Shader: Add fallback for LDG from "ube" buffer ranges. (#4027)

We have a conversion from LDG on the compute shader to a special constant buffer binding that's used to exceed hardware limits on compute, but it was only running if the byte offset could be identified. The fallback that checks all of the bindings at runtime only checks the storage buffers.

This PR adds checking ube ranges to the LoadGlobal fallback. This extends the changes in #4011 to only check ube entries which are accessed by the shader.

Fixes particles affected by the wind in The Legend of Zelda: Breath of the Wild. May fix other weird issues with compute shaders in some games.

Try a bunch of games and drivers to make sure they don't blow up loading constants willynilly from searchable buffers.

Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs

@@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         private const ushort FileFormatVersionMajor = 1;
         private const ushort FileFormatVersionMinor = 2;
         private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
-        private const uint CodeGenVersion = 4037;
+        private const uint CodeGenVersion = 4028;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

Ryujinx.Graphics.Shader/Translation/GlobalMemory.cs

@@ -48,5 +48,10 @@ namespace Ryujinx.Graphics.Shader.Translation
                 _ => 0
             };
         }
+
+        public static int GetConstantUbeOffset(int slot)
+        {
+            return UbeBaseOffset + slot * StorageDescSize;
+        }
     }
 }

Ryujinx.Graphics.Shader/Translation/Optimizations/GlobalToStorage.cs

@@ -8,11 +8,14 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 {
     static class GlobalToStorage
     {
-        public static void RunPass(BasicBlock block, ShaderConfig config, ref int sbUseMask)
+        public static void RunPass(BasicBlock block, ShaderConfig config, ref int sbUseMask, ref int ubeUseMask)
         {
             int sbStart = GetStorageBaseCbOffset(config.Stage);
             int sbEnd = sbStart + StorageDescsSize;
 
+            int ubeStart = UbeBaseOffset;
+            int ubeEnd = UbeBaseOffset + UbeDescsSize;
+
             for (LinkedListNode<INode> node = block.Operations.First; node != null; node = node.Next)
             {
                 for (int index = 0; index < node.Value.SourcesCount; index++)
@@ -25,6 +28,16 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     {
                         sbUseMask |= 1 << storageIndex;
                     }
+
+                    if (config.Stage == ShaderStage.Compute)
+                    {
+                        int constantIndex = GetStorageIndex(src, ubeStart, ubeEnd);
+
+                        if (constantIndex >= 0)
+                        {
+                            ubeUseMask |= 1 << constantIndex;
+                        }
+                    }
                 }
 
                 if (!(node.Value is Operation operation))
@@ -54,7 +67,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                         // so NVN "emulates" more constant buffers using global memory access.
                         // Here we try to replace the global access back to a constant buffer
                         // load.
-                        storageIndex = SearchForStorageBase(block, source, UbeBaseOffset, UbeBaseOffset + UbeDescsSize);
+                        storageIndex = SearchForStorageBase(block, source, ubeStart, ubeStart + ubeEnd);
 
                         if (storageIndex >= 0)
                         {
@@ -64,7 +77,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 }
             }
 
-            config.SetAccessibleStorageBuffersMask(sbUseMask);
+            config.SetAccessibleBufferMasks(sbUseMask, ubeUseMask);
         }
 
         private static LinkedListNode<INode> ReplaceGlobalWithStorage(BasicBlock block, LinkedListNode<INode> node, ShaderConfig config, int storageIndex)

Ryujinx.Graphics.Shader/Translation/Optimizations/Optimizer.cs

@@ -12,16 +12,17 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             RunOptimizationPasses(blocks);
 
             int sbUseMask = 0;
+            int ubeUseMask = 0;
 
             // Those passes are looking for specific patterns and only needs to run once.
             for (int blkIndex = 0; blkIndex < blocks.Length; blkIndex++)
             {
-                GlobalToStorage.RunPass(blocks[blkIndex], config, ref sbUseMask);
+                GlobalToStorage.RunPass(blocks[blkIndex], config, ref sbUseMask, ref ubeUseMask);
                 BindlessToIndexed.RunPass(blocks[blkIndex], config);
                 BindlessElimination.RunPass(blocks[blkIndex], config);
             }
 
-            config.SetAccessibleStorageBuffersMask(sbUseMask);
+            config.SetAccessibleBufferMasks(sbUseMask, ubeUseMask);
 
             // Run optimizations one last time to remove any code that is now optimizable after above passes.
             RunOptimizationPasses(blocks);

Ryujinx.Graphics.Shader/Translation/Rewriter.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Graphics.Shader.Decoders;
 using Ryujinx.Graphics.Shader.IntermediateRepresentation;
 using System.Collections.Generic;
 using System.Diagnostics;
@@ -89,12 +90,42 @@ namespace Ryujinx.Graphics.Shader.Translation
                 return local;
             }
 
+            Operand PrependExistingOperation(Operation operation)
+            {
+                Operand local = Local();
+
+                operation.Dest = local;
+                node.List.AddBefore(node, operation);
+
+                return local;
+            }
+
             Operand addrLow  = operation.GetSource(0);
             Operand addrHigh = operation.GetSource(1);
 
             Operand sbBaseAddrLow = Const(0);
             Operand sbSlot        = Const(0);
 
+            Operand alignMask = Const(-config.GpuAccessor.QueryHostStorageBufferOffsetAlignment());
+
+            Operand BindingRangeCheck(int cbOffset, out Operand baseAddrLow)
+            {
+                baseAddrLow = Cbuf(0, cbOffset);
+                Operand baseAddrHigh = Cbuf(0, cbOffset + 1);
+                Operand size = Cbuf(0, cbOffset + 2);
+
+                Operand offset = PrependOperation(Instruction.Subtract, addrLow, baseAddrLow);
+                Operand borrow = PrependOperation(Instruction.CompareLessU32, addrLow, baseAddrLow);
+
+                Operand inRangeLow = PrependOperation(Instruction.CompareLessU32, offset, size);
+
+                Operand addrHighBorrowed = PrependOperation(Instruction.Add, addrHigh, borrow);
+
+                Operand inRangeHigh = PrependOperation(Instruction.CompareEqual, addrHighBorrowed, baseAddrHigh);
+
+                return PrependOperation(Instruction.BitwiseAnd, inRangeLow, inRangeHigh);
+            }
+
             int sbUseMask = config.AccessibleStorageBuffersMask;
 
             while (sbUseMask != 0)
@@ -107,20 +138,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 int cbOffset = GetStorageCbOffset(config.Stage, slot);
 
-                Operand baseAddrLow  = Cbuf(0, cbOffset);
+                Operand inRange = BindingRangeCheck(cbOffset, out Operand baseAddrLow);
-                Operand baseAddrHigh = Cbuf(0, cbOffset + 1);
-                Operand size         = Cbuf(0, cbOffset + 2);
-
-                Operand offset = PrependOperation(Instruction.Subtract,       addrLow, baseAddrLow);
-                Operand borrow = PrependOperation(Instruction.CompareLessU32, addrLow, baseAddrLow);
-
-                Operand inRangeLow = PrependOperation(Instruction.CompareLessU32, offset, size);
-
-                Operand addrHighBorrowed = PrependOperation(Instruction.Add, addrHigh, borrow);
-
-                Operand inRangeHigh = PrependOperation(Instruction.CompareEqual, addrHighBorrowed, baseAddrHigh);
-
-                Operand inRange = PrependOperation(Instruction.BitwiseAnd, inRangeLow, inRangeHigh);
 
                 sbBaseAddrLow = PrependOperation(Instruction.ConditionalSelect, inRange, baseAddrLow, sbBaseAddrLow);
                 sbSlot        = PrependOperation(Instruction.ConditionalSelect, inRange, Const(slot), sbSlot);
@@ -128,8 +146,6 @@ namespace Ryujinx.Graphics.Shader.Translation
 
             if (config.AccessibleStorageBuffersMask != 0)
             {
-                Operand alignMask = Const(-config.GpuAccessor.QueryHostStorageBufferOffsetAlignment());
-
                 Operand baseAddrTrunc = PrependOperation(Instruction.BitwiseAnd, sbBaseAddrLow, alignMask);
                 Operand byteOffset    = PrependOperation(Instruction.Subtract, addrLow, baseAddrTrunc);
 
@@ -178,6 +194,46 @@ namespace Ryujinx.Graphics.Shader.Translation
                 storageOp = new Operation(Instruction.Copy, operation.Dest, Const(0));
             }
 
+            if (operation.Inst == Instruction.LoadGlobal)
+            {
+                int cbeUseMask = config.AccessibleConstantBuffersMask;
+
+                while (cbeUseMask != 0)
+                {
+                    int slot = BitOperations.TrailingZeroCount(cbeUseMask);
+                    int cbSlot = UbeFirstCbuf + slot;
+
+                    cbeUseMask &= ~(1 << slot);
+
+                    config.SetUsedConstantBuffer(cbSlot);
+
+                    Operand previousResult = PrependExistingOperation(storageOp);
+
+                    int cbOffset = GetConstantUbeOffset(slot);
+
+                    Operand inRange = BindingRangeCheck(cbOffset, out Operand baseAddrLow);
+
+                    Operand baseAddrTruncConst = PrependOperation(Instruction.BitwiseAnd, baseAddrLow, alignMask);
+                    Operand byteOffsetConst = PrependOperation(Instruction.Subtract, addrLow, baseAddrTruncConst);
+
+                    Operand cbIndex = PrependOperation(Instruction.ShiftRightU32, byteOffsetConst, Const(2));
+
+                    Operand[] sourcesCb = new Operand[operation.SourcesCount];
+
+                    sourcesCb[0] = Const(cbSlot);
+                    sourcesCb[1] = cbIndex;
+
+                    for (int index = 2; index < operation.SourcesCount; index++)
+                    {
+                        sourcesCb[index] = operation.GetSource(index);
+                    }
+
+                    Operand ldcResult = PrependOperation(Instruction.LoadConstant, sourcesCb);
+
+                    storageOp = new Operation(Instruction.ConditionalSelect, operation.Dest, inRange, ldcResult, previousResult);
+                }
+            }
+
             for (int index = 0; index < operation.SourcesCount; index++)
             {
                 operation.SetSource(index, null);

Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs

@@ -66,6 +66,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         public UInt128 ThisInputAttributesComponents { get; private set; }
 
         public int AccessibleStorageBuffersMask { get; private set; }
+        public int AccessibleConstantBuffersMask { get; private set; }
 
         private int _usedConstantBuffers;
         private int _usedStorageBuffers;
@@ -100,7 +101,8 @@ namespace Ryujinx.Graphics.Shader.Translation
             GpuAccessor = gpuAccessor;
             Options     = options;
 
-            AccessibleStorageBuffersMask = (1 << GlobalMemory.StorageMaxCount) - 1;
+            AccessibleStorageBuffersMask  = (1 << GlobalMemory.StorageMaxCount) - 1;
+            AccessibleConstantBuffersMask = (1 << GlobalMemory.UbeMaxCount) - 1;
 
             UsedInputAttributesPerPatch  = new HashSet<int>();
             UsedOutputAttributesPerPatch = new HashSet<int>();
@@ -121,6 +123,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             OutputTopology           = outputTopology;
             MaxOutputVertices        = maxOutputVertices;
             TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
+
+            if (Stage != ShaderStage.Compute)
+            {
+                AccessibleConstantBuffersMask = 0;
+            }
         }
 
         public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(gpuAccessor, options)
@@ -404,9 +411,10 @@ namespace Ryujinx.Graphics.Shader.Translation
             UsedFeatures |= flags;
         }
 
-        public void SetAccessibleStorageBuffersMask(int mask)
+        public void SetAccessibleBufferMasks(int sbMask, int ubeMask)
         {
-            AccessibleStorageBuffersMask = mask;
+            AccessibleStorageBuffersMask = sbMask;
+            AccessibleConstantBuffersMask = ubeMask;
         }
 
         public void SetUsedConstantBuffer(int slot)