-		DeclareBufferVariable(psDecl, 0, 1);
+		DeclareBufferVariable(psDecl, false, true);
 		break;
 	}
 	case OPCODE_DCL_UNORDERED_ACCESS_VIEW_RAW:
 	{
-		if (psDecl->sUAV.bCounter)
-		{
-			std::ostringstream oss;
-			std::string bufName = ResourceName(RGROUP_UAV, psDecl->asOperands[0].ui32RegisterNumber);
-			oss << "device atomic_uint *" << bufName << "_counter = reinterpret_cast<atomic_uint *> (" << bufName << ") - 1";
-			bformata(psContext->psShader->asPhases[psContext->currentPhase].earlyMain, "\t%s;\n", oss.str().c_str());		}
-
-		DeclareBufferVariable(psDecl, 1, 1);
-
+		DeclareBufferVariable(psDecl, true, true);
 		break;
 	}
 	case OPCODE_DCL_RESOURCE_STRUCTURED:
 	{
-		DeclareBufferVariable(psDecl, 0, 0);
+		DeclareBufferVariable(psDecl, false, false);
 		break;
 	}
 	case OPCODE_DCL_RESOURCE_RAW:
 	{
-		DeclareBufferVariable(psDecl, 1, 0);
+		DeclareBufferVariable(psDecl, true, false);
 		break;
 	}
 	case OPCODE_DCL_THREAD_GROUP_SHARED_MEMORY_STRUCTURED:
@@ -1718,22 +1956,19 @@ std::string ToMetal::ResourceName(ResourceGroup group, const uint32_t ui32Regist
 void ToMetal::TranslateResourceTexture(const Declaration* psDecl, uint32_t samplerCanDoShadowCmp, HLSLCC_TEX_DIMENSION texDim)
 {
 
+	std::string texName = ResourceName(RGROUP_TEXTURE, psDecl->asOperands[0].ui32RegisterNumber);
 	std::string samplerTypeName = TranslateResourceDeclaration(psContext,
-		psDecl, (samplerCanDoShadowCmp && psDecl->ui32IsShadowTex), false);
+		psDecl, texName, (samplerCanDoShadowCmp && psDecl->ui32IsShadowTex), false);
 
 	uint32_t slot = m_TextureSlots.GetBindingSlot(psDecl->asOperands[0].ui32RegisterNumber, BindingSlotAllocator::Texture);
-	std::string texName = ResourceName(RGROUP_TEXTURE, psDecl->asOperands[0].ui32RegisterNumber);
 	std::ostringstream oss;
 	oss << samplerTypeName << " " << texName
 		<< " [[ texture (" << slot << ") ]] ";
 
 	m_StructDefinitions[""].m_Members.push_back(oss.str());
-	psContext->m_Reflection.OnTextureBinding(texName, slot, texDim, false);
-	oss.str("");
-	// the default sampler for a texture is named after the texture with a "sampler" prefix
-	oss << "sampler sampler" << texName
-		<< " [[ sampler (" << slot << ") ]] ";
-	m_StructDefinitions[""].m_Members.push_back(oss.str());
+
+	TextureSamplerDesc desc = {texName, (int)slot, -1, texDim, false};
+	m_Textures.push_back(desc);
 
 	if (samplerCanDoShadowCmp && psDecl->ui32IsShadowTex)
 		EnsureShadowSamplerDeclared();
@@ -1746,17 +1981,21 @@ void ToMetal::DeclareResource(const Declaration *psDecl)
 	{
 		case RESOURCE_DIMENSION_BUFFER:
 		{
+            // Fake single comp 32bit texel buffers by using raw buffer
+            DeclareBufferVariable(psDecl, true, false);
+            break;
+            
+			/* TODO: re-enable this code for buffer textures when sripting API has proper support for it
             uint32_t slot = m_TextureSlots.GetBindingSlot(psDecl->asOperands[0].ui32RegisterNumber, BindingSlotAllocator::Texture);
 			std::string texName = TranslateOperand(&psDecl->asOperands[0], TO_FLAG_NAME_ONLY);
 			std::ostringstream oss;
-			oss << "device " << TranslateResourceDeclaration(psContext,
-				psDecl, false, false);
+			oss << "device " << TranslateResourceDeclaration(psContext, psDecl, texName, false, false);
 
 			oss << texName << " [[ texture(" << slot << ") ]]";
 
 			m_StructDefinitions[""].m_Members.push_back(oss.str());
 			psContext->m_Reflection.OnTextureBinding(texName, slot, TD_2D, false); //TODO: correct HLSLCC_TEX_DIMENSION?
-			break;
+			break;*/
 
