GPU: Add fallback when 16-bit formats are not supported (#4108)1.1.492

* Add conversion for 16 bit RGBA formats (not supported in Rosetta)

* Rebase fix

Rebase fix

* Forgot to remove this

* Fix RGBA16 format conversion

* Add RGBA4 -> RGBA8 conversion

* Handle host stride alignment

* Address Feedback Part 1

* Can't count

* Don't zero out rgb when alpha is 0

* Separate RGBA4 and 5-bit component formats

Not sure of a better way to name them...

* Add A1B5G5R5 conversion

* Put this in the right place.

* Make format naming consistent for capabilities

* Change method names

8 files changed, 285 insertions, 14 deletions

diff --git a/Ryujinx.Graphics.GAL/Capabilities.cs b/Ryujinx.Graphics.GAL/Capabilities.cs
index bc93908b1..abacdcfa6 100644
--- a/Ryujinx.Graphics.GAL/Capabilities.cs
+++ b/Ryujinx.Graphics.GAL/Capabilities.cs
@@ -18,7 +18,9 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool Supports3DTextureCompression;
         public readonly bool SupportsBgraFormat;
         public readonly bool SupportsR4G4Format;
+        public readonly bool SupportsR4G4B4A4Format;
         public readonly bool SupportsSnormBufferTextureFormat;
+        public readonly bool Supports5BitComponentFormat;
         public readonly bool SupportsFragmentShaderInterlock;
         public readonly bool SupportsFragmentShaderOrderingIntel;
         public readonly bool SupportsGeometryShaderPassthrough;
@@ -55,7 +57,9 @@ namespace Ryujinx.Graphics.GAL
             bool supports3DTextureCompression,
             bool supportsBgraFormat,
             bool supportsR4G4Format,
+            bool supportsR4G4B4A4Format,
             bool supportsSnormBufferTextureFormat,
+            bool supports5BitComponentFormat,
             bool supportsFragmentShaderInterlock,
             bool supportsFragmentShaderOrderingIntel,
             bool supportsGeometryShaderPassthrough,
@@ -89,7 +93,9 @@ namespace Ryujinx.Graphics.GAL
             Supports3DTextureCompression = supports3DTextureCompression;
             SupportsBgraFormat = supportsBgraFormat;
             SupportsR4G4Format = supportsR4G4Format;
+            SupportsR4G4B4A4Format = supportsR4G4B4A4Format;
             SupportsSnormBufferTextureFormat = supportsSnormBufferTextureFormat;
+            Supports5BitComponentFormat = supports5BitComponentFormat;
             SupportsFragmentShaderInterlock = supportsFragmentShaderInterlock;
             SupportsFragmentShaderOrderingIntel = supportsFragmentShaderOrderingIntel;
             SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
diff --git a/Ryujinx.Graphics.GAL/Format.cs b/Ryujinx.Graphics.GAL/Format.cs
index 87d08803d..5e0274e58 100644
--- a/Ryujinx.Graphics.GAL/Format.cs
+++ b/Ryujinx.Graphics.GAL/Format.cs
@@ -449,6 +449,27 @@ namespace Ryujinx.Graphics.GAL
         }
 
         /// <summary>
+        /// Checks if the texture format is 16 bit packed.
+        /// </summary>
+        /// <param name="format">Texture format</param>
+        /// <returns>True if the texture format is 16 bit packed, false otherwise</returns>
+        public static bool Is16BitPacked(this Format format)
+        {
+            switch (format)
+            {
+                case Format.B5G6R5Unorm:
+                case Format.B5G5R5A1Unorm:
+                case Format.R5G5B5X1Unorm:
+                case Format.R5G5B5A1Unorm:
+                case Format.R5G6B5Unorm:
+                case Format.R4G4B4A4Unorm:
+                    return true;
+            }
+
+            return false;
+        }
+
+        /// <summary>
         /// Checks if the texture format is an ASTC format.
         /// </summary>
         /// <param name="format">Texture format</param>
diff --git a/Ryujinx.Graphics.Gpu/Image/Texture.cs b/Ryujinx.Graphics.Gpu/Image/Texture.cs
index 904c908f2..0995314d0 100644
--- a/Ryujinx.Graphics.Gpu/Image/Texture.cs
+++ b/Ryujinx.Graphics.Gpu/Image/Texture.cs
@@ -911,7 +911,40 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
             else if (!_context.Capabilities.SupportsR4G4Format && Format == Format.R4G4Unorm)
             {
-                result = PixelConverter.ConvertR4G4ToR4G4B4A4(result);
+                result = PixelConverter.ConvertR4G4ToR4G4B4A4(result, width);
+
+                if (!_context.Capabilities.SupportsR4G4B4A4Format)
+                {
+                    result = PixelConverter.ConvertR4G4B4A4ToR8G8B8A8(result, width);
+                }
+            }
+            else if (Format == Format.R4G4B4A4Unorm)
+            {
+                if (!_context.Capabilities.SupportsR4G4B4A4Format)
+                {
+                    result = PixelConverter.ConvertR4G4B4A4ToR8G8B8A8(result, width);
+                }
+            }
+            else if (!_context.Capabilities.Supports5BitComponentFormat && Format.Is16BitPacked())
+            {
+                switch (Format)
+                {
+                    case Format.B5G6R5Unorm:
+                    case Format.R5G6B5Unorm:
+                        result = PixelConverter.ConvertR5G6B5ToR8G8B8A8(result, width);
+                        break;
+                    case Format.B5G5R5A1Unorm:
+                    case Format.R5G5B5X1Unorm:
+                    case Format.R5G5B5A1Unorm:
+                        result = PixelConverter.ConvertR5G5B5ToR8G8B8A8(result, width, Format == Format.R5G5B5X1Unorm);
+                        break;
+                    case Format.A1B5G5R5Unorm:
+                        result = PixelConverter.ConvertA1B5G5R5ToR8G8B8A8(result, width);
+                        break;
+                    case Format.R4G4B4A4Unorm:
+                        result = PixelConverter.ConvertR4G4B4A4ToR8G8B8A8(result, width);
+                        break;
+                }
             }
 
             return result;
diff --git a/Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs b/Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs
index 642e03b68..7ec4c7ac2 100644
--- a/Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs
+++ b/Ryujinx.Graphics.Gpu/Image/TextureCompatibility.cs
@@ -132,7 +132,26 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             if (!caps.SupportsR4G4Format && info.FormatInfo.Format == Format.R4G4Unorm)
             {
-                return new FormatInfo(Format.R4G4B4A4Unorm, 1, 1, 2, 4);
+                if (caps.SupportsR4G4B4A4Format)
+                {
+                    return new FormatInfo(Format.R4G4B4A4Unorm, 1, 1, 2, 4);
+                }
+                else
+                {
+                    return new FormatInfo(Format.R8G8B8A8Unorm, 1, 1, 4, 4);
+                }
+            }
+
+            if (info.FormatInfo.Format == Format.R4G4B4A4Unorm)
+            {
+                if (!caps.SupportsR4G4B4A4Format)
+                {
+                    return new FormatInfo(Format.R8G8B8A8Unorm, 1, 1, 4, 4);
+                }
+            }
+            else if (!caps.Supports5BitComponentFormat && info.FormatInfo.Format.Is16BitPacked())
+            {
+                return new FormatInfo(info.FormatInfo.Format.IsBgr() ? Format.B8G8R8A8Unorm : Format.R8G8B8A8Unorm, 1, 1, 4, 4);
             }
 
             return info.FormatInfo;
diff --git a/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs b/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
index 59ca6afdc..de1ce4a34 100644
--- a/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
+++ b/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs
@@ -114,7 +114,9 @@ namespace Ryujinx.Graphics.OpenGL
                 supports3DTextureCompression: false,
                 supportsBgraFormat: false,
                 supportsR4G4Format: false,
+                supportsR4G4B4A4Format: true,
                 supportsSnormBufferTextureFormat: false,
+                supports5BitComponentFormat: true,
                 supportsFragmentShaderInterlock: HwCapabilities.SupportsFragmentShaderInterlock,
                 supportsFragmentShaderOrderingIntel: HwCapabilities.SupportsFragmentShaderOrdering,
                 supportsGeometryShaderPassthrough: HwCapabilities.SupportsGeometryShaderPassthrough,
diff --git a/Ryujinx.Graphics.Texture/LayoutConverter.cs b/Ryujinx.Graphics.Texture/LayoutConverter.cs
index 188ae0c15..b8ec9748b 100644
--- a/Ryujinx.Graphics.Texture/LayoutConverter.cs
+++ b/Ryujinx.Graphics.Texture/LayoutConverter.cs
@@ -7,7 +7,7 @@ namespace Ryujinx.Graphics.Texture
 {
     public static class LayoutConverter
     {
-        private const int HostStrideAlignment = 4;
+        public const int HostStrideAlignment = 4;
 
         public static void ConvertBlockLinearToLinear(
             Span<byte> dst,
diff --git a/Ryujinx.Graphics.Texture/PixelConverter.cs b/Ryujinx.Graphics.Texture/PixelConverter.cs
index d7e45a693..add25cd35 100644
--- a/Ryujinx.Graphics.Texture/PixelConverter.cs
+++ b/Ryujinx.Graphics.Texture/PixelConverter.cs
@@ -1,3 +1,4 @@
+using Ryujinx.Common;
 using System;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics;
@@ -7,30 +8,206 @@ namespace Ryujinx.Graphics.Texture
 {
     public static class PixelConverter
     {
-        public unsafe static byte[] ConvertR4G4ToR4G4B4A4(ReadOnlySpan<byte> data)
+        private static (int remainder, int outRemainder, int height) GetLineRemainders(int length, int width, int bpp, int outBpp)
+        {
+            int stride = BitUtils.AlignUp(width * bpp, LayoutConverter.HostStrideAlignment);
+            int remainder = stride / bpp - width;
+
+            int outStride = BitUtils.AlignUp(width * outBpp, LayoutConverter.HostStrideAlignment);
+            int outRemainder = outStride / outBpp - width;
+
+            return (remainder, outRemainder, length / stride);
+        }
+
+        public unsafe static byte[] ConvertR4G4ToR4G4B4A4(ReadOnlySpan<byte> data, int width)
         {
             byte[] output = new byte[data.Length * 2];
-            int start = 0;
 
-            if (Sse41.IsSupported)
+            (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 1, 2);
+
+            Span<ushort> outputSpan = MemoryMarshal.Cast<byte, ushort>(output);
+
+            if (remainder == 0)
+            {
+                int start = 0;
+
+                if (Sse41.IsSupported)
+                {
+                    int sizeTrunc = data.Length & ~7;
+                    start = sizeTrunc;
+
+                    fixed (byte* inputPtr = data, outputPtr = output)
+                    {
+                        for (ulong offset = 0; offset < (ulong)sizeTrunc; offset += 8)
+                        {
+                            Sse2.Store(outputPtr + offset * 2, Sse41.ConvertToVector128Int16(inputPtr + offset).AsByte());
+                        }
+                    }
+                }
+
+                for (int i = start; i < data.Length; i++)
+                {
+                    outputSpan[i] = (ushort)data[i];
+                }
+            }
+            else
             {
-                int sizeTrunc = data.Length & ~7;
-                start = sizeTrunc;
+                int offset = 0;
+                int outOffset = 0;
 
-                fixed (byte* inputPtr = data, outputPtr = output)
+                for (int y = 0; y < height; y++)
                 {
-                    for (ulong offset = 0; offset < (ulong)sizeTrunc; offset += 8)
+                    for (int x = 0; x < width; x++)
                     {
-                        Sse2.Store(outputPtr + offset * 2, Sse41.ConvertToVector128Int16(inputPtr + offset).AsByte());
+                        outputSpan[outOffset++] = data[offset++];
                     }
+
+                    offset += remainder;
+                    outOffset += outRemainder;
                 }
             }
 
-            Span<ushort> outputSpan = MemoryMarshal.Cast<byte, ushort>(output);
+            return output;
+        }
+
+        public unsafe static byte[] ConvertR5G6B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        {
+            byte[] output = new byte[data.Length * 2];
+            int offset = 0;
+            int outOffset = 0;
+
+            (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
+
+            ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output);
 
-            for (int i = start; i < data.Length; i++)
+            for (int y = 0; y < height; y++)
             {
-                outputSpan[i] = (ushort)data[i];
+                for (int x = 0; x < width; x++)
+                {
+                    uint packed = inputSpan[offset++];
+
+                    uint outputPacked =  0xff000000;
+                         outputPacked |= (packed << 3) & 0x000000f8;
+                         outputPacked |= (packed << 8) & 0x00f80000;
+
+                    // Replicate 5 bit components.
+                    outputPacked |= (outputPacked >> 5) & 0x00070007;
+
+                    // Include and replicate 6 bit component.
+                    outputPacked |= ((packed << 5) & 0x0000fc00) | ((packed >> 1) & 0x00000300);
+
+                    outputSpan[outOffset++] = outputPacked;
+                }
+
+                offset += remainder;
+                outOffset += outRemainder;
+            }
+
+            return output;
+        }
+
+        public unsafe static byte[] ConvertR5G5B5ToR8G8B8A8(ReadOnlySpan<byte> data, int width, bool forceAlpha)
+        {
+            byte[] output = new byte[data.Length * 2];
+            int offset = 0;
+            int outOffset = 0;
+
+            (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
+
+            ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output);
+
+            for (int y = 0; y < height; y++)
+            {
+                for (int x = 0; x < width; x++)
+                {
+                    uint packed = inputSpan[offset++];
+
+                    uint a = forceAlpha ? 1 : (packed >> 15);
+
+                    uint outputPacked =  a * 0xff000000;
+                         outputPacked |= (packed << 3) & 0x000000f8;
+                         outputPacked |= (packed << 6) & 0x0000f800;
+                         outputPacked |= (packed << 9) & 0x00f80000;
+
+                    // Replicate 5 bit components.
+                    outputPacked |= (outputPacked >> 5) & 0x00070707;
+
+                    outputSpan[outOffset++] = outputPacked;
+                }
+
+                offset += remainder;
+                outOffset += outRemainder;
+            }
+
+            return output;
+        }
+
+        public unsafe static byte[] ConvertA1B5G5R5ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        {
+            byte[] output = new byte[data.Length * 2];
+            int offset = 0;
+            int outOffset = 0;
+
+            (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
+
+            ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output);
+
+            for (int y = 0; y < height; y++)
+            {
+                for (int x = 0; x < width; x++)
+                {
+                    uint packed = inputSpan[offset++];
+
+                    uint a = packed >> 15;
+
+                    uint outputPacked = a * 0xff000000;
+                    outputPacked |= (packed >> 8) & 0x000000f8;
+                    outputPacked |= (packed << 5) & 0x0000f800;
+                    outputPacked |= (packed << 18) & 0x00f80000;
+
+                    // Replicate 5 bit components.
+                    outputPacked |= (outputPacked >> 5) & 0x00070707;
+
+                    outputSpan[outOffset++] = outputPacked;
+                }
+
+                offset += remainder;
+                outOffset += outRemainder;
+            }
+
+            return output;
+        }
+
+        public unsafe static byte[] ConvertR4G4B4A4ToR8G8B8A8(ReadOnlySpan<byte> data, int width)
+        {
+            byte[] output = new byte[data.Length * 2];
+            int offset = 0;
+            int outOffset = 0;
+
+            (int remainder, int outRemainder, int height) = GetLineRemainders(data.Length, width, 2, 4);
+
+            ReadOnlySpan<ushort> inputSpan = MemoryMarshal.Cast<byte, ushort>(data);
+            Span<uint> outputSpan = MemoryMarshal.Cast<byte, uint>(output);
+
+            for (int y = 0; y < height; y++)
+            {
+                for (int x = 0; x < width; x++)
+                {
+                    uint packed = inputSpan[offset++];
+
+                    uint outputPacked =  packed         & 0x0000000f;
+                         outputPacked |= (packed << 4)  & 0x00000f00;
+                         outputPacked |= (packed << 8)  & 0x000f0000;
+                         outputPacked |= (packed << 12) & 0x0f000000;
+
+                    outputSpan[outOffset++] = outputPacked * 0x11;
+                }
+
+                offset += remainder;
+                outOffset += outRemainder;
             }
 
             return output;
diff --git a/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs b/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
index a32400f56..3c446abf3 100644
--- a/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
+++ b/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
@@ -396,6 +396,17 @@ namespace Ryujinx.Graphics.Vulkan
                 GAL.Format.Etc2RgbSrgb,
                 GAL.Format.Etc2RgbUnorm);
 
+            bool supports5BitComponentFormat = FormatCapabilities.OptimalFormatsSupport(compressedFormatFeatureFlags,
+                GAL.Format.R5G6B5Unorm,
+                GAL.Format.R5G5B5A1Unorm,
+                GAL.Format.R5G5B5X1Unorm,
+                GAL.Format.B5G6R5Unorm,
+                GAL.Format.B5G5R5A1Unorm,
+                GAL.Format.A1B5G5R5Unorm);
+
+            bool supportsR4G4B4A4Format = FormatCapabilities.OptimalFormatsSupport(compressedFormatFeatureFlags,
+                GAL.Format.R4G4B4A4Unorm);
+
             PhysicalDeviceVulkan12Features featuresVk12 = new PhysicalDeviceVulkan12Features()
             {
                 SType = StructureType.PhysicalDeviceVulkan12Features
@@ -425,7 +436,9 @@ namespace Ryujinx.Graphics.Vulkan
                 supports3DTextureCompression: true,
                 supportsBgraFormat: true,
                 supportsR4G4Format: false,
+                supportsR4G4B4A4Format: supportsR4G4B4A4Format,
                 supportsSnormBufferTextureFormat: true,
+                supports5BitComponentFormat: supports5BitComponentFormat,
                 supportsFragmentShaderInterlock: Capabilities.SupportsFragmentShaderInterlock,
                 supportsFragmentShaderOrderingIntel: false,
                 supportsGeometryShaderPassthrough: Capabilities.SupportsGeometryShaderPassthrough,