Common: Add D3D12 wrapper/helper classes

This commit is contained in:
Connor McLaughlin 2022-03-19 22:17:18 +10:00 committed by refractionpcsx2
parent 3255422836
commit e767fb8d35
17 changed files with 3624 additions and 3 deletions

View File

@ -173,13 +173,27 @@ endif()
if(WIN32)
enable_language(ASM_MASM)
target_sources(common PRIVATE FastJmp.asm)
target_link_libraries(common PUBLIC WIL::WIL pthreads4w Winmm.lib)
target_link_libraries(common PUBLIC WIL::WIL D3D12MemAlloc pthreads4w Winmm.lib)
target_sources(common PRIVATE
FastJmp.asm
D3D11/ShaderCache.cpp
D3D11/ShaderCache.h
D3D11/ShaderCompiler.cpp
D3D11/ShaderCompiler.h
D3D12/Builders.cpp
D3D12/Builders.h
D3D12/Context.cpp
D3D12/Context.h
D3D12/DescriptorHeapManager.cpp
D3D12/DescriptorHeapManager.h
D3D12/ShaderCache.cpp
D3D12/ShaderCache.h
D3D12/StreamBuffer.cpp
D3D12/StreamBuffer.h
D3D12/Texture.cpp
D3D12/Texture.h
D3D12/Util.cpp
D3D12/Util.h
)
endif()

312
common/D3D12/Builders.cpp Normal file
View File

@ -0,0 +1,312 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/Builders.h"
#include "common/D3D12/Context.h"
#include "common/D3D12/ShaderCache.h"
#include "common/Console.h"
#include <cstdarg>
#include <limits>
using namespace D3D12;
GraphicsPipelineBuilder::GraphicsPipelineBuilder()
{
Clear();
}
void GraphicsPipelineBuilder::Clear()
{
std::memset(&m_desc, 0, sizeof(m_desc));
std::memset(m_input_elements.data(), 0, sizeof(D3D12_INPUT_ELEMENT_DESC) * m_input_elements.size());
m_desc.NodeMask = 1;
m_desc.SampleMask = 0xFFFFFFFF;
m_desc.SampleDesc.Count = 1;
}
wil::com_ptr_nothrow<ID3D12PipelineState> GraphicsPipelineBuilder::Create(ID3D12Device* device, bool clear /*= true*/)
{
wil::com_ptr_nothrow<ID3D12PipelineState> ps;
HRESULT hr = device->CreateGraphicsPipelineState(&m_desc, IID_PPV_ARGS(ps.put()));
if (FAILED(hr))
{
Console.Error("CreateGraphicsPipelineState() failed: %08X", hr);
return {};
}
if (clear)
Clear();
return ps;
}
wil::com_ptr_nothrow<ID3D12PipelineState> GraphicsPipelineBuilder::Create(ID3D12Device* device, ShaderCache& cache,
bool clear /*= true*/)
{
wil::com_ptr_nothrow<ID3D12PipelineState> pso = cache.GetPipelineState(device, m_desc);
if (!pso)
return {};
if (clear)
Clear();
return pso;
}
void GraphicsPipelineBuilder::SetRootSignature(ID3D12RootSignature* rs)
{
m_desc.pRootSignature = rs;
}
void GraphicsPipelineBuilder::SetVertexShader(const ID3DBlob* blob)
{
SetVertexShader(const_cast<ID3DBlob*>(blob)->GetBufferPointer(), static_cast<u32>(const_cast<ID3DBlob*>(blob)->GetBufferSize()));
}
void GraphicsPipelineBuilder::SetVertexShader(const void* data, u32 data_size)
{
m_desc.VS.pShaderBytecode = data;
m_desc.VS.BytecodeLength = data_size;
}
void GraphicsPipelineBuilder::SetGeometryShader(const ID3DBlob* blob)
{
SetGeometryShader(const_cast<ID3DBlob*>(blob)->GetBufferPointer(), static_cast<u32>(const_cast<ID3DBlob*>(blob)->GetBufferSize()));
}
void GraphicsPipelineBuilder::SetGeometryShader(const void* data, u32 data_size)
{
m_desc.GS.pShaderBytecode = data;
m_desc.GS.BytecodeLength = data_size;
}
void GraphicsPipelineBuilder::SetPixelShader(const ID3DBlob* blob)
{
SetPixelShader(const_cast<ID3DBlob*>(blob)->GetBufferPointer(), static_cast<u32>(const_cast<ID3DBlob*>(blob)->GetBufferSize()));
}
void GraphicsPipelineBuilder::SetPixelShader(const void* data, u32 data_size)
{
m_desc.PS.pShaderBytecode = data;
m_desc.PS.BytecodeLength = data_size;
}
void GraphicsPipelineBuilder::AddVertexAttribute(const char* semantic_name, u32 semantic_index, DXGI_FORMAT format,
u32 buffer, u32 offset)
{
const u32 index = m_desc.InputLayout.NumElements;
m_input_elements[index].SemanticIndex = semantic_index;
m_input_elements[index].SemanticName = semantic_name;
m_input_elements[index].Format = format;
m_input_elements[index].AlignedByteOffset = offset;
m_input_elements[index].InputSlot = buffer;
m_input_elements[index].InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA;
m_input_elements[index].InstanceDataStepRate = 0;
m_desc.InputLayout.pInputElementDescs = m_input_elements.data();
m_desc.InputLayout.NumElements++;
}
void GraphicsPipelineBuilder::SetPrimitiveTopologyType(D3D12_PRIMITIVE_TOPOLOGY_TYPE type)
{
m_desc.PrimitiveTopologyType = type;
}
void GraphicsPipelineBuilder::SetRasterizationState(D3D12_FILL_MODE polygon_mode, D3D12_CULL_MODE cull_mode,
bool front_face_ccw)
{
m_desc.RasterizerState.FillMode = polygon_mode;
m_desc.RasterizerState.CullMode = cull_mode;
m_desc.RasterizerState.FrontCounterClockwise = front_face_ccw;
}
void GraphicsPipelineBuilder::SetMultisamples(u32 multisamples)
{
m_desc.RasterizerState.MultisampleEnable = multisamples > 1;
m_desc.SampleDesc.Count = multisamples;
}
void GraphicsPipelineBuilder::SetNoCullRasterizationState()
{
SetRasterizationState(D3D12_FILL_MODE_SOLID, D3D12_CULL_MODE_NONE, false);
}
void GraphicsPipelineBuilder::SetDepthState(bool depth_test, bool depth_write, D3D12_COMPARISON_FUNC compare_op)
{
m_desc.DepthStencilState.DepthEnable = depth_test;
m_desc.DepthStencilState.DepthWriteMask = depth_write ? D3D12_DEPTH_WRITE_MASK_ALL : D3D12_DEPTH_WRITE_MASK_ZERO;
m_desc.DepthStencilState.DepthFunc = compare_op;
}
void GraphicsPipelineBuilder::SetStencilState(bool stencil_test, u8 read_mask, u8 write_mask,
const D3D12_DEPTH_STENCILOP_DESC& front, const D3D12_DEPTH_STENCILOP_DESC& back)
{
m_desc.DepthStencilState.StencilEnable = stencil_test;
m_desc.DepthStencilState.StencilReadMask = read_mask;
m_desc.DepthStencilState.StencilWriteMask = write_mask;
m_desc.DepthStencilState.FrontFace = front;
m_desc.DepthStencilState.BackFace = back;
}
void GraphicsPipelineBuilder::SetNoDepthTestState()
{
SetDepthState(false, false, D3D12_COMPARISON_FUNC_ALWAYS);
}
void GraphicsPipelineBuilder::SetNoStencilState()
{
D3D12_DEPTH_STENCILOP_DESC empty = {};
SetStencilState(false, 0, 0, empty, empty);
}
void GraphicsPipelineBuilder::SetBlendState(u32 rt, bool blend_enable, D3D12_BLEND src_factor, D3D12_BLEND dst_factor,
D3D12_BLEND_OP op, D3D12_BLEND alpha_src_factor,
D3D12_BLEND alpha_dst_factor, D3D12_BLEND_OP alpha_op,
u8 write_mask /*= 0xFF*/)
{
m_desc.BlendState.RenderTarget[rt].BlendEnable = blend_enable;
m_desc.BlendState.RenderTarget[rt].SrcBlend = src_factor;
m_desc.BlendState.RenderTarget[rt].DestBlend = dst_factor;
m_desc.BlendState.RenderTarget[rt].BlendOp = op;
m_desc.BlendState.RenderTarget[rt].SrcBlendAlpha = alpha_src_factor;
m_desc.BlendState.RenderTarget[rt].DestBlendAlpha = alpha_dst_factor;
m_desc.BlendState.RenderTarget[rt].BlendOpAlpha = alpha_op;
m_desc.BlendState.RenderTarget[rt].RenderTargetWriteMask = write_mask;
if (rt > 0)
m_desc.BlendState.IndependentBlendEnable = TRUE;
}
void GraphicsPipelineBuilder::SetNoBlendingState()
{
SetBlendState(0, false, D3D12_BLEND_ONE, D3D12_BLEND_ZERO, D3D12_BLEND_OP_ADD, D3D12_BLEND_ONE, D3D12_BLEND_ZERO,
D3D12_BLEND_OP_ADD, D3D12_COLOR_WRITE_ENABLE_ALL);
m_desc.BlendState.IndependentBlendEnable = FALSE;
}
void GraphicsPipelineBuilder::ClearRenderTargets()
{
m_desc.NumRenderTargets = 0;
for (u32 i = 0; i < sizeof(m_desc.RTVFormats) / sizeof(m_desc.RTVFormats[0]); i++)
m_desc.RTVFormats[i] = DXGI_FORMAT_UNKNOWN;
}
void GraphicsPipelineBuilder::SetRenderTarget(u32 rt, DXGI_FORMAT format)
{
m_desc.RTVFormats[rt] = format;
if (rt >= m_desc.NumRenderTargets)
m_desc.NumRenderTargets = rt + 1;
}
void GraphicsPipelineBuilder::ClearDepthStencilFormat()
{
m_desc.DSVFormat = DXGI_FORMAT_UNKNOWN;
}
void GraphicsPipelineBuilder::SetDepthStencilFormat(DXGI_FORMAT format)
{
m_desc.DSVFormat = format;
}
RootSignatureBuilder::RootSignatureBuilder()
{
Clear();
}
void RootSignatureBuilder::Clear()
{
m_desc = {};
m_desc.pParameters = m_params.data();
m_params = {};
m_descriptor_ranges = {};
m_num_descriptor_ranges = 0;
}
wil::com_ptr_nothrow<ID3D12RootSignature> RootSignatureBuilder::Create(bool clear /*= true*/)
{
wil::com_ptr_nothrow<ID3D12RootSignature> rs = g_d3d12_context->CreateRootSignature(&m_desc);
if (!rs)
return {};
if (clear)
Clear();
return rs;
}
void RootSignatureBuilder::SetInputAssemblerFlag()
{
m_desc.Flags |= D3D12_ROOT_SIGNATURE_FLAG_ALLOW_INPUT_ASSEMBLER_INPUT_LAYOUT;
}
u32 RootSignatureBuilder::Add32BitConstants(u32 shader_reg, u32 num_values, D3D12_SHADER_VISIBILITY visibility)
{
const u32 index = m_desc.NumParameters++;
m_params[index].ParameterType = D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
m_params[index].ShaderVisibility = visibility;
m_params[index].Constants.ShaderRegister = shader_reg;
m_params[index].Constants.RegisterSpace = 0;
m_params[index].Constants.Num32BitValues = num_values;
return index;
}
u32 RootSignatureBuilder::AddCBVParameter(u32 shader_reg, D3D12_SHADER_VISIBILITY visibility)
{
const u32 index = m_desc.NumParameters++;
m_params[index].ParameterType = D3D12_ROOT_PARAMETER_TYPE_CBV;
m_params[index].ShaderVisibility = visibility;
m_params[index].Descriptor.ShaderRegister = shader_reg;
m_params[index].Descriptor.RegisterSpace = 0;
return index;
}
u32 RootSignatureBuilder::AddSRVParameter(u32 shader_reg, D3D12_SHADER_VISIBILITY visibility)
{
const u32 index = m_desc.NumParameters++;
m_params[index].ParameterType = D3D12_ROOT_PARAMETER_TYPE_SRV;
m_params[index].ShaderVisibility = visibility;
m_params[index].Descriptor.ShaderRegister = shader_reg;
m_params[index].Descriptor.RegisterSpace = 0;
return index;
}
u32 RootSignatureBuilder::AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE rt, u32 start_shader_reg, u32 num_shader_regs,
D3D12_SHADER_VISIBILITY visibility)
{
const u32 index = m_desc.NumParameters++;
const u32 dr_index = m_num_descriptor_ranges++;
m_descriptor_ranges[dr_index].RangeType = rt;
m_descriptor_ranges[dr_index].NumDescriptors = num_shader_regs;
m_descriptor_ranges[dr_index].BaseShaderRegister = start_shader_reg;
m_descriptor_ranges[dr_index].RegisterSpace = 0;
m_descriptor_ranges[dr_index].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
m_params[index].ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
m_params[index].DescriptorTable.pDescriptorRanges = &m_descriptor_ranges[dr_index];
m_params[index].DescriptorTable.NumDescriptorRanges = 1;
m_params[index].ShaderVisibility = visibility;
return index;
}

121
common/D3D12/Builders.h Normal file
View File

@ -0,0 +1,121 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/RedtapeWindows.h"
#include <array>
#include <d3d12.h>
#include <wil/com.h>
namespace D3D12
{
class ShaderCache;
class RootSignatureBuilder
{
public:
enum : u32
{
MAX_PARAMETERS = 16,
MAX_DESCRIPTOR_RANGES = 16
};
RootSignatureBuilder();
void Clear();
wil::com_ptr_nothrow<ID3D12RootSignature> Create(bool clear = true);
void SetInputAssemblerFlag();
u32 Add32BitConstants(u32 shader_reg, u32 num_values, D3D12_SHADER_VISIBILITY visibility);
u32 AddCBVParameter(u32 shader_reg, D3D12_SHADER_VISIBILITY visibility);
u32 AddSRVParameter(u32 shader_reg, D3D12_SHADER_VISIBILITY visibility);
u32 AddDescriptorTable(D3D12_DESCRIPTOR_RANGE_TYPE rt, u32 start_shader_reg, u32 num_shader_regs,
D3D12_SHADER_VISIBILITY visibility);
private:
D3D12_ROOT_SIGNATURE_DESC m_desc{};
std::array<D3D12_ROOT_PARAMETER, MAX_PARAMETERS> m_params{};
std::array<D3D12_DESCRIPTOR_RANGE, MAX_DESCRIPTOR_RANGES> m_descriptor_ranges{};
u32 m_num_descriptor_ranges = 0;
};
class GraphicsPipelineBuilder
{
public:
enum : u32
{
MAX_VERTEX_ATTRIBUTES = 16,
};
GraphicsPipelineBuilder();
~GraphicsPipelineBuilder() = default;
void Clear();
wil::com_ptr_nothrow<ID3D12PipelineState> Create(ID3D12Device* device, bool clear = true);
wil::com_ptr_nothrow<ID3D12PipelineState> Create(ID3D12Device* device, ShaderCache& cache, bool clear = true);
void SetRootSignature(ID3D12RootSignature* rs);
void SetVertexShader(const void* data, u32 data_size);
void SetGeometryShader(const void* data, u32 data_size);
void SetPixelShader(const void* data, u32 data_size);
void SetVertexShader(const ID3DBlob* blob);
void SetGeometryShader(const ID3DBlob* blob);
void SetPixelShader(const ID3DBlob* blob);
void AddVertexAttribute(const char* semantic_name, u32 semantic_index, DXGI_FORMAT format, u32 buffer, u32 offset);
void SetPrimitiveTopologyType(D3D12_PRIMITIVE_TOPOLOGY_TYPE type);
void SetRasterizationState(D3D12_FILL_MODE polygon_mode, D3D12_CULL_MODE cull_mode, bool front_face_ccw);
void SetMultisamples(u32 multisamples);
void SetNoCullRasterizationState();
void SetDepthState(bool depth_test, bool depth_write, D3D12_COMPARISON_FUNC compare_op);
void SetStencilState(bool stencil_test, u8 read_mask, u8 write_mask, const D3D12_DEPTH_STENCILOP_DESC& front, const D3D12_DEPTH_STENCILOP_DESC& back);
void SetNoDepthTestState();
void SetNoStencilState();
void SetBlendState(u32 rt, bool blend_enable, D3D12_BLEND src_factor, D3D12_BLEND dst_factor, D3D12_BLEND_OP op,
D3D12_BLEND alpha_src_factor, D3D12_BLEND alpha_dst_factor, D3D12_BLEND_OP alpha_op,
u8 write_mask = D3D12_COLOR_WRITE_ENABLE_ALL);
void SetNoBlendingState();
void ClearRenderTargets();
void SetRenderTarget(u32 rt, DXGI_FORMAT format);
void ClearDepthStencilFormat();
void SetDepthStencilFormat(DXGI_FORMAT format);
private:
D3D12_GRAPHICS_PIPELINE_STATE_DESC m_desc{};
std::array<D3D12_INPUT_ELEMENT_DESC, MAX_VERTEX_ATTRIBUTES> m_input_elements{};
};
} // namespace D3D12

679
common/D3D12/Context.cpp Normal file
View File

@ -0,0 +1,679 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/Context.h"
#include "common/Assertions.h"
#include "common/ScopedGuard.h"
#include "common/Console.h"
#include "D3D12MemAlloc.h"
#include <algorithm>
#include <array>
#include <dxgi1_4.h>
#include <queue>
#include <vector>
std::unique_ptr<D3D12::Context> g_d3d12_context;
using namespace D3D12;
#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP)
// Private D3D12 state
static HMODULE s_d3d12_library;
static PFN_D3D12_CREATE_DEVICE s_d3d12_create_device;
static PFN_D3D12_GET_DEBUG_INTERFACE s_d3d12_get_debug_interface;
static PFN_D3D12_SERIALIZE_ROOT_SIGNATURE s_d3d12_serialize_root_signature;
static bool LoadD3D12Library()
{
if ((s_d3d12_library = LoadLibraryW(L"d3d12.dll")) == nullptr ||
(s_d3d12_create_device =
reinterpret_cast<PFN_D3D12_CREATE_DEVICE>(GetProcAddress(s_d3d12_library, "D3D12CreateDevice"))) == nullptr ||
(s_d3d12_get_debug_interface = reinterpret_cast<PFN_D3D12_GET_DEBUG_INTERFACE>(
GetProcAddress(s_d3d12_library, "D3D12GetDebugInterface"))) == nullptr ||
(s_d3d12_serialize_root_signature = reinterpret_cast<PFN_D3D12_SERIALIZE_ROOT_SIGNATURE>(
GetProcAddress(s_d3d12_library, "D3D12SerializeRootSignature"))) == nullptr)
{
Console.Error("d3d12.dll could not be loaded.");
s_d3d12_create_device = nullptr;
s_d3d12_get_debug_interface = nullptr;
s_d3d12_serialize_root_signature = nullptr;
if (s_d3d12_library)
FreeLibrary(s_d3d12_library);
s_d3d12_library = nullptr;
return false;
}
return true;
}
static void UnloadD3D12Library()
{
s_d3d12_serialize_root_signature = nullptr;
s_d3d12_get_debug_interface = nullptr;
s_d3d12_create_device = nullptr;
if (s_d3d12_library)
{
FreeLibrary(s_d3d12_library);
s_d3d12_library = nullptr;
}
}
#else
static const PFN_D3D12_CREATE_DEVICE s_d3d12_create_device = D3D12CreateDevice;
static const PFN_D3D12_GET_DEBUG_INTERFACE s_d3d12_get_debug_interface = D3D12GetDebugInterface;
static const PFN_D3D12_SERIALIZE_ROOT_SIGNATURE s_d3d12_serialize_root_signature = D3D12SerializeRootSignature;
static bool LoadD3D12Library()
{
return true;
}
static void UnloadD3D12Library() {}
#endif
Context::Context() = default;
Context::~Context()
{
DestroyResources();
}
Context::ComPtr<ID3DBlob> Context::SerializeRootSignature(const D3D12_ROOT_SIGNATURE_DESC* desc)
{
ComPtr<ID3DBlob> blob;
ComPtr<ID3DBlob> error_blob;
const HRESULT hr = s_d3d12_serialize_root_signature(desc, D3D_ROOT_SIGNATURE_VERSION_1, blob.put(),
error_blob.put());
if (FAILED(hr))
{
Console.Error("D3D12SerializeRootSignature() failed: %08X", hr);
if (error_blob)
Console.Error("%s", error_blob->GetBufferPointer());
return {};
}
return blob;
}
D3D12::Context::ComPtr<ID3D12RootSignature> Context::CreateRootSignature(const D3D12_ROOT_SIGNATURE_DESC* desc)
{
ComPtr<ID3DBlob> blob = SerializeRootSignature(desc);
if (!blob)
return {};
ComPtr<ID3D12RootSignature> rs;
const HRESULT hr =
m_device->CreateRootSignature(0, blob->GetBufferPointer(), blob->GetBufferSize(), IID_PPV_ARGS(rs.put()));
if (FAILED(hr))
{
Console.Error("CreateRootSignature() failed: %08X", hr);
return {};
}
return rs;
}
bool Context::SupportsTextureFormat(DXGI_FORMAT format)
{
constexpr u32 required = D3D12_FORMAT_SUPPORT1_TEXTURE2D | D3D12_FORMAT_SUPPORT1_SHADER_SAMPLE;
D3D12_FEATURE_DATA_FORMAT_SUPPORT support = {format};
return SUCCEEDED(m_device->CheckFeatureSupport(D3D12_FEATURE_FORMAT_SUPPORT, &support, sizeof(support))) &&
(support.Support1 & required) == required;
}
bool Context::Create(IDXGIFactory* dxgi_factory, u32 adapter_index, bool enable_debug_layer)
{
pxAssertRel(!g_d3d12_context, "No context exists");
if (!LoadD3D12Library())
return false;
g_d3d12_context.reset(new Context());
if (!g_d3d12_context->CreateDevice(dxgi_factory, adapter_index, enable_debug_layer) ||
!g_d3d12_context->CreateCommandQueue() || !g_d3d12_context->CreateAllocator() ||
!g_d3d12_context->CreateFence() || !g_d3d12_context->CreateDescriptorHeaps() ||
!g_d3d12_context->CreateCommandLists() || !g_d3d12_context->CreateTimestampQuery() ||
!g_d3d12_context->CreateTextureStreamBuffer())
{
Destroy();
return false;
}
return true;
}
void Context::Destroy()
{
if (g_d3d12_context)
g_d3d12_context.reset();
UnloadD3D12Library();
}
u32 Context::GetAdapterVendorID() const
{
if (!m_adapter)
return 0;
DXGI_ADAPTER_DESC desc;
if (FAILED(m_adapter->GetDesc(&desc)))
return 0;
return desc.VendorId;
}
bool Context::CreateDevice(IDXGIFactory* dxgi_factory, u32 adapter_index, bool enable_debug_layer)
{
ComPtr<IDXGIAdapter> adapter;
HRESULT hr = dxgi_factory->EnumAdapters(adapter_index, &adapter);
if (FAILED(hr))
{
Console.Error("Adapter %u not found, using default", adapter_index);
adapter = nullptr;
}
else
{
DXGI_ADAPTER_DESC adapter_desc;
if (SUCCEEDED(adapter->GetDesc(&adapter_desc)))
{
char adapter_name_buffer[128];
const int name_length = WideCharToMultiByte(CP_UTF8, 0, adapter_desc.Description,
static_cast<int>(std::wcslen(adapter_desc.Description)),
adapter_name_buffer, std::size(adapter_name_buffer), 0, nullptr);
if (name_length >= 0)
{
adapter_name_buffer[name_length] = 0;
Console.WriteLn("D3D Adapter: %s", adapter_name_buffer);
}
}
}
// Enabling the debug layer will fail if the Graphics Tools feature is not installed.
if (enable_debug_layer)
{
hr = s_d3d12_get_debug_interface(IID_PPV_ARGS(&m_debug_interface));
if (SUCCEEDED(hr))
{
m_debug_interface->EnableDebugLayer();
}
else
{
Console.Error("Debug layer requested but not available.");
enable_debug_layer = false;
}
}
// Create the actual device.
hr = s_d3d12_create_device(adapter.get(), D3D_FEATURE_LEVEL_11_0, IID_PPV_ARGS(&m_device));
pxAssertRel(SUCCEEDED(hr), "Create D3D12 device");
if (FAILED(hr))
return false;
// get adapter
ComPtr<IDXGIFactory4> dxgi_factory4;
if (SUCCEEDED(dxgi_factory->QueryInterface<IDXGIFactory4>(dxgi_factory4.put())))
{
const LUID luid(m_device->GetAdapterLuid());
if (FAILED(dxgi_factory4->EnumAdapterByLuid(luid, IID_PPV_ARGS(m_adapter.put()))))
Console.Error("Failed to get lookup adapter by device LUID");
}
if (enable_debug_layer)
{
ComPtr<ID3D12InfoQueue> info_queue = m_device.try_query<ID3D12InfoQueue>();
if (info_queue)
{
if (IsDebuggerPresent())
{
info_queue->SetBreakOnSeverity(D3D12_MESSAGE_SEVERITY_ERROR, TRUE);
info_queue->SetBreakOnSeverity(D3D12_MESSAGE_SEVERITY_WARNING, TRUE);
}
D3D12_INFO_QUEUE_FILTER filter = {};
std::array<D3D12_MESSAGE_ID, 5> id_list{
D3D12_MESSAGE_ID_CLEARRENDERTARGETVIEW_MISMATCHINGCLEARVALUE,
D3D12_MESSAGE_ID_CLEARDEPTHSTENCILVIEW_MISMATCHINGCLEARVALUE,
D3D12_MESSAGE_ID_CREATEGRAPHICSPIPELINESTATE_RENDERTARGETVIEW_NOT_SET,
D3D12_MESSAGE_ID_CREATEINPUTLAYOUT_TYPE_MISMATCH,
D3D12_MESSAGE_ID_DRAW_EMPTY_SCISSOR_RECTANGLE,
};
filter.DenyList.NumIDs = static_cast<UINT>(id_list.size());
filter.DenyList.pIDList = id_list.data();
info_queue->PushStorageFilter(&filter);
}
}
return true;
}
bool Context::CreateCommandQueue()
{
const D3D12_COMMAND_QUEUE_DESC queue_desc = {D3D12_COMMAND_LIST_TYPE_DIRECT, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL,
D3D12_COMMAND_QUEUE_FLAG_NONE};
HRESULT hr = m_device->CreateCommandQueue(&queue_desc, IID_PPV_ARGS(&m_command_queue));
pxAssertRel(SUCCEEDED(hr), "Create command queue");
return SUCCEEDED(hr);
}
bool Context::CreateAllocator()
{
D3D12MA::ALLOCATOR_DESC allocatorDesc = {};
allocatorDesc.pDevice = m_device.get();
allocatorDesc.pAdapter = m_adapter.get();
allocatorDesc.Flags = D3D12MA::ALLOCATOR_FLAG_SINGLETHREADED | D3D12MA::ALLOCATOR_FLAG_DEFAULT_POOLS_NOT_ZEROED /* | D3D12MA::ALLOCATOR_FLAG_ALWAYS_COMMITTED*/;
const HRESULT hr = D3D12MA::CreateAllocator(&allocatorDesc, m_allocator.put());
if (FAILED(hr))
{
Console.Error("D3D12MA::CreateAllocator() failed with HRESULT %08X", hr);
return false;
}
return true;
}
bool Context::CreateFence()
{
HRESULT hr = m_device->CreateFence(m_completed_fence_value, D3D12_FENCE_FLAG_NONE, IID_PPV_ARGS(&m_fence));
pxAssertRel(SUCCEEDED(hr), "Create fence");
if (FAILED(hr))
return false;
m_fence_event = CreateEvent(nullptr, FALSE, FALSE, nullptr);
pxAssertRel(m_fence_event != NULL, "Create fence event");
if (!m_fence_event)
return false;
return true;
}
bool Context::CreateDescriptorHeaps()
{
static constexpr size_t MAX_SRVS = 32768;
static constexpr size_t MAX_RTVS = 16384;
static constexpr size_t MAX_DSVS = 16384;
static constexpr size_t MAX_CPU_SAMPLERS = 1024;
if (!m_descriptor_heap_manager.Create(m_device.get(), D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, MAX_SRVS, false) ||
!m_rtv_heap_manager.Create(m_device.get(), D3D12_DESCRIPTOR_HEAP_TYPE_RTV, MAX_RTVS, false) ||
!m_dsv_heap_manager.Create(m_device.get(), D3D12_DESCRIPTOR_HEAP_TYPE_DSV, MAX_DSVS, false) ||
!m_sampler_heap_manager.Create(m_device.get(), D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER, MAX_CPU_SAMPLERS, false))
{
return false;
}
// Allocate null SRV descriptor for unbound textures.
constexpr D3D12_SHADER_RESOURCE_VIEW_DESC null_srv_desc = {DXGI_FORMAT_R8G8B8A8_UNORM, D3D12_SRV_DIMENSION_TEXTURE2D,
D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING};
if (!m_descriptor_heap_manager.Allocate(&m_null_srv_descriptor))
{
pxFailRel("Failed to allocate null descriptor");
return false;
}
m_device->CreateShaderResourceView(nullptr, &null_srv_desc, m_null_srv_descriptor.cpu_handle);
return true;
}
bool Context::CreateCommandLists()
{
static constexpr size_t MAX_GPU_SRVS = 32768;
static constexpr size_t MAX_GPU_SAMPLERS = 2048;
for (u32 i = 0; i < NUM_COMMAND_LISTS; i++)
{
CommandListResources& res = m_command_lists[i];
HRESULT hr;
for (u32 i = 0; i < 2; i++)
{
hr = m_device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT,
IID_PPV_ARGS(res.command_allocators[i].put()));
pxAssertRel(SUCCEEDED(hr), "Create command allocator");
if (FAILED(hr))
return false;
hr = m_device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT,
res.command_allocators[i].get(), nullptr,
IID_PPV_ARGS(res.command_lists[i].put()));
if (FAILED(hr))
{
Console.Error("Failed to create command list: %08X", hr);
return false;
}
// Close the command lists, since the first thing we do is reset them.
hr = res.command_lists[i]->Close();
pxAssertRel(SUCCEEDED(hr), "Closing new command list failed");
if (FAILED(hr))
return false;
}
if (!res.descriptor_allocator.Create(m_device.get(), D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, MAX_GPU_SRVS))
{
Console.Error("Failed to create per frame descriptor allocator");
return false;
}
if (!res.sampler_allocator.Create(m_device.get(), MAX_GPU_SAMPLERS))
{
Console.Error("Failed to create per frame sampler allocator");
return false;
}
}
MoveToNextCommandList();
return true;
}
bool Context::CreateTextureStreamBuffer()
{
return m_texture_stream_buffer.Create(TEXTURE_UPLOAD_BUFFER_SIZE);
}
void Context::MoveToNextCommandList()
{
m_current_command_list = (m_current_command_list + 1) % NUM_COMMAND_LISTS;
m_current_fence_value++;
// We may have to wait if this command list hasn't finished on the GPU.
CommandListResources& res = m_command_lists[m_current_command_list];
WaitForFence(res.ready_fence_value);
res.ready_fence_value = m_current_fence_value;
res.init_command_list_used = false;
// Begin command list.
res.command_allocators[1]->Reset();
res.command_lists[1]->Reset(res.command_allocators[1].get(), nullptr);
res.descriptor_allocator.Reset();
if (res.sampler_allocator.ShouldReset())
res.sampler_allocator.Reset();
if (res.has_timestamp_query)
{
// readback timestamp from the last time this cmdlist was used.
// we don't need to worry about disjoint in dx12, the frequency is reliable within a single cmdlist.
const u32 offset = (m_current_command_list * (sizeof(u64) * NUM_TIMESTAMP_QUERIES_PER_CMDLIST));
const D3D12_RANGE read_range = {offset, offset + (sizeof(u64) * NUM_TIMESTAMP_QUERIES_PER_CMDLIST)};
void* map;
HRESULT hr = m_timestamp_query_buffer->Map(0, &read_range, &map);
if (SUCCEEDED(hr))
{
u64 timestamps[2];
std::memcpy(timestamps, static_cast<const u8*>(map) + offset, sizeof(timestamps));
m_accumulated_gpu_time += static_cast<float>(static_cast<double>(timestamps[1] - timestamps[0]) / m_timestamp_frequency);
const D3D12_RANGE write_range = {};
m_timestamp_query_buffer->Unmap(0, &write_range);
}
else
{
Console.Warning("Map() for timestamp query failed: %08X", hr);
}
}
res.has_timestamp_query = m_gpu_timing_enabled;
if (m_gpu_timing_enabled)
{
res.command_lists[1]->EndQuery(m_timestamp_query_heap.get(), D3D12_QUERY_TYPE_TIMESTAMP,
m_current_command_list * NUM_TIMESTAMP_QUERIES_PER_CMDLIST);
}
ID3D12DescriptorHeap* heaps[2] = {res.descriptor_allocator.GetDescriptorHeap(), res.sampler_allocator.GetDescriptorHeap()};
res.command_lists[1]->SetDescriptorHeaps(std::size(heaps), heaps);
m_allocator->SetCurrentFrameIndex(static_cast<UINT>(m_current_fence_value));
}
ID3D12GraphicsCommandList4* Context::GetInitCommandList()
{
CommandListResources& res = m_command_lists[m_current_command_list];
if (!res.init_command_list_used)
{
HRESULT hr = res.command_allocators[0]->Reset();
pxAssertMsg(SUCCEEDED(hr), "Reset init command allocator failed");
res.command_lists[0]->Reset(res.command_allocators[0].get(), nullptr);
pxAssertMsg(SUCCEEDED(hr), "Reset init command list failed");
res.init_command_list_used = true;
}
return res.command_lists[0].get();
}
void Context::ExecuteCommandList(bool wait_for_completion)
{
CommandListResources& res = m_command_lists[m_current_command_list];
HRESULT hr;
if (res.has_timestamp_query)
{
// write the timestamp back at the end of the cmdlist
res.command_lists[1]->EndQuery(m_timestamp_query_heap.get(), D3D12_QUERY_TYPE_TIMESTAMP,
(m_current_command_list * NUM_TIMESTAMP_QUERIES_PER_CMDLIST) + 1);
res.command_lists[1]->ResolveQueryData(m_timestamp_query_heap.get(), D3D12_QUERY_TYPE_TIMESTAMP,
m_current_command_list * NUM_TIMESTAMP_QUERIES_PER_CMDLIST, NUM_TIMESTAMP_QUERIES_PER_CMDLIST,
m_timestamp_query_buffer.get(), m_current_command_list * (sizeof(u64) * NUM_TIMESTAMP_QUERIES_PER_CMDLIST));
}
if (res.init_command_list_used)
{
hr = res.command_lists[0]->Close();
pxAssertRel(SUCCEEDED(hr), "Close init command list");
}
// Close and queue command list.
hr = res.command_lists[1]->Close();
pxAssertRel(SUCCEEDED(hr), "Close command list");
if (res.init_command_list_used)
{
const std::array<ID3D12CommandList*, 2> execute_lists{res.command_lists[0].get(), res.command_lists[1].get()};
m_command_queue->ExecuteCommandLists(static_cast<UINT>(execute_lists.size()), execute_lists.data());
}
else
{
const std::array<ID3D12CommandList*, 1> execute_lists{res.command_lists[1].get()};
m_command_queue->ExecuteCommandLists(static_cast<UINT>(execute_lists.size()), execute_lists.data());
}
// Update fence when GPU has completed.
hr = m_command_queue->Signal(m_fence.get(), m_current_fence_value);
pxAssertRel(SUCCEEDED(hr), "Signal fence");
MoveToNextCommandList();
if (wait_for_completion)
WaitForFence(res.ready_fence_value);
}
void Context::InvalidateSamplerGroups()
{
for (CommandListResources& res : m_command_lists)
res.sampler_allocator.InvalidateCache();
}
void Context::DeferObjectDestruction(ID3D12DeviceChild* resource)
{
if (!resource)
return;
resource->AddRef();
m_command_lists[m_current_command_list].pending_resources.emplace_back(nullptr, resource);
}
void Context::DeferResourceDestruction(D3D12MA::Allocation* allocation, ID3D12Resource* resource)
{
if (!resource)
return;
if (allocation)
allocation->AddRef();
resource->AddRef();
m_command_lists[m_current_command_list].pending_resources.emplace_back(allocation, resource);
}
void Context::DeferDescriptorDestruction(DescriptorHeapManager& manager, u32 index)
{
m_command_lists[m_current_command_list].pending_descriptors.emplace_back(manager, index);
}
void Context::DeferDescriptorDestruction(DescriptorHeapManager& manager, DescriptorHandle* handle)
{
if (handle->index == DescriptorHandle::INVALID_INDEX)
return;
m_command_lists[m_current_command_list].pending_descriptors.emplace_back(manager, handle->index);
handle->Clear();
}
void Context::DestroyPendingResources(CommandListResources& cmdlist)
{
for (const auto& dd : cmdlist.pending_descriptors)
dd.first.Free(dd.second);
cmdlist.pending_descriptors.clear();
for (const auto& it : cmdlist.pending_resources)
{
if (it.first)
it.first->Release();
it.second->Release();
}
cmdlist.pending_resources.clear();
}
void Context::DestroyResources()
{
ExecuteCommandList(true);
m_texture_stream_buffer.Destroy(false);
m_descriptor_heap_manager.Free(&m_null_srv_descriptor);
m_timestamp_query_buffer.reset();
m_timestamp_query_allocation.reset();
m_sampler_heap_manager.Destroy();
m_dsv_heap_manager.Destroy();
m_rtv_heap_manager.Destroy();
m_descriptor_heap_manager.Destroy();
m_command_lists = {};
m_current_command_list = 0;
m_completed_fence_value = 0;
m_current_fence_value = 0;
if (m_fence_event)
{
CloseHandle(m_fence_event);
m_fence_event = {};
}
m_allocator.reset();
m_command_queue.reset();
m_debug_interface.reset();
m_device.reset();
}
void Context::WaitForFence(u64 fence)
{
if (m_completed_fence_value >= fence)
return;
// Try non-blocking check.
m_completed_fence_value = m_fence->GetCompletedValue();
if (m_completed_fence_value < fence)
{
// Fall back to event.
HRESULT hr = m_fence->SetEventOnCompletion(fence, m_fence_event);
pxAssertRel(SUCCEEDED(hr), "Set fence event on completion");
WaitForSingleObject(m_fence_event, INFINITE);
m_completed_fence_value = m_fence->GetCompletedValue();
}
// Release resources for as many command lists which have completed.
u32 index = (m_current_command_list + 1) % NUM_COMMAND_LISTS;
for (u32 i = 0; i < NUM_COMMAND_LISTS; i++)
{
CommandListResources& res = m_command_lists[index];
if (m_completed_fence_value < res.ready_fence_value)
break;
DestroyPendingResources(res);
index = (index + 1) % NUM_COMMAND_LISTS;
}
}
void Context::WaitForGPUIdle()
{
u32 index = (m_current_command_list + 1) % NUM_COMMAND_LISTS;
for (u32 i = 0; i < (NUM_COMMAND_LISTS - 1); i++)
{
WaitForFence(m_command_lists[index].ready_fence_value);
index = (index + 1) % NUM_COMMAND_LISTS;
}
}
bool Context::CreateTimestampQuery()
{
constexpr u32 QUERY_COUNT = NUM_TIMESTAMP_QUERIES_PER_CMDLIST * NUM_COMMAND_LISTS;
constexpr u32 BUFFER_SIZE = sizeof(u64) * QUERY_COUNT;
const D3D12_QUERY_HEAP_DESC desc = {D3D12_QUERY_HEAP_TYPE_TIMESTAMP, QUERY_COUNT};
HRESULT hr = m_device->CreateQueryHeap(&desc, IID_PPV_ARGS(m_timestamp_query_heap.put()));
if (FAILED(hr))
{
Console.Error("CreateQueryHeap() for timestamp failed with %08X", hr);
return false;
}
const D3D12MA::ALLOCATION_DESC allocation_desc = {D3D12MA::ALLOCATION_FLAG_NONE, D3D12_HEAP_TYPE_READBACK};
const D3D12_RESOURCE_DESC resource_desc = {
D3D12_RESOURCE_DIMENSION_BUFFER, 0, BUFFER_SIZE, 1, 1, 1, DXGI_FORMAT_UNKNOWN, {1, 0}, D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
D3D12_RESOURCE_FLAG_NONE};
hr = m_allocator->CreateResource(&allocation_desc, &resource_desc, D3D12_RESOURCE_STATE_COPY_DEST, nullptr,
m_timestamp_query_allocation.put(), IID_PPV_ARGS(m_timestamp_query_buffer.put()));
if (FAILED(hr))
{
Console.Error("CreateResource() for timestamp failed with %08X", hr);
return false;
}
u64 frequency;
hr = m_command_queue->GetTimestampFrequency(&frequency);
if (FAILED(hr))
{
Console.Error("GetTimestampFrequency() failed: %08X", hr);
return false;
}
m_timestamp_frequency = static_cast<double>(frequency) / 1000.0;
return true;
}
float Context::GetAndResetAccumulatedGPUTime()
{
const float time = m_accumulated_gpu_time;
m_accumulated_gpu_time = 0.0f;
return time;
}
void Context::SetEnableGPUTiming(bool enabled)
{
m_gpu_timing_enabled = enabled;
}

