ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #11551 from iwubcode/d3d11_sampler_bitset 

VideoBackends: use a bitset for D3D11's state management
This commit is contained in:
Pierre Bourdon 2023-02-25 09:32:10 +09:00 committed by GitHub
parent 88fc431dce 5d571b068f
commit f682225c15
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 68 additions and 88 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

70
Source/Core/VideoBackends/D3D/D3DState.cpp
View File

@ -28,12 +28,12 @@ StateManager::~StateManager() = default;
void StateManager::Apply()
{
if (!m_dirtyFlags)
if (m_dirtyFlags.none())
return;
// Framebuffer changes must occur before texture changes, otherwise the D3D runtime messes with
// our bindings and sets them to null to prevent hazards.
if (m_dirtyFlags & DirtyFlag_Framebuffer)
if (m_dirtyFlags.test(DirtyFlag_Framebuffer))
{
if (g_ActiveConfig.backend_info.bSupportsBBox)
{
@ -56,11 +56,14 @@ void StateManager::Apply()
m_current.use_integer_rtv = m_pending.use_integer_rtv;
}
u32 dirtyConstants = m_dirtyFlags & (DirtyFlag_PixelConstants | DirtyFlag_VertexConstants |
DirtyFlag_GeometryConstants);
u32 dirtyShaders =
m_dirtyFlags & (DirtyFlag_PixelShader | DirtyFlag_VertexShader | DirtyFlag_GeometryShader);
u32 dirtyBuffers = m_dirtyFlags & (DirtyFlag_VertexBuffer | DirtyFlag_IndexBuffer);
const bool dirtyConstants = m_dirtyFlags.test(DirtyFlag_PixelConstants) ||
m_dirtyFlags.test(DirtyFlag_VertexConstants) ||
m_dirtyFlags.test(DirtyFlag_GeometryConstants);
const bool dirtyShaders = m_dirtyFlags.test(DirtyFlag_PixelShader) ||
m_dirtyFlags.test(DirtyFlag_VertexShader) ||
m_dirtyFlags.test(DirtyFlag_GeometryShader);
const bool dirtyBuffers =
m_dirtyFlags.test(DirtyFlag_VertexBuffer) || m_dirtyFlags.test(DirtyFlag_IndexBuffer);
if (dirtyConstants)
{
@ -87,7 +90,7 @@ void StateManager::Apply()
}
}
if (dirtyBuffers || (m_dirtyFlags & DirtyFlag_InputAssembler))
if (dirtyBuffers || (m_dirtyFlags.test(DirtyFlag_InputAssembler)))
{
if (m_current.vertexBuffer != m_pending.vertexBuffer ||
m_current.vertexBufferStride != m_pending.vertexBufferStride ||
@ -140,17 +143,17 @@ void StateManager::Apply()
}
}
if (m_dirtyFlags & DirtyFlag_BlendState)
if (m_dirtyFlags.test(DirtyFlag_BlendState))
{
D3D::context->OMSetBlendState(m_pending.blendState, nullptr, 0xFFFFFFFF);
m_current.blendState = m_pending.blendState;
}
if (m_dirtyFlags & DirtyFlag_DepthState)
if (m_dirtyFlags.test(DirtyFlag_DepthState))
{
D3D::context->OMSetDepthStencilState(m_pending.depthState, 0);
m_current.depthState = m_pending.depthState;
}
if (m_dirtyFlags & DirtyFlag_RasterizerState)
if (m_dirtyFlags.test(DirtyFlag_RasterizerState))
{
D3D::context->RSSetState(m_pending.rasterizerState);
m_current.rasterizerState = m_pending.rasterizerState;
@ -163,41 +166,32 @@ void StateManager::Apply()
void StateManager::ApplyTextures()
{
const int textureMaskShift = std::countr_zero((u32)DirtyFlag_Texture0);
const int samplerMaskShift = std::countr_zero((u32)DirtyFlag_Sampler0);
u32 dirtyTextures =
(m_dirtyFlags &
(DirtyFlag_Texture0 | DirtyFlag_Texture1 | DirtyFlag_Texture2 | DirtyFlag_Texture3 |
DirtyFlag_Texture4 | DirtyFlag_Texture5 | DirtyFlag_Texture6 | DirtyFlag_Texture7)) >>
textureMaskShift;
u32 dirtySamplers =
(m_dirtyFlags &
(DirtyFlag_Sampler0 | DirtyFlag_Sampler1 | DirtyFlag_Sampler2 | DirtyFlag_Sampler3 |
DirtyFlag_Sampler4 | DirtyFlag_Sampler5 | DirtyFlag_Sampler6 | DirtyFlag_Sampler7)) >>
samplerMaskShift;
while (dirtyTextures)
for (u32 i = 0; i < VideoCommon::MAX_PIXEL_SHADER_SAMPLERS; i++)
{
const int index = std::countr_zero(dirtyTextures);
if (m_current.textures[index] != m_pending.textures[index])
const u32 flag = i;
if (m_dirtyFlags.test(flag))
{
D3D::context->PSSetShaderResources(index, 1, &m_pending.textures[index]);
m_current.textures[index] = m_pending.textures[index];
if (m_current.textures[i] != m_pending.textures[i])
{
D3D::context->PSSetShaderResources(i, 1, &m_pending.textures[i]);
m_current.textures[i] = m_pending.textures[i];
}
m_dirtyFlags.reset(flag);
}
dirtyTextures &= ~(1 << index);
}
while (dirtySamplers)
for (u32 i = 0; i < VideoCommon::MAX_PIXEL_SHADER_SAMPLERS; i++)
{
const int index = std::countr_zero(dirtySamplers);
if (m_current.samplers[index] != m_pending.samplers[index])
const u32 flag = i + VideoCommon::MAX_PIXEL_SHADER_SAMPLERS;
if (m_dirtyFlags.test(flag))
{
D3D::context->PSSetSamplers(index, 1, &m_pending.samplers[index]);
m_current.samplers[index] = m_pending.samplers[index];
if (m_current.samplers[i] != m_pending.samplers[i])
{
D3D::context->PSSetSamplers(i, 1, &m_pending.samplers[i]);
m_current.samplers[i] = m_pending.samplers[i];
}
m_dirtyFlags.reset(flag);
}
dirtySamplers &= ~(1 << index);
}
}

86
Source/Core/VideoBackends/D3D/D3DState.h
View File

@ -4,6 +4,7 @@
#pragma once
#include <array>
#include <bitset>
#include <cstddef>
#include <memory>
#include <mutex>
@ -53,7 +54,7 @@ public:
void SetBlendState(ID3D11BlendState* state)
{
if (m_current.blendState != state)
m_dirtyFlags |= DirtyFlag_BlendState;
m_dirtyFlags.set(DirtyFlag_BlendState);
m_pending.blendState = state;
}
@ -61,7 +62,7 @@ public:
void SetDepthState(ID3D11DepthStencilState* state)
{
if (m_current.depthState != state)
m_dirtyFlags |= DirtyFlag_DepthState;
m_dirtyFlags.set(DirtyFlag_DepthState);
m_pending.depthState = state;
}
@ -69,7 +70,7 @@ public:
void SetRasterizerState(ID3D11RasterizerState* state)
{
if (m_current.rasterizerState != state)
m_dirtyFlags |= DirtyFlag_RasterizerState;
m_dirtyFlags.set(DirtyFlag_RasterizerState);
m_pending.rasterizerState = state;
}
@ -77,7 +78,7 @@ public:
void SetTexture(size_t index, ID3D11ShaderResourceView* texture)
{
if (m_current.textures[index] != texture)
m_dirtyFlags |= DirtyFlag_Texture0 << index;
m_dirtyFlags.set(DirtyFlag_Texture0 + index);
m_pending.textures[index] = texture;
}
@ -85,7 +86,7 @@ public:
void SetSampler(size_t index, ID3D11SamplerState* sampler)
{
if (m_current.samplers[index] != sampler)
m_dirtyFlags |= DirtyFlag_Sampler0 << index;
m_dirtyFlags.set(DirtyFlag_Sampler0 + index);
m_pending.samplers[index] = sampler;
}
@ -93,7 +94,7 @@ public:
void SetPixelConstants(ID3D11Buffer* buffer0, ID3D11Buffer* buffer1 = nullptr)
{
if (m_current.pixelConstants[0] != buffer0 || m_current.pixelConstants[1] != buffer1)
m_dirtyFlags |= DirtyFlag_PixelConstants;
m_dirtyFlags.set(DirtyFlag_PixelConstants);
m_pending.pixelConstants[0] = buffer0;
m_pending.pixelConstants[1] = buffer1;
@ -102,7 +103,7 @@ public:
void SetVertexConstants(ID3D11Buffer* buffer)
{
if (m_current.vertexConstants != buffer)
m_dirtyFlags |= DirtyFlag_VertexConstants;
m_dirtyFlags.set(DirtyFlag_VertexConstants);
m_pending.vertexConstants = buffer;
}
@ -110,7 +111,7 @@ public:
void SetGeometryConstants(ID3D11Buffer* buffer)
{
if (m_current.geometryConstants != buffer)
m_dirtyFlags |= DirtyFlag_GeometryConstants;
m_dirtyFlags.set(DirtyFlag_GeometryConstants);
m_pending.geometryConstants = buffer;
}
@ -119,7 +120,7 @@ public:
{
if (m_current.vertexBuffer != buffer || m_current.vertexBufferStride != stride ||
m_current.vertexBufferOffset != offset)
m_dirtyFlags |= DirtyFlag_VertexBuffer;
m_dirtyFlags.set(DirtyFlag_VertexBuffer);
m_pending.vertexBuffer = buffer;
m_pending.vertexBufferStride = stride;
@ -129,7 +130,7 @@ public:
void SetIndexBuffer(ID3D11Buffer* buffer)
{
if (m_current.indexBuffer != buffer)
m_dirtyFlags |= DirtyFlag_IndexBuffer;
m_dirtyFlags.set(DirtyFlag_IndexBuffer);
m_pending.indexBuffer = buffer;
}
@ -137,7 +138,7 @@ public:
void SetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY topology)
{
if (m_current.topology != topology)
m_dirtyFlags |= DirtyFlag_InputAssembler;
m_dirtyFlags.set(DirtyFlag_InputAssembler);
m_pending.topology = topology;
}
@ -145,7 +146,7 @@ public:
void SetInputLayout(ID3D11InputLayout* layout)
{
if (m_current.inputLayout != layout)
m_dirtyFlags |= DirtyFlag_InputAssembler;
m_dirtyFlags.set(DirtyFlag_InputAssembler);
m_pending.inputLayout = layout;
}
@ -153,7 +154,7 @@ public:
void SetPixelShader(ID3D11PixelShader* shader)
{
if (m_current.pixelShader != shader)
m_dirtyFlags |= DirtyFlag_PixelShader;
m_dirtyFlags.set(DirtyFlag_PixelShader);
m_pending.pixelShader = shader;
}
@ -169,7 +170,7 @@ public:
void SetVertexShader(ID3D11VertexShader* shader)
{
if (m_current.vertexShader != shader)
m_dirtyFlags |= DirtyFlag_VertexShader;
m_dirtyFlags.set(DirtyFlag_VertexShader);
m_pending.vertexShader = shader;
}
@ -177,7 +178,7 @@ public:
void SetGeometryShader(ID3D11GeometryShader* shader)
{
if (m_current.geometryShader != shader)
m_dirtyFlags |= DirtyFlag_GeometryShader;
m_dirtyFlags.set(DirtyFlag_GeometryShader);
m_pending.geometryShader = shader;
}
@ -185,7 +186,7 @@ public:
void SetFramebuffer(DXFramebuffer* fb)
{
if (m_current.framebuffer != fb)
m_dirtyFlags |= DirtyFlag_Framebuffer;
m_dirtyFlags.set(DirtyFlag_Framebuffer);
m_pending.framebuffer = fb;
}
@ -193,7 +194,7 @@ public:
void SetOMUAV(ID3D11UnorderedAccessView* uav)
{
if (m_current.uav != uav)
m_dirtyFlags |= DirtyFlag_Framebuffer;
m_dirtyFlags.set(DirtyFlag_Framebuffer);
m_pending.uav = uav;
}
@ -201,7 +202,7 @@ public:
void SetIntegerRTV(bool enable)
{
if (m_current.use_integer_rtv != enable)
m_dirtyFlags |= DirtyFlag_Framebuffer;
m_dirtyFlags.set(DirtyFlag_Framebuffer);
m_pending.use_integer_rtv = enable;
}
@ -224,43 +225,28 @@ public:
private:
enum DirtyFlags
{
DirtyFlag_Texture0 = 1 << 0,
DirtyFlag_Texture1 = 1 << 1,
DirtyFlag_Texture2 = 1 << 2,
DirtyFlag_Texture3 = 1 << 3,
DirtyFlag_Texture4 = 1 << 4,
DirtyFlag_Texture5 = 1 << 5,
DirtyFlag_Texture6 = 1 << 6,
DirtyFlag_Texture7 = 1 << 7,
DirtyFlag_Texture0 = 0,
DirtyFlag_Sampler0 = DirtyFlag_Texture0 + VideoCommon::MAX_PIXEL_SHADER_SAMPLERS,
DirtyFlag_PixelConstants = DirtyFlag_Sampler0 + VideoCommon::MAX_PIXEL_SHADER_SAMPLERS,
DirtyFlag_VertexConstants,
DirtyFlag_GeometryConstants,
DirtyFlag_Sampler0 = 1 << 8,
DirtyFlag_Sampler1 = 1 << 9,
DirtyFlag_Sampler2 = 1 << 10,
DirtyFlag_Sampler3 = 1 << 11,
DirtyFlag_Sampler4 = 1 << 12,
DirtyFlag_Sampler5 = 1 << 13,
DirtyFlag_Sampler6 = 1 << 14,
DirtyFlag_Sampler7 = 1 << 15,
DirtyFlag_VertexBuffer,
DirtyFlag_IndexBuffer,
DirtyFlag_PixelConstants = 1 << 16,
DirtyFlag_VertexConstants = 1 << 17,
DirtyFlag_GeometryConstants = 1 << 18,
DirtyFlag_PixelShader,
DirtyFlag_VertexShader,
DirtyFlag_GeometryShader,
DirtyFlag_VertexBuffer = 1 << 19,
DirtyFlag_IndexBuffer = 1 << 20,
DirtyFlag_PixelShader = 1 << 21,
DirtyFlag_VertexShader = 1 << 22,
DirtyFlag_GeometryShader = 1 << 23,
DirtyFlag_InputAssembler = 1 << 24,
DirtyFlag_BlendState = 1 << 25,
DirtyFlag_DepthState = 1 << 26,
DirtyFlag_RasterizerState = 1 << 27,
DirtyFlag_Framebuffer = 1 << 28
DirtyFlag_InputAssembler,
DirtyFlag_BlendState,
DirtyFlag_DepthState,
DirtyFlag_RasterizerState,
DirtyFlag_Framebuffer,
DirtyFlag_Max
};
u32 m_dirtyFlags = ~0u;
std::bitset<DirtyFlags::DirtyFlag_Max> m_dirtyFlags;
struct Resources
{