Generate scaling helper functions on IR (#4714)

* Generate scaling helper functions on IR

* Delete unused code

* Split RewriteTextureSample and move gather bias add to an earlier pass

* Remove using

* Shader cache version bump

5 files changed, 20 insertions, 269 deletions

diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
index 0ef89b39..7af6d316 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/CodeGenContext.cs
@@ -36,6 +36,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
         public Dictionary<IoDefinition, Instruction> OutputsPerPatch { get; } = new Dictionary<IoDefinition, Instruction>();
 
         public Instruction CoordTemp { get; set; }
+        public StructuredFunction CurrentFunction { get; set; }
         private readonly Dictionary<AstOperand, Instruction> _locals = new Dictionary<AstOperand, Instruction>();
         private readonly Dictionary<int, Instruction[]> _localForArgs = new Dictionary<int, Instruction[]>();
         private readonly Dictionary<int, Instruction> _funcArgs = new Dictionary<int, Instruction>();
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs
index fda0dc47..eb64f824 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs
@@ -4,7 +4,6 @@ using Ryujinx.Graphics.Shader.Translation;
 using System;
 using System.Collections.Generic;
 using System.Diagnostics;
-using System.Linq;
 using System.Numerics;
 using static Spv.Specification;
 
@@ -114,6 +113,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             Add(Instruction.MemoryBarrier,            GenerateMemoryBarrier);
             Add(Instruction.Minimum,                  GenerateMinimum);
             Add(Instruction.MinimumU32,               GenerateMinimumU32);
+            Add(Instruction.Modulo,                   GenerateModulo);
             Add(Instruction.Multiply,                 GenerateMultiply);
             Add(Instruction.MultiplyHighS32,          GenerateMultiplyHighS32);
             Add(Instruction.MultiplyHighU32,          GenerateMultiplyHighU32);
@@ -744,8 +744,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 pCoords = Src(AggregateType.S32);
             }
 
-            pCoords = ScalingHelpers.ApplyScaling(context, texOp, pCoords, intCoords: true, isBindless, isIndexed, isArray, pCount);
-
             (var imageType, var imageVariable) = context.Images[new TextureMeta(texOp.CbufSlot, texOp.Handle, texOp.Format)];
 
             var image = context.Load(imageType, imageVariable);
@@ -1040,6 +1038,11 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             return GenerateBinaryU32(context, operation, context.Delegates.GlslUMin);
         }
 
+        private static OperationResult GenerateModulo(CodeGenContext context, AstOperation operation)
+        {
+            return GenerateBinary(context, operation, context.Delegates.FMod, null);
+        }
+
         private static OperationResult GenerateMultiply(CodeGenContext context, AstOperation operation)
         {
             return GenerateBinary(context, operation, context.Delegates.FMul, context.Delegates.IMul);
@@ -1101,7 +1104,15 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         private static OperationResult GenerateReturn(CodeGenContext context, AstOperation operation)
         {
-            context.Return();
+            if (operation.SourcesCount != 0)
+            {
+                context.ReturnValue(context.Get(context.CurrentFunction.ReturnType, operation.GetSource(0)));
+            }
+            else
+            {
+                context.Return();
+            }
+
             return OperationResult.Invalid;
         }
 
@@ -1439,35 +1450,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 }
             }
 
-            SpvInstruction ApplyBias(SpvInstruction vector, SpvInstruction image)
-            {
-                int gatherBiasPrecision = context.Config.GpuAccessor.QueryHostGatherBiasPrecision();
-                if (isGather && gatherBiasPrecision != 0)
-                {
-                    // GPU requires texture gather to be slightly offset to match NVIDIA behaviour when point is exactly between two texels.
-                    // Offset by the gather precision divided by 2 to correct for rounding.
-                    var sizeType = pCount == 1 ? context.TypeS32() : context.TypeVector(context.TypeS32(), pCount);
-                    var pVectorType = pCount == 1 ? context.TypeFP32() : context.TypeVector(context.TypeFP32(), pCount);
-
-                    var bias = context.Constant(context.TypeFP32(), (float)(1 << (gatherBiasPrecision + 1)));
-                    var biasVector = context.CompositeConstruct(pVectorType, Enumerable.Repeat(bias, pCount).ToArray());
-
-                    var one = context.Constant(context.TypeFP32(), 1f);
-                    var oneVector = context.CompositeConstruct(pVectorType, Enumerable.Repeat(one, pCount).ToArray());
-
-                    var divisor = context.FMul(
-                        pVectorType,
-                        context.ConvertSToF(pVectorType, context.ImageQuerySize(sizeType, image)),
-                        biasVector);
-
-                    vector = context.FAdd(pVectorType, vector, context.FDiv(pVectorType, oneVector, divisor));
-                }
-
-                return vector;
-            }
-
             SpvInstruction pCoords = AssemblePVector(pCount);
-            pCoords = ScalingHelpers.ApplyScaling(context, texOp, pCoords, intCoords, isBindless, isIndexed, isArray, pCount);
 
             SpvInstruction AssembleDerivativesVector(int count)
             {
@@ -1638,8 +1621,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 image = context.Image(imageType, image);
             }
 
-            pCoords = ApplyBias(pCoords, image);
-
             var operands = operandsList.ToArray();
 
             SpvInstruction result;
@@ -1755,11 +1736,6 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                     result = context.CompositeExtract(context.TypeS32(), result, (SpvLiteralInteger)texOp.Index);
                 }
 
-                if (texOp.Index < 2 || (type & SamplerType.Mask) == SamplerType.Texture3D)
-                {
-                    result = ScalingHelpers.ApplyUnscaling(context, texOp.WithType(type), result, isBindless, isIndexed);
-                }
-
                 return new OperationResult(AggregateType.S32, result);
             }
         }
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/ScalingHelpers.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/ScalingHelpers.cs
deleted file mode 100644
index c8b21e88..00000000
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/ScalingHelpers.cs
+++ /dev/null
@@ -1,227 +0,0 @@
-using Ryujinx.Graphics.Shader.IntermediateRepresentation;
-using Ryujinx.Graphics.Shader.StructuredIr;
-using Ryujinx.Graphics.Shader.Translation;
-using static Spv.Specification;
-
-namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
-{
-    using SpvInstruction = Spv.Generator.Instruction;
-
-    static class ScalingHelpers
-    {
-        public static SpvInstruction ApplyScaling(
-            CodeGenContext context,
-            AstTextureOperation texOp,
-            SpvInstruction vector,
-            bool intCoords,
-            bool isBindless,
-            bool isIndexed,
-            bool isArray,
-            int pCount)
-        {
-            if (intCoords)
-            {
-                if (context.Config.Stage.SupportsRenderScale() &&
-                    !isBindless &&
-                    !isIndexed)
-                {
-                    int index = texOp.Inst == Instruction.ImageLoad
-                        ? context.Config.GetTextureDescriptors().Length + context.Config.FindImageDescriptorIndex(texOp)
-                        : context.Config.FindTextureDescriptorIndex(texOp);
-
-                    if (pCount == 3 && isArray)
-                    {
-                        return ApplyScaling2DArray(context, vector, index);
-                    }
-                    else if (pCount == 2 && !isArray)
-                    {
-                        return ApplyScaling2D(context, vector, index);
-                    }
-                }
-            }
-
-            return vector;
-        }
-
-        private static SpvInstruction ApplyScaling2DArray(CodeGenContext context, SpvInstruction vector, int index)
-        {
-            // The array index is not scaled, just x and y.
-            var vectorXY = context.VectorShuffle(context.TypeVector(context.TypeS32(), 2), vector, vector, 0, 1);
-            var vectorZ = context.CompositeExtract(context.TypeS32(), vector, 2);
-            var vectorXYScaled = ApplyScaling2D(context, vectorXY, index);
-            var vectorScaled = context.CompositeConstruct(context.TypeVector(context.TypeS32(), 3), vectorXYScaled, vectorZ);
-
-            return vectorScaled;
-        }
-
-        private static SpvInstruction ApplyScaling2D(CodeGenContext context, SpvInstruction vector, int index)
-        {
-            var pointerType = context.TypePointer(StorageClass.Uniform, context.TypeFP32());
-            var fieldIndex = context.Constant(context.TypeU32(), 4);
-            var scaleIndex = context.Constant(context.TypeU32(), index);
-
-            if (context.Config.Stage == ShaderStage.Vertex)
-            {
-                var scaleCountPointerType = context.TypePointer(StorageClass.Uniform, context.TypeS32());
-                var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.ConstantBuffers[0], context.Constant(context.TypeU32(), 3));
-                var scaleCount = context.Load(context.TypeS32(), scaleCountElemPointer);
-
-                scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, scaleCount);
-            }
-
-            scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, context.Constant(context.TypeU32(), 1));
-
-            var scaleElemPointer = context.AccessChain(pointerType, context.ConstantBuffers[0], fieldIndex, scaleIndex);
-            var scale = context.Load(context.TypeFP32(), scaleElemPointer);
-
-            var ivector2Type = context.TypeVector(context.TypeS32(), 2);
-            var localVector = context.CoordTemp;
-
-            var passthrough = context.FOrdEqual(context.TypeBool(), scale, context.Constant(context.TypeFP32(), 1f));
-
-            var mergeLabel = context.Label();
-
-            if (context.Config.Stage == ShaderStage.Fragment)
-            {
-                var scaledInterpolatedLabel = context.Label();
-                var scaledNoInterpolationLabel = context.Label();
-
-                var needsInterpolation = context.FOrdLessThan(context.TypeBool(), scale, context.Constant(context.TypeFP32(), 0f));
-
-                context.SelectionMerge(mergeLabel, SelectionControlMask.MaskNone);
-                context.BranchConditional(needsInterpolation, scaledInterpolatedLabel, scaledNoInterpolationLabel);
-
-                // scale < 0.0
-                context.AddLabel(scaledInterpolatedLabel);
-
-                ApplyScalingInterpolated(context, localVector, vector, scale);
-                context.Branch(mergeLabel);
-
-                // scale >= 0.0
-                context.AddLabel(scaledNoInterpolationLabel);
-
-                ApplyScalingNoInterpolation(context, localVector, vector, scale);
-                context.Branch(mergeLabel);
-
-                context.AddLabel(mergeLabel);
-
-                var passthroughLabel = context.Label();
-                var finalMergeLabel = context.Label();
-
-                context.SelectionMerge(finalMergeLabel, SelectionControlMask.MaskNone);
-                context.BranchConditional(passthrough, passthroughLabel, finalMergeLabel);
-
-                context.AddLabel(passthroughLabel);
-
-                context.Store(localVector, vector);
-                context.Branch(finalMergeLabel);
-
-                context.AddLabel(finalMergeLabel);
-
-                return context.Load(ivector2Type, localVector);
-            }
-            else
-            {
-                var passthroughLabel = context.Label();
-                var scaledLabel = context.Label();
-
-                context.SelectionMerge(mergeLabel, SelectionControlMask.MaskNone);
-                context.BranchConditional(passthrough, passthroughLabel, scaledLabel);
-
-                // scale == 1.0
-                context.AddLabel(passthroughLabel);
-
-                context.Store(localVector, vector);
-                context.Branch(mergeLabel);
-
-                // scale != 1.0
-                context.AddLabel(scaledLabel);
-
-                ApplyScalingNoInterpolation(context, localVector, vector, scale);
-                context.Branch(mergeLabel);
-
-                context.AddLabel(mergeLabel);
-
-                return context.Load(ivector2Type, localVector);
-            }
-        }
-
-        private static void ApplyScalingInterpolated(CodeGenContext context, SpvInstruction output, SpvInstruction vector, SpvInstruction scale)
-        {
-            var vector2Type = context.TypeVector(context.TypeFP32(), 2);
-
-            var scaleNegated = context.FNegate(context.TypeFP32(), scale);
-            var scaleVector = context.CompositeConstruct(vector2Type, scaleNegated, scaleNegated);
-
-            var vectorFloat = context.ConvertSToF(vector2Type, vector);
-            var vectorScaled = context.VectorTimesScalar(vector2Type, vectorFloat, scaleNegated);
-
-            var fragCoordPointer = context.Inputs[new IoDefinition(StorageKind.Input, IoVariable.FragmentCoord)];
-            var fragCoord = context.Load(context.TypeVector(context.TypeFP32(), 4), fragCoordPointer);
-            var fragCoordXY = context.VectorShuffle(vector2Type, fragCoord, fragCoord, 0, 1);
-
-            var scaleMod = context.FMod(vector2Type, fragCoordXY, scaleVector);
-            var vectorInterpolated = context.FAdd(vector2Type, vectorScaled, scaleMod);
-
-            context.Store(output, context.ConvertFToS(context.TypeVector(context.TypeS32(), 2), vectorInterpolated));
-        }
-
-        private static void ApplyScalingNoInterpolation(CodeGenContext context, SpvInstruction output, SpvInstruction vector, SpvInstruction scale)
-        {
-            if (context.Config.Stage == ShaderStage.Vertex)
-            {
-                scale = context.GlslFAbs(context.TypeFP32(), scale);
-            }
-
-            var vector2Type = context.TypeVector(context.TypeFP32(), 2);
-
-            var vectorFloat = context.ConvertSToF(vector2Type, vector);
-            var vectorScaled = context.VectorTimesScalar(vector2Type, vectorFloat, scale);
-
-            context.Store(output, context.ConvertFToS(context.TypeVector(context.TypeS32(), 2), vectorScaled));
-        }
-
-        public static SpvInstruction ApplyUnscaling(
-            CodeGenContext context,
-            AstTextureOperation texOp,
-            SpvInstruction size,
-            bool isBindless,
-            bool isIndexed)
-        {
-            if (context.Config.Stage.SupportsRenderScale() &&
-                !isBindless &&
-                !isIndexed)
-            {
-                int index = context.Config.FindTextureDescriptorIndex(texOp);
-
-                var pointerType = context.TypePointer(StorageClass.Uniform, context.TypeFP32());
-                var fieldIndex = context.Constant(context.TypeU32(), 4);
-                var scaleIndex = context.Constant(context.TypeU32(), index);
-
-                if (context.Config.Stage == ShaderStage.Vertex)
-                {
-                    var scaleCountPointerType = context.TypePointer(StorageClass.Uniform, context.TypeS32());
-                    var scaleCountElemPointer = context.AccessChain(scaleCountPointerType, context.ConstantBuffers[0], context.Constant(context.TypeU32(), 3));
-                    var scaleCount = context.Load(context.TypeS32(), scaleCountElemPointer);
-
-                    scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, scaleCount);
-                }
-
-                scaleIndex = context.IAdd(context.TypeU32(), scaleIndex, context.Constant(context.TypeU32(), 1));
-
-                var scaleElemPointer = context.AccessChain(pointerType, context.ConstantBuffers[0], fieldIndex, scaleIndex);
-                var scale = context.GlslFAbs(context.TypeFP32(), context.Load(context.TypeFP32(), scaleElemPointer));
-
-                var passthrough = context.FOrdEqual(context.TypeBool(), scale, context.Constant(context.TypeFP32(), 1f));
-
-                var sizeFloat = context.ConvertSToF(context.TypeFP32(), size);
-                var sizeUnscaled = context.FDiv(context.TypeFP32(), sizeFloat, scale);
-                var sizeUnscaledInt = context.ConvertFToS(context.TypeS32(), sizeUnscaled);
-
-                return context.Select(context.TypeS32(), passthrough, size, sizeUnscaledInt);
-            }
-
-            return size;
-        }
-    }
-}
\ No newline at end of file
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvDelegates.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvDelegates.cs
index 3ccfd7f5..0fa954e1 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvDelegates.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvDelegates.cs
@@ -67,6 +67,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
         public readonly FuncBinaryInstruction GlslSMax;
         public readonly FuncBinaryInstruction GlslFMin;
         public readonly FuncBinaryInstruction GlslSMin;
+        public readonly FuncBinaryInstruction FMod;
         public readonly FuncBinaryInstruction FMul;
         public readonly FuncBinaryInstruction IMul;
         public readonly FuncBinaryInstruction FSub;
@@ -174,6 +175,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             GlslSMax = context.GlslSMax;
             GlslFMin = context.GlslFMin;
             GlslSMin = context.GlslSMin;
+            FMod = context.FMod;
             FMul = context.FMul;
             IMul = context.IMul;
             FSub = context.FSub;
diff --git a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
index 3e11a974..a55e09fd 100644
--- a/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
+++ b/src/Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs
@@ -144,10 +144,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
         private static void Generate(CodeGenContext context, StructuredProgramInfo info, int funcIndex)
         {
-            var function = info.Functions[funcIndex];
-
-            (_, var spvFunc) = context.GetFunction(funcIndex);
+            (var function, var spvFunc) = context.GetFunction(funcIndex);
 
+            context.CurrentFunction = function;
             context.AddFunction(spvFunc);
             context.StartFunction();