 		}
 		default:
@@ -1816,7 +2055,6 @@ void ToMetal::DeclareResource(const Declaration *psDecl)
 
 void ToMetal::DeclareOutput(const Declaration *psDecl)
 {
-	bstring glsl = *psContext->currentGLSLString;
 	Shader* psShader = psContext->psShader;
 
 	if (!psContext->OutputNeedsDeclaring(&psDecl->asOperands[0], 1))
@@ -1934,7 +2172,8 @@ void ToMetal::DeclareOutput(const Declaration *psDecl)
 		default:
 		{
 			std::ostringstream oss;
-			oss << type << " " << name << " [[ color(" << psSignature->ui32SemanticIndex << ") ]]";
+			oss << type << " " << name << " [[ color(xlt_remap_o[" << psSignature->ui32SemanticIndex << "]) ]]";
+            m_NeedFBOutputRemapDecl = true;
 			m_StructDefinitions[GetOutputStructName()].m_Members.push_back(oss.str());
 		}
 		}

src/toMetalInstruction.cpp

@@ -111,7 +111,7 @@ void ToMetal::AddOpAssignToDestWithMask(const Operand* psDest,
 	case SVT_FLOAT:
 	case SVT_FLOAT10:
 	case SVT_FLOAT16:
-		ASSERT(eSrcType != SVT_INT12 || eSrcType != SVT_INT16 && eSrcType != SVT_UINT16);
+		ASSERT(eSrcType != SVT_INT12 || (eSrcType != SVT_INT16 && eSrcType != SVT_UINT16));
 		if (psContext->psShader->ui32MajorVersion > 3)
 		{
 			if (ui32DestElementCount > 1)
@@ -646,7 +646,6 @@ void ToMetal::TranslateTexelFetch(
 	bstring glsl)
 {
 	int numParenthesis = 0;
-	uint32_t destCount = psInst->asOperands[0].GetNumSwizzleElements();
 	psContext->AddIndentation();
 	AddAssignToDest(&psInst->asOperands[0], psContext->psShader->sInfo.GetTextureDataType(psInst->asOperands[2].ui32RegisterNumber), 4, &numParenthesis);
 	glsl << TranslateOperand(&psInst->asOperands[2], TO_FLAG_NONE);
@@ -714,7 +713,7 @@ void ToMetal::TranslateTexelFetch(
 	}
 	bcatcstr(glsl, ")");
 
-	AddSwizzleUsingElementCount(glsl, destCount);
+	glsl << TranslateOperandSwizzle(&psInst->asOperands[2], psInst->asOperands[0].GetAccessMask(), 0);
 	AddAssignPrologue(numParenthesis);
 }
 
@@ -724,7 +723,6 @@ void ToMetal::TranslateTexelFetchOffset(
 	bstring glsl)
 {
 	int numParenthesis = 0;
-	uint32_t destCount = psInst->asOperands[0].GetNumSwizzleElements();
 	psContext->AddIndentation();
 	AddAssignToDest(&psInst->asOperands[0], psContext->psShader->sInfo.GetTextureDataType(psInst->asOperands[2].ui32RegisterNumber), 4, &numParenthesis);
 
@@ -799,7 +797,7 @@ void ToMetal::TranslateTexelFetchOffset(
 	}
 	bcatcstr(glsl, ")");
 
-	AddSwizzleUsingElementCount(glsl, destCount);
+	glsl << TranslateOperandSwizzle(&psInst->asOperands[2], psInst->asOperands[0].GetAccessMask(), 0);
 	AddAssignPrologue(numParenthesis);
 }
 
@@ -813,6 +811,7 @@ void ToMetal::TranslateTexCoord(
 {
 	uint32_t flags = TO_AUTO_BITCAST_TO_FLOAT;
 	uint32_t opMask = OPERAND_4_COMPONENT_MASK_ALL;
+	bool isArray = false;
 
 	switch (eResDim)
 	{
@@ -822,8 +821,21 @@ void ToMetal::TranslateTexCoord(
 		opMask = OPERAND_4_COMPONENT_MASK_X;
 		break;
 	}
-	case RESOURCE_DIMENSION_TEXTURE2D:
 	case RESOURCE_DIMENSION_TEXTURE1DARRAY:
+	{
+		// x for coord, y for array element
+		opMask = OPERAND_4_COMPONENT_MASK_X;
+		bstring glsl = *psContext->currentGLSLString;
+		glsl << TranslateOperand(psTexCoordOperand, flags, opMask);
+		
+		bcatcstr(glsl, ", round(");
+		
+		opMask = OPERAND_4_COMPONENT_MASK_Y;
+		flags = TO_AUTO_BITCAST_TO_FLOAT;
+		isArray = true;
+		break;
+	}
+	case RESOURCE_DIMENSION_TEXTURE2D:
 	{
 		//Vec2 texcoord. Mask out the other components.
 		opMask = OPERAND_4_COMPONENT_MASK_X | OPERAND_4_COMPONENT_MASK_Y;
@@ -840,22 +852,34 @@ void ToMetal::TranslateTexCoord(
 	}
 	case RESOURCE_DIMENSION_TEXTURE2DARRAY:
 	{
+		// xy for coord, z for array element
 		opMask = OPERAND_4_COMPONENT_MASK_X | OPERAND_4_COMPONENT_MASK_Y;
 		flags |= TO_AUTO_EXPAND_TO_VEC2;
 		
 		bstring glsl = *psContext->currentGLSLString;
 		glsl << TranslateOperand(psTexCoordOperand, flags, opMask);
 		
-		bcatcstr(glsl, ", ");
+		bcatcstr(glsl, ", round(");
 
 		opMask = OPERAND_4_COMPONENT_MASK_Z;
 		flags = TO_AUTO_BITCAST_TO_FLOAT;
-
+		isArray = true;
 		break;
 	}
 	case RESOURCE_DIMENSION_TEXTURECUBEARRAY:
 	{
-		flags |= TO_AUTO_EXPAND_TO_VEC4;
+		// xyz for coord, w for array element
+		opMask = OPERAND_4_COMPONENT_MASK_X | OPERAND_4_COMPONENT_MASK_Y | OPERAND_4_COMPONENT_MASK_Z;
+		flags |= TO_AUTO_EXPAND_TO_VEC3;
+		
+		bstring glsl = *psContext->currentGLSLString;
+		glsl << TranslateOperand(psTexCoordOperand, flags, opMask);
+		
+		bcatcstr(glsl, ", round(");
+		
+		opMask = OPERAND_4_COMPONENT_MASK_W;
+		flags = TO_AUTO_BITCAST_TO_FLOAT;
+		isArray = true;
 		break;
 	}
 	default:
@@ -868,6 +892,10 @@ void ToMetal::TranslateTexCoord(
 	//FIXME detect when integer coords are needed.
 	bstring glsl = *psContext->currentGLSLString;
 	glsl << TranslateOperand(psTexCoordOperand, flags, opMask);
+	
+	if (isArray)
+		bcatcstr(glsl, ")");
+
 }
 
 void ToMetal::GetResInfoData(Instruction* psInst, int index, int destElem)
@@ -875,12 +903,11 @@ void ToMetal::GetResInfoData(Instruction* psInst, int index, int destElem)
 	bstring glsl = *psContext->currentGLSLString;
 	int numParenthesis = 0;
 	const RESINFO_RETURN_TYPE eResInfoReturnType = psInst->eResInfoReturnType;
-	const int isUAV = (psInst->asOperands[2].eType == OPERAND_TYPE_UNORDERED_ACCESS_VIEW);
 
 	psContext->AddIndentation();
 	AddOpAssignToDestWithMask(&psInst->asOperands[0], eResInfoReturnType == RESINFO_INSTRUCTION_RETURN_UINT ? SVT_UINT : SVT_FLOAT, 1, "=", &numParenthesis, 1 << destElem);
 