207
common/D3D12/Context.h Normal file
View File

@ -0,0 +1,207 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/RedtapeWindows.h"
#include "common/D3D12/DescriptorHeapManager.h"
#include "common/D3D12/StreamBuffer.h"
#include <array>
#include <d3d12.h>
#include <memory>
#include <vector>
#include <wil/com.h>
struct IDXGIAdapter;
struct IDXGIFactory;
namespace D3D12MA
{
class Allocator;
class Allocation;
} // namespace D3D12MA
namespace D3D12
{
class Context
{
public:
template <typename T>
using ComPtr = wil::com_ptr_nothrow<T>;
enum : u32
{
/// Number of command lists. One is being built while the other(s) are executed.
NUM_COMMAND_LISTS = 3,
/// Textures that don't fit into this buffer will be uploaded with a staging buffer.
TEXTURE_UPLOAD_BUFFER_SIZE = 64 * 1024 * 1024,
/// Maximum number of samples in a single allocation group.
SAMPLER_GROUP_SIZE = 2,
/// Start/End timestamp queries.
NUM_TIMESTAMP_QUERIES_PER_CMDLIST = 2,
};
~Context();
/// Creates new device and context.
static bool Create(IDXGIFactory* dxgi_factory, u32 adapter_index, bool enable_debug_layer);
/// Destroys active context.
static void Destroy();
__fi IDXGIAdapter* GetAdapter() const { return m_adapter.get(); }
__fi ID3D12Device* GetDevice() const { return m_device.get(); }
__fi ID3D12CommandQueue* GetCommandQueue() const { return m_command_queue.get(); }
__fi D3D12MA::Allocator* GetAllocator() const { return m_allocator.get(); }
/// Returns the PCI vendor ID of the device, if known.
u32 GetAdapterVendorID() const;
/// Returns the current command list, commands can be recorded directly.
ID3D12GraphicsCommandList4* GetCommandList() const
{
return m_command_lists[m_current_command_list].command_lists[1].get();
}
/// Returns the init command list for uploading.
ID3D12GraphicsCommandList4* GetInitCommandList();
/// Returns the per-frame SRV/CBV/UAV allocator.
DescriptorAllocator& GetDescriptorAllocator()
{
return m_command_lists[m_current_command_list].descriptor_allocator;
}
/// Returns the per-frame sampler allocator.
GroupedSamplerAllocator<SAMPLER_GROUP_SIZE>& GetSamplerAllocator()
{
return m_command_lists[m_current_command_list].sampler_allocator;
}
/// Invalidates GPU-side sampler caches for all command lists. Call after you've freed samplers,
/// and are going to re-use the handles from GetSamplerHeapManager().
void InvalidateSamplerGroups();
// Descriptor manager access.
DescriptorHeapManager& GetDescriptorHeapManager() { return m_descriptor_heap_manager; }
DescriptorHeapManager& GetRTVHeapManager() { return m_rtv_heap_manager; }
DescriptorHeapManager& GetDSVHeapManager() { return m_dsv_heap_manager; }
DescriptorHeapManager& GetSamplerHeapManager() { return m_sampler_heap_manager; }
const DescriptorHandle& GetNullSRVDescriptor() const { return m_null_srv_descriptor; }
StreamBuffer& GetTextureStreamBuffer() { return m_texture_stream_buffer; }
// Root signature access.
ComPtr<ID3DBlob> SerializeRootSignature(const D3D12_ROOT_SIGNATURE_DESC* desc);
ComPtr<ID3D12RootSignature> CreateRootSignature(const D3D12_ROOT_SIGNATURE_DESC* desc);
/// Fence value for current command list.
u64 GetCurrentFenceValue() const { return m_current_fence_value; }
/// Last "completed" fence.
u64 GetCompletedFenceValue() const { return m_completed_fence_value; }
/// Feature level to use when compiling shaders.
D3D_FEATURE_LEVEL GetFeatureLevel() const { return m_feature_level; }
/// Test for support for the specified texture format.
bool SupportsTextureFormat(DXGI_FORMAT format);
/// Executes the current command list.
void ExecuteCommandList(bool wait_for_completion);
/// Waits for a specific fence.
void WaitForFence(u64 fence);
/// Waits for any in-flight command buffers to complete.
void WaitForGPUIdle();
/// Defers destruction of a D3D resource (associates it with the current list).
void DeferObjectDestruction(ID3D12DeviceChild* resource);
/// Defers destruction of a D3D resource (associates it with the current list).
void DeferResourceDestruction(D3D12MA::Allocation* allocation, ID3D12Resource* resource);
/// Defers destruction of a descriptor handle (associates it with the current list).
void DeferDescriptorDestruction(DescriptorHeapManager& manager, u32 index);
void DeferDescriptorDestruction(DescriptorHeapManager& manager, DescriptorHandle* handle);
float GetAndResetAccumulatedGPUTime();
void SetEnableGPUTiming(bool enabled);
private:
struct CommandListResources
{
std::array<ComPtr<ID3D12CommandAllocator>, 2> command_allocators;
std::array<ComPtr<ID3D12GraphicsCommandList4>, 2> command_lists;
DescriptorAllocator descriptor_allocator;
GroupedSamplerAllocator<SAMPLER_GROUP_SIZE> sampler_allocator;
std::vector<std::pair<D3D12MA::Allocation*, ID3D12DeviceChild*>> pending_resources;
std::vector<std::pair<DescriptorHeapManager&, u32>> pending_descriptors;
u64 ready_fence_value = 0;
bool init_command_list_used = false;
bool has_timestamp_query = false;
};
Context();
bool CreateDevice(IDXGIFactory* dxgi_factory, u32 adapter_index, bool enable_debug_layer);
bool CreateCommandQueue();
bool CreateAllocator();
bool CreateFence();
bool CreateDescriptorHeaps();
bool CreateCommandLists();
bool CreateTextureStreamBuffer();
bool CreateTimestampQuery();
void MoveToNextCommandList();
void DestroyPendingResources(CommandListResources& cmdlist);
void DestroyResources();
ComPtr<IDXGIAdapter> m_adapter;
ComPtr<ID3D12Debug> m_debug_interface;
ComPtr<ID3D12Device> m_device;
ComPtr<ID3D12CommandQueue> m_command_queue;
ComPtr<D3D12MA::Allocator> m_allocator;
ComPtr<ID3D12Fence> m_fence;
HANDLE m_fence_event = {};
u32 m_current_fence_value = 0;
u64 m_completed_fence_value = 0;
std::array<CommandListResources, NUM_COMMAND_LISTS> m_command_lists;
u32 m_current_command_list = NUM_COMMAND_LISTS - 1;
ComPtr<ID3D12QueryHeap> m_timestamp_query_heap;
ComPtr<ID3D12Resource> m_timestamp_query_buffer;
ComPtr<D3D12MA::Allocation> m_timestamp_query_allocation;
double m_timestamp_frequency = 0.0;
float m_accumulated_gpu_time = 0.0f;
bool m_gpu_timing_enabled = false;
DescriptorHeapManager m_descriptor_heap_manager;
DescriptorHeapManager m_rtv_heap_manager;
DescriptorHeapManager m_dsv_heap_manager;
DescriptorHeapManager m_sampler_heap_manager;
DescriptorHandle m_null_srv_descriptor;
StreamBuffer m_texture_stream_buffer;
D3D_FEATURE_LEVEL m_feature_level = D3D_FEATURE_LEVEL_11_0;
};
} // namespace D3D12
extern std::unique_ptr<D3D12::Context> g_d3d12_context;

View File

@ -0,0 +1,161 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/DescriptorHeapManager.h"
#include "common/Assertions.h"
using namespace D3D12;
DescriptorHeapManager::DescriptorHeapManager() = default;
DescriptorHeapManager::~DescriptorHeapManager() = default;
bool DescriptorHeapManager::Create(ID3D12Device* device, D3D12_DESCRIPTOR_HEAP_TYPE type, u32 num_descriptors,
bool shader_visible)
{
D3D12_DESCRIPTOR_HEAP_DESC desc = {type, static_cast<UINT>(num_descriptors),
shader_visible ? D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE :
D3D12_DESCRIPTOR_HEAP_FLAG_NONE};
HRESULT hr = device->CreateDescriptorHeap(&desc, IID_PPV_ARGS(m_descriptor_heap.put()));
pxAssertRel(SUCCEEDED(hr), "Create descriptor heap");
if (FAILED(hr))
return false;
m_heap_base_cpu = m_descriptor_heap->GetCPUDescriptorHandleForHeapStart();
m_heap_base_gpu = m_descriptor_heap->GetGPUDescriptorHandleForHeapStart();
m_num_descriptors = num_descriptors;
m_descriptor_increment_size = device->GetDescriptorHandleIncrementSize(type);
// Set all slots to unallocated (1)
const u32 bitset_count = num_descriptors / BITSET_SIZE + (((num_descriptors % BITSET_SIZE) != 0) ? 1 : 0);
m_free_slots.resize(bitset_count);
for (BitSetType& bs : m_free_slots)
bs.flip();
return true;
}
void DescriptorHeapManager::Destroy()
{
for (BitSetType& bs : m_free_slots)
{
pxAssert(bs.all());
}
m_num_descriptors = 0;
m_descriptor_increment_size = 0;
m_heap_base_cpu = {};
m_heap_base_gpu = {};
m_descriptor_heap.reset();
m_free_slots.clear();
}
bool DescriptorHeapManager::Allocate(DescriptorHandle* handle)
{
// Start past the temporary slots, no point in searching those.
for (u32 group = 0; group < m_free_slots.size(); group++)
{
BitSetType& bs = m_free_slots[group];
if (bs.none())
continue;
u32 bit = 0;
for (; bit < BITSET_SIZE; bit++)
{
if (bs[bit])
break;
}
u32 index = group * BITSET_SIZE + bit;
bs[bit] = false;
handle->index = index;
handle->cpu_handle.ptr = m_heap_base_cpu.ptr + index * m_descriptor_increment_size;
handle->gpu_handle.ptr = m_heap_base_gpu.ptr + index * m_descriptor_increment_size;
return true;
}
pxFailRel("Out of fixed descriptors");
return false;
}
void DescriptorHeapManager::Free(u32 index)
{
pxAssert(index < m_num_descriptors);
u32 group = index / BITSET_SIZE;
u32 bit = index % BITSET_SIZE;
m_free_slots[group][bit] = true;
}
void DescriptorHeapManager::Free(DescriptorHandle* handle)
{
if (handle->index == DescriptorHandle::INVALID_INDEX)
return;
Free(handle->index);
handle->Clear();
}
DescriptorAllocator::DescriptorAllocator() = default;
DescriptorAllocator::~DescriptorAllocator() = default;
bool DescriptorAllocator::Create(ID3D12Device* device, D3D12_DESCRIPTOR_HEAP_TYPE type,
u32 num_descriptors)
{
const D3D12_DESCRIPTOR_HEAP_DESC desc = {type, static_cast<UINT>(num_descriptors),
D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE};
const HRESULT hr = device->CreateDescriptorHeap(&desc, IID_PPV_ARGS(&m_descriptor_heap));
pxAssertRel(SUCCEEDED(hr), "Creating descriptor heap for linear allocator");
if (FAILED(hr))
return false;
m_num_descriptors = num_descriptors;
m_descriptor_increment_size = device->GetDescriptorHandleIncrementSize(type);
m_heap_base_cpu = m_descriptor_heap->GetCPUDescriptorHandleForHeapStart();
m_heap_base_gpu = m_descriptor_heap->GetGPUDescriptorHandleForHeapStart();
return true;
}
void DescriptorAllocator::Destroy()
{
m_descriptor_heap.reset();
m_descriptor_increment_size = 0;
m_num_descriptors = 0;
m_current_offset = 0;
m_heap_base_cpu = {};
m_heap_base_gpu = {};
}
bool DescriptorAllocator::Allocate(u32 num_handles, DescriptorHandle* out_base_handle)
{
if ((m_current_offset + num_handles) > m_num_descriptors)
return false;
out_base_handle->index = m_current_offset;
out_base_handle->cpu_handle.ptr =
m_heap_base_cpu.ptr + m_current_offset * m_descriptor_increment_size;
out_base_handle->gpu_handle.ptr =
m_heap_base_gpu.ptr + m_current_offset * m_descriptor_increment_size;
m_current_offset += num_handles;
return true;
}
void DescriptorAllocator::Reset()
{
m_current_offset = 0;
}

View File

@ -0,0 +1,268 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/HashCombine.h"
#include "common/RedtapeWindows.h"
#include <bitset>
#include <cstring>
#include <d3d12.h>
#include <unordered_map>
#include <vector>
#include <wil/com.h>
namespace D3D12
{
// This class provides an abstraction for D3D12 descriptor heaps.
struct DescriptorHandle final
{
enum : u32
{
INVALID_INDEX = 0xFFFFFFFF
};
D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle{};
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle{};
u32 index = INVALID_INDEX;
__fi operator bool() const { return index != INVALID_INDEX; }
__fi operator D3D12_CPU_DESCRIPTOR_HANDLE() const { return cpu_handle; }
__fi operator D3D12_GPU_DESCRIPTOR_HANDLE() const { return gpu_handle; }
__fi bool operator==(const DescriptorHandle& rhs) const { return (index == rhs.index); }
__fi bool operator!=(const DescriptorHandle& rhs) const { return (index != rhs.index); }
__fi bool operator<(const DescriptorHandle& rhs) const { return (index < rhs.index); }
__fi bool operator<=(const DescriptorHandle& rhs) const { return (index <= rhs.index); }
__fi bool operator>(const DescriptorHandle& rhs) const { return (index > rhs.index); }
__fi bool operator>=(const DescriptorHandle& rhs) const { return (index >= rhs.index); }
__fi void Clear()
{
cpu_handle = {};
gpu_handle = {};
index = INVALID_INDEX;
}
};
class DescriptorHeapManager final
{
public:
DescriptorHeapManager();
~DescriptorHeapManager();
ID3D12DescriptorHeap* GetDescriptorHeap() const { return m_descriptor_heap.get(); }
u32 GetDescriptorIncrementSize() const { return m_descriptor_increment_size; }
bool Create(ID3D12Device* device, D3D12_DESCRIPTOR_HEAP_TYPE type, u32 num_descriptors, bool shader_visible);
void Destroy();
bool Allocate(DescriptorHandle* handle);
void Free(DescriptorHandle* handle);
void Free(u32 index);
private:
wil::com_ptr_nothrow<ID3D12DescriptorHeap> m_descriptor_heap;
u32 m_num_descriptors = 0;
u32 m_descriptor_increment_size = 0;
D3D12_CPU_DESCRIPTOR_HANDLE m_heap_base_cpu = {};
D3D12_GPU_DESCRIPTOR_HANDLE m_heap_base_gpu = {};
static constexpr u32 BITSET_SIZE = 1024;
using BitSetType = std::bitset<BITSET_SIZE>;
std::vector<BitSetType> m_free_slots = {};
};
class DescriptorAllocator
{
public:
DescriptorAllocator();
~DescriptorAllocator();
__fi ID3D12DescriptorHeap* GetDescriptorHeap() const { return m_descriptor_heap.get(); }
__fi u32 GetDescriptorIncrementSize() const { return m_descriptor_increment_size; }
bool Create(ID3D12Device* device, D3D12_DESCRIPTOR_HEAP_TYPE type, u32 num_descriptors);
void Destroy();
bool Allocate(u32 num_handles, DescriptorHandle* out_base_handle);
void Reset();
private:
wil::com_ptr_nothrow<ID3D12DescriptorHeap> m_descriptor_heap;
u32 m_descriptor_increment_size = 0;
u32 m_num_descriptors = 0;
u32 m_current_offset = 0;
D3D12_CPU_DESCRIPTOR_HANDLE m_heap_base_cpu = {};
D3D12_GPU_DESCRIPTOR_HANDLE m_heap_base_gpu = {};
};
template <u32 NumSamplers>
class GroupedSamplerAllocator : private DescriptorAllocator
{
struct Key
{
u32 idx[NumSamplers];
__fi bool operator==(const Key& rhs) const
{
return (std::memcmp(idx, rhs.idx, sizeof(idx)) == 0);
}
__fi bool operator!=(const Key& rhs) const
{
return (std::memcmp(idx, rhs.idx, sizeof(idx)) != 0);
}
};
struct KeyHash
{
__fi std::size_t operator()(const Key& key) const
{
size_t seed = 0;
for (u32 key : key.idx)
HashCombine(seed, key);
return seed;
}
};
public:
GroupedSamplerAllocator();
~GroupedSamplerAllocator();
using DescriptorAllocator::GetDescriptorHeap;
using DescriptorAllocator::GetDescriptorIncrementSize;
bool Create(ID3D12Device* device, u32 num_descriptors);
void Destroy();
bool LookupSingle(DescriptorHandle* gpu_handle, const DescriptorHandle& cpu_handle);
bool LookupGroup(DescriptorHandle* gpu_handle, const DescriptorHandle* cpu_handles);
// Clears cache but doesn't reset allocator.
void InvalidateCache();
void Reset();
bool ShouldReset() const;
private:
wil::com_ptr_nothrow<ID3D12Device> m_device;
std::unordered_map<Key, D3D12::DescriptorHandle, KeyHash> m_groups;
};
template <u32 NumSamplers>
GroupedSamplerAllocator<NumSamplers>::GroupedSamplerAllocator() = default;
template <u32 NumSamplers>
GroupedSamplerAllocator<NumSamplers>::~GroupedSamplerAllocator() = default;
template <u32 NumSamplers>
bool GroupedSamplerAllocator<NumSamplers>::Create(ID3D12Device* device, u32 num_descriptors)
{
if (!DescriptorAllocator::Create(device, D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER, num_descriptors))
return false;
m_device = device;
return true;
}
template <u32 NumSamplers>
void GroupedSamplerAllocator<NumSamplers>::Destroy()
{
DescriptorAllocator::Destroy();
m_device.reset();
}
template <u32 NumSamplers>
void GroupedSamplerAllocator<NumSamplers>::Reset()
{
m_groups.clear();
DescriptorAllocator::Reset();
}
template <u32 NumSamplers>
void GroupedSamplerAllocator<NumSamplers>::InvalidateCache()
{
m_groups.clear();
}
template <u32 NumSamplers>
bool GroupedSamplerAllocator<NumSamplers>::LookupSingle(DescriptorHandle* gpu_handle, const DescriptorHandle& cpu_handle)
{
Key key;
key.idx[0] = cpu_handle.index;
for (u32 i = 1; i < NumSamplers; i++)
key.idx[i] = 0;
auto it = m_groups.find(key);
if (it != m_groups.end())
{
*gpu_handle = it->second;
return true;
}
if (!Allocate(1, gpu_handle))
return false;
m_device->CopyDescriptorsSimple(1, *gpu_handle, cpu_handle, D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER);
m_groups.emplace(key, *gpu_handle);
return true;
}
template <u32 NumSamplers>
bool GroupedSamplerAllocator<NumSamplers>::LookupGroup(DescriptorHandle* gpu_handle, const DescriptorHandle* cpu_handles)
{
Key key;
for (u32 i = 0; i < NumSamplers; i++)
key.idx[i] = cpu_handles[i].index;
auto it = m_groups.find(key);
if (it != m_groups.end())
{
*gpu_handle = it->second;
return true;
}
if (!Allocate(NumSamplers, gpu_handle))
return false;
D3D12_CPU_DESCRIPTOR_HANDLE dst_handle = *gpu_handle;
UINT dst_size = NumSamplers;
D3D12_CPU_DESCRIPTOR_HANDLE src_handles[NumSamplers];
UINT src_sizes[NumSamplers];
for (u32 i = 0; i < NumSamplers; i++)
{
src_handles[i] = cpu_handles[i];
src_sizes[i] = 1;
}
m_device->CopyDescriptors(1, &dst_handle, &dst_size, NumSamplers, src_handles, src_sizes, D3D12_DESCRIPTOR_HEAP_TYPE_SAMPLER);
m_groups.emplace(key, *gpu_handle);
return true;
}
template <u32 NumSamplers>
bool GroupedSamplerAllocator<NumSamplers>::ShouldReset() const
{
// We only reset the sampler heap if more than half of the descriptors are used.
// This saves descriptor copying when there isn't a large number of sampler configs per frame.
return m_groups.size() >= (D3D12_MAX_SHADER_VISIBLE_SAMPLER_HEAP_SIZE / 2);
}
} // namespace D3D12

