R/Ryujinx.Graphics/Graphics3d/NvGpuEngine3d.cs

1177 lines
44 KiB
C#
Raw Normal View History

using Ryujinx.Common;
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 01:12:24 +00:00
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.Gal;
using Ryujinx.Graphics.Memory;
using Ryujinx.Graphics.Texture;
using System;
using System.Collections.Generic;
namespace Ryujinx.Graphics.Graphics3d
{
class NvGpuEngine3d : INvGpuEngine
{
public int[] Registers { get; private set; }
private NvGpu Gpu;
private Dictionary<int, NvGpuMethod> Methods;
private struct ConstBuffer
{
public bool Enabled;
public long Position;
public int Size;
}
private ConstBuffer[][] ConstBuffers;
// Viewport dimensions kept for scissor test limits
private int ViewportX0 = 0;
private int ViewportY0 = 0;
private int ViewportX1 = 0;
private int ViewportY1 = 0;
private int ViewportWidth = 0;
private int ViewportHeight = 0;
2018-08-25 04:16:58 +00:00
private int CurrentInstance = 0;
public NvGpuEngine3d(NvGpu Gpu)
{
this.Gpu = Gpu;
Registers = new int[0xe00];
Methods = new Dictionary<int, NvGpuMethod>();
void AddMethod(int Meth, int Count, int Stride, NvGpuMethod Method)
{
while (Count-- > 0)
{
Methods.Add(Meth, Method);
Meth += Stride;
}
}
AddMethod(0x585, 1, 1, VertexEndGl);
AddMethod(0x674, 1, 1, ClearBuffers);
AddMethod(0x6c3, 1, 1, QueryControl);
AddMethod(0x8e4, 16, 1, CbData);
AddMethod(0x904, 5, 8, CbBind);
ConstBuffers = new ConstBuffer[6][];
for (int Index = 0; Index < ConstBuffers.Length; Index++)
{
ConstBuffers[Index] = new ConstBuffer[18];
}
//Ensure that all components are enabled by default.
//FIXME: Is this correct?
WriteRegister(NvGpuEngine3dReg.ColorMaskN, 0x1111);
WriteRegister(NvGpuEngine3dReg.FrameBufferSrgb, 1);
WriteRegister(NvGpuEngine3dReg.FrontFace, (int)GalFrontFace.CW);
for (int Index = 0; Index < GalPipelineState.RenderTargetsCount; Index++)
{
WriteRegister(NvGpuEngine3dReg.IBlendNEquationRgb + Index * 8, (int)GalBlendEquation.FuncAdd);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncSrcRgb + Index * 8, (int)GalBlendFactor.One);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncDstRgb + Index * 8, (int)GalBlendFactor.Zero);
WriteRegister(NvGpuEngine3dReg.IBlendNEquationAlpha + Index * 8, (int)GalBlendEquation.FuncAdd);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncSrcAlpha + Index * 8, (int)GalBlendFactor.One);
WriteRegister(NvGpuEngine3dReg.IBlendNFuncDstAlpha + Index * 8, (int)GalBlendFactor.Zero);
}
}
public void CallMethod(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
if (Methods.TryGetValue(MethCall.Method, out NvGpuMethod Method))
{
Method(Vmm, MethCall);
}
else
{
WriteRegister(MethCall);
}
}
private void VertexEndGl(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
LockCaches();
GalPipelineState State = new GalPipelineState();
// Framebuffer must be run configured because viewport dimensions may be used in other methods
SetFrameBuffer(State);
for (int FbIndex = 0; FbIndex < 8; FbIndex++)
{
SetFrameBuffer(Vmm, FbIndex);
}
SetFrontFace(State);
SetCullFace(State);
SetDepth(State);
SetStencil(State);
SetScissor(State);
SetBlending(State);
SetColorMask(State);
SetPrimitiveRestart(State);
SetZeta(Vmm);
SetRenderTargets();
long[] Keys = UploadShaders(Vmm);
Gpu.Renderer.Shader.BindProgram();
UploadTextures(Vmm, State, Keys);
UploadConstBuffers(Vmm, State, Keys);
UploadVertexArrays(Vmm, State);
DispatchRender(Vmm, State);
UnlockCaches();
}
private void LockCaches()
{
Gpu.Renderer.Buffer.LockCache();
Gpu.Renderer.Rasterizer.LockCaches();
Gpu.Renderer.Texture.LockCache();
}
private void UnlockCaches()
{
Gpu.Renderer.Buffer.UnlockCache();
Gpu.Renderer.Rasterizer.UnlockCaches();
Gpu.Renderer.Texture.UnlockCache();
}
private void ClearBuffers(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
int Attachment = (MethCall.Argument >> 6) & 0xf;
GalClearBufferFlags Flags = (GalClearBufferFlags)(MethCall.Argument & 0x3f);
float Red = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 0);
float Green = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 1);
float Blue = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 2);
float Alpha = ReadRegisterFloat(NvGpuEngine3dReg.ClearNColor + 3);
float Depth = ReadRegisterFloat(NvGpuEngine3dReg.ClearDepth);
int Stencil = ReadRegister(NvGpuEngine3dReg.ClearStencil);
SetFrameBuffer(Vmm, Attachment);
SetZeta(Vmm);
SetRenderTargets();
Gpu.Renderer.RenderTarget.Bind();
Gpu.Renderer.Rasterizer.ClearBuffers(Flags, Attachment, Red, Green, Blue, Alpha, Depth, Stencil);
Gpu.Renderer.Pipeline.ResetDepthMask();
Gpu.Renderer.Pipeline.ResetColorMask(Attachment);
}
private void SetFrameBuffer(NvGpuVmm Vmm, int FbIndex)
{
long VA = MakeInt64From2xInt32(NvGpuEngine3dReg.FrameBufferNAddress + FbIndex * 0x10);
int SurfFormat = ReadRegister(NvGpuEngine3dReg.FrameBufferNFormat + FbIndex * 0x10);
if (VA == 0 || SurfFormat == 0)
{
Gpu.Renderer.RenderTarget.UnbindColor(FbIndex);
return;
}
long Key = Vmm.GetPhysicalAddress(VA);
int Width = ReadRegister(NvGpuEngine3dReg.FrameBufferNWidth + FbIndex * 0x10);
int Height = ReadRegister(NvGpuEngine3dReg.FrameBufferNHeight + FbIndex * 0x10);
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 01:12:24 +00:00
int ArrayMode = ReadRegister(NvGpuEngine3dReg.FrameBufferNArrayMode + FbIndex * 0x10);
int LayerCount = ArrayMode & 0xFFFF;
int LayerStride = ReadRegister(NvGpuEngine3dReg.FrameBufferNLayerStride + FbIndex * 0x10);
int BaseLayer = ReadRegister(NvGpuEngine3dReg.FrameBufferNBaseLayer + FbIndex * 0x10);
int BlockDim = ReadRegister(NvGpuEngine3dReg.FrameBufferNBlockDim + FbIndex * 0x10);
int GobBlockHeight = 1 << ((BlockDim >> 4) & 7);
GalMemoryLayout Layout = (GalMemoryLayout)((BlockDim >> 12) & 1);
float TX = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNTranslateX + FbIndex * 8);
float TY = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNTranslateY + FbIndex * 8);
float SX = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNScaleX + FbIndex * 8);
float SY = ReadRegisterFloat(NvGpuEngine3dReg.ViewportNScaleY + FbIndex * 8);
ViewportX0 = (int)MathF.Max(0, TX - MathF.Abs(SX));
ViewportY0 = (int)MathF.Max(0, TY - MathF.Abs(SY));
ViewportX1 = (int)(TX + MathF.Abs(SX));
ViewportY1 = (int)(TY + MathF.Abs(SY));
GalImageFormat Format = ImageUtils.ConvertSurface((GalSurfaceFormat)SurfFormat);
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 01:12:24 +00:00
GalImage Image = new GalImage(Width, Height, 1, 1, 1, GobBlockHeight, 1, Layout, Format, GalTextureTarget.TwoD);
Gpu.ResourceManager.SendColorBuffer(Vmm, Key, FbIndex, Image);
Gpu.Renderer.RenderTarget.SetViewport(FbIndex, ViewportX0, ViewportY0, ViewportX1 - ViewportX0, ViewportY1 - ViewportY0);
}
private void SetFrameBuffer(GalPipelineState State)
{
State.FramebufferSrgb = ReadRegisterBool(NvGpuEngine3dReg.FrameBufferSrgb);
State.FlipX = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleX);
State.FlipY = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleY);
int ScreenYControl = ReadRegister(NvGpuEngine3dReg.ScreenYControl);
bool NegateY = (ScreenYControl & 1) != 0;
if (NegateY)
{
State.FlipY = -State.FlipY;
}
}
private void SetZeta(NvGpuVmm Vmm)
{
long VA = MakeInt64From2xInt32(NvGpuEngine3dReg.ZetaAddress);
int ZetaFormat = ReadRegister(NvGpuEngine3dReg.ZetaFormat);
int BlockDim = ReadRegister(NvGpuEngine3dReg.ZetaBlockDimensions);
int GobBlockHeight = 1 << ((BlockDim >> 4) & 7);
GalMemoryLayout Layout = (GalMemoryLayout)((BlockDim >> 12) & 1); //?
bool ZetaEnable = ReadRegisterBool(NvGpuEngine3dReg.ZetaEnable);
if (VA == 0 || ZetaFormat == 0 || !ZetaEnable)
{
Gpu.Renderer.RenderTarget.UnbindZeta();
return;
}
long Key = Vmm.GetPhysicalAddress(VA);
int Width = ReadRegister(NvGpuEngine3dReg.ZetaHoriz);
int Height = ReadRegister(NvGpuEngine3dReg.ZetaVert);
GalImageFormat Format = ImageUtils.ConvertZeta((GalZetaFormat)ZetaFormat);
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 01:12:24 +00:00
// TODO: Support non 2D?
GalImage Image = new GalImage(Width, Height, 1, 1, 1, GobBlockHeight, 1, Layout, Format, GalTextureTarget.TwoD);
Gpu.ResourceManager.SendZetaBuffer(Vmm, Key, Image);
}
private long[] UploadShaders(NvGpuVmm Vmm)
{
long[] Keys = new long[5];
long BasePosition = MakeInt64From2xInt32(NvGpuEngine3dReg.ShaderAddress);
int Index = 1;
int VpAControl = ReadRegister(NvGpuEngine3dReg.ShaderNControl);
bool VpAEnable = (VpAControl & 1) != 0;
if (VpAEnable)
{
//Note: The maxwell supports 2 vertex programs, usually
//only VP B is used, but in some cases VP A is also used.
//In this case, it seems to function as an extra vertex
//shader stage.
//The graphics abstraction layer has a special overload for this
//case, which should merge the two shaders into one vertex shader.
int VpAOffset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset);
int VpBOffset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset + 0x10);
long VpAPos = BasePosition + (uint)VpAOffset;
long VpBPos = BasePosition + (uint)VpBOffset;
Keys[(int)GalShaderType.Vertex] = VpBPos;
Gpu.Renderer.Shader.Create(Vmm, VpAPos, VpBPos, GalShaderType.Vertex);
Gpu.Renderer.Shader.Bind(VpBPos);
Index = 2;
}
for (; Index < 6; Index++)
{
GalShaderType Type = GetTypeFromProgram(Index);
int Control = ReadRegister(NvGpuEngine3dReg.ShaderNControl + Index * 0x10);
int Offset = ReadRegister(NvGpuEngine3dReg.ShaderNOffset + Index * 0x10);
//Note: Vertex Program (B) is always enabled.
bool Enable = (Control & 1) != 0 || Index == 1;
if (!Enable)
{
Gpu.Renderer.Shader.Unbind(Type);
continue;
}
long Key = BasePosition + (uint)Offset;
Keys[(int)Type] = Key;
Gpu.Renderer.Shader.Create(Vmm, Key, Type);
Gpu.Renderer.Shader.Bind(Key);
}
return Keys;
}
private static GalShaderType GetTypeFromProgram(int Program)
{
switch (Program)
{
case 0:
case 1: return GalShaderType.Vertex;
case 2: return GalShaderType.TessControl;
case 3: return GalShaderType.TessEvaluation;
case 4: return GalShaderType.Geometry;
case 5: return GalShaderType.Fragment;
}
throw new ArgumentOutOfRangeException(nameof(Program));
}
private void SetFrontFace(GalPipelineState State)
{
float SignX = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleX);
float SignY = GetFlipSign(NvGpuEngine3dReg.ViewportNScaleY);
GalFrontFace FrontFace = (GalFrontFace)ReadRegister(NvGpuEngine3dReg.FrontFace);
//Flipping breaks facing. Flipping front facing too fixes it
if (SignX != SignY)
{
switch (FrontFace)
{
case GalFrontFace.CW: FrontFace = GalFrontFace.CCW; break;
case GalFrontFace.CCW: FrontFace = GalFrontFace.CW; break;
}
}
State.FrontFace = FrontFace;
}
private void SetCullFace(GalPipelineState State)
{
State.CullFaceEnabled = ReadRegisterBool(NvGpuEngine3dReg.CullFaceEnable);
if (State.CullFaceEnabled)
{
State.CullFace = (GalCullFace)ReadRegister(NvGpuEngine3dReg.CullFace);
}
}
private void SetDepth(GalPipelineState State)
{
State.DepthTestEnabled = ReadRegisterBool(NvGpuEngine3dReg.DepthTestEnable);
State.DepthWriteEnabled = ReadRegisterBool(NvGpuEngine3dReg.DepthWriteEnable);
if (State.DepthTestEnabled)
{
State.DepthFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.DepthTestFunction);
}
State.DepthRangeNear = ReadRegisterFloat(NvGpuEngine3dReg.DepthRangeNNear);
State.DepthRangeFar = ReadRegisterFloat(NvGpuEngine3dReg.DepthRangeNFar);
}
private void SetStencil(GalPipelineState State)
{
State.StencilTestEnabled = ReadRegisterBool(NvGpuEngine3dReg.StencilEnable);
if (State.StencilTestEnabled)
{
State.StencilBackFuncFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.StencilBackFuncFunc);
State.StencilBackFuncRef = ReadRegister(NvGpuEngine3dReg.StencilBackFuncRef);
State.StencilBackFuncMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilBackFuncMask);
State.StencilBackOpFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpFail);
State.StencilBackOpZFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpZFail);
State.StencilBackOpZPass = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilBackOpZPass);
State.StencilBackMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilBackMask);
State.StencilFrontFuncFunc = (GalComparisonOp)ReadRegister(NvGpuEngine3dReg.StencilFrontFuncFunc);
State.StencilFrontFuncRef = ReadRegister(NvGpuEngine3dReg.StencilFrontFuncRef);
State.StencilFrontFuncMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilFrontFuncMask);
State.StencilFrontOpFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpFail);
State.StencilFrontOpZFail = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpZFail);
State.StencilFrontOpZPass = (GalStencilOp)ReadRegister(NvGpuEngine3dReg.StencilFrontOpZPass);
State.StencilFrontMask = (uint)ReadRegister(NvGpuEngine3dReg.StencilFrontMask);
}
}
private void SetScissor(GalPipelineState State)
{
int count = 0;
for (int Index = 0; Index < GalPipelineState.RenderTargetsCount; Index++)
{
State.ScissorTestEnabled[Index] = ReadRegisterBool(NvGpuEngine3dReg.ScissorEnable + Index * 4);
if (State.ScissorTestEnabled[Index])
{
uint ScissorHorizontal = (uint)ReadRegister(NvGpuEngine3dReg.ScissorHorizontal + Index * 4);
uint ScissorVertical = (uint)ReadRegister(NvGpuEngine3dReg.ScissorVertical + Index * 4);
int left = (int)(ScissorHorizontal & 0xFFFF); // Left, lower 16 bits
int right = (int)(ScissorHorizontal >> 16); // Right, upper 16 bits
int bottom = (int)(ScissorVertical & 0xFFFF); // Bottom, lower 16 bits
int top = (int)(ScissorVertical >> 16); // Top, upper 16 bits
int width = Math.Abs(right - left);
int height = Math.Abs(top - bottom);
// If the scissor test covers the whole possible viewport, i.e. uninititalized, disable scissor test
if ((width > NvGpu.MaxViewportSize && height > NvGpu.MaxViewportSize) || width <= 0 || height <= 0)
{
State.ScissorTestEnabled[Index] = false;
continue;
}
// Keep track of how many scissor tests are active.
// If only 1, and it's the first user should apply to all viewports
count++;
// Flip X
if (State.FlipX == -1)
{
left = ViewportX1 - (left - ViewportX0);
right = ViewportX1 - (right - ViewportX0);
}
// Ensure X is in the right order
if (left > right)
{
int temp = left;
left = right;
right = temp;
}
// Flip Y
if (State.FlipY == -1)
{
bottom = ViewportY1 - (bottom - ViewportY0);
top = ViewportY1 - (top - ViewportY0);
}
// Ensure Y is in the right order
if (bottom > top)
{
int temp = top;
top = bottom;
bottom = temp;
}
// Handle out of active viewport dimensions
left = Math.Clamp(left, ViewportX0, ViewportX1);
right = Math.Clamp(right, ViewportX0, ViewportX1);
top = Math.Clamp(top, ViewportY0, ViewportY1);
bottom = Math.Clamp(bottom, ViewportY0, ViewportY1);
// Save values to state
State.ScissorTestX[Index] = left;
State.ScissorTestY[Index] = bottom;
State.ScissorTestWidth[Index] = right - left;
State.ScissorTestHeight[Index] = top - bottom;
}
}
State.ScissorTestCount = count;
}
private void SetBlending(GalPipelineState State)
{
bool BlendIndependent = ReadRegisterBool(NvGpuEngine3dReg.BlendIndependent);
State.BlendIndependent = BlendIndependent;
for (int Index = 0; Index < GalPipelineState.RenderTargetsCount; Index++)
{
if (BlendIndependent)
{
State.Blends[Index].Enabled = ReadRegisterBool(NvGpuEngine3dReg.IBlendNEnable + Index);
if (State.Blends[Index].Enabled)
{
State.Blends[Index].SeparateAlpha = ReadRegisterBool(NvGpuEngine3dReg.IBlendNSeparateAlpha + Index * 8);
State.Blends[Index].EquationRgb = ReadBlendEquation(NvGpuEngine3dReg.IBlendNEquationRgb + Index * 8);
State.Blends[Index].FuncSrcRgb = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncSrcRgb + Index * 8);
State.Blends[Index].FuncDstRgb = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncDstRgb + Index * 8);
State.Blends[Index].EquationAlpha = ReadBlendEquation(NvGpuEngine3dReg.IBlendNEquationAlpha + Index * 8);
State.Blends[Index].FuncSrcAlpha = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncSrcAlpha + Index * 8);
State.Blends[Index].FuncDstAlpha = ReadBlendFactor (NvGpuEngine3dReg.IBlendNFuncDstAlpha + Index * 8);
}
}
else
{
//It seems that even when independent blend is disabled, the first IBlend enable
//register is still set to indicate whenever blend is enabled or not (?).
State.Blends[Index].Enabled = ReadRegisterBool(NvGpuEngine3dReg.IBlendNEnable);
if (State.Blends[Index].Enabled)
{
State.Blends[Index].SeparateAlpha = ReadRegisterBool(NvGpuEngine3dReg.BlendSeparateAlpha);
State.Blends[Index].EquationRgb = ReadBlendEquation(NvGpuEngine3dReg.BlendEquationRgb);
State.Blends[Index].FuncSrcRgb = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncSrcRgb);
State.Blends[Index].FuncDstRgb = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncDstRgb);
State.Blends[Index].EquationAlpha = ReadBlendEquation(NvGpuEngine3dReg.BlendEquationAlpha);
State.Blends[Index].FuncSrcAlpha = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncSrcAlpha);
State.Blends[Index].FuncDstAlpha = ReadBlendFactor (NvGpuEngine3dReg.BlendFuncDstAlpha);
}
}
}
}
private GalBlendEquation ReadBlendEquation(NvGpuEngine3dReg Register)
{
return (GalBlendEquation)ReadRegister(Register);
}
private GalBlendFactor ReadBlendFactor(NvGpuEngine3dReg Register)
{
return (GalBlendFactor)ReadRegister(Register);
}
private void SetColorMask(GalPipelineState State)
{
bool ColorMaskCommon = ReadRegisterBool(NvGpuEngine3dReg.ColorMaskCommon);
State.ColorMaskCommon = ColorMaskCommon;
for (int Index = 0; Index < GalPipelineState.RenderTargetsCount; Index++)
{
int ColorMask = ReadRegister(NvGpuEngine3dReg.ColorMaskN + (ColorMaskCommon ? 0 : Index));
State.ColorMasks[Index].Red = ((ColorMask >> 0) & 0xf) != 0;
State.ColorMasks[Index].Green = ((ColorMask >> 4) & 0xf) != 0;
State.ColorMasks[Index].Blue = ((ColorMask >> 8) & 0xf) != 0;
State.ColorMasks[Index].Alpha = ((ColorMask >> 12) & 0xf) != 0;
}
}
private void SetPrimitiveRestart(GalPipelineState State)
2018-07-08 16:14:35 +00:00
{
State.PrimitiveRestartEnabled = ReadRegisterBool(NvGpuEngine3dReg.PrimRestartEnable);
2018-07-08 16:14:35 +00:00
if (State.PrimitiveRestartEnabled)
2018-07-08 16:14:35 +00:00
{
State.PrimitiveRestartIndex = (uint)ReadRegister(NvGpuEngine3dReg.PrimRestartIndex);
2018-07-08 16:14:35 +00:00
}
}
private void SetRenderTargets()
{
//Commercial games do not seem to
//bool SeparateFragData = ReadRegisterBool(NvGpuEngine3dReg.RTSeparateFragData);
uint Control = (uint)(ReadRegister(NvGpuEngine3dReg.RTControl));
uint Count = Control & 0xf;
if (Count > 0)
{
int[] Map = new int[Count];
for (int Index = 0; Index < Count; Index++)
{
int Shift = 4 + Index * 3;
Map[Index] = (int)((Control >> Shift) & 7);
}
Gpu.Renderer.RenderTarget.SetMap(Map);
}
else
{
Gpu.Renderer.RenderTarget.SetMap(null);
}
}
private void UploadTextures(NvGpuVmm Vmm, GalPipelineState State, long[] Keys)
{
long BaseShPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.ShaderAddress);
int TextureCbIndex = ReadRegister(NvGpuEngine3dReg.TextureCbIndex);
List<(long, GalImage, GalTextureSampler)> UnboundTextures = new List<(long, GalImage, GalTextureSampler)>();
for (int Index = 0; Index < Keys.Length; Index++)
{
foreach (ShaderDeclInfo DeclInfo in Gpu.Renderer.Shader.GetTextureUsage(Keys[Index]))
{
long Position;
if (DeclInfo.IsCb)
{
Position = ConstBuffers[Index][DeclInfo.Cbuf].Position;
}
else
{
Position = ConstBuffers[Index][TextureCbIndex].Position;
}
int TextureHandle = Vmm.ReadInt32(Position + DeclInfo.Index * 4);
UnboundTextures.Add(UploadTexture(Vmm, TextureHandle));
}
}
for (int Index = 0; Index < UnboundTextures.Count; Index++)
{
(long Key, GalImage Image, GalTextureSampler Sampler) = UnboundTextures[Index];
if (Key == 0)
{
continue;
}
Gpu.Renderer.Texture.Bind(Key, Index, Image);
Initial non 2D textures support (#525) * Initial non 2D textures support - Shaders still need to be changed - Some types aren't yet implemented * Start implementing texture instructions suffixes Fix wrong texture type with cube and TEXS Also support array textures in TEX and TEX.B Clean up TEX and TEXS coords managment Fix TEXS.LL with non-2d textures Implement TEX.AOFFI Get the right arguments for TEX, TEXS and TLDS Also, store suffix operands in appropriate values to support multiple suffix combinaisons * Support depth in read/writeTexture Also support WrapR and detect mipmap * Proper cube map textures support + fix TEXS.LZ * Implement depth compare * some code clean up * Implement CubeMap textures in OGLTexture.Create * Implement TLD4 and TLD4S * Add Texture 1D support * updates comments * fix some code style issues * Fix some nits + rename some things to be less confusing * Remove GetSuffix local functions * AOFFI => AOffI * TextureType => GalTextureTarget * finish renaming TextureType to TextureTarget * Disable LL, LZ and LB support in the decompiler This needs more work at the GL level (GLSL implementation should be right) * Revert "Disable LL, LZ and LB support in the decompiler" This reverts commit 64536c3d9f673645faff3152838d1413c3203395. * Fix TEXS ARRAY_2D index * ImageFormat depth should be 1 for all image format * Fix shader build issues with sampler1DShadow and texture * Fix DC & AOFFI combinaison with TEX/TEXS * Support AOFFI with TLD4 and TLD4S * Fix shader compilation error for TLD4.AOFFI with no DC * Fix binding isuses on the 2d copy engine TODO: support 2d array copy * Support 2D array copy operation in the 2D engine This make every copy right in the GPU side. Thie CPU copy probably needs to be updated * Implement GetGpuSize + fix somes issues with 2d engine copies TODO: mipmap level in it * Don't throw an exception in the layer handling * Fix because of rebase * Reject 2d layers of non textures in 2d copy engine * Add 3D textures and mipmap support on BlockLinearSwizzle * Fix naming on new BitUtils methods * gpu cache: Make sure to invalidate textures that doesn't have the same target * Add the concept of layer count for array instead of using depth Also cleanup GetGpuSize as Swizzle can compute the size with mipmap * Support multi layer with mip map in ReadTexture * Add more check for cache invalidation & remove cubemap and cubemap array code for now Also fix compressed 2d array * Fix texelFetchOffset shader build error * Start looking into cube map again Also add some way to log write in register in engines * fix write register log levles * Remove debug logs in WriteRegister * Disable AOFFI support on non NVIDIA drivers * Fix code align
2019-02-28 01:12:24 +00:00
Gpu.Renderer.Texture.SetSampler(Image, Sampler);
}
}
private (long, GalImage, GalTextureSampler) UploadTexture(NvGpuVmm Vmm, int TextureHandle)
{
if (TextureHandle == 0)
{
//FIXME: Some games like puyo puyo will use handles with the value 0.
//This is a bug, most likely caused by sync issues.
return (0, default(GalImage), default(GalTextureSampler));
}
bool LinkedTsc = ReadRegisterBool(NvGpuEngine3dReg.LinkedTsc);
int TicIndex = (TextureHandle >> 0) & 0xfffff;
int TscIndex = LinkedTsc ? TicIndex : (TextureHandle >> 20) & 0xfff;
long TicPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.TexHeaderPoolOffset);
long TscPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.TexSamplerPoolOffset);
TicPosition += TicIndex * 0x20;
TscPosition += TscIndex * 0x20;
GalImage Image = TextureFactory.MakeTexture(Vmm, TicPosition);
GalTextureSampler Sampler = TextureFactory.MakeSampler(Gpu, Vmm, TscPosition);
long Key = Vmm.ReadInt64(TicPosition + 4) & 0xffffffffffff;
if (Image.Layout == GalMemoryLayout.BlockLinear)
{
Key &= ~0x1ffL;
}
else if (Image.Layout == GalMemoryLayout.Pitch)
{
Key &= ~0x1fL;
}
Key = Vmm.GetPhysicalAddress(Key);
if (Key == -1)
{
//FIXME: Shouldn't ignore invalid addresses.
return (0, default(GalImage), default(GalTextureSampler));
}
Gpu.ResourceManager.SendTexture(Vmm, Key, Image);
return (Key, Image, Sampler);
}
private void UploadConstBuffers(NvGpuVmm Vmm, GalPipelineState State, long[] Keys)
{
for (int Stage = 0; Stage < Keys.Length; Stage++)
{
foreach (ShaderDeclInfo DeclInfo in Gpu.Renderer.Shader.GetConstBufferUsage(Keys[Stage]))
{
ConstBuffer Cb = ConstBuffers[Stage][DeclInfo.Cbuf];
if (!Cb.Enabled)
{
continue;
}
long Key = Vmm.GetPhysicalAddress(Cb.Position);
if (Gpu.ResourceManager.MemoryRegionModified(Vmm, Key, Cb.Size, NvGpuBufferType.ConstBuffer))
{
if (Vmm.TryGetHostAddress(Cb.Position, Cb.Size, out IntPtr CbPtr))
{
Gpu.Renderer.Buffer.SetData(Key, Cb.Size, CbPtr);
}
else
{
Gpu.Renderer.Buffer.SetData(Key, Vmm.ReadBytes(Cb.Position, Cb.Size));
}
}
State.ConstBufferKeys[Stage][DeclInfo.Cbuf] = Key;
}
}
}
private void UploadVertexArrays(NvGpuVmm Vmm, GalPipelineState State)
{
long IbPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.IndexArrayAddress);
long IboKey = Vmm.GetPhysicalAddress(IbPosition);
int IndexEntryFmt = ReadRegister(NvGpuEngine3dReg.IndexArrayFormat);
int IndexCount = ReadRegister(NvGpuEngine3dReg.IndexBatchCount);
int PrimCtrl = ReadRegister(NvGpuEngine3dReg.VertexBeginGl);
GalPrimitiveType PrimType = (GalPrimitiveType)(PrimCtrl & 0xffff);
GalIndexFormat IndexFormat = (GalIndexFormat)IndexEntryFmt;
int IndexEntrySize = 1 << IndexEntryFmt;
if (IndexEntrySize > 4)
{
throw new InvalidOperationException("Invalid index entry size \"" + IndexEntrySize + "\"!");
}
if (IndexCount != 0)
{
int IbSize = IndexCount * IndexEntrySize;
bool IboCached = Gpu.Renderer.Rasterizer.IsIboCached(IboKey, (uint)IbSize);
bool UsesLegacyQuads =
PrimType == GalPrimitiveType.Quads ||
PrimType == GalPrimitiveType.QuadStrip;
if (!IboCached || Gpu.ResourceManager.MemoryRegionModified(Vmm, IboKey, (uint)IbSize, NvGpuBufferType.Index))
{
if (!UsesLegacyQuads)
{
if (Vmm.TryGetHostAddress(IbPosition, IbSize, out IntPtr IbPtr))
{
Gpu.Renderer.Rasterizer.CreateIbo(IboKey, IbSize, IbPtr);
}
else
{
Gpu.Renderer.Rasterizer.CreateIbo(IboKey, IbSize, Vmm.ReadBytes(IbPosition, IbSize));
}
}
else
{
byte[] Buffer = Vmm.ReadBytes(IbPosition, IbSize);
if (PrimType == GalPrimitiveType.Quads)
{
Buffer = QuadHelper.ConvertQuadsToTris(Buffer, IndexEntrySize, IndexCount);
}
else /* if (PrimType == GalPrimitiveType.QuadStrip) */
{
Buffer = QuadHelper.ConvertQuadStripToTris(Buffer, IndexEntrySize, IndexCount);
}
Gpu.Renderer.Rasterizer.CreateIbo(IboKey, IbSize, Buffer);
}
}
if (!UsesLegacyQuads)
{
Gpu.Renderer.Rasterizer.SetIndexArray(IbSize, IndexFormat);
}
else
{
if (PrimType == GalPrimitiveType.Quads)
{
Gpu.Renderer.Rasterizer.SetIndexArray(QuadHelper.ConvertSizeQuadsToTris(IbSize), IndexFormat);
}
else /* if (PrimType == GalPrimitiveType.QuadStrip) */
{
Gpu.Renderer.Rasterizer.SetIndexArray(QuadHelper.ConvertSizeQuadStripToTris(IbSize), IndexFormat);
}
}
}
List<GalVertexAttrib>[] Attribs = new List<GalVertexAttrib>[32];
for (int Attr = 0; Attr < 16; Attr++)
{
int Packed = ReadRegister(NvGpuEngine3dReg.VertexAttribNFormat + Attr);
int ArrayIndex = Packed & 0x1f;
if (Attribs[ArrayIndex] == null)
{
Attribs[ArrayIndex] = new List<GalVertexAttrib>();
}
long VbPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNAddress + ArrayIndex * 4);
if (VbPosition == 0)
{
continue;
}
bool IsConst = ((Packed >> 6) & 1) != 0;
int Offset = (Packed >> 7) & 0x3fff;
GalVertexAttribSize Size = (GalVertexAttribSize)((Packed >> 21) & 0x3f);
GalVertexAttribType Type = (GalVertexAttribType)((Packed >> 27) & 0x7);
bool IsRgba = ((Packed >> 31) & 1) != 0;
// Check vertex array is enabled to avoid out of bounds exception when reading bytes
bool Enable = (ReadRegister(NvGpuEngine3dReg.VertexArrayNControl + ArrayIndex * 4) & 0x1000) != 0;
//Note: 16 is the maximum size of an attribute,
//having a component size of 32-bits with 4 elements (a vec4).
if (Enable)
{
byte[] Data = Vmm.ReadBytes(VbPosition + Offset, 16);
Attribs[ArrayIndex].Add(new GalVertexAttrib(Attr, IsConst, Offset, Data, Size, Type, IsRgba));
}
}
State.VertexBindings = new GalVertexBinding[32];
for (int Index = 0; Index < 32; Index++)
{
if (Attribs[Index] == null)
{
continue;
}
int Control = ReadRegister(NvGpuEngine3dReg.VertexArrayNControl + Index * 4);
bool Enable = (Control & 0x1000) != 0;
if (!Enable)
{
continue;
}
long VbPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNAddress + Index * 4);
long VbEndPos = MakeInt64From2xInt32(NvGpuEngine3dReg.VertexArrayNEndAddr + Index * 2);
2018-08-25 04:16:58 +00:00
int VertexDivisor = ReadRegister(NvGpuEngine3dReg.VertexArrayNDivisor + Index * 4);
bool Instanced = ReadRegisterBool(NvGpuEngine3dReg.VertexArrayNInstance + Index);
int Stride = Control & 0xfff;
2018-08-25 04:16:58 +00:00
if (Instanced && VertexDivisor != 0)
{
VbPosition += Stride * (CurrentInstance / VertexDivisor);
2018-08-25 04:16:58 +00:00
}
if (VbPosition > VbEndPos)
2018-08-25 04:16:58 +00:00
{
//Instance is invalid, ignore the draw call
continue;
}
long VboKey = Vmm.GetPhysicalAddress(VbPosition);
2018-08-25 04:16:58 +00:00
long VbSize = (VbEndPos - VbPosition) + 1;
int ModifiedVbSize = (int)VbSize;
// If quads convert size to triangle length
if (Stride == 0)
{
if (PrimType == GalPrimitiveType.Quads)
{
ModifiedVbSize = QuadHelper.ConvertSizeQuadsToTris(ModifiedVbSize);
}
else if (PrimType == GalPrimitiveType.QuadStrip)
{
ModifiedVbSize = QuadHelper.ConvertSizeQuadStripToTris(ModifiedVbSize);
}
}
bool VboCached = Gpu.Renderer.Rasterizer.IsVboCached(VboKey, ModifiedVbSize);
if (!VboCached || Gpu.ResourceManager.MemoryRegionModified(Vmm, VboKey, VbSize, NvGpuBufferType.Vertex))
{
if ((PrimType == GalPrimitiveType.Quads | PrimType == GalPrimitiveType.QuadStrip) && Stride != 0)
{
// Convert quad buffer to triangles
byte[] data = Vmm.ReadBytes(VbPosition, VbSize);
if (PrimType == GalPrimitiveType.Quads)
{
data = QuadHelper.ConvertQuadsToTris(data, Stride, (int)(VbSize / Stride));
}
else
{
data = QuadHelper.ConvertQuadStripToTris(data, Stride, (int)(VbSize / Stride));
}
Gpu.Renderer.Rasterizer.CreateVbo(VboKey, data);
}
else if (Vmm.TryGetHostAddress(VbPosition, VbSize, out IntPtr VbPtr))
{
Gpu.Renderer.Rasterizer.CreateVbo(VboKey, (int)VbSize, VbPtr);
}
else
{
Gpu.Renderer.Rasterizer.CreateVbo(VboKey, Vmm.ReadBytes(VbPosition, VbSize));
}
}
2018-08-25 04:16:58 +00:00
State.VertexBindings[Index].Enabled = true;
State.VertexBindings[Index].Stride = Stride;
State.VertexBindings[Index].VboKey = VboKey;
State.VertexBindings[Index].Instanced = Instanced;
State.VertexBindings[Index].Divisor = VertexDivisor;
State.VertexBindings[Index].Attribs = Attribs[Index].ToArray();
}
}
private void DispatchRender(NvGpuVmm Vmm, GalPipelineState State)
{
int IndexCount = ReadRegister(NvGpuEngine3dReg.IndexBatchCount);
int PrimCtrl = ReadRegister(NvGpuEngine3dReg.VertexBeginGl);
GalPrimitiveType PrimType = (GalPrimitiveType)(PrimCtrl & 0xffff);
2018-08-25 04:16:58 +00:00
bool InstanceNext = ((PrimCtrl >> 26) & 1) != 0;
bool InstanceCont = ((PrimCtrl >> 27) & 1) != 0;
if (InstanceNext && InstanceCont)
{
throw new InvalidOperationException("GPU tried to increase and reset instance count at the same time");
}
if (InstanceNext)
{
CurrentInstance++;
}
else if (!InstanceCont)
{
CurrentInstance = 0;
}
State.Instance = CurrentInstance;
Gpu.Renderer.Pipeline.Bind(State);
Gpu.Renderer.RenderTarget.Bind();
if (IndexCount != 0)
{
int IndexEntryFmt = ReadRegister(NvGpuEngine3dReg.IndexArrayFormat);
int IndexFirst = ReadRegister(NvGpuEngine3dReg.IndexBatchFirst);
int VertexBase = ReadRegister(NvGpuEngine3dReg.VertexArrayElemBase);
long IndexPosition = MakeInt64From2xInt32(NvGpuEngine3dReg.IndexArrayAddress);
long IboKey = Vmm.GetPhysicalAddress(IndexPosition);
//Quad primitive types were deprecated on OpenGL 3.x,
//they are converted to a triangles index buffer on IB creation,
//so we should use the triangles type here too.
if (PrimType == GalPrimitiveType.Quads || PrimType == GalPrimitiveType.QuadStrip)
{
//Note: We assume that index first points to the first
//vertex of a quad, if it points to the middle of a
//quad (First % 4 != 0 for Quads) then it will not work properly.
if (PrimType == GalPrimitiveType.Quads)
{
IndexFirst = QuadHelper.ConvertSizeQuadsToTris(IndexFirst);
}
else // QuadStrip
{
IndexFirst = QuadHelper.ConvertSizeQuadStripToTris(IndexFirst);
}
PrimType = GalPrimitiveType.Triangles;
}
Gpu.Renderer.Rasterizer.DrawElements(IboKey, IndexFirst, VertexBase, PrimType);
}
else
{
int VertexFirst = ReadRegister(NvGpuEngine3dReg.VertexArrayFirst);
int VertexCount = ReadRegister(NvGpuEngine3dReg.VertexArrayCount);
//Quad primitive types were deprecated on OpenGL 3.x,
//they are converted to a triangles index buffer on IB creation,
//so we should use the triangles type here too.
if (PrimType == GalPrimitiveType.Quads || PrimType == GalPrimitiveType.QuadStrip)
{
//Note: We assume that index first points to the first
//vertex of a quad, if it points to the middle of a
//quad (First % 4 != 0 for Quads) then it will not work properly.
if (PrimType == GalPrimitiveType.Quads)
{
VertexFirst = QuadHelper.ConvertSizeQuadsToTris(VertexFirst);
}
else // QuadStrip
{
VertexFirst = QuadHelper.ConvertSizeQuadStripToTris(VertexFirst);
}
PrimType = GalPrimitiveType.Triangles;
VertexCount = QuadHelper.ConvertSizeQuadsToTris(VertexCount);
}
Gpu.Renderer.Rasterizer.DrawArrays(VertexFirst, VertexCount, PrimType);
}
// Reset pipeline for host OpenGL calls
Gpu.Renderer.Pipeline.Unbind(State);
//Is the GPU really clearing those registers after draw?
WriteRegister(NvGpuEngine3dReg.IndexBatchFirst, 0);
WriteRegister(NvGpuEngine3dReg.IndexBatchCount, 0);
}
private enum QueryMode
{
WriteSeq,
Sync,
WriteCounterAndTimestamp
}
private void QueryControl(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
WriteRegister(MethCall);
long Position = MakeInt64From2xInt32(NvGpuEngine3dReg.QueryAddress);
int Seq = Registers[(int)NvGpuEngine3dReg.QuerySequence];
int Ctrl = Registers[(int)NvGpuEngine3dReg.QueryControl];
QueryMode Mode = (QueryMode)(Ctrl & 3);
switch (Mode)
{
case QueryMode.WriteSeq: Vmm.WriteInt32(Position, Seq); break;
case QueryMode.WriteCounterAndTimestamp:
{
//TODO: Implement counters.
long Counter = 1;
long Timestamp = PerformanceCounter.ElapsedMilliseconds;
Vmm.WriteInt64(Position + 0, Counter);
Vmm.WriteInt64(Position + 8, Timestamp);
break;
}
}
}
private void CbData(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
long Position = MakeInt64From2xInt32(NvGpuEngine3dReg.ConstBufferAddress);
int Offset = ReadRegister(NvGpuEngine3dReg.ConstBufferOffset);
Vmm.WriteInt32(Position + Offset, MethCall.Argument);
WriteRegister(NvGpuEngine3dReg.ConstBufferOffset, Offset + 4);
Gpu.ResourceManager.ClearPbCache(NvGpuBufferType.ConstBuffer);
}
private void CbBind(NvGpuVmm Vmm, GpuMethodCall MethCall)
{
int Stage = (MethCall.Method - 0x904) >> 3;
int Index = MethCall.Argument;
bool Enabled = (Index & 1) != 0;
Index = (Index >> 4) & 0x1f;
long Position = MakeInt64From2xInt32(NvGpuEngine3dReg.ConstBufferAddress);
long CbKey = Vmm.GetPhysicalAddress(Position);
int Size = ReadRegister(NvGpuEngine3dReg.ConstBufferSize);
if (!Gpu.Renderer.Buffer.IsCached(CbKey, Size))
{
Gpu.Renderer.Buffer.Create(CbKey, Size);
}
ConstBuffer Cb = ConstBuffers[Stage][Index];
if (Cb.Position != Position || Cb.Enabled != Enabled || Cb.Size != Size)
{
ConstBuffers[Stage][Index].Position = Position;
ConstBuffers[Stage][Index].Enabled = Enabled;
ConstBuffers[Stage][Index].Size = Size;
}
}
private float GetFlipSign(NvGpuEngine3dReg Reg)
{
return MathF.Sign(ReadRegisterFloat(Reg));
}
private long MakeInt64From2xInt32(NvGpuEngine3dReg Reg)
{
return
(long)Registers[(int)Reg + 0] << 32 |
(uint)Registers[(int)Reg + 1];
}
private void WriteRegister(GpuMethodCall MethCall)
{
Registers[MethCall.Method] = MethCall.Argument;
}
private int ReadRegister(NvGpuEngine3dReg Reg)
{
return Registers[(int)Reg];
}
private float ReadRegisterFloat(NvGpuEngine3dReg Reg)
{
return BitConverter.Int32BitsToSingle(ReadRegister(Reg));
}
private bool ReadRegisterBool(NvGpuEngine3dReg Reg)
{
return (ReadRegister(Reg) & 1) != 0;
}
private void WriteRegister(NvGpuEngine3dReg Reg, int Value)
{
Registers[(int)Reg] = Value;
}
}
}