-	const char *metalGetters[] = { ".get_width()", ".get_height()", ".get_depth()", ".get_num_mip_levels()" };
+	const char *metalGetters[] = { ".get_width(", ".get_height(", ".get_depth(", ".get_num_mip_levels()" };
 	int dim = GetNumTextureDimensions(psInst->eResDim);
 	if (dim < (index + 1) && index != 3)
 	{
@@ -899,17 +926,25 @@ void ToMetal::GetResInfoData(Instruction* psInst, int index, int destElem)
 			numParenthesis++;
 		}
 		glsl << TranslateOperand(&psInst->asOperands[2], TO_FLAG_NONE);
-		if (index == 2 &&
-			(psInst->eResDim == RESOURCE_DIMENSION_TEXTURE1DARRAY ||
-				psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DARRAY ||
-				psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DMSARRAY))
+		if ((index == 1 && psInst->eResDim == RESOURCE_DIMENSION_TEXTURE1DARRAY) ||
+			(index == 2 && (psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DARRAY ||
+				psInst->eResDim == RESOURCE_DIMENSION_TEXTURE2DMSARRAY)))
 		{
 			bcatcstr(glsl, ".get_array_size()");
 		}
 		else
+		{
 			bcatcstr(glsl, metalGetters[index]);
 			
-		// TODO Metal has no way to query for info on lower mip levels, now always returns info for highest.
+			if (index < 3)
+			{
+				if (psInst->eResDim != RESOURCE_DIMENSION_TEXTURE2DMS &&
+					psInst->eResDim != RESOURCE_DIMENSION_TEXTURE2DMSARRAY)
+					glsl << TranslateOperand(&psInst->asOperands[1], TO_FLAG_INTEGER); //mip level
+			
+				bcatcstr(glsl, ")");
+			}
+		}
 	}
 	AddAssignPrologue(numParenthesis);
 }
@@ -933,7 +968,6 @@ void ToMetal::TranslateTextureSample(Instruction* psInst,
 	Operand* psSrcBias = (ui32Flags & TEXSMP_FLAG_BIAS) ? &psInst->asOperands[4] : 0;
 
 	const char *funcName = "";
-	const char* offset = "";
 	const char* gradSwizzle = "";
 	const char *gradientName = "";
 
@@ -1165,7 +1199,6 @@ void ToMetal::TranslateTextureSample(Instruction* psInst,
 	AddAssignPrologue(numParenthesis);
 }
 
-static const char* swizzleString[] = { ".x", ".y", ".z", ".w" };
 
 // Handle cases where vector components are accessed with dynamic index ([] notation).
 // A bit ugly hack because compiled HLSL uses byte offsets to access data in structs => we are converting
@@ -1212,6 +1245,7 @@ void ToMetal::TranslateShaderStorageStore(Instruction* psInst)
 
 		break;
 	case OPCODE_STORE_RAW:
+	case OPCODE_STORE_UAV_TYPED: // Hack typed buffer as raw buf
 		psDest = &psInst->asOperands[0];
 		psDestByteOff = &psInst->asOperands[1];
 		psSrc = &psInst->asOperands[2];
@@ -1243,6 +1277,12 @@ void ToMetal::TranslateShaderStorageStore(Instruction* psInst)
 
 			bcatcstr(glsl, "[(");
 			glsl << TranslateOperand(psDestByteOff, dstOffFlag);
+			if (psInst->eOpcode == OPCODE_STORE_UAV_TYPED)
+			{
+				bcatcstr(glsl, ")");
+			}
+			else
+			{
 				bcatcstr(glsl, " >> 2");
 				if (dstOffFlag == TO_FLAG_UNSIGNED_INTEGER)
 					bcatcstr(glsl, "u");
@@ -1254,7 +1294,7 @@ void ToMetal::TranslateShaderStorageStore(Instruction* psInst)
 					if (dstOffFlag == TO_FLAG_UNSIGNED_INTEGER)
 						bcatcstr(glsl, "u");
 				}
-
+			}
 			bcatcstr(glsl, "]");
 