View File

@ -0,0 +1,565 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/ShaderCache.h"
#include "common/D3D11/ShaderCompiler.h"
#include "common/FileSystem.h"
#include "common/Console.h"
#include "common/MD5Digest.h"
#include <d3dcompiler.h>
#ifdef _UWP
#include <winrt/Windows.System.Profile.h>
#endif
using namespace D3D12;
#pragma pack(push, 1)
struct CacheIndexEntry
{
u64 source_hash_low;
u64 source_hash_high;
u64 macro_hash_low;
u64 macro_hash_high;
u64 entry_point_low;
u64 entry_point_high;
u32 source_length;
u32 shader_type;
u32 file_offset;
u32 blob_size;
};
#pragma pack(pop)
static bool CanUsePipelineCache()
{
#ifdef _UWP
// GetCachedBlob crashes on XBox UWP for some reason...
const auto version_info = winrt::Windows::System::Profile::AnalyticsInfo::VersionInfo();
const auto device_family = version_info.DeviceFamily();
return (device_family != L"Windows.Xbox");
#else
return true;
#endif
}
ShaderCache::ShaderCache()
: m_use_pipeline_cache(CanUsePipelineCache())
{
}
ShaderCache::~ShaderCache()
{
if (m_pipeline_index_file)
std::fclose(m_pipeline_index_file);
if (m_pipeline_blob_file)
std::fclose(m_pipeline_blob_file);
if (m_shader_index_file)
std::fclose(m_shader_index_file);
if (m_shader_blob_file)
std::fclose(m_shader_blob_file);
}
bool ShaderCache::CacheIndexKey::operator==(const CacheIndexKey& key) const
{
return (source_hash_low == key.source_hash_low && source_hash_high == key.source_hash_high &&
macro_hash_low == key.macro_hash_low && macro_hash_high == key.macro_hash_high &&
entry_point_low == key.entry_point_low && entry_point_high == key.entry_point_high &&
type == key.type && source_length == key.source_length);
}
bool ShaderCache::CacheIndexKey::operator!=(const CacheIndexKey& key) const
{
return (source_hash_low != key.source_hash_low || source_hash_high != key.source_hash_high ||
macro_hash_low != key.macro_hash_low || macro_hash_high != key.macro_hash_high ||
entry_point_low != key.entry_point_low || entry_point_high != key.entry_point_high ||
type != key.type || source_length != key.source_length);
}
bool ShaderCache::Open(std::string_view base_path, D3D_FEATURE_LEVEL feature_level, u32 version, bool debug)
{
m_base_path = base_path;
m_feature_level = feature_level;
m_data_version = version;
m_debug = debug;
bool result = true;
if (!base_path.empty())
{
const std::string base_shader_filename = GetCacheBaseFileName(base_path, "shaders", feature_level, debug);
const std::string shader_index_filename = base_shader_filename + ".idx";
const std::string shader_blob_filename = base_shader_filename + ".bin";
if (!ReadExisting(shader_index_filename, shader_blob_filename, m_shader_index_file, m_shader_blob_file,
m_shader_index))
{
result = CreateNew(shader_index_filename, shader_blob_filename, m_shader_index_file, m_shader_blob_file);
}
if (m_use_pipeline_cache && result)
{
const std::string base_pipelines_filename = GetCacheBaseFileName(base_path, "pipelines", feature_level, debug);
const std::string pipelines_index_filename = base_pipelines_filename + ".idx";
const std::string pipelines_blob_filename = base_pipelines_filename + ".bin";
if (!ReadExisting(pipelines_index_filename, pipelines_blob_filename, m_pipeline_index_file, m_pipeline_blob_file,
m_pipeline_index))
{
result = CreateNew(pipelines_index_filename, pipelines_blob_filename, m_pipeline_index_file, m_pipeline_blob_file);
}
}
}
return result;
}
void ShaderCache::InvalidatePipelineCache()
{
m_pipeline_index.clear();
if (m_pipeline_blob_file)
{
std::fclose(m_pipeline_blob_file);
m_pipeline_blob_file = nullptr;
}
if (m_pipeline_index_file)
{
std::fclose(m_pipeline_index_file);
m_pipeline_index_file = nullptr;
}
if (m_use_pipeline_cache)
{
const std::string base_pipelines_filename =
GetCacheBaseFileName(m_base_path, "pipelines", m_feature_level, m_debug);
const std::string pipelines_index_filename = base_pipelines_filename + ".idx";
const std::string pipelines_blob_filename = base_pipelines_filename + ".bin";
CreateNew(pipelines_index_filename, pipelines_blob_filename, m_pipeline_index_file, m_pipeline_blob_file);
}
}
bool ShaderCache::CreateNew(const std::string& index_filename, const std::string& blob_filename, std::FILE*& index_file,
std::FILE*& blob_file)
{
if (FileSystem::FileExists(index_filename.c_str()))
{
Console.Warning("Removing existing index file '%s'", index_filename.c_str());
FileSystem::DeleteFilePath(index_filename.c_str());
}
if (FileSystem::FileExists(blob_filename.c_str()))
{
Console.Warning("Removing existing blob file '%s'", blob_filename.c_str());
FileSystem::DeleteFilePath(blob_filename.c_str());
}
index_file = FileSystem::OpenCFile(index_filename.c_str(), "wb");
if (!index_file)
{
Console.Error("Failed to open index file '%s' for writing", index_filename.c_str());
return false;
}
const u32 index_version = FILE_VERSION;
if (std::fwrite(&index_version, sizeof(index_version), 1, index_file) != 1 ||
std::fwrite(&m_data_version, sizeof(m_data_version), 1, index_file) != 1)
{
Console.Error("Failed to write version to index file '%s'", index_filename.c_str());
std::fclose(index_file);
index_file = nullptr;
FileSystem::DeleteFilePath(index_filename.c_str());
return false;
}
blob_file = FileSystem::OpenCFile(blob_filename.c_str(), "w+b");
if (!blob_file)
{
Console.Error("Failed to open blob file '%s' for writing", blob_filename.c_str());
std::fclose(blob_file);
blob_file = nullptr;
FileSystem::DeleteFilePath(index_filename.c_str());
return false;
}
return true;
}
bool ShaderCache::ReadExisting(const std::string& index_filename, const std::string& blob_filename,
std::FILE*& index_file, std::FILE*& blob_file, CacheIndex& index)
{
index_file = FileSystem::OpenCFile(index_filename.c_str(), "r+b");
if (!index_file)
{
// special case here: when there's a sharing violation (i.e. two instances running),
// we don't want to blow away the cache. so just continue without a cache.
if (errno == EACCES)
{
Console.WriteLn("Failed to open shader cache index with EACCES, are you running two instances?");
return true;
}
return false;
}
u32 file_version;
u32 data_version;
if (std::fread(&file_version, sizeof(file_version), 1, index_file) != 1 || file_version != FILE_VERSION ||
std::fread(&data_version, sizeof(data_version), 1, index_file) != 1 || data_version != m_data_version)
{
Console.Error("Bad file version in '%s'", index_filename.c_str());
std::fclose(index_file);
index_file = nullptr;
return false;
}
blob_file = FileSystem::OpenCFile(blob_filename.c_str(), "a+b");
if (!blob_file)
{
Console.Error("Blob file '%s' is missing", blob_filename.c_str());
std::fclose(index_file);
index_file = nullptr;
return false;
}
std::fseek(blob_file, 0, SEEK_END);
const u32 blob_file_size = static_cast<u32>(std::ftell(blob_file));
for (;;)
{
CacheIndexEntry entry;
if (std::fread(&entry, sizeof(entry), 1, index_file) != 1 || (entry.file_offset + entry.blob_size) > blob_file_size)
{
if (std::feof(index_file))
break;
Console.Error("Failed to read entry from '%s', corrupt file?", index_filename.c_str());
index.clear();
std::fclose(blob_file);
blob_file = nullptr;
std::fclose(index_file);
index_file = nullptr;
return false;
}
const CacheIndexKey key{
entry.source_hash_low, entry.source_hash_high,
entry.macro_hash_low, entry.macro_hash_high,
entry.entry_point_low, entry.entry_point_high,
entry.source_length, static_cast<EntryType>(entry.shader_type)};
const CacheIndexData data{entry.file_offset, entry.blob_size};
index.emplace(key, data);
}
// ensure we don't write before seeking
std::fseek(index_file, 0, SEEK_END);
DevCon.WriteLn("Read %zu entries from '%s'", index.size(), index_filename.c_str());
return true;
}
std::string ShaderCache::GetCacheBaseFileName(const std::string_view& base_path, const std::string_view& type,
D3D_FEATURE_LEVEL feature_level, bool debug)
{
std::string base_filename(base_path);
base_filename += FS_OSPATH_SEPARATOR_STR "d3d12_";
base_filename += type;
base_filename += "_";
switch (feature_level)
{
case D3D_FEATURE_LEVEL_10_0:
base_filename += "sm40";
break;
case D3D_FEATURE_LEVEL_10_1:
base_filename += "sm41";
break;
case D3D_FEATURE_LEVEL_11_0:
base_filename += "sm50";
break;
default:
base_filename += "unk";
break;
}
if (debug)
base_filename += "_debug";
return base_filename;
}
union MD5Hash
{
struct
{
u64 low;
u64 high;
};
u8 hash[16];
};
ShaderCache::CacheIndexKey ShaderCache::GetShaderCacheKey(EntryType type, const std::string_view& shader_code,
const D3D_SHADER_MACRO* macros, const char* entry_point)
{
union
{
struct
{
u64 hash_low;
u64 hash_high;
};
u8 hash[16];
};
CacheIndexKey key = {};
key.type = type;
MD5Digest digest;
digest.Update(shader_code.data(), static_cast<u32>(shader_code.length()));
digest.Final(hash);
key.source_hash_low = hash_low;
key.source_hash_high = hash_high;
key.source_length = static_cast<u32>(shader_code.length());
if (macros)
{
digest.Reset();
for (const D3D_SHADER_MACRO* macro = macros; macro->Name != nullptr; macro++)
{
digest.Update(macro->Name, std::strlen(macro->Name));
digest.Update(macro->Definition, std::strlen(macro->Definition));
}
digest.Final(hash);
key.macro_hash_low = hash_low;
key.macro_hash_high = hash_high;
}
digest.Reset();
digest.Update(entry_point, static_cast<u32>(std::strlen(entry_point)));
digest.Final(hash);
key.entry_point_low = hash_low;
key.entry_point_high = hash_high;
return key;
}
ShaderCache::CacheIndexKey ShaderCache::GetPipelineCacheKey(const D3D12_GRAPHICS_PIPELINE_STATE_DESC& gpdesc)
{
MD5Digest digest;
u32 length = sizeof(D3D12_GRAPHICS_PIPELINE_STATE_DESC);
if (gpdesc.VS.BytecodeLength > 0)
{
digest.Update(gpdesc.VS.pShaderBytecode, static_cast<u32>(gpdesc.VS.BytecodeLength));
length += static_cast<u32>(gpdesc.VS.BytecodeLength);
}
if (gpdesc.GS.BytecodeLength > 0)
{
digest.Update(gpdesc.GS.pShaderBytecode, static_cast<u32>(gpdesc.GS.BytecodeLength));
length += static_cast<u32>(gpdesc.GS.BytecodeLength);
}
if (gpdesc.PS.BytecodeLength > 0)
{
digest.Update(gpdesc.PS.pShaderBytecode, static_cast<u32>(gpdesc.PS.BytecodeLength));
length += static_cast<u32>(gpdesc.PS.BytecodeLength);
}
digest.Update(&gpdesc.BlendState, sizeof(gpdesc.BlendState));
digest.Update(&gpdesc.SampleMask, sizeof(gpdesc.SampleMask));
digest.Update(&gpdesc.RasterizerState, sizeof(gpdesc.RasterizerState));
digest.Update(&gpdesc.DepthStencilState, sizeof(gpdesc.DepthStencilState));
for (u32 i = 0; i < gpdesc.InputLayout.NumElements; i++)
{
const D3D12_INPUT_ELEMENT_DESC& ie = gpdesc.InputLayout.pInputElementDescs[i];
digest.Update(ie.SemanticName, static_cast<u32>(std::strlen(ie.SemanticName)));
digest.Update(&ie.SemanticIndex, sizeof(ie.SemanticIndex));
digest.Update(&ie.Format, sizeof(ie.Format));
digest.Update(&ie.InputSlot, sizeof(ie.InputSlot));
digest.Update(&ie.AlignedByteOffset, sizeof(ie.AlignedByteOffset));
digest.Update(&ie.InputSlotClass, sizeof(ie.InputSlotClass));
digest.Update(&ie.InstanceDataStepRate, sizeof(ie.InstanceDataStepRate));
length += sizeof(D3D12_INPUT_ELEMENT_DESC);
}
digest.Update(&gpdesc.IBStripCutValue, sizeof(gpdesc.IBStripCutValue));
digest.Update(&gpdesc.PrimitiveTopologyType, sizeof(gpdesc.PrimitiveTopologyType));
digest.Update(&gpdesc.NumRenderTargets, sizeof(gpdesc.NumRenderTargets));
digest.Update(gpdesc.RTVFormats, sizeof(gpdesc.RTVFormats));
digest.Update(&gpdesc.DSVFormat, sizeof(gpdesc.DSVFormat));
digest.Update(&gpdesc.SampleDesc, sizeof(gpdesc.SampleDesc));
digest.Update(&gpdesc.Flags, sizeof(gpdesc.Flags));
MD5Hash h;
digest.Final(h.hash);
return CacheIndexKey{h.low, h.high, 0, 0, 0, 0, length, EntryType::GraphicsPipeline};
}
ShaderCache::ComPtr<ID3DBlob> ShaderCache::GetShaderBlob(EntryType type, std::string_view shader_code,
const D3D_SHADER_MACRO* macros /* = nullptr */, const char* entry_point /* = "main" */)
{
const auto key = GetShaderCacheKey(type, shader_code, macros, entry_point);
auto iter = m_shader_index.find(key);
if (iter == m_shader_index.end())
return CompileAndAddShaderBlob(key, shader_code, macros, entry_point);
ComPtr<ID3DBlob> blob;
HRESULT hr = D3DCreateBlob(iter->second.blob_size, blob.put());
if (FAILED(hr) || std::fseek(m_shader_blob_file, iter->second.file_offset, SEEK_SET) != 0 ||
std::fread(blob->GetBufferPointer(), 1, iter->second.blob_size, m_shader_blob_file) != iter->second.blob_size)
{
Console.Error("Read blob from file failed");
return {};
}
return blob;
}
ShaderCache::ComPtr<ID3D12PipelineState> ShaderCache::GetPipelineState(ID3D12Device* device,
const D3D12_GRAPHICS_PIPELINE_STATE_DESC& desc)
{
const auto key = GetPipelineCacheKey(desc);
auto iter = m_pipeline_index.find(key);
if (iter == m_pipeline_index.end())
return CompileAndAddPipeline(device, key, desc);
ComPtr<ID3DBlob> blob;
HRESULT hr = D3DCreateBlob(iter->second.blob_size, blob.put());
if (FAILED(hr) || std::fseek(m_pipeline_blob_file, iter->second.file_offset, SEEK_SET) != 0 ||
std::fread(blob->GetBufferPointer(), 1, iter->second.blob_size, m_pipeline_blob_file) != iter->second.blob_size)
{
Console.Error("Read blob from file failed");
return {};
}
D3D12_GRAPHICS_PIPELINE_STATE_DESC desc_with_blob(desc);
desc_with_blob.CachedPSO.pCachedBlob = blob->GetBufferPointer();
desc_with_blob.CachedPSO.CachedBlobSizeInBytes = blob->GetBufferSize();
ComPtr<ID3D12PipelineState> pso;
hr = device->CreateGraphicsPipelineState(&desc_with_blob, IID_PPV_ARGS(pso.put()));
if (FAILED(hr))
{
Console.Warning("Creating cached PSO failed: %08X. Invalidating cache.", hr);
InvalidatePipelineCache();
pso = CompileAndAddPipeline(device, key, desc);
}
return pso;
}
ShaderCache::ComPtr<ID3DBlob> ShaderCache::CompileAndAddShaderBlob(const CacheIndexKey& key, std::string_view shader_code,
const D3D_SHADER_MACRO* macros, const char* entry_point)
{
ComPtr<ID3DBlob> blob;
switch (key.type)
{
case EntryType::VertexShader:
blob = D3D11::ShaderCompiler::CompileShader(D3D11::ShaderCompiler::Type::Vertex, m_feature_level, m_debug, shader_code, macros, entry_point);
break;
case EntryType::GeometryShader:
blob = D3D11::ShaderCompiler::CompileShader(D3D11::ShaderCompiler::Type::Geometry, m_feature_level, m_debug, shader_code, macros, entry_point);
break;
case EntryType::PixelShader:
blob = D3D11::ShaderCompiler::CompileShader(D3D11::ShaderCompiler::Type::Pixel, m_feature_level, m_debug, shader_code, macros, entry_point);
break;
default:
break;
}
if (!blob)
return {};
if (!m_shader_blob_file || std::fseek(m_shader_blob_file, 0, SEEK_END) != 0)
return blob;
CacheIndexData data;
data.file_offset = static_cast<u32>(std::ftell(m_shader_blob_file));
data.blob_size = static_cast<u32>(blob->GetBufferSize());
CacheIndexEntry entry = {};
entry.source_hash_low = key.source_hash_low;
entry.source_hash_high = key.source_hash_high;
entry.macro_hash_low = key.macro_hash_low;
entry.macro_hash_high = key.macro_hash_high;
entry.entry_point_low = key.entry_point_low;
entry.entry_point_high = key.entry_point_high;
entry.source_length = key.source_length;
entry.shader_type = static_cast<u32>(key.type);
entry.blob_size = data.blob_size;
entry.file_offset = data.file_offset;
if (std::fwrite(blob->GetBufferPointer(), 1, entry.blob_size, m_shader_blob_file) != entry.blob_size ||
std::fflush(m_shader_blob_file) != 0 || std::fwrite(&entry, sizeof(entry), 1, m_shader_index_file) != 1 ||
std::fflush(m_shader_index_file) != 0)
{
Console.Error("Failed to write shader blob to file");
return blob;
}
m_shader_index.emplace(key, data);
return blob;
}
ShaderCache::ComPtr<ID3D12PipelineState>
ShaderCache::CompileAndAddPipeline(ID3D12Device* device, const CacheIndexKey& key,
const D3D12_GRAPHICS_PIPELINE_STATE_DESC& gpdesc)
{
ComPtr<ID3D12PipelineState> pso;
HRESULT hr = device->CreateGraphicsPipelineState(&gpdesc, IID_PPV_ARGS(pso.put()));
if (FAILED(hr))
{
Console.Error("Creating cached PSO failed: %08X", hr);
return {};
}
if (!m_pipeline_blob_file || std::fseek(m_pipeline_blob_file, 0, SEEK_END) != 0)
return pso;
ComPtr<ID3DBlob> blob;
hr = pso->GetCachedBlob(blob.put());
if (FAILED(hr))
{
Console.Warning("Failed to get cached PSO data: %08X", hr);
return pso;
}
CacheIndexData data;
data.file_offset = static_cast<u32>(std::ftell(m_pipeline_blob_file));
data.blob_size = static_cast<u32>(blob->GetBufferSize());
CacheIndexEntry entry = {};
entry.source_hash_low = key.source_hash_low;
entry.source_hash_high = key.source_hash_high;
entry.source_length = key.source_length;
entry.shader_type = static_cast<u32>(key.type);
entry.blob_size = data.blob_size;
entry.file_offset = data.file_offset;
if (std::fwrite(blob->GetBufferPointer(), 1, entry.blob_size, m_pipeline_blob_file) != entry.blob_size ||
std::fflush(m_pipeline_blob_file) != 0 || std::fwrite(&entry, sizeof(entry), 1, m_pipeline_index_file) != 1 ||
std::fflush(m_pipeline_index_file) != 0)
{
Console.Error("Failed to write pipeline blob to file");
return pso;
}
m_shader_index.emplace(key, data);
return pso;
}

152
common/D3D12/ShaderCache.h Normal file
View File

@ -0,0 +1,152 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/HashCombine.h"
#include "common/RedtapeWindows.h"
#include <cstdio>
#include <d3d12.h>
#include <string_view>
#include <unordered_map>
#include <vector>
#include <wil/com.h>
namespace D3D12
{
class ShaderCache
{
public:
template <typename T>
using ComPtr = wil::com_ptr_nothrow<T>;
enum class EntryType
{
VertexShader,
GeometryShader,
PixelShader,
ComputeShader,
GraphicsPipeline,
};
ShaderCache();
~ShaderCache();
__fi D3D_FEATURE_LEVEL GetFeatureLevel() const { return m_feature_level; }
__fi u32 GetDataVersion() const { return m_data_version; }
__fi bool UsingPipelineCache() const { return m_use_pipeline_cache; }
__fi bool UsingDebugShaders() const { return m_debug; }
bool Open(std::string_view base_path, D3D_FEATURE_LEVEL feature_level, u32 version, bool debug);
__fi ComPtr<ID3DBlob> GetVertexShader(std::string_view shader_code,
const D3D_SHADER_MACRO* macros = nullptr, const char* entry_point = "main")
{
return GetShaderBlob(EntryType::VertexShader, shader_code, macros, entry_point);
}
__fi ComPtr<ID3DBlob> GetGeometryShader(std::string_view shader_code,
const D3D_SHADER_MACRO* macros = nullptr, const char* entry_point = "main")
{
return GetShaderBlob(EntryType::GeometryShader, shader_code, macros, entry_point);
}
__fi ComPtr<ID3DBlob> GetPixelShader(std::string_view shader_code,
const D3D_SHADER_MACRO* macros = nullptr, const char* entry_point = "main")
{
return GetShaderBlob(EntryType::PixelShader, shader_code, macros, entry_point);
}
__fi ComPtr<ID3DBlob> GetComputeShader(std::string_view shader_code,
const D3D_SHADER_MACRO* macros = nullptr, const char* entry_point = "main")
{
return GetShaderBlob(EntryType::ComputeShader, shader_code, macros, entry_point);
}
ComPtr<ID3DBlob> GetShaderBlob(EntryType type, std::string_view shader_code,
const D3D_SHADER_MACRO* macros = nullptr, const char* entry_point = "main");
ComPtr<ID3D12PipelineState> GetPipelineState(ID3D12Device* device, const D3D12_GRAPHICS_PIPELINE_STATE_DESC& desc);
private:
static constexpr u32 FILE_VERSION = 1;
struct CacheIndexKey
{
u64 source_hash_low;
u64 source_hash_high;
u64 macro_hash_low;
u64 macro_hash_high;
u64 entry_point_low;
u64 entry_point_high;
u32 source_length;
EntryType type;
bool operator==(const CacheIndexKey& key) const;
bool operator!=(const CacheIndexKey& key) const;
};
struct CacheIndexEntryHasher
{
std::size_t operator()(const CacheIndexKey& e) const noexcept
{
std::size_t h = 0;
HashCombine(h, e.entry_point_low, e.entry_point_high, e.macro_hash_low, e.macro_hash_high,
e.source_hash_low, e.source_hash_high, e.source_length, e.type);
return h;
}
};
struct CacheIndexData
{
u32 file_offset;
u32 blob_size;
};
using CacheIndex = std::unordered_map<CacheIndexKey, CacheIndexData, CacheIndexEntryHasher>;
static std::string GetCacheBaseFileName(const std::string_view& base_path, const std::string_view& type,
D3D_FEATURE_LEVEL feature_level, bool debug);
static CacheIndexKey GetShaderCacheKey(EntryType type, const std::string_view& shader_code,
const D3D_SHADER_MACRO* macros, const char* entry_point);
static CacheIndexKey GetPipelineCacheKey(const D3D12_GRAPHICS_PIPELINE_STATE_DESC& gpdesc);
bool CreateNew(const std::string& index_filename, const std::string& blob_filename, std::FILE*& index_file,
std::FILE*& blob_file);
bool ReadExisting(const std::string& index_filename, const std::string& blob_filename, std::FILE*& index_file,
std::FILE*& blob_file, CacheIndex& index);
void InvalidatePipelineCache();
void Close();
ComPtr<ID3DBlob> CompileAndAddShaderBlob(const CacheIndexKey& key, std::string_view shader_code,
const D3D_SHADER_MACRO* macros, const char* entry_point);
ComPtr<ID3D12PipelineState> CompileAndAddPipeline(ID3D12Device* device, const CacheIndexKey& key,
const D3D12_GRAPHICS_PIPELINE_STATE_DESC& gpdesc);
std::string m_base_path;
std::FILE* m_shader_index_file = nullptr;
std::FILE* m_shader_blob_file = nullptr;
CacheIndex m_shader_index;
std::FILE* m_pipeline_index_file = nullptr;
std::FILE* m_pipeline_blob_file = nullptr;
CacheIndex m_pipeline_index;
D3D_FEATURE_LEVEL m_feature_level = D3D_FEATURE_LEVEL_11_0;
u32 m_data_version = 0;
bool m_use_pipeline_cache = false;
bool m_debug = false;
};
} // namespace D3D12

