New NVDEC and VIC implementation (#1384)

* Initial NVDEC and VIC implementation

* Update FFmpeg.AutoGen to 4.3.0

* Add nvdec dependencies for Windows

* Unify some VP9 structures

* Rename VP9 structure fields

* Improvements to Video API

* XML docs for Common.Memory

* Remove now unused or redundant overloads from MemoryAccessor

* NVDEC UV surface read/write scalar paths

* Add FIXME comments about hacky things/stuff that will need to be fixed in the future

* Cleaned up VP9 memory allocation

* Remove some debug logs

* Rename some VP9 structs

* Remove unused struct

* No need to compile Ryujinx.Graphics.Host1x with unsafe anymore

* Name AsyncWorkQueue threads to make debugging easier

* Make Vp9PictureInfo a ref struct

* LayoutConverter no longer needs the depth argument (broken by rebase)

* Pooling of VP9 buffers, plus fix a memory leak on VP9

* Really wish VS could rename projects properly...

* Address feedback

* Remove using

* Catch OperationCanceledException

* Add licensing informations

* Add THIRDPARTY.md to release too

Co-authored-by: Thog <me@thog.eu>

8 files changed, 114 insertions, 29 deletions

diff --git a/Ryujinx.HLE/HOS/Horizon.cs b/Ryujinx.HLE/HOS/Horizon.cs
index b3af3290..f302e98a 100644
--- a/Ryujinx.HLE/HOS/Horizon.cs
+++ b/Ryujinx.HLE/HOS/Horizon.cs
@@ -293,8 +293,6 @@ namespace Ryujinx.HLE.HOS
                 KernelContext.ThreadCounter.Wait();
 
                 KernelContext.Dispose();
-
-                Device.Unload();
             }
         }
     }
diff --git a/Ryujinx.HLE/HOS/Services/Nv/INvDrvServices.cs b/Ryujinx.HLE/HOS/Services/Nv/INvDrvServices.cs
index ce7314f4..d6cc85e9 100644
--- a/Ryujinx.HLE/HOS/Services/Nv/INvDrvServices.cs
+++ b/Ryujinx.HLE/HOS/Services/Nv/INvDrvServices.cs
@@ -60,6 +60,8 @@ namespace Ryujinx.HLE.HOS.Services.Nv
 
                     NvDeviceFile deviceFile = (NvDeviceFile)constructor.Invoke(new object[] { context });
 
+                    deviceFile.Path = path;
+
                     return _deviceFileIdRegistry.Add(deviceFile);
                 }
                 else
diff --git a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvDeviceFile.cs b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvDeviceFile.cs
index fe3ae652..e426945d 100644
--- a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvDeviceFile.cs
+++ b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvDeviceFile.cs
@@ -14,6 +14,8 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices
         public readonly ServiceCtx Context;
         public readonly KProcess   Owner;
 
+        public string Path;
+
         public NvDeviceFile(ServiceCtx context)
         {
             Context = context;
diff --git a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/NvHostChannelDeviceFile.cs b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/NvHostChannelDeviceFile.cs
index 208bec3b..70c9a47b 100644
--- a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/NvHostChannelDeviceFile.cs
+++ b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/NvHostChannelDeviceFile.cs
@@ -1,10 +1,10 @@
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.Gpu.Memory;
-using Ryujinx.HLE.HOS.Services.Nv.Types;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostAsGpu;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel.Types;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostCtrl;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvMap;
+using Ryujinx.HLE.HOS.Services.Nv.Types;
 using System;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
@@ -130,28 +130,56 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
 
         private NvInternalResult Submit(Span<byte> arguments)
         {
-            int                 headerSize           = Unsafe.SizeOf<SubmitArguments>();
-            SubmitArguments     submitHeader         = MemoryMarshal.Cast<byte, SubmitArguments>(arguments)[0];
-            Span<CommandBuffer> commandBufferEntries = MemoryMarshal.Cast<byte, CommandBuffer>(arguments.Slice(headerSize)).Slice(0, submitHeader.CmdBufsCount);
-            MemoryManager       gmm                  = NvHostAsGpuDeviceFile.GetAddressSpaceContext(Context).Gmm;
-
-            foreach (CommandBuffer commandBufferEntry in commandBufferEntries)
+            SubmitArguments     submitHeader   = GetSpanAndSkip<SubmitArguments>(ref arguments, 1)[0];
+            Span<CommandBuffer> commandBuffers = GetSpanAndSkip<CommandBuffer>(ref arguments, submitHeader.CmdBufsCount);
+            Span<Reloc>         relocs         = GetSpanAndSkip<Reloc>(ref arguments, submitHeader.RelocsCount);
+            Span<uint>          relocShifts    = GetSpanAndSkip<uint>(ref arguments, submitHeader.RelocsCount);
+            Span<SyncptIncr>    syncptIncrs    = GetSpanAndSkip<SyncptIncr>(ref arguments, submitHeader.SyncptIncrsCount);
+            Span<SyncptIncr>    waitChecks     = GetSpanAndSkip<SyncptIncr>(ref arguments, submitHeader.SyncptIncrsCount); // ?
+            Span<Fence>         fences         = GetSpanAndSkip<Fence>(ref arguments, submitHeader.FencesCount);
+
+            lock (_device)
             {
-                NvMapHandle map = NvMapDeviceFile.GetMapFromHandle(Owner, commandBufferEntry.MemoryId);
+                for (int i = 0; i < syncptIncrs.Length; i++)
+                {
+                    SyncptIncr syncptIncr = syncptIncrs[i];
 
-                int[] commandBufferData = new int[commandBufferEntry.WordsCount];
+                    uint id = syncptIncr.Id;
 
-                for (int offset = 0; offset < commandBufferData.Length; offset++)
-                {
-                    commandBufferData[offset] = _memory.Read<int>((ulong)(map.Address + commandBufferEntry.Offset + offset * 4));
+                    fences[i].Id = id;
+                    fences[i].Thresh = Context.Device.System.HostSyncpoint.IncrementSyncpointMax(id, syncptIncr.Incrs);
                 }
 
-                // TODO: Submit command to engines.
+                foreach (CommandBuffer commandBuffer in commandBuffers)
+                {
+                    NvMapHandle map = NvMapDeviceFile.GetMapFromHandle(Owner, commandBuffer.Mem);
+
+                    var data = _memory.GetSpan((ulong)map.Address + commandBuffer.Offset, commandBuffer.WordsCount * 4);
+
+                    _device.Host1x.Submit(MemoryMarshal.Cast<byte, int>(data));
+                }
             }
 
+            fences[0].Thresh = Context.Device.System.HostSyncpoint.IncrementSyncpointMax(fences[0].Id, 1);
+
+            Span<int> tmpCmdBuff = stackalloc int[1];
+
+            tmpCmdBuff[0] = (4 << 28) | (int)fences[0].Id;
+
+            _device.Host1x.Submit(tmpCmdBuff);
+
             return NvInternalResult.Success;
         }
 
+        private Span<T> GetSpanAndSkip<T>(ref Span<byte> arguments, int count) where T : unmanaged
+        {
+            Span<T> output = MemoryMarshal.Cast<byte, T>(arguments).Slice(0, count);
+
+            arguments = arguments.Slice(Unsafe.SizeOf<T>() * count);
+
+            return output;
+        }
+
         private NvInternalResult GetSyncpoint(ref GetParameterArguments arguments)
         {
             if (arguments.Parameter >= MaxModuleSyncpoint)
@@ -248,9 +276,13 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel
                 {
                     if (map.DmaMapAddress != 0)
                     {
-                        gmm.Free((ulong)map.DmaMapAddress, (uint)map.Size);
+                        // FIXME:
+                        // To make unmapping work, we need separate address space per channel.
+                        // Right now NVDEC and VIC share the GPU address space which is not correct at all.
+
+                        // gmm.Free((ulong)map.DmaMapAddress, (uint)map.Size);
 
-                        map.DmaMapAddress = 0;
+                        // map.DmaMapAddress = 0;
                     }
                 }
             }
diff --git a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/Types/SubmitArguments.cs b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/Types/SubmitArguments.cs
index bb2fd1cc..7ef7e39e 100644
--- a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/Types/SubmitArguments.cs
+++ b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostChannel/Types/SubmitArguments.cs
@@ -5,17 +5,40 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel.Types
     [StructLayout(LayoutKind.Sequential)]
     struct CommandBuffer
     {
-        public int MemoryId;
-        public int Offset;
+        public int Mem;
+        public uint Offset;
         public int WordsCount;
     }
 
     [StructLayout(LayoutKind.Sequential)]
+    struct Reloc
+    {
+        public int CmdbufMem;
+        public int CmdbufOffset;
+        public int Target;
+        public int TargetOffset;
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    struct SyncptIncr
+    {
+        public uint Id;
+        public uint Incrs;
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
+    struct Fence
+    {
+        public uint Id;
+        public uint Thresh;
+    }
+
+    [StructLayout(LayoutKind.Sequential)]
     struct SubmitArguments
     {
         public int CmdBufsCount;
         public int RelocsCount;
         public int SyncptIncrsCount;
-        public int WaitchecksCount;
+        public int FencesCount;
     }
 }
diff --git a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostCtrl/Types/NvHostSyncPt.cs b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostCtrl/Types/NvHostSyncPt.cs
index ff56fbf5..aa730b57 100644
--- a/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostCtrl/Types/NvHostSyncPt.cs
+++ b/Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostCtrl/Types/NvHostSyncPt.cs
@@ -1,6 +1,5 @@
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.Gpu.Synchronization;
-using Ryujinx.HLE.HOS.Kernel.Threading;
 using System;
 using System.Threading;
 
@@ -172,6 +171,11 @@ namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostCtrl
             return (uint)Interlocked.Increment(ref _counterMax[id]);
         }
 
+        public uint IncrementSyncpointMax(uint id, uint incrs)
+        {
+            return (uint)Interlocked.Add(ref _counterMax[id], (int)incrs);
+        }
+
         public bool IsSyncpointExpired(uint id, uint threshold)
         {
             return MinCompare(id, _counterMin[id], _counterMax[id], (int)threshold);
diff --git a/Ryujinx.HLE/Ryujinx.HLE.csproj b/Ryujinx.HLE/Ryujinx.HLE.csproj
index b05a9087..01e20792 100644
--- a/Ryujinx.HLE/Ryujinx.HLE.csproj
+++ b/Ryujinx.HLE/Ryujinx.HLE.csproj
@@ -47,6 +47,9 @@
     <ProjectReference Include="..\Ryujinx.Common\Ryujinx.Common.csproj" />
     <ProjectReference Include="..\Ryujinx.Cpu\Ryujinx.Cpu.csproj" />
     <ProjectReference Include="..\Ryujinx.Debugger\Ryujinx.Debugger.csproj" />
+    <ProjectReference Include="..\Ryujinx.Graphics.Host1x\Ryujinx.Graphics.Host1x.csproj" />
+    <ProjectReference Include="..\Ryujinx.Graphics.Nvdec\Ryujinx.Graphics.Nvdec.csproj" />
+    <ProjectReference Include="..\Ryujinx.Graphics.Vic\Ryujinx.Graphics.Vic.csproj" />
     <ProjectReference Include="..\Ryujinx.Memory\Ryujinx.Memory.csproj" />
     <ProjectReference Include="..\ARMeilleure\ARMeilleure.csproj" />
     <ProjectReference Include="..\Ryujinx.Graphics.Gpu\Ryujinx.Graphics.Gpu.csproj" />
diff --git a/Ryujinx.HLE/Switch.cs b/Ryujinx.HLE/Switch.cs
index 5713bd9e..9defe25d 100644
--- a/Ryujinx.HLE/Switch.cs
+++ b/Ryujinx.HLE/Switch.cs
@@ -3,6 +3,9 @@ using Ryujinx.Audio;
 using Ryujinx.Configuration;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu;
+using Ryujinx.Graphics.Host1x;
+using Ryujinx.Graphics.Nvdec;
+using Ryujinx.Graphics.Vic;
 using Ryujinx.HLE.FileSystem;
 using Ryujinx.HLE.FileSystem.Content;
 using Ryujinx.HLE.HOS;
@@ -11,7 +14,6 @@ using Ryujinx.HLE.HOS.Services.Hid;
 using Ryujinx.HLE.HOS.SystemState;
 using Ryujinx.Memory;
 using System;
-using System.Threading;
 
 namespace Ryujinx.HLE
 {
@@ -23,6 +25,8 @@ namespace Ryujinx.HLE
 
         public GpuContext Gpu { get; private set; }
 
+        internal Host1xDevice Host1x { get; }
+
         public VirtualFileSystem FileSystem { get; private set; }
 
         public Horizon System { get; private set; }
@@ -53,6 +57,27 @@ namespace Ryujinx.HLE
 
             Gpu = new GpuContext(renderer);
 
+            Host1x = new Host1xDevice(Gpu.Synchronization);
+            var nvdec = new NvdecDevice(Gpu.MemoryManager);
+            var vic = new VicDevice(Gpu.MemoryManager);
+            Host1x.RegisterDevice(ClassId.Nvdec, nvdec);
+            Host1x.RegisterDevice(ClassId.Vic, vic);
+
+            nvdec.FrameDecoded += (FrameDecodedEventArgs e) =>
+            {
+                // FIXME:
+                // Figure out what is causing frame ordering issues on H264.
+                // For now this is needed as workaround.
+                if (e.CodecId == CodecId.H264)
+                {
+                    vic.SetSurfaceOverride(e.LumaOffset, e.ChromaOffset, 0);
+                }
+                else
+                {
+                    vic.DisableSurfaceOverride();
+                }
+            };
+
             FileSystem = fileSystem;
 
             System = new Horizon(this, contentManager);
@@ -136,13 +161,6 @@ namespace Ryujinx.HLE
             Gpu.Window.Present(swapBuffersCallback);
         }
 
-        internal void Unload()
-        {
-            FileSystem.Unload();
-
-            Memory.Dispose();
-        }
-
         public void DisposeGpu()
         {
             Gpu.Dispose();
@@ -158,7 +176,10 @@ namespace Ryujinx.HLE
             if (disposing)
             {
                 System.Dispose();
+                Host1x.Dispose();
                 AudioOut.Dispose();
+                FileSystem.Unload();
+                Memory.Dispose();
             }
         }
     }