PreAllocator: Check if instruction supports a Vex prefix in IsVexSameOperandDestSrc1 (#3587)

* Limit vertex buffer range for non-indexed draws

* Fix typo

ARMeilleure/CodeGen/X86/AssemblerTable.cs

@@ -4,6 +4,11 @@ namespace ARMeilleure.CodeGen.X86
 {
     partial class Assembler
     {
+        public static bool SupportsVexPrefix(X86Instruction inst)
+        {
+            return _instTable[(int)inst].Flags.HasFlag(InstructionFlags.Vex);
+        }
+
         private const int BadOp = 0;
 
         [Flags]

ARMeilleure/CodeGen/X86/PreAllocator.cs

@@ -1297,11 +1297,15 @@ namespace ARMeilleure.CodeGen.X86
         {
             if (IsIntrinsic(operation.Instruction))
             {
+                IntrinsicInfo info = IntrinsicTable.GetInfo(operation.Intrinsic);
+
+                bool hasVex = HardwareCapabilities.SupportsVexEncoding && Assembler.SupportsVexPrefix(info.Inst);
+
                 bool isUnary = operation.SourcesCount < 2;
 
                 bool hasVecDest = operation.Destination != default && operation.Destination.Type == OperandType.V128;
 
-                return !HardwareCapabilities.SupportsVexEncoding && !isUnary && hasVecDest;
+                return !hasVex && !isUnary && hasVecDest;
             }
 
             return false;

Ryujinx.Graphics.Gpu/Image/TextureBindingsManager.cs

@@ -553,7 +553,7 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                 specStateMatches &= specState.MatchesTexture(stage, index, descriptor);
 
-                Sampler sampler = _samplerPool?.Get(samplerId);
+                Sampler sampler = samplerPool?.Get(samplerId);
 
                 ITexture hostTexture = texture?.GetTargetTexture(bindingInfo.Target);
                 ISampler hostSampler = sampler?.GetHostSampler(texture);

Ryujinx.Graphics.OpenGL/Pipeline.cs

@@ -236,7 +236,7 @@ namespace Ryujinx.Graphics.OpenGL
                 return;
             }
 
-            PreDraw();
+            PreDraw(vertexCount);
 
             if (_primitiveType == PrimitiveType.Quads && !HwCapabilities.SupportsQuads)
             {
@@ -354,7 +354,7 @@ namespace Ryujinx.Graphics.OpenGL
                 return;
             }
 
-            PreDraw();
+            PreDrawVbUnbounded();
 
             int indexElemSize = 1;
 
@@ -686,7 +686,7 @@ namespace Ryujinx.Graphics.OpenGL
                 return;
             }
 
-            PreDraw();
+            PreDrawVbUnbounded();
 
             GL.BindBuffer((BufferTarget)All.DrawIndirectBuffer, indirectBuffer.Handle.ToInt32());
             GL.BindBuffer((BufferTarget)All.ParameterBuffer, parameterBuffer.Handle.ToInt32());
@@ -709,7 +709,7 @@ namespace Ryujinx.Graphics.OpenGL
                 return;
             }
 
-            PreDraw();
+            PreDrawVbUnbounded();
 
             _vertexArray.SetRangeOfIndexBuffer();
 
@@ -1515,11 +1515,22 @@ namespace Ryujinx.Graphics.OpenGL
             _supportBuffer.Commit();
         }
 
+        private void PreDraw(int vertexCount)
+        {
+            _vertexArray.PreDraw(vertexCount);
+            PreDraw();
+        }
+
+        private void PreDrawVbUnbounded()
+        {
+            _vertexArray.PreDrawVbUnbounded();
+            PreDraw();
+        }
+
         private void PreDraw()
         {
             DrawCount++;
 
-            _vertexArray.Validate();
             _unit0Texture?.Bind(0);
             _supportBuffer.Commit();
         }

Ryujinx.Graphics.OpenGL/VertexArray.cs

@@ -1,6 +1,7 @@
 using OpenTK.Graphics.OpenGL;
 using Ryujinx.Graphics.GAL;
 using System;
+using System.Numerics;
 using System.Runtime.CompilerServices;
 
 namespace Ryujinx.Graphics.OpenGL
@@ -16,12 +17,16 @@ namespace Ryujinx.Graphics.OpenGL
 
         private int _vertexAttribsCount;
         private int _vertexBuffersCount;
+        private int _minVertexCount;
 
         private uint _vertexAttribsInUse;
         private uint _vertexBuffersInUse;
+        private uint _vertexBuffersLimited;
 
         private BufferRange _indexBuffer;
         private BufferHandle _tempIndexBuffer;
+        private BufferHandle _tempVertexBuffer;
+        private int _tempVertexBufferSize;
 
         public VertexArray()
         {
@@ -40,6 +45,8 @@ namespace Ryujinx.Graphics.OpenGL
 
         public void SetVertexBuffers(ReadOnlySpan<VertexBufferDescriptor> vertexBuffers)
         {
+            int minVertexCount = int.MaxValue;
+
             int bindingIndex;
             for (bindingIndex = 0; bindingIndex < vertexBuffers.Length; bindingIndex++)
             {
@@ -47,6 +54,12 @@ namespace Ryujinx.Graphics.OpenGL
 