View File

@ -0,0 +1,282 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/StreamBuffer.h"
#include "common/D3D12/Context.h"
#include "common/Align.h"
#include "common/Assertions.h"
#include "common/Console.h"
#include "D3D12MemAlloc.h"
#include <algorithm>
#include <functional>
using namespace D3D12;
StreamBuffer::StreamBuffer() = default;
StreamBuffer::~StreamBuffer()
{
Destroy();
}
bool StreamBuffer::Create(u32 size)
{
const D3D12_RESOURCE_DESC resource_desc = {
D3D12_RESOURCE_DIMENSION_BUFFER, 0, size, 1, 1, 1, DXGI_FORMAT_UNKNOWN, {1, 0}, D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
D3D12_RESOURCE_FLAG_NONE};
D3D12MA::ALLOCATION_DESC allocationDesc = {};
allocationDesc.Flags = D3D12MA::ALLOCATION_FLAG_COMMITTED;
allocationDesc.HeapType = D3D12_HEAP_TYPE_UPLOAD;
wil::com_ptr_nothrow<ID3D12Resource> buffer;
wil::com_ptr_nothrow<D3D12MA::Allocation> allocation;
HRESULT hr = g_d3d12_context->GetAllocator()->CreateResource(&allocationDesc,
&resource_desc, D3D12_RESOURCE_STATE_GENERIC_READ,
nullptr, allocation.put(), IID_PPV_ARGS(buffer.put()));
pxAssertMsg(SUCCEEDED(hr), "Allocate buffer");
if (FAILED(hr))
return false;
static const D3D12_RANGE read_range = {};
u8* host_pointer;
hr = buffer->Map(0, &read_range, reinterpret_cast<void**>(&host_pointer));
pxAssertMsg(SUCCEEDED(hr), "Map buffer");
if (FAILED(hr))
return false;
Destroy(true);
m_buffer = std::move(buffer);
m_allocation = std::move(allocation);
m_host_pointer = host_pointer;
m_size = size;
m_gpu_pointer = m_buffer->GetGPUVirtualAddress();
return true;
}
bool StreamBuffer::ReserveMemory(u32 num_bytes, u32 alignment)
{
const u32 required_bytes = num_bytes + alignment;
// Check for sane allocations
if (num_bytes > m_size)
{
Console.Error("Attempting to allocate %u bytes from a %u byte stream buffer", static_cast<u32>(num_bytes),
static_cast<u32>(m_size));
pxFailRel("Stream buffer overflow");
return false;
}
// Is the GPU behind or up to date with our current offset?
UpdateCurrentFencePosition();
if (m_current_offset >= m_current_gpu_position)
{
const u32 aligned_required_bytes = (m_current_offset > 0) ? required_bytes : num_bytes;
const u32 remaining_bytes = m_size - m_current_offset;
if (aligned_required_bytes <= remaining_bytes)
{
// Place at the current position, after the GPU position.
m_current_offset = Common::AlignUp(m_current_offset, alignment);
m_current_space = m_size - m_current_offset;
return true;
}
// Check for space at the start of the buffer
// We use < here because we don't want to have the case of m_current_offset ==
// m_current_gpu_position. That would mean the code above would assume the
// GPU has caught up to us, which it hasn't.
if (required_bytes < m_current_gpu_position)
{
// Reset offset to zero, since we're allocating behind the gpu now
m_current_offset = 0;
m_current_space = m_current_gpu_position;
return true;
}
}
// Is the GPU ahead of our current offset?
if (m_current_offset < m_current_gpu_position)
{
// We have from m_current_offset..m_current_gpu_position space to use.
const u32 remaining_bytes = m_current_gpu_position - m_current_offset;
if (required_bytes < remaining_bytes)
{
// Place at the current position, since this is still behind the GPU.
m_current_offset = Common::AlignUp(m_current_offset, alignment);
m_current_space = m_current_gpu_position - m_current_offset;
return true;
}
}
// Can we find a fence to wait on that will give us enough memory?
if (WaitForClearSpace(required_bytes))
{
const u32 align_diff = Common::AlignUp(m_current_offset, alignment) - m_current_offset;
m_current_offset += align_diff;
m_current_space -= align_diff;
return true;
}
// We tried everything we could, and still couldn't get anything. This means that too much space
// in the buffer is being used by the command buffer currently being recorded. Therefore, the
// only option is to execute it, and wait until it's done.
return false;
}
void StreamBuffer::CommitMemory(u32 final_num_bytes)
{
pxAssert((m_current_offset + final_num_bytes) <= m_size);
pxAssert(final_num_bytes <= m_current_space);
m_current_offset += final_num_bytes;
m_current_space -= final_num_bytes;
}
void StreamBuffer::Destroy(bool defer)
{
if (m_host_pointer)
{
const D3D12_RANGE written_range = {0, m_size};
m_buffer->Unmap(0, &written_range);
m_host_pointer = nullptr;
}
if (m_buffer && defer)
g_d3d12_context->DeferResourceDestruction(m_allocation.get(), m_buffer.get());
m_buffer.reset();
m_allocation.reset();
m_current_offset = 0;
m_current_space = 0;
m_current_gpu_position = 0;
m_tracked_fences.clear();
}
void StreamBuffer::UpdateCurrentFencePosition()
{
// Don't create a tracking entry if the GPU is caught up with the buffer.
if (m_current_offset == m_current_gpu_position)
return;
// Has the offset changed since the last fence?
const u64 fence = g_d3d12_context->GetCurrentFenceValue();
if (!m_tracked_fences.empty() && m_tracked_fences.back().first == fence)
{
// Still haven't executed a command buffer, so just update the offset.
m_tracked_fences.back().second = m_current_offset;
return;
}
UpdateGPUPosition();
m_tracked_fences.emplace_back(fence, m_current_offset);
}
void StreamBuffer::UpdateGPUPosition()
{
auto start = m_tracked_fences.begin();
auto end = start;
const u64 completed_counter = g_d3d12_context->GetCompletedFenceValue();
while (end != m_tracked_fences.end() && completed_counter >= end->first)
{
m_current_gpu_position = end->second;
++end;
}
if (start != end)
m_tracked_fences.erase(start, end);
}
bool StreamBuffer::WaitForClearSpace(u32 num_bytes)
{
u32 new_offset = 0;
u32 new_space = 0;
u32 new_gpu_position = 0;
auto iter = m_tracked_fences.begin();
for (; iter != m_tracked_fences.end(); ++iter)
{
// Would this fence bring us in line with the GPU?
// This is the "last resort" case, where a command buffer execution has been forced
// after no additional data has been written to it, so we can assume that after the
// fence has been signaled the entire buffer is now consumed.
u32 gpu_position = iter->second;
if (m_current_offset == gpu_position)
{
new_offset = 0;
new_space = m_size;
new_gpu_position = 0;
break;
}
// Assuming that we wait for this fence, are we allocating in front of the GPU?
if (m_current_offset > gpu_position)
{
// This would suggest the GPU has now followed us and wrapped around, so we have from
// m_current_position..m_size free, as well as and 0..gpu_position.
const u32 remaining_space_after_offset = m_size - m_current_offset;
if (remaining_space_after_offset >= num_bytes)
{
// Switch to allocating in front of the GPU, using the remainder of the buffer.
new_offset = m_current_offset;
new_space = m_size - m_current_offset;
new_gpu_position = gpu_position;
break;
}
// We can wrap around to the start, behind the GPU, if there is enough space.
// We use > here because otherwise we'd end up lining up with the GPU, and then the
// allocator would assume that the GPU has consumed what we just wrote.
if (gpu_position > num_bytes)
{
new_offset = 0;
new_space = gpu_position;
new_gpu_position = gpu_position;
break;
}
}
else
{
// We're currently allocating behind the GPU. This would give us between the current
// offset and the GPU position worth of space to work with. Again, > because we can't
// align the GPU position with the buffer offset.
u32 available_space_inbetween = gpu_position - m_current_offset;
if (available_space_inbetween > num_bytes)
{
// Leave the offset as-is, but update the GPU position.
new_offset = m_current_offset;
new_space = gpu_position - m_current_offset;
new_gpu_position = gpu_position;
break;
}
}
}
// Did any fences satisfy this condition?
// Has the command buffer been executed yet? If not, the caller should execute it.
if (iter == m_tracked_fences.end() || iter->first == g_d3d12_context->GetCurrentFenceValue())
return false;
// Wait until this fence is signaled. This will fire the callback, updating the GPU position.
g_d3d12_context->WaitForFence(iter->first);
m_tracked_fences.erase(m_tracked_fences.begin(), m_current_offset == iter->second ? m_tracked_fences.end() : ++iter);
m_current_offset = new_offset;
m_current_space = new_space;
m_current_gpu_position = new_gpu_position;
return true;
}

View File

@ -0,0 +1,77 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/RedtapeWindows.h"
#include <d3d12.h>
#include <deque>
#include <utility>
#include <wil/com.h>
namespace D3D12MA
{
class Allocation;
}
namespace D3D12
{
class StreamBuffer
{
public:
StreamBuffer();
~StreamBuffer();
bool Create(u32 size);
__fi bool IsValid() const { return static_cast<bool>(m_buffer); }
__fi ID3D12Resource* GetBuffer() const { return m_buffer.get(); }
__fi D3D12_GPU_VIRTUAL_ADDRESS GetGPUPointer() const { return m_gpu_pointer; }
__fi void* GetHostPointer() const { return m_host_pointer; }
__fi void* GetCurrentHostPointer() const { return m_host_pointer + m_current_offset; }
__fi D3D12_GPU_VIRTUAL_ADDRESS GetCurrentGPUPointer() const { return m_gpu_pointer + m_current_offset; }
__fi u32 GetSize() const { return m_size; }
__fi u32 GetCurrentOffset() const { return m_current_offset; }
__fi u32 GetCurrentSpace() const { return m_current_space; }
bool ReserveMemory(u32 num_bytes, u32 alignment);
void CommitMemory(u32 final_num_bytes);
void Destroy(bool defer = true);
private:
void UpdateCurrentFencePosition();
void UpdateGPUPosition();
// Waits for as many fences as needed to allocate num_bytes bytes from the buffer.
bool WaitForClearSpace(u32 num_bytes);
u32 m_size = 0;
u32 m_current_offset = 0;
u32 m_current_space = 0;
u32 m_current_gpu_position = 0;
wil::com_ptr_nothrow<ID3D12Resource> m_buffer;
wil::com_ptr_nothrow<D3D12MA::Allocation> m_allocation;
D3D12_GPU_VIRTUAL_ADDRESS m_gpu_pointer = {};
u8* m_host_pointer = nullptr;
// List of fences and the corresponding positions in the buffer
std::deque<std::pair<u64, u32>> m_tracked_fences;
};
} // namespace D3D12

448
common/D3D12/Texture.cpp Normal file
View File

@ -0,0 +1,448 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/Texture.h"
#include "common/D3D12/Context.h"
#include "common/D3D12/Util.h"
#include "common/Align.h"
#include "common/Assertions.h"
#include "common/Console.h"
#include "common/StringUtil.h"
#include "D3D12MemAlloc.h"
using namespace D3D12;
// Somehow NOMINMAX isn't getting set for CMake builds...
#ifdef min
#undef min
#endif
Texture::Texture() = default;
Texture::Texture(ID3D12Resource* resource, D3D12_RESOURCE_STATES state)
: m_resource(std::move(resource))
{
const D3D12_RESOURCE_DESC desc = GetDesc();
m_width = static_cast<u32>(desc.Width);
m_height = desc.Height;
m_levels = desc.MipLevels;
m_format = desc.Format;
}
Texture::Texture(Texture&& texture)
: m_resource(std::move(texture.m_resource))
, m_allocation(std::move(texture.m_allocation))
, m_srv_descriptor(texture.m_srv_descriptor)
, m_rtv_or_dsv_descriptor(texture.m_rtv_or_dsv_descriptor)
, m_width(texture.m_width)
, m_height(texture.m_height)
, m_levels(texture.m_levels)
, m_format(texture.m_format)
, m_state(texture.m_state)
, m_is_depth_view(texture.m_is_depth_view)
{
texture.m_srv_descriptor = {};
texture.m_rtv_or_dsv_descriptor = {};
texture.m_width = 0;
texture.m_height = 0;
texture.m_levels = 0;
texture.m_format = DXGI_FORMAT_UNKNOWN;
texture.m_state = D3D12_RESOURCE_STATE_COMMON;
texture.m_is_depth_view = false;
}
Texture::~Texture()
{
Destroy();
}
Texture& Texture::operator=(Texture&& texture)
{
Destroy();
m_resource = std::move(texture.m_resource);
m_allocation = std::move(texture.m_allocation);
m_srv_descriptor = texture.m_srv_descriptor;
m_rtv_or_dsv_descriptor = texture.m_rtv_or_dsv_descriptor;
m_width = texture.m_width;
m_height = texture.m_height;
m_levels = texture.m_levels;
m_format = texture.m_format;
m_state = texture.m_state;
m_is_depth_view = texture.m_is_depth_view;
texture.m_srv_descriptor = {};
texture.m_rtv_or_dsv_descriptor = {};
texture.m_width = 0;
texture.m_height = 0;
texture.m_levels = 0;
texture.m_format = DXGI_FORMAT_UNKNOWN;
texture.m_state = D3D12_RESOURCE_STATE_COMMON;
texture.m_is_depth_view = false;
return *this;
}
D3D12_RESOURCE_DESC Texture::GetDesc() const
{
return m_resource->GetDesc();
}
bool Texture::Create(u32 width, u32 height, u32 levels, DXGI_FORMAT format, DXGI_FORMAT srv_format,
DXGI_FORMAT rtv_format, DXGI_FORMAT dsv_format, D3D12_RESOURCE_FLAGS flags, u32 alloc_flags)
{
D3D12_RESOURCE_DESC desc = {};
desc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D;
desc.Width = width;
desc.Height = height;
desc.DepthOrArraySize = 1;
desc.MipLevels = levels;
desc.Format = format;
desc.SampleDesc.Count = 1;
desc.Layout = D3D12_TEXTURE_LAYOUT_UNKNOWN;
desc.Flags = flags;
D3D12_CLEAR_VALUE optimized_clear_value = {};
D3D12_RESOURCE_STATES state;
if (rtv_format != DXGI_FORMAT_UNKNOWN)
{
optimized_clear_value.Format = rtv_format;