 			//Dest type is currently always a uint array.
@@ -1287,6 +1327,7 @@ void ToMetal::TranslateShaderStorageLoad(Instruction* psInst)
 		psSrc = &psInst->asOperands[3];
 		break;
 	case OPCODE_LD_RAW:
+	case OPCODE_LD_UAV_TYPED: // Hack typed buffer as raw buf
 		psDest = &psInst->asOperands[0];
 		psSrcByteOff = &psInst->asOperands[1];
 		psSrc = &psInst->asOperands[2];
@@ -1348,6 +1389,12 @@ void ToMetal::TranslateShaderStorageLoad(Instruction* psInst)
 		}
 		bcatcstr(glsl, "[(");
 		glsl << TranslateOperand(psSrcByteOff, srcOffFlag);
+		if (psInst->eOpcode == OPCODE_LD_UAV_TYPED)
+		{
+			bcatcstr(glsl, ")");
+		}
+		else
+		{
 			bcatcstr(glsl, " >> 2");
 			if (srcOffFlag == TO_FLAG_UNSIGNED_INTEGER)
 				bcatcstr(glsl, "u");
@@ -1355,7 +1402,7 @@ void ToMetal::TranslateShaderStorageLoad(Instruction* psInst)
 			bformata(glsl, ") + %d", psSrc->eSelMode == OPERAND_4_COMPONENT_SWIZZLE_MODE ? psSrc->aui32Swizzle[component] : component);
 			if (srcOffFlag == TO_FLAG_UNSIGNED_INTEGER)
 				bcatcstr(glsl, "u");
-
+		}
 		bcatcstr(glsl, "]");
 		
 		if (addedBitcast)
@@ -1635,9 +1682,9 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 
 	psContext->AddIndentation();
 
+	const ResourceBinding* psBinding = 0;
 	if (dest->eType != OPERAND_TYPE_THREAD_GROUP_SHARED_MEMORY)
 	{
-		const ResourceBinding* psBinding = 0;
 		psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_UAV, dest->ui32RegisterNumber, &psBinding);
 
 		if (psBinding->eType == RTYPE_UAV_RWTYPED)
@@ -1662,6 +1709,8 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 			case REFLECT_RESOURCE_DIMENSION_TEXTURECUBEARRAY:
 				texDim = 3;
 				break;
+			case REFLECT_RESOURCE_DIMENSION_BUFFER: // Hack typed buffer as raw buf
+				break;
             default:
                 ASSERT(0);
                 break;
@@ -1683,14 +1732,14 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 	if (shouldExtractCompare)
 	{
 		bcatcstr(glsl, "{\n");
-		psContext->AddIndentation();
+		++psContext->indent;
 		psContext->AddIndentation();
 		bcatcstr(glsl, "uint compare_value = ");
 		glsl << TranslateOperand(compare, ui32DataTypeFlag);
-		bcatcstr(glsl, "; ");
+		bcatcstr(glsl, ";\n");
+		psContext->AddIndentation();
 	}
-
-	if (previousValue)
+	else if (previousValue)
 		AddAssignToDest(previousValue, isUint ? SVT_UINT : SVT_INT, 1, &numParenthesis);
 
 	bcatcstr(glsl, func);
@@ -1709,6 +1758,8 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 	bcatcstr(glsl, "[");
 	glsl << TranslateOperand(destAddr, destAddrFlag, OPERAND_4_COMPONENT_MASK_X);
 	
+	if (!psBinding || psBinding->eType != RTYPE_UAV_RWTYPED)
+	{
 		// Structured buf if we have both x & y swizzles. Raw buf has only x -> no .value[]
 		if (destAddr->GetNumSwizzleElements(OPERAND_4_COMPONENT_MASK_X | OPERAND_4_COMPONENT_MASK_Y) == 2)
 		{
@@ -1720,7 +1771,7 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 		bcatcstr(glsl, " >> 2");//bytes to floats
 		if (destAddrFlag == TO_FLAG_UNSIGNED_INTEGER)
 			bcatcstr(glsl, "u");
-
+	}
 	bcatcstr(glsl, "]), ");
 
 	if (compare)
@@ -1750,6 +1801,14 @@ void ToMetal::TranslateAtomicMemOp(Instruction* psInst)
 
 	if (shouldExtractCompare)
 	{
+		if (previousValue)
+		{
+			psContext->AddIndentation();
+			AddAssignToDest(previousValue, SVT_UINT, 1, &numParenthesis);
+			bcatcstr(glsl, "compare_value");
+			AddAssignPrologue(numParenthesis);
+		}
+		--psContext->indent;
 		psContext->AddIndentation();
 		bcatcstr(glsl, "}\n");
 	}
@@ -1834,8 +1893,9 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 	int numParenthesis = 0;
 
 #ifdef _DEBUG
-	psContext->AddIndentation();
-	bformata(glsl, "//Instruction %d\n", psInst->id);
+	// Uncomment to print instruction IDs
+	//psContext->AddIndentation();
+	//bformata(glsl, "//Instruction %d\n", psInst->id);
 #if 0
 	if(psInst->id == 73)
 	{
@@ -2003,7 +2063,6 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 		psContext->AddIndentation();
 		bcatcstr(glsl, "//OR\n");
 #endif
-		uint32_t destMask = psInst->asOperands[0].GetAccessMask();
 		if (psInst->asOperands[0].GetDataType(psContext) == SVT_BOOL)
 		{
 			uint32_t destMask = psInst->asOperands[0].GetAccessMask();
@@ -3011,6 +3070,16 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 
 		psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_TEXTURE, psInst->asOperands[2].ui32RegisterNumber, &psBinding);
         
+        if (psInst->eResDim == RESOURCE_DIMENSION_BUFFER) // Hack typed buffer as raw buf
+		{
+            psInst->eOpcode = OPCODE_LD_UAV_TYPED;
+			psInst->asOperands[1].eSelMode = OPERAND_4_COMPONENT_SELECT_1_MODE;
+			if (psInst->asOperands[1].eType == OPERAND_TYPE_IMMEDIATE32)
+				psInst->asOperands[1].iNumComponents = 1;
+			TranslateShaderStorageLoad(psInst);
+		break;
+		}
+
 		if (psInst->bAddressOffset)
 		{
 			TranslateTexelFetchOffset(psInst, psBinding, glsl);
@@ -3153,6 +3222,20 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 								Operand* psSrc = &psInst->asOperands[2];
 								Operand* psSrcAddr = &psInst->asOperands[1];
 
+								const ResourceBinding* psRes = 0;
+								psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_UAV, psSrc->ui32RegisterNumber, &psRes);
+								SHADER_VARIABLE_TYPE srcDataType = ResourceReturnTypeToSVTType(psRes->ui32ReturnType, psRes->ePrecision);
+
+								if (psInst->eResDim == RESOURCE_DIMENSION_BUFFER) // Hack typed buffer as raw buf
+								{
+									psSrc->aeDataType[0] = srcDataType;
+                                    psSrcAddr->eSelMode = OPERAND_4_COMPONENT_SELECT_1_MODE;
+                                    if (psSrcAddr->eType == OPERAND_TYPE_IMMEDIATE32)
+                                        psSrcAddr->iNumComponents = 1;
+									TranslateShaderStorageLoad(psInst);
+									break;
+								}
+
 								int srcCount = psSrc->GetNumSwizzleElements();
 								int numParenthesis = 0;
 								uint32_t compMask = 0;
@@ -3177,25 +3260,6 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 									break;
 								}
 
-								SHADER_VARIABLE_TYPE srcDataType;
-								const ResourceBinding* psBinding = 0;
-								psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_UAV, psSrc->ui32RegisterNumber, &psBinding);
-								switch (psBinding->ui32ReturnType)
-								{
-								case RETURN_TYPE_FLOAT:
-									srcDataType = SVT_FLOAT;
-									break;
-								case RETURN_TYPE_SINT:
-									srcDataType = SVT_INT;
-									break;
-								case RETURN_TYPE_UINT:
-									srcDataType = SVT_UINT;
-									break;
-								default:
-									ASSERT(0);
-									break;
-								}
-
 								psContext->AddIndentation();
 								AddAssignToDest(psDest, srcDataType, srcCount, &numParenthesis);
 								glsl << TranslateOperand(psSrc, TO_FLAG_NAME_ONLY);
@@ -3235,14 +3299,23 @@ void ToMetal::TranslateInstruction(Instruction* psInst)
 								   psContext->AddIndentation();
 								   bcatcstr(glsl, "//STORE_UAV_TYPED\n");
 #endif
-								   psContext->AddIndentation();
-
 								   foundResource = psContext->psShader->sInfo.GetResourceFromBindingPoint(RGROUP_UAV,
 									   psInst->asOperands[0].ui32RegisterNumber,
 									   &psRes);
-
 								   ASSERT(foundResource);
 
+								   if (psRes->eDimension == REFLECT_RESOURCE_DIMENSION_BUFFER) // Hack typed buffer as raw buf
+								   {
+									   psInst->asOperands[0].aeDataType[0] = ResourceReturnTypeToSVTType(psRes->ui32ReturnType, psRes->ePrecision);
+                                       psInst->asOperands[1].eSelMode = OPERAND_4_COMPONENT_SELECT_1_MODE;
+                                       if (psInst->asOperands[1].eType == OPERAND_TYPE_IMMEDIATE32)
+                                           psInst->asOperands[1].iNumComponents = 1;
+									   TranslateShaderStorageStore(psInst);
+									   break;
+								   }
+
+								   psContext->AddIndentation();
+
 								   glsl << TranslateOperand(&psInst->asOperands[0], TO_FLAG_NAME_ONLY);
 								   bcatcstr(glsl, ".write(");
 
@@ -3685,6 +3758,21 @@ template <int N> vec<int, N> bitFieldExtractI(const vec<uint, N> width, const ve
 		break;
 	}
 			
+	case OPCODE_SAMPLE_INFO:
+	{
+#ifdef _DEBUG
+		psContext->AddIndentation();
+		bcatcstr(glsl, "//SAMPLE_INFO\n");
+#endif
+		const RESINFO_RETURN_TYPE eResInfoReturnType = psInst->eResInfoReturnType;
+		psContext->AddIndentation();
+		AddAssignToDest(&psInst->asOperands[0], eResInfoReturnType == RESINFO_INSTRUCTION_RETURN_UINT ? SVT_UINT : SVT_FLOAT, 1, &numParenthesis);
+		TranslateOperand(&psInst->asOperands[1], TO_FLAG_NAME_ONLY);
+		bcatcstr(glsl, ".get_num_samples()");
+		AddAssignPrologue(numParenthesis);
+		break;
+	}
+
 	case OPCODE_DMAX:
 	case OPCODE_DMIN:
 	case OPCODE_DMUL:

src/toMetalOperand.cpp

@@ -14,7 +14,11 @@
 using namespace HLSLcc;
 
 #ifdef _MSC_VER
+
+#if _MSC_VER < 1900
 #define snprintf _snprintf
+#endif
+
 #define fpcheck(x) (_isnan(x) || !_finite(x))
 #else
 #define fpcheck(x) (std::isnan(x) || std::isinf(x))
@@ -379,6 +383,7 @@ std::string ToMetal::TranslateVariableName(const Operand* psOperand, uint32_t ui
 			((Operand *)psOperand)->aeDataType[3] = requestedType;
 		}
 
+		bool bitcast = false;
 		if (AreTypesCompatible(eType, ui32TOFlag) == 0)
 		{
 			if (CanDoDirectCast(eType, requestedType))
@@ -394,15 +399,16 @@ std::string ToMetal::TranslateVariableName(const Operand* psOperand, uint32_t ui
 				// Direct cast not possible, need to do bitcast.
 				oss << "as_type<"<< GetConstructorForTypeMetal(requestedType, requestedComponents) << ">(";
 				hasCtor = 1;
+				bitcast = true;
 				numParenthesis++;
 			}
 		}
 
 		// Add ctor if needed (upscaling). Type conversion is already handled above, so here we must
 		// use the original type to not make type conflicts in bitcasts
-		if (((numComponents < requestedComponents)||(scalarWithSwizzle != 0)) && (hasCtor == 0))
+		if (((numComponents < requestedComponents)||(scalarWithSwizzle != 0)) && (hasCtor == 0 || bitcast))
 		{
-			oss << GetConstructorForType(psContext, requestedType, requestedComponents, false) << "(";
+			oss << GetConstructorForType(psContext, eType, requestedComponents, false) << "(";
 
 			numParenthesis++;
 			hasCtor = 1;
@@ -623,7 +629,7 @@ std::string ToMetal::TranslateVariableName(const Operand* psOperand, uint32_t ui
 			const ShaderVarType* psVarType = NULL;
 			int32_t index = -1;
 			std::vector<uint32_t> arrayIndices;
-			bool isArray;
+			bool isArray = false;
 			psContext->psShader->sInfo.GetConstantBufferFromBindingPoint(RGROUP_CBUFFER, psOperand->aui32ArraySizes[0], &psCBuf);
 			ASSERT(psCBuf != NULL);
 
@@ -644,6 +650,9 @@ std::string ToMetal::TranslateVariableName(const Operand* psOperand, uint32_t ui
 					cbName = psCBuf->name;
 				}
 				cbName += ".";
+				// Drop the constant buffer name from subpass inputs
+				if (cbName.substr(0, 19) == "hlslcc_SubpassInput")
+					cbName = "";
 			}
 
 			if((ui32TOFlag & TO_FLAG_DECLARATION_NAME) != TO_FLAG_DECLARATION_NAME)