More flexible memory manager (#307)

* Keep track mapped buffers with fixed offsets

* Started rewriting the memory manager

* Initial support for MapPhysicalMemory and UnmapPhysicalMemory, other tweaks

* MapPhysicalMemory/UnmapPhysicalMemory support, other tweaks

* Rebased

* Optimize the map/unmap physical memory svcs

* Integrate shared font support

* Fix address space reserve alignment

* Some fixes related to gpu memory mapping

* Some cleanup

* Only try uploading const buffers that are really used

* Check if memory region is contiguous

* Rebased

* Add missing count increment on IsRegionModified

* Check for reads/writes outside of the address space, optimize translation with a tail call

5 files changed, 47 insertions, 73 deletions

diff --git a/Ryujinx.HLE/Gpu/Engines/NvGpuEngine3d.cs b/Ryujinx.HLE/Gpu/Engines/NvGpuEngine3d.cs
index cb1514b5..6f601244 100644
--- a/Ryujinx.HLE/Gpu/Engines/NvGpuEngine3d.cs
+++ b/Ryujinx.HLE/Gpu/Engines/NvGpuEngine3d.cs
@@ -109,7 +109,7 @@ namespace Ryujinx.HLE.Gpu.Engines
             Gpu.Renderer.Shader.BindProgram();
 
             UploadTextures(Vmm, State, Keys);
-            UploadConstBuffers(Vmm, State);
+            UploadConstBuffers(Vmm, State, Keys);
             UploadVertexArrays(Vmm, State);
 
             DispatchRender(Vmm, State);
@@ -426,6 +426,12 @@ namespace Ryujinx.HLE.Gpu.Engines
 
             Key = Vmm.GetPhysicalAddress(Key);
 
+            if (Key == -1)
+            {
+                //FIXME: Should'nt ignore invalid addresses.
+                return;
+            }
+
             if (IsFrameBufferPosition(Key))
             {
                 //This texture is a frame buffer texture,
@@ -465,24 +471,29 @@ namespace Ryujinx.HLE.Gpu.Engines
             Gpu.Renderer.Texture.SetSampler(Sampler);
         }
 
-        private void UploadConstBuffers(NvGpuVmm Vmm, GalPipelineState State)
+        private void UploadConstBuffers(NvGpuVmm Vmm, GalPipelineState State, long[] Keys)
         {
-            for (int Stage = 0; Stage < State.ConstBufferKeys.Length; Stage++)
+            for (int Stage = 0; Stage < Keys.Length; Stage++)
             {
-                for (int Index = 0; Index < State.ConstBufferKeys[Stage].Length; Index++)
+                foreach (ShaderDeclInfo DeclInfo in Gpu.Renderer.Shader.GetConstBufferUsage(Keys[Stage]))
                 {
-                    ConstBuffer Cb = ConstBuffers[Stage][Index];
+                    ConstBuffer Cb = ConstBuffers[Stage][DeclInfo.Cbuf];
 
-                    long Key = Cb.Position;
+                    if (!Cb.Enabled)
+                    {
+                        continue;
+                    }
 
-                    if (Cb.Enabled && QueryKeyUpload(Vmm, Key, Cb.Size, NvGpuBufferType.ConstBuffer))
+                    long Key = Vmm.GetPhysicalAddress(Cb.Position);
+
+                    if (QueryKeyUpload(Vmm, Key, Cb.Size, NvGpuBufferType.ConstBuffer))
                     {
-                        IntPtr Source = Vmm.GetHostAddress(Key, Cb.Size);
+                        IntPtr Source = Vmm.GetHostAddress(Cb.Position, Cb.Size);
 
                         Gpu.Renderer.Buffer.SetData(Key, Cb.Size, Source);
                     }
 
-                    State.ConstBufferKeys[Stage][Index] = Key;
+                    State.ConstBufferKeys[Stage][DeclInfo.Cbuf] = Key;
                 }
             }
         }
@@ -668,11 +679,13 @@ namespace Ryujinx.HLE.Gpu.Engines
 
             long Position = MakeInt64From2xInt32(NvGpuEngine3dReg.ConstBufferAddress);
 
+            long CbKey = Vmm.GetPhysicalAddress(Position);
+
             int Size = ReadRegister(NvGpuEngine3dReg.ConstBufferSize);
 
-            if (!Gpu.Renderer.Buffer.IsCached(Position, Size))
+            if (!Gpu.Renderer.Buffer.IsCached(CbKey, Size))
             {
-                Gpu.Renderer.Buffer.Create(Position, Size);
+                Gpu.Renderer.Buffer.Create(CbKey, Size);
             }
 
             ConstBuffer Cb = ConstBuffers[Stage][Index];
diff --git a/Ryujinx.HLE/Gpu/Memory/NvGpuVmm.cs b/Ryujinx.HLE/Gpu/Memory/NvGpuVmm.cs
index 7b23e49f..e7e18064 100644
--- a/Ryujinx.HLE/Gpu/Memory/NvGpuVmm.cs
+++ b/Ryujinx.HLE/Gpu/Memory/NvGpuVmm.cs
@@ -1,7 +1,6 @@
 using ChocolArm64.Memory;
 using Ryujinx.Graphics.Gal;
 using System;
-using System.Collections.Concurrent;
 
 namespace Ryujinx.HLE.Gpu.Memory
 {
@@ -26,18 +25,6 @@ namespace Ryujinx.HLE.Gpu.Memory
 
         public AMemory Memory { get; private set; }
 
-        private struct MappedMemory
-        {
-            public long Size;
-
-            public MappedMemory(long Size)
-            {
-                this.Size = Size;
-            }
-        }
-
-        private ConcurrentDictionary<long, MappedMemory> Maps;
-
         private NvGpuVmmCache Cache;
 
         private const long PteUnmapped = -1;
@@ -49,8 +36,6 @@ namespace Ryujinx.HLE.Gpu.Memory
         {
             this.Memory = Memory;
 
-            Maps = new ConcurrentDictionary<long, MappedMemory>();
-
             Cache = new NvGpuVmmCache();
 
             PageTable = new long[PTLvl0Size][];
@@ -62,14 +47,6 @@ namespace Ryujinx.HLE.Gpu.Memory
             {
                 for (long Offset = 0; Offset < Size; Offset += PageSize)
                 {
-                    if (GetPte(VA + Offset) != PteReserved)
-                    {
-                        return Map(PA, Size);
-                    }
-                }
-
-                for (long Offset = 0; Offset < Size; Offset += PageSize)
-                {
                     SetPte(VA + Offset, PA + Offset);
                 }
             }
@@ -85,10 +62,6 @@ namespace Ryujinx.HLE.Gpu.Memory
 
                 if (VA != -1)
                 {
-                    MappedMemory Map = new MappedMemory(Size);
-
-                    Maps.AddOrUpdate(VA, Map, (Key, Old) => Map);
-
                     for (long Offset = 0; Offset < Size; Offset += PageSize)
                     {
                         SetPte(VA + Offset, PA + Offset);
@@ -99,19 +72,7 @@ namespace Ryujinx.HLE.Gpu.Memory
             }
         }
 
-        public bool Unmap(long VA)
-        {
-            if (Maps.TryRemove(VA, out MappedMemory Map))
-            {
-                Free(VA, Map.Size);
-
-                return true;
-            }
-
-            return false;
-        }
-
-        public long Reserve(long VA, long Size, long Align)
+        public long ReserveFixed(long VA, long Size)
         {
             lock (PageTable)
             {
@@ -119,7 +80,7 @@ namespace Ryujinx.HLE.Gpu.Memory
                 {
                     if (IsPageInUse(VA + Offset))
                     {
-                        return Reserve(Size, Align);
+                        return -1;
                     }
                 }
 
@@ -163,7 +124,9 @@ namespace Ryujinx.HLE.Gpu.Memory
 
         private long GetFreePosition(long Size, long Align = 1)
         {
-            long Position = 0;
+            //Note: Address 0 is not considered valid by the driver,
+            //when 0 is returned it's considered a mapping error.
+            long Position = PageSize;
             long FreeSize = 0;
 
             if (Align < 1)
diff --git a/Ryujinx.HLE/Gpu/Memory/NvGpuVmmCache.cs b/Ryujinx.HLE/Gpu/Memory/NvGpuVmmCache.cs
index 3c256044..b3f253b3 100644
--- a/Ryujinx.HLE/Gpu/Memory/NvGpuVmmCache.cs
+++ b/Ryujinx.HLE/Gpu/Memory/NvGpuVmmCache.cs
@@ -1,4 +1,5 @@
 using ChocolArm64.Memory;
+using Ryujinx.HLE.Memory;
 using System;
 using System.Collections.Generic;
 
@@ -129,14 +130,11 @@ namespace Ryujinx.HLE.Gpu.Memory
         {
             (bool[] Modified, long ModifiedCount) = Memory.IsRegionModified(PA, Size);
 
-            if (Modified == null)
-            {
-                return true;
-            }
+            PA = Memory.GetPhysicalAddress(PA);
 
             ClearCachedPagesIfNeeded();
 
-            long PageSize = Memory.GetHostPageSize();
+            long PageSize = AMemory.PageSize;
 
             EnsureResidencyInitialized(PageSize);
 
@@ -159,9 +157,9 @@ namespace Ryujinx.HLE.Gpu.Memory
 
             while (PA < PAEnd)
             {
-                long Key = PA & ~Mask;
+                long Key = PA & ~AMemory.PageMask;
 
-                long PAPgEnd = Math.Min((PA + PageSize) & ~Mask, PAEnd);
+                long PAPgEnd = Math.Min((PA + AMemory.PageSize) & ~AMemory.PageMask, PAEnd);
 
                 bool IsCached = Cache.TryGetValue(Key, out CachedPage Cp);
 
@@ -228,7 +226,7 @@ namespace Ryujinx.HLE.Gpu.Memory
         {
             if (Residency == null)
             {
-                Residency = new HashSet<long>[AMemoryMgr.RamSize / PageSize];
+                Residency = new HashSet<long>[DeviceMemory.RamSize / PageSize];
 
                 for (int i = 0; i < Residency.Length; i++)
                 {
diff --git a/Ryujinx.HLE/Gpu/Texture/TextureReader.cs b/Ryujinx.HLE/Gpu/Texture/TextureReader.cs
index f8cd6765..0c6103af 100644
--- a/Ryujinx.HLE/Gpu/Texture/TextureReader.cs
+++ b/Ryujinx.HLE/Gpu/Texture/TextureReader.cs
@@ -72,7 +72,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    byte Pixel = CpuMem.ReadByteUnchecked(Position + Offset);
+                    byte Pixel = CpuMem.ReadByte(Position + Offset);
 
                     *(BuffPtr + OutOffs) = Pixel;
 
@@ -105,7 +105,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    uint Pixel = (uint)CpuMem.ReadInt16Unchecked(Position + Offset);
+                    uint Pixel = (uint)CpuMem.ReadInt16(Position + Offset);
 
                     Pixel = (Pixel & 0x001f) << 11 |
                             (Pixel & 0x03e0) << 1  |
@@ -143,7 +143,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    uint Pixel = (uint)CpuMem.ReadInt16Unchecked(Position + Offset);
+                    uint Pixel = (uint)CpuMem.ReadInt16(Position + Offset);
 
                     Pixel = (Pixel & 0x001f) << 11 |
                             (Pixel & 0x07e0)       |
@@ -180,7 +180,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    short Pixel = CpuMem.ReadInt16Unchecked(Position + Offset);
+                    short Pixel = CpuMem.ReadInt16(Position + Offset);
 
                     *(short*)(BuffPtr + OutOffs) = Pixel;
 
@@ -213,7 +213,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    int Pixel = CpuMem.ReadInt32Unchecked(Position + Offset);
+                    int Pixel = CpuMem.ReadInt32(Position + Offset);
 
                     *(int*)(BuffPtr + OutOffs) = Pixel;
 
@@ -246,7 +246,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    long Pixel = CpuMem.ReadInt64Unchecked(Position + Offset);
+                    long Pixel = CpuMem.ReadInt64(Position + Offset);
 
                     *(long*)(BuffPtr + OutOffs) = Pixel;
 
@@ -279,8 +279,8 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    long PxLow  = CpuMem.ReadInt64Unchecked(Position + Offset + 0);
-                    long PxHigh = CpuMem.ReadInt64Unchecked(Position + Offset + 8);
+                    long PxLow  = CpuMem.ReadInt64(Position + Offset + 0);
+                    long PxHigh = CpuMem.ReadInt64(Position + Offset + 8);
 
                     *(long*)(BuffPtr + OutOffs + 0) = PxLow;
                     *(long*)(BuffPtr + OutOffs + 8) = PxHigh;
@@ -314,7 +314,7 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    long Tile = CpuMem.ReadInt64Unchecked(Position + Offset);
+                    long Tile = CpuMem.ReadInt64(Position + Offset);
 
                     *(long*)(BuffPtr + OutOffs) = Tile;
 
@@ -347,8 +347,8 @@ namespace Ryujinx.HLE.Gpu.Texture
                 {
                     long Offset = (uint)Swizzle.GetSwizzleOffset(X, Y);
 
-                    long Tile0 = CpuMem.ReadInt64Unchecked(Position + Offset + 0);
-                    long Tile1 = CpuMem.ReadInt64Unchecked(Position + Offset + 8);
+                    long Tile0 = CpuMem.ReadInt64(Position + Offset + 0);
+                    long Tile1 = CpuMem.ReadInt64(Position + Offset + 8);
 
                     *(long*)(BuffPtr + OutOffs + 0) = Tile0;
                     *(long*)(BuffPtr + OutOffs + 8) = Tile1;
diff --git a/Ryujinx.HLE/Gpu/Texture/TextureWriter.cs b/Ryujinx.HLE/Gpu/Texture/TextureWriter.cs
index a87d4545..113dc6f6 100644
--- a/Ryujinx.HLE/Gpu/Texture/TextureWriter.cs
+++ b/Ryujinx.HLE/Gpu/Texture/TextureWriter.cs
@@ -25,7 +25,7 @@ namespace Ryujinx.HLE.Gpu.Texture
 
                     int Pixel = *(int*)(BuffPtr + InOffs);
 
-                    CpuMem.WriteInt32Unchecked(Position + Offset, Pixel);
+                    CpuMem.WriteInt32(Position + Offset, Pixel);
 
                     InOffs += 4;
                 }