Implement Load/Store Local/Shared and Atomic shared using new instructions (#5241)

* Implement Load/Store Local/Shared and Atomic shared using new instructions

* Remove now unused code

* Fix base offset register overwrite

* Fix missing storage buffer set index when generating GLSL for Vulkan

* Shader cache version bump

* Remove more unused code

* Some PR feedback

9 files changed, 61 insertions, 215 deletions

diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
index 958f1cef..08e8eb19 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
@@ -71,40 +71,10 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             context.AppendLine($"const int {DefaultNames.UndefinedName} = 0;");
             context.AppendLine();
 
-            if (context.Config.Stage == ShaderStage.Compute)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeLocalMemorySize(), 4);
-
-                if (localMemorySize != 0)
-                {
-                    string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
-
-                    context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
-                    context.AppendLine();
-                }
-
-                int sharedMemorySize = BitUtils.DivRoundUp(context.Config.GpuAccessor.QueryComputeSharedMemorySize(), 4);
-
-                if (sharedMemorySize != 0)
-                {
-                    string sharedMemorySizeStr = NumberFormatter.FormatInt(sharedMemorySize);
-
-                    context.AppendLine($"shared uint {DefaultNames.SharedMemoryName}[{sharedMemorySizeStr}];");
-                    context.AppendLine();
-                }
-            }
-            else if (context.Config.LocalMemorySize != 0)
-            {
-                int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
-
-                string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
-
-                context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
-                context.AppendLine();
-            }
-
             DeclareConstantBuffers(context, context.Config.Properties.ConstantBuffers.Values);
             DeclareStorageBuffers(context, context.Config.Properties.StorageBuffers.Values);
+            DeclareMemories(context, context.Config.Properties.LocalMemories.Values, isShared: false);
+            DeclareMemories(context, context.Config.Properties.SharedMemories.Values, isShared: true);
 
             var textureDescriptors = context.Config.GetTextureDescriptors();
             if (textureDescriptors.Length != 0)
@@ -238,11 +208,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 context.AppendLine();
             }
 
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.AtomicMinMaxS32Shared) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl");
-            }
-
             if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighS32) != 0)
             {
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/MultiplyHighS32.glsl");
@@ -273,11 +238,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/ShuffleXor.glsl");
             }
 
-            if ((info.HelperFunctionsMask & HelperFunctionsMask.StoreSharedSmallInt) != 0)
-            {
-                AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl");
-            }
-
             if ((info.HelperFunctionsMask & HelperFunctionsMask.SwizzleAdd) != 0)
             {
                 AppendHelperFunction(context, "Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/SwizzleAdd.glsl");
@@ -358,7 +318,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                     _ => "std430"
                 };
 
-                context.AppendLine($"layout (binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
+                string set = string.Empty;
+
+                if (context.Config.Options.TargetApi == TargetApi.Vulkan)
+                {
+                    set = $"set = {buffer.Set}, ";
+                }
+
+                context.AppendLine($"layout ({set}binding = {buffer.Binding}, {layout}) {declType} _{buffer.Name}");
                 context.EnterScope();
 
                 foreach (StructureField field in buffer.Type.Fields)
@@ -391,6 +358,27 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             }
         }
 
+        private static void DeclareMemories(CodeGenContext context, IEnumerable<MemoryDefinition> memories, bool isShared)
+        {
+            string prefix = isShared ? "shared " : string.Empty;
+
+            foreach (MemoryDefinition memory in memories)
+            {
+                string typeName = GetVarTypeName(context, memory.Type & ~AggregateType.Array);
+
+                if (memory.ArrayLength > 0)
+                {
+                    string arraySize = memory.ArrayLength.ToString(CultureInfo.InvariantCulture);
+
+                    context.AppendLine($"{prefix}{typeName} {memory.Name}[{arraySize}];");
+                }
+                else
+                {
+                    context.AppendLine($"{prefix}{typeName} {memory.Name}[];");
+                }
+            }
+        }
+
         private static void DeclareSamplers(CodeGenContext context, TextureDescriptor[] descriptors)
         {
             int arraySize = 0;
@@ -717,7 +705,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             string code = EmbeddedResources.ReadAllText(filename);
 
             code = code.Replace("\t", CodeGenContext.Tab);
-            code = code.Replace("$SHARED_MEM$", DefaultNames.SharedMemoryName);
 
             if (context.Config.GpuAccessor.QueryHostSupportsShaderBallot())
             {
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
index 5ee8259c..e909dcf0 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/DefaultNames.cs
@@ -11,9 +11,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
         public const string IAttributePrefix = "in_attr";
         public const string OAttributePrefix = "out_attr";
 
-        public const string LocalMemoryName  = "local_mem";
-        public const string SharedMemoryName = "shared_mem";
-
         public const string ArgumentNamePrefix = "a";
 
         public const string UndefinedName = "undef";
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl
deleted file mode 100644
index 82b76bcc..00000000
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/AtomicMinMaxS32Shared.glsl
+++ /dev/null
@@ -1,21 +0,0 @@
-int Helper_AtomicMaxS32(int offset, int value)
-{
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[offset];
-        newValue = uint(max(int(oldValue), value));
-    } while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
-    return int(oldValue);
-}
-
-int Helper_AtomicMinS32(int offset, int value)
-{
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[offset];
-        newValue = uint(min(int(oldValue), value));
-    } while (atomicCompSwap($SHARED_MEM$[offset], oldValue, newValue) != oldValue);
-    return int(oldValue);
-}
\ No newline at end of file
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs
index 54f35b15..21c43547 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/HelperFunctionNames.cs
@@ -2,9 +2,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 {
     static class HelperFunctionNames
     {
-        public static string AtomicMaxS32 = "Helper_AtomicMaxS32";
-        public static string AtomicMinS32 = "Helper_AtomicMinS32";
-
         public static string MultiplyHighS32 = "Helper_MultiplyHighS32";
         public static string MultiplyHighU32 = "Helper_MultiplyHighU32";
 
@@ -13,10 +10,5 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
         public static string ShuffleUp   = "Helper_ShuffleUp";
         public static string ShuffleXor  = "Helper_ShuffleXor";
         public static string SwizzleAdd  = "Helper_SwizzleAdd";
-
-        public static string StoreShared16  = "Helper_StoreShared16";
-        public static string StoreShared8   = "Helper_StoreShared8";
-        public static string StoreStorage16 = "Helper_StoreStorage16";
-        public static string StoreStorage8  = "Helper_StoreStorage8";
     }
 }
\ No newline at end of file
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl
deleted file mode 100644
index 2f57b5ff..00000000
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/HelperFunctions/StoreSharedSmallInt.glsl
+++ /dev/null
@@ -1,23 +0,0 @@
-void Helper_StoreShared16(int offset, uint value)
-{
-    int wordOffset = offset >> 2;
-    int bitOffset = (offset & 3) * 8;
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[wordOffset];
-        newValue = bitfieldInsert(oldValue, value, bitOffset, 16);
-    } while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
-}
-
-void Helper_StoreShared8(int offset, uint value)
-{
-    int wordOffset = offset >> 2;
-    int bitOffset = (offset & 3) * 8;
-    uint oldValue, newValue;
-    do
-    {
-        oldValue = $SHARED_MEM$[wordOffset];
-        newValue = bitfieldInsert(oldValue, value, bitOffset, 8);
-    } while (atomicCompSwap($SHARED_MEM$[wordOffset], oldValue, newValue) != oldValue);
-}
\ No newline at end of file
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
index 01d8a6e7..b2577a99 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGen.cs
@@ -68,7 +68,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
 
                 string args = string.Empty;
 
-                if (atomic && operation.StorageKind == StorageKind.StorageBuffer)
+                if (atomic && (operation.StorageKind == StorageKind.StorageBuffer || operation.StorageKind == StorageKind.SharedMemory))
                 {
                     args = GenerateLoadOrStore(context, operation, isStore: false);
 
@@ -81,23 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                         args += ", " + GetSoureExpr(context, operation.GetSource(argIndex), dstType);
                     }
                 }
-                else if (atomic && operation.StorageKind == StorageKind.SharedMemory)
-                {
-                    args = LoadShared(context, operation);
-
-                    // For shared memory access, the second argument is unused and should be ignored.
-                    // It is there to make both storage and shared access have the same number of arguments.
-                    // For storage, both inputs are consumed when the argument index is 0, so we should skip it here.
-
-                    for (int argIndex = 2; argIndex < arity; argIndex++)
-                    {
-                        args += ", ";
-
-                        AggregateType dstType = GetSrcVarType(inst, argIndex);
-
-                        args += GetSoureExpr(context, operation.GetSource(argIndex), dstType);
-                    }
-                }
                 else
                 {
                     for (int argIndex = 0; argIndex < arity; argIndex++)
@@ -179,12 +162,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     case Instruction.Load:
                         return Load(context, operation);
 
-                    case Instruction.LoadLocal:
-                        return LoadLocal(context, operation);
-
-                    case Instruction.LoadShared:
-                        return LoadShared(context, operation);
-
                     case Instruction.Lod:
                         return Lod(context, operation);
 
@@ -200,18 +177,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     case Instruction.Store:
                         return Store(context, operation);
 
-                    case Instruction.StoreLocal:
-                        return StoreLocal(context, operation);
-
-                    case Instruction.StoreShared:
-                        return StoreShared(context, operation);
-
-                    case Instruction.StoreShared16:
-                        return StoreShared16(context, operation);
-
-                    case Instruction.StoreShared8:
-                        return StoreShared8(context, operation);
-
                     case Instruction.TextureSample:
                         return TextureSample(context, operation);
 
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
index f42d9898..8b0b744a 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenHelper.cs
@@ -17,9 +17,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.AtomicAdd,                InstType.AtomicBinary,   "atomicAdd");
             Add(Instruction.AtomicAnd,                InstType.AtomicBinary,   "atomicAnd");
             Add(Instruction.AtomicCompareAndSwap,     InstType.AtomicTernary,  "atomicCompSwap");
-            Add(Instruction.AtomicMaxS32,             InstType.CallTernary,    HelperFunctionNames.AtomicMaxS32);
             Add(Instruction.AtomicMaxU32,             InstType.AtomicBinary,   "atomicMax");
-            Add(Instruction.AtomicMinS32,             InstType.CallTernary,    HelperFunctionNames.AtomicMinS32);
             Add(Instruction.AtomicMinU32,             InstType.AtomicBinary,   "atomicMin");
             Add(Instruction.AtomicOr,                 InstType.AtomicBinary,   "atomicOr");
             Add(Instruction.AtomicSwap,               InstType.AtomicBinary,   "atomicExchange");
@@ -83,8 +81,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.ImageAtomic,              InstType.Special);
             Add(Instruction.IsNan,                    InstType.CallUnary,      "isnan");
             Add(Instruction.Load,                     InstType.Special);
-            Add(Instruction.LoadLocal,                InstType.Special);
-            Add(Instruction.LoadShared,               InstType.Special);
             Add(Instruction.Lod,                      InstType.Special);
             Add(Instruction.LogarithmB2,              InstType.CallUnary,      "log2");
             Add(Instruction.LogicalAnd,               InstType.OpBinaryCom,    "&&",              9);
@@ -118,10 +114,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             Add(Instruction.Sine,                     InstType.CallUnary,      "sin");
             Add(Instruction.SquareRoot,               InstType.CallUnary,      "sqrt");
             Add(Instruction.Store,                    InstType.Special);
-            Add(Instruction.StoreLocal,               InstType.Special);
-            Add(Instruction.StoreShared,              InstType.Special);
-            Add(Instruction.StoreShared16,            InstType.Special);
-            Add(Instruction.StoreShared8,             InstType.Special);
             Add(Instruction.Subtract,                 InstType.OpBinary,       "-",               2);
             Add(Instruction.SwizzleAdd,               InstType.CallTernary,    HelperFunctionNames.SwizzleAdd);
             Add(Instruction.TextureSample,            InstType.Special);
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
index c8084d9d..99376ffb 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/Instructions/InstGenMemory.cs
@@ -191,25 +191,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             return GenerateLoadOrStore(context, operation, isStore: false);
         }
 
-        public static string LoadLocal(CodeGenContext context, AstOperation operation)
-        {
-            return LoadLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
-        }
-
-        public static string LoadShared(CodeGenContext context, AstOperation operation)
-        {
-            return LoadLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
-        }
-
-        private static string LoadLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
-        {
-            IAstNode src1 = operation.GetSource(0);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            return $"{arrayName}[{offsetExpr}]";
-        }
-
         public static string Lod(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -263,58 +244,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
             return GenerateLoadOrStore(context, operation, isStore: true);
         }
 
-        public static string StoreLocal(CodeGenContext context, AstOperation operation)
-        {
-            return StoreLocalOrShared(context, operation, DefaultNames.LocalMemoryName);
-        }
-
-        public static string StoreShared(CodeGenContext context, AstOperation operation)
-        {
-            return StoreLocalOrShared(context, operation, DefaultNames.SharedMemoryName);
-        }
-
-        private static string StoreLocalOrShared(CodeGenContext context, AstOperation operation, string arrayName)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{arrayName}[{offsetExpr}] = {src}";
-        }
-
-        public static string StoreShared16(CodeGenContext context, AstOperation operation)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{HelperFunctionNames.StoreShared16}({offsetExpr}, {src})";
-        }
-
-        public static string StoreShared8(CodeGenContext context, AstOperation operation)
-        {
-            IAstNode src1 = operation.GetSource(0);
-            IAstNode src2 = operation.GetSource(1);
-
-            string offsetExpr = GetSoureExpr(context, src1, GetSrcVarType(operation.Inst, 0));
-
-            AggregateType srcType = OperandManager.GetNodeDestType(context, src2);
-
-            string src = TypeConversion.ReinterpretCast(context, src2, srcType, AggregateType.U32);
-
-            return $"{HelperFunctionNames.StoreShared8}({offsetExpr}, {src})";
-        }
-
         public static string TextureSample(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -675,6 +604,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl.Instructions
                     varType = field.Type;
                     break;
 
+                case StorageKind.LocalMemory:
+                case StorageKind.SharedMemory:
+                    if (!(operation.GetSource(srcIndex++) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
+                    {
+                        throw new InvalidOperationException($"First input of {operation.Inst} with {storageKind} storage must be a constant operand.");
+                    }
+
+                    MemoryDefinition memory = storageKind == StorageKind.LocalMemory
+                        ? context.Config.Properties.LocalMemories[bindingId.Value]
+                        : context.Config.Properties.SharedMemories[bindingId.Value];
+
+                    varName = memory.Name;
+                    varType = memory.Type;
+                    break;
+
                 case StorageKind.Input:
                 case StorageKind.InputPerPatch:
                 case StorageKind.Output:
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs
index 4fd1d17c..4f6ca642 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Glsl/OperandManager.cs
@@ -113,7 +113,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
             if (node is AstOperation operation)
             {
-                if (operation.Inst == Instruction.Load)
+                if (operation.Inst == Instruction.Load || operation.Inst.IsAtomic())
                 {
                     switch (operation.StorageKind)
                     {
@@ -136,6 +136,19 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
 
                             return field.Type & AggregateType.ElementTypeMask;
 
+                        case StorageKind.LocalMemory:
+                        case StorageKind.SharedMemory:
+                            if (!(operation.GetSource(0) is AstOperand bindingId) || bindingId.Type != OperandType.Constant)
+                            {
+                                throw new InvalidOperationException($"First input of {operation.Inst} with {operation.StorageKind} storage must be a constant operand.");
+                            }
+
+                            MemoryDefinition memory = operation.StorageKind == StorageKind.LocalMemory
+                                ? context.Config.Properties.LocalMemories[bindingId.Value]
+                                : context.Config.Properties.SharedMemories[bindingId.Value];
+
+                            return memory.Type & AggregateType.ElementTypeMask;
+
                         case StorageKind.Input:
                         case StorageKind.InputPerPatch:
                         case StorageKind.Output: