From cee712105850ac3385cd0091a923438167433f9f Mon Sep 17 00:00:00 2001 From: TSR Berry <20988865+TSRBerry@users.noreply.github.com> Date: Sat, 8 Apr 2023 01:22:00 +0200 Subject: Move solution and projects to src --- .../Engine/InlineToMemory/InlineToMemoryClass.cs | 273 +++++++++++++++++++++ 1 file changed, 273 insertions(+) create mode 100644 src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClass.cs (limited to 'src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClass.cs') diff --git a/src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClass.cs b/src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClass.cs new file mode 100644 index 00000000..e1d7e940 --- /dev/null +++ b/src/Ryujinx.Graphics.Gpu/Engine/InlineToMemory/InlineToMemoryClass.cs @@ -0,0 +1,273 @@ +using Ryujinx.Common; +using Ryujinx.Graphics.Device; +using Ryujinx.Graphics.Texture; +using System; +using System.Collections.Generic; +using System.Runtime.InteropServices; +using System.Runtime.Intrinsics; + +namespace Ryujinx.Graphics.Gpu.Engine.InlineToMemory +{ + /// + /// Represents a Inline-to-Memory engine class. + /// + class InlineToMemoryClass : IDeviceState + { + private readonly GpuContext _context; + private readonly GpuChannel _channel; + private readonly DeviceState _state; + + private bool _isLinear; + + private int _offset; + private int _size; + + private ulong _dstGpuVa; + private int _dstX; + private int _dstY; + private int _dstWidth; + private int _dstHeight; + private int _dstStride; + private int _dstGobBlocksInY; + private int _dstGobBlocksInZ; + private int _lineLengthIn; + private int _lineCount; + + private bool _finished; + + private int[] _buffer; + + /// + /// Creates a new instance of the Inline-to-Memory engine class. + /// + /// GPU context + /// GPU channel + /// Indicates if the internal state should be initialized. Set to false if part of another engine + public InlineToMemoryClass(GpuContext context, GpuChannel channel, bool initializeState) + { + _context = context; + _channel = channel; + + if (initializeState) + { + _state = new DeviceState(new Dictionary + { + { nameof(InlineToMemoryClassState.LaunchDma), new RwCallback(LaunchDma, null) }, + { nameof(InlineToMemoryClassState.LoadInlineData), new RwCallback(LoadInlineData, null) } + }); + } + } + + /// + /// Creates a new instance of the inline-to-memory engine class. + /// + /// GPU context + /// GPU channel + public InlineToMemoryClass(GpuContext context, GpuChannel channel) : this(context, channel, true) + { + } + + /// + /// Reads data from the class registers. + /// + /// Register byte offset + /// Data at the specified offset + public int Read(int offset) => _state.Read(offset); + + /// + /// Writes data to the class registers. + /// + /// Register byte offset + /// Data to be written + public void Write(int offset, int data) => _state.Write(offset, data); + + /// + /// Launches Inline-to-Memory engine DMA copy. + /// + /// Method call argument + private void LaunchDma(int argument) + { + LaunchDma(ref _state.State, argument); + } + + /// + /// Launches Inline-to-Memory engine DMA copy. + /// + /// Current class state + /// Method call argument + public void LaunchDma(ref InlineToMemoryClassState state, int argument) + { + _isLinear = (argument & 1) != 0; + + _offset = 0; + _size = (int)(BitUtils.AlignUp(state.LineLengthIn, 4) * state.LineCount); + + int count = _size / 4; + + if (_buffer == null || _buffer.Length < count) + { + _buffer = new int[count]; + } + + ulong dstGpuVa = ((ulong)state.OffsetOutUpperValue << 32) | state.OffsetOut; + + _dstGpuVa = dstGpuVa; + _dstX = state.SetDstOriginBytesXV; + _dstY = state.SetDstOriginSamplesYV; + _dstWidth = (int)state.SetDstWidth; + _dstHeight = (int)state.SetDstHeight; + _dstStride = (int)state.PitchOut; + _dstGobBlocksInY = 1 << (int)state.SetDstBlockSizeHeight; + _dstGobBlocksInZ = 1 << (int)state.SetDstBlockSizeDepth; + _lineLengthIn = (int)state.LineLengthIn; + _lineCount = (int)state.LineCount; + + _finished = false; + } + + /// + /// Pushes a block of data to the Inline-to-Memory engine. + /// + /// Data to push + public void LoadInlineData(ReadOnlySpan data) + { + if (!_finished) + { + int copySize = Math.Min(data.Length, _buffer.Length - _offset); + data.Slice(0, copySize).CopyTo(new Span(_buffer).Slice(_offset, copySize)); + + _offset += copySize; + + if (_offset * 4 >= _size) + { + FinishTransfer(); + } + } + } + + /// + /// Pushes a word of data to the Inline-to-Memory engine. + /// + /// Method call argument + public void LoadInlineData(int argument) + { + if (!_finished) + { + _buffer[_offset++] = argument; + + if (_offset * 4 >= _size) + { + FinishTransfer(); + } + } + } + + /// + /// Performs actual copy of the inline data after the transfer is finished. + /// + private void FinishTransfer() + { + var memoryManager = _channel.MemoryManager; + + var data = MemoryMarshal.Cast(_buffer).Slice(0, _size); + + if (_isLinear && _lineCount == 1) + { + memoryManager.WriteTrackedResource(_dstGpuVa, data.Slice(0, _lineLengthIn)); + _context.AdvanceSequence(); + } + else + { + // TODO: Verify if the destination X/Y and width/height are taken into account + // for linear texture transfers. If not, we can use the fast path for that aswell. + // Right now the copy code at the bottom assumes that it is used on both which might be incorrect. + if (!_isLinear) + { + var target = memoryManager.Physical.TextureCache.FindTexture( + memoryManager, + _dstGpuVa, + 1, + _dstStride, + _dstHeight, + _lineLengthIn, + _lineCount, + _isLinear, + _dstGobBlocksInY, + _dstGobBlocksInZ); + + if (target != null) + { + target.SynchronizeMemory(); + target.SetData(data, 0, 0, new GAL.Rectangle(_dstX, _dstY, _lineLengthIn / target.Info.FormatInfo.BytesPerPixel, _lineCount)); + target.SignalModified(); + + return; + } + } + + var dstCalculator = new OffsetCalculator( + _dstWidth, + _dstHeight, + _dstStride, + _isLinear, + _dstGobBlocksInY, + 1); + + int srcOffset = 0; + + for (int y = _dstY; y < _dstY + _lineCount; y++) + { + int x1 = _dstX; + int x2 = _dstX + _lineLengthIn; + int x1Round = BitUtils.AlignUp(_dstX, 16); + int x2Trunc = BitUtils.AlignDown(x2, 16); + + int x = x1; + + if (x1Round <= x2) + { + for (; x < x1Round; x++, srcOffset++) + { + int dstOffset = dstCalculator.GetOffset(x, y); + + ulong dstAddress = _dstGpuVa + (uint)dstOffset; + + memoryManager.Write(dstAddress, data[srcOffset]); + } + } + + for (; x < x2Trunc; x += 16, srcOffset += 16) + { + int dstOffset = dstCalculator.GetOffset(x, y); + + ulong dstAddress = _dstGpuVa + (uint)dstOffset; + + memoryManager.Write(dstAddress, MemoryMarshal.Cast>(data.Slice(srcOffset, 16))[0]); + } + + for (; x < x2; x++, srcOffset++) + { + int dstOffset = dstCalculator.GetOffset(x, y); + + ulong dstAddress = _dstGpuVa + (uint)dstOffset; + + memoryManager.Write(dstAddress, data[srcOffset]); + } + + // All lines must be aligned to 4 bytes, as the data is pushed one word at a time. + // If our copy length is not a multiple of 4, then we need to skip the padding bytes here. + int misalignment = _lineLengthIn & 3; + + if (misalignment != 0) + { + srcOffset += 4 - misalignment; + } + } + + _context.AdvanceSequence(); + } + + _finished = true; + } + } +} -- cgit v1.2.3