                 if (vb.Buffer.Handle != BufferHandle.Null)
                 {
+                    int vertexCount = vb.Stride <= 0 ? 0 : vb.Buffer.Size / vb.Stride;
+                    if (minVertexCount > vertexCount)
+                    {
+                        minVertexCount = vertexCount;
+                    }
+
                     GL.BindVertexBuffer(bindingIndex, vb.Buffer.Handle.ToInt32(), (IntPtr)vb.Buffer.Offset, vb.Stride);
                     GL.VertexBindingDivisor(bindingIndex, vb.Divisor);
                     _vertexBuffersInUse |= 1u << bindingIndex;
@@ -64,6 +77,7 @@ namespace Ryujinx.Graphics.OpenGL
             }
 
             _vertexBuffersCount = bindingIndex;
+            _minVertexCount = minVertexCount;
             _needsAttribsUpdate = true;
         }
 
@@ -143,6 +157,101 @@ namespace Ryujinx.Graphics.OpenGL
             GL.BindBuffer(BufferTarget.ElementArrayBuffer, _indexBuffer.Handle.ToInt32());
         }
 
+        public void PreDraw(int vertexCount)
+        {
+            LimitVertexBuffers(vertexCount);
+            Validate();
+        }
+
+        public void PreDrawVbUnbounded()
+        {
+            UnlimitVertexBuffers();
+            Validate();
+        }
+
+        public void LimitVertexBuffers(int vertexCount)
+        {
+            // Is it possible for the draw to fetch outside the bounds of any vertex buffer currently bound?
+
+            if (vertexCount <= _minVertexCount)
+            {
+                return;
+            }
+
+            // If the draw can fetch out of bounds, let's ensure that it will only fetch zeros rather than memory garbage.
+
+            int currentTempVbOffset = 0;
+            uint buffersInUse = _vertexBuffersInUse;
+
+            while (buffersInUse != 0)
+            {
+                int vbIndex = BitOperations.TrailingZeroCount(buffersInUse);
+
+                ref var vb = ref _vertexBuffers[vbIndex];
+
+                int requiredSize = vertexCount * vb.Stride;
+
+                if (vb.Buffer.Size < requiredSize)
+                {
+                    BufferHandle tempVertexBuffer = EnsureTempVertexBufferSize(currentTempVbOffset + requiredSize);
+
+                    Buffer.Copy(vb.Buffer.Handle, tempVertexBuffer, vb.Buffer.Offset, currentTempVbOffset, vb.Buffer.Size);
+                    Buffer.Clear(tempVertexBuffer, currentTempVbOffset + vb.Buffer.Size, requiredSize - vb.Buffer.Size, 0);
+
+                    GL.BindVertexBuffer(vbIndex, tempVertexBuffer.ToInt32(), (IntPtr)currentTempVbOffset, vb.Stride);
+
+                    currentTempVbOffset += requiredSize;
+                    _vertexBuffersLimited |= 1u << vbIndex;
+                }
+
+                buffersInUse &= ~(1u << vbIndex);
+            }
+        }
+
+        private BufferHandle EnsureTempVertexBufferSize(int size)
+        {
+            BufferHandle tempVertexBuffer = _tempVertexBuffer;
+
+            if (_tempVertexBufferSize < size)
+            {
+                _tempVertexBufferSize = size;
+
+                if (tempVertexBuffer == BufferHandle.Null)
+                {
+                    tempVertexBuffer = Buffer.Create(size);
+                    _tempVertexBuffer = tempVertexBuffer;
+                    return tempVertexBuffer;
+                }
+
+                Buffer.Resize(_tempVertexBuffer, size);
+            }
+
+            return tempVertexBuffer;
+        }
+
+        public void UnlimitVertexBuffers()
+        {
+            uint buffersLimited = _vertexBuffersLimited;
+
+            if (buffersLimited == 0)
+            {
+                return;
+            }
+
+            while (buffersLimited != 0)
+            {
+                int vbIndex = BitOperations.TrailingZeroCount(buffersLimited);
+
+                ref var vb = ref _vertexBuffers[vbIndex];
+
+                GL.BindVertexBuffer(vbIndex, vb.Buffer.Handle.ToInt32(), (IntPtr)vb.Buffer.Offset, vb.Stride);
+
+                buffersLimited &= ~(1u << vbIndex);
+            }
+
+            _vertexBuffersLimited = 0;
+        }
+
         public void Validate()
         {
             for (int attribIndex = 0; attribIndex < _vertexAttribsCount; attribIndex++)

Ryujinx.Graphics.OpenGL/VertexBuffer.cs

@@ -1,19 +0,0 @@
-using Ryujinx.Graphics.GAL;
-
-namespace Ryujinx.Graphics.OpenGL
-{
-    struct VertexBuffer
-    {
-        public BufferRange Range { get; }
-
-        public int Divisor { get; }
-        public int Stride  { get; }
-
-        public VertexBuffer(BufferRange range, int divisor, int stride)
-        {
-            Range   = range;
-            Divisor = divisor;
-            Stride  = stride;
-        }
-    }
-}