Support shared color mask, implement more shader instructions

Support shared color masks (used by Nouveau and maybe the NVIDIA
driver).
Support draw buffers (also required by OpenGL).
Support viewport transform disable (disabled for now as it breaks some
games).
Fix instanced rendering draw being ignored for multi draw.
Fix IADD and IADD3 immediate shader encodings, that was not matching
some ops.
Implement FFMA32I shader instruction.
Implement IMAD shader instruction.

4 files changed, 51 insertions, 28 deletions

diff --git a/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs b/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
index 3995d430..7ba7b697 100644
--- a/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
+++ b/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs
@@ -11,18 +11,25 @@ namespace Ryujinx.Graphics.Shader.Translation
         public Block  CurrBlock { get; set; }
         public OpCode CurrOp    { get; set; }
 
-        private ShaderStage _stage;
-
-        private ShaderHeader _header;
+        private ShaderStage        _stage;
+        private ShaderHeader       _header;
+        private ShaderCapabilities _capabilities;
+        private TranslationFlags   _flags;
 
         private List<Operation> _operations;
 
         private Dictionary<ulong, Operand> _labels;
 
-        public EmitterContext(ShaderStage stage, ShaderHeader header)
+        public EmitterContext(
+            ShaderStage        stage,
+            ShaderHeader       header,
+            ShaderCapabilities capabilities,
+            TranslationFlags   flags)
         {
-            _stage  = stage;
-            _header = header;
+            _stage        = stage;
+            _header       = header;
+            _capabilities = capabilities;
+            _flags        = flags;
 
             _operations = new List<Operation>();
 
@@ -62,7 +69,18 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public void PrepareForReturn()
         {
-            if (_stage == ShaderStage.Fragment)
+            if (_stage == ShaderStage.Vertex)
+            {
+                if ((_flags & TranslationFlags.DividePosXY) != 0)
+                {
+                    Operand posX = Attribute(AttributeConsts.PositionX);
+                    Operand posY = Attribute(AttributeConsts.PositionY);
+
+                    this.Copy(posX, this.FPDivide(posX, ConstF(_capabilities.MaximumViewportDimensions / 2)));
+                    this.Copy(posY, this.FPDivide(posY, ConstF(_capabilities.MaximumViewportDimensions / 2)));
+                }
+            }
+            else if (_stage == ShaderStage.Fragment)
             {
                 if (_header.OmapDepth)
                 {
diff --git a/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs b/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
index d884cfdb..e39d8c64 100644
--- a/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
+++ b/Ryujinx.Graphics.Shader/Translation/EmitterContextInsts.cs
@@ -476,6 +476,16 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.LoadShared, Local(), a);
         }
 
+        public static Operand MultiplyHighS32(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.MultiplyHighS32, Local(), a, b);
+        }
+
+        public static Operand MultiplyHighU32(this EmitterContext context, Operand a, Operand b)
+        {
+            return context.Add(Instruction.MultiplyHighU32, Local(), a, b);
+        }
+
         public static Operand PackHalf2x16(this EmitterContext context, Operand a, Operand b)
         {
             return context.Add(Instruction.PackHalf2x16, Local(), a, b);
diff --git a/Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs b/Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs
index 99b6107a..8faa4383 100644
--- a/Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs
+++ b/Ryujinx.Graphics.Shader/Translation/TranslationFlags.cs
@@ -6,6 +6,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         Compute       = 1 << 0,
         DebugMode     = 1 << 1,
-        Unspecialized = 1 << 2
+        Unspecialized = 1 << 2,
+        DividePosXY   = 1 << 3
     }
 }
\ No newline at end of file
diff --git a/Ryujinx.Graphics.Shader/Translation/Translator.cs b/Ryujinx.Graphics.Shader/Translation/Translator.cs
index b129be93..69e63ae1 100644
--- a/Ryujinx.Graphics.Shader/Translation/Translator.cs
+++ b/Ryujinx.Graphics.Shader/Translation/Translator.cs
@@ -49,15 +49,9 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public static ShaderProgram Translate(Span<byte> code, ShaderCapabilities capabilities, TranslationFlags flags)
         {
-            bool compute   = (flags & TranslationFlags.Compute)   != 0;
-            bool debugMode = (flags & TranslationFlags.DebugMode) != 0;
+            bool compute = (flags & TranslationFlags.Compute) != 0;
 
-            Operation[] ops = DecodeShader(
-                code,
-                compute,
-                debugMode,
-                out ShaderHeader header,
-                out int size);
+            Operation[] ops = DecodeShader(code, capabilities, flags, out ShaderHeader header, out int size);
 
             ShaderStage stage;
 
@@ -94,8 +88,8 @@ namespace Ryujinx.Graphics.Shader.Translation
         {
             bool debugMode = (flags & TranslationFlags.DebugMode) != 0;
 
-            Operation[] vpAOps = DecodeShader(vpACode, compute: false, debugMode, out _, out _);
-            Operation[] vpBOps = DecodeShader(vpBCode, compute: false, debugMode, out ShaderHeader header, out int sizeB);
+            Operation[] vpAOps = DecodeShader(vpACode, capabilities, flags, out _, out _);
+            Operation[] vpBOps = DecodeShader(vpBCode, capabilities, flags, out ShaderHeader header, out int sizeB);
 
             ShaderConfig config = new ShaderConfig(
                 header.Stage,
@@ -142,23 +136,23 @@ namespace Ryujinx.Graphics.Shader.Translation
         }
 
         private static Operation[] DecodeShader(
-            Span<byte>       code,
-            bool             compute,
-            bool             debugMode,
-            out ShaderHeader header,
-            out int          size)
+            Span<byte>         code,
+            ShaderCapabilities capabilities,
+            TranslationFlags   flags,
+            out ShaderHeader   header,
+            out int            size)
         {
             Block[] cfg;
 
             EmitterContext context;
 
-            if (compute)
+            if ((flags & TranslationFlags.Compute) != 0)
             {
                 header = null;
 
                 cfg = Decoder.Decode(code, 0);
 
-                context = new EmitterContext(ShaderStage.Compute, header);
+                context = new EmitterContext(ShaderStage.Compute, header, capabilities, flags);
             }
             else
             {
@@ -166,7 +160,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 cfg = Decoder.Decode(code, HeaderSize);
 
-                context = new EmitterContext(header.Stage, header);
+                context = new EmitterContext(header.Stage, header, capabilities, flags);
             }
 
             if (cfg == null)
@@ -197,7 +191,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 {
                     OpCode op = block.OpCodes[opIndex];
 
-                    if (debugMode)
+                    if ((flags & TranslationFlags.DebugMode) != 0)
                     {
                         string instName;
 
@@ -274,7 +268,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 }
             }
 
-            size = (int)maxEndAddress + (compute ? 0 : HeaderSize);
+            size = (int)maxEndAddress + (((flags & TranslationFlags.Compute) != 0) ? 0 : HeaderSize);
 
             return context.GetOperations();
         }