Use correct shader local memory size instead of a hardcoded size (#914)

* Use correct shader local size instead of a hardcoded size

* Remove unused uniform block

* Update XML doc

* Local memory size has 23 bits on maxwell

* Generate compute QMD struct from nv open doc header

* Remove dummy arrays when shared or local memory is not used, other improvements

3 files changed, 29 insertions, 12 deletions

diff --git a/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs b/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
index 200569c4..2e7f9f1b 100644
--- a/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
+++ b/Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs
@@ -47,25 +47,35 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 context.AppendLine();
             }
 
-            context.AppendLine("layout (std140) uniform Extra");
-
-            context.EnterScope();
+            if (context.Config.Stage == ShaderStage.Compute)
+            {
+                int localMemorySize = BitUtils.DivRoundUp(context.Config.QueryInfo(QueryInfoName.ComputeLocalMemorySize), 4);
 
-            context.AppendLine("vec2 flip;");
-            context.AppendLine("int instance;");
+                if (localMemorySize != 0)
+                {
+                    string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
 
-            context.LeaveScope(";");
+                    context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
+                    context.AppendLine();
+                }
 
-            context.AppendLine();
+                int sharedMemorySize = BitUtils.DivRoundUp(context.Config.QueryInfo(QueryInfoName.ComputeSharedMemorySize), 4);
 
-            context.AppendLine($"uint {DefaultNames.LocalMemoryName}[0x100];");
-            context.AppendLine();
+                if (sharedMemorySize != 0)
+                {
+                    string sharedMemorySizeStr = NumberFormatter.FormatInt(sharedMemorySize);
 
-            if (context.Config.Stage == ShaderStage.Compute)
+                    context.AppendLine($"shared uint {DefaultNames.SharedMemoryName}[{sharedMemorySizeStr}];");
+                    context.AppendLine();
+                }
+            }
+            else if (context.Config.LocalMemorySize != 0)
             {
-                string size = NumberFormatter.FormatInt(BitUtils.DivRoundUp(context.Config.QueryInfo(QueryInfoName.ComputeSharedMemorySize), 4));
+                int localMemorySize = BitUtils.DivRoundUp(context.Config.LocalMemorySize, 4);
+
+                string localMemorySizeStr = NumberFormatter.FormatInt(localMemorySize);
 
-                context.AppendLine($"shared uint {DefaultNames.SharedMemoryName}[{size}];");
+                context.AppendLine($"uint {DefaultNames.LocalMemoryName}[{localMemorySizeStr}];");
                 context.AppendLine();
             }
 
diff --git a/Ryujinx.Graphics.Shader/QueryInfoName.cs b/Ryujinx.Graphics.Shader/QueryInfoName.cs
index c4f2cb6c..887c0d7d 100644
--- a/Ryujinx.Graphics.Shader/QueryInfoName.cs
+++ b/Ryujinx.Graphics.Shader/QueryInfoName.cs
@@ -5,6 +5,7 @@ namespace Ryujinx.Graphics.Shader
         ComputeLocalSizeX,
         ComputeLocalSizeY,
         ComputeLocalSizeZ,
+        ComputeLocalMemorySize,
         ComputeSharedMemorySize,
         IsTextureBuffer,
         IsTextureRectangle,
diff --git a/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs b/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs
index 8a0f25fe..e3708b41 100644
--- a/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs
+++ b/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs
@@ -10,6 +10,8 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public int MaxOutputVertices { get; }
 
+        public int LocalMemorySize { get; }
+
         public OutputMapTarget[] OmapTargets    { get; }
         public bool              OmapSampleMask { get; }
         public bool              OmapDepth      { get; }
@@ -23,6 +25,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             Stage             = ShaderStage.Compute;
             OutputTopology    = OutputTopology.PointList;
             MaxOutputVertices = 0;
+            LocalMemorySize   = 0;
             OmapTargets       = null;
             OmapSampleMask    = false;
             OmapDepth         = false;
@@ -35,6 +38,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             Stage             = header.Stage;
             OutputTopology    = header.OutputTopology;
             MaxOutputVertices = header.MaxOutputVertexCount;
+            LocalMemorySize   = header.ShaderLocalMemoryLowSize + header.ShaderLocalMemoryHighSize;
             OmapTargets       = header.OmapTargets;
             OmapSampleMask    = header.OmapSampleMask;
             OmapDepth         = header.OmapDepth;
@@ -80,6 +84,8 @@ namespace Ryujinx.Graphics.Shader.Translation
                     case QueryInfoName.ComputeLocalSizeY:
                     case QueryInfoName.ComputeLocalSizeZ:
                         return 1;
+                    case QueryInfoName.ComputeLocalMemorySize:
+                        return 0x1000;
                     case QueryInfoName.ComputeSharedMemorySize:
                         return 0xc000;
                     case QueryInfoName.IsTextureBuffer: