[SpirvTranslator] Use a flat input for v_interp_mov P0 on MoltenVK (#910)

MoltenVK advertises VK_KHR_fragment_shader_barycentric, but SPIRV-Cross
cannot translate PerVertexKHR inputs to MSL, so every pixel shader with
v_interp_mov_f32 failed pipeline creation on macOS (Poppy Playtime).
Reading only P0 is the provoking vertex value, which a flat input gives.
Attributes that read P10 or P20 still use per-vertex inputs.

src/SharpEmu.Libs/Gpu/Pipelines/IShaderPipelineHost.cs

@@ -38,6 +38,8 @@ internal interface IShaderPipelineHost
 
     bool GraphicsSubgroupOperationsEnabled { get; }
 
+    bool PerVertexPixelInputsSupported => true;
+
     RenderHostLimits Limits { get; }
 
     // The sample counts a pipeline without attachments can rasterize at.

src/SharpEmu.Libs/Gpu/Pipelines/ShaderProgramCache.cs

@@ -560,6 +560,7 @@ internal sealed class ShaderProgramCache
                     PixelOutputs = options.PixelOutputs,
                     PixelInputEnable = options.PixelInputEnable,
                     PixelCustomInterpolationMask = info.CustomInterpolationMask,
+                    SupportsPerVertexPixelInputs = _host.PerVertexPixelInputsSupported,
                     PixelInputAddress = options.PixelInputAddress,
                     PixelInputCntl = interpolators,
                 };

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.Device.Setup.cs

@@ -645,6 +645,7 @@ internal static unsafe partial class VulkanVideoPresenter
         }
 
         private bool _supportsFragmentShaderBarycentric;
+        private bool _supportsPerVertexPixelInputs;
         private const string FragmentShaderBarycentricExtensionName = "VK_KHR_fragment_shader_barycentric";
 
         private void CreateDevice()
@@ -742,6 +743,10 @@ internal static unsafe partial class VulkanVideoPresenter
                 _supportsFragmentShaderBarycentric = barycentricFeatures.FragmentShaderBarycentric;
             }
 
+            // MoltenVK exposes the barycentric builtins, but SPIRV-Cross rejects PerVertexKHR inputs.
+            _supportsPerVertexPixelInputs = _supportsFragmentShaderBarycentric &&
+                !IsDeviceExtensionAvailable(PortabilitySubsetExtensionName);
+
             var depthClipEnableFeatures = new PhysicalDeviceDepthClipEnableFeaturesEXT
             {
                 SType = StructureType.PhysicalDeviceDepthClipEnableFeaturesExt,

src/SharpEmu.Libs/VideoOut/VulkanVideoPresenter.Pipelines.cs

@@ -66,6 +66,8 @@ internal static unsafe partial class VulkanVideoPresenter
 
         bool IShaderPipelineHost.GraphicsSubgroupOperationsEnabled => GraphicsSubgroupOperationsEnabled;
 
+        bool IShaderPipelineHost.PerVertexPixelInputsSupported => _supportsPerVertexPixelInputs;
+
         RenderHostLimits IShaderPipelineHost.Limits => _renderHostLimits;
 
         SampleCountFlags IShaderPipelineHost.NoAttachmentSampleCounts => _noAttachmentSampleCounts;

src/SharpEmu.ShaderCompiler.Vulkan/Gen5SpirvTranslator.Interpolation.cs

@@ -10,6 +10,7 @@ public static partial class Gen5SpirvTranslator
     private sealed partial class CompilationContext
     {
         private readonly HashSet<uint> _perVertexAttributes = [];
+        private readonly HashSet<uint> _flatParameterAttributes = [];
         private readonly Dictionary<int, uint> _barycentricInputs = [];
         private uint _interpolationSampleId;
         private const uint InterpolateAtCentroid = 76;
@@ -27,6 +28,23 @@ public static partial class Gen5SpirvTranslator
                 }
             }
 
+            if (!_request.SupportsPerVertexPixelInputs)
+            {
+                // Reading P0 alone is the provoking vertex value, which a flat input gives.
+                // Other parameters still need the per-vertex values.
+                foreach (var attribute in _perVertexAttributes.ToArray())
+                {
+                    if (_request.Program.Instructions.All(instruction =>
+                            instruction.Control is not Gen5InterpolationControl control ||
+                            control.Attribute != attribute ||
+                            (instruction.Opcode == "VInterpMovF32" && (instruction.Words[0] & 0xFFu) == 2)))
+                    {
+                        _perVertexAttributes.Remove(attribute);
+                        _flatParameterAttributes.Add(attribute);
+                    }
+                }
+            }
+
             if (_perVertexAttributes.Count == 0)
             {
                 return;

src/SharpEmu.ShaderCompiler.Vulkan/Gen5SpirvTranslator.cs

@@ -873,7 +873,7 @@ public static partial class Gen5SpirvTranslator
                     {
                         _module.AddDecoration(variable, SpirvDecoration.PerVertexKhr);
                     }
-                    else if ((cntl & 0x400u) != 0)
+                    else if ((cntl & 0x400u) != 0 || _flatParameterAttributes.Contains(attribute))
                     {
                         _module.AddDecoration(variable, SpirvDecoration.Flat);
                     }

src/SharpEmu.ShaderCompiler/ShaderCompileRequest.cs

@@ -138,6 +138,10 @@ public sealed class ShaderCompileRequest
     public IReadOnlyList<Gen5PixelOutputBinding> PixelOutputs { get; init; } = [];
     public uint PixelInputEnable { get; init; }
     public uint PixelCustomInterpolationMask { get; init; }
+
+    // False when the device cannot read one vertex's value of a pixel input (PerVertexKHR).
+    // MoltenVK advertises fragment shader barycentrics but cannot translate PerVertexKHR to MSL.
+    public bool SupportsPerVertexPixelInputs { get; init; } = true;
     public uint PixelInputAddress { get; init; }
     public IReadOnlyList<uint>? PixelInputCntl { get; init; }
 

tests/SharpEmu.ShaderCompiler.Tests/Gen5InterpolationParameterTests.cs

@@ -79,8 +79,37 @@ public sealed class Gen5InterpolationParameterTests
         Assert.Contains("Pull-model interpolation", error);
     }
 
+    [Fact]
+    public void ProvokingVertexMove_UsesAFlatInputWithoutPerVertexSupport()
+    {
+        var request = Request(2, false, inputCntl: 0x1, supportsPerVertex: false);
+        Assert.True(Gen5SpirvTranslator.TryCompileProgram(request, out var shader, out var error), error);
+        var instructions = Instructions(shader.Spirv);
+        Assert.DoesNotContain(instructions, instruction => instruction.Opcode == SpirvOp.Capability &&
+            instruction.Operands[0] == (uint)SpirvCapability.FragmentBarycentricKhr);
+        Assert.DoesNotContain(instructions, instruction => instruction.Opcode == SpirvOp.Decorate &&
+            instruction.Operands[1] == (uint)SpirvDecoration.PerVertexKhr);
+        var input = Assert.Single(instructions, instruction => instruction.Opcode == SpirvOp.Decorate &&
+            instruction.Operands[1] == (uint)SpirvDecoration.Location).Operands[0];
+        Assert.Contains(instructions, instruction => instruction.Opcode == SpirvOp.Decorate &&
+            instruction.Operands[0] == input && instruction.Operands[1] == (uint)SpirvDecoration.Flat);
+        ValidateWhenAvailable(shader.Spirv);
+    }
+
+    [Theory]
+    [InlineData(0u)]
+    [InlineData(1u)]
+    public void VertexDifferenceMove_KeepsPerVertexInputWithoutPerVertexSupport(uint selector)
+    {
+        var request = Request(selector, false, inputCntl: 0x1, supportsPerVertex: false);
+        Assert.True(Gen5SpirvTranslator.TryCompileProgram(request, out var shader, out var error), error);
+        Assert.Contains(Instructions(shader.Spirv), instruction => instruction.Opcode == SpirvOp.Decorate &&
+            instruction.Operands[1] == (uint)SpirvDecoration.PerVertexKhr);
+    }
+
     private static ShaderCompileRequest Request(
-        uint selector, bool custom, uint inputs = 2, string opcode = "VInterpMovF32")
+        uint selector, bool custom, uint inputs = 2, string opcode = "VInterpMovF32",
+        uint inputCntl = 0x401, bool supportsPerVertex = true)
     {
         var interpolation = new Gen5ShaderInstruction(0, Gen5ShaderEncoding.Vintrp, opcode,
             [selector], [Gen5Operand.Vector(selector)], [Gen5Operand.Vector(4)], new Gen5InterpolationControl(1, 2));
@@ -90,8 +119,9 @@ public sealed class Gen5InterpolationParameterTests
         {
             PixelInputAddress = inputs,
             PixelInputEnable = inputs,
-            PixelInputCntl = [0, 0x401],
+            PixelInputCntl = [0, inputCntl],
             PixelCustomInterpolationMask = custom ? 2u : 0u,
+            SupportsPerVertexPixelInputs = supportsPerVertex,
         };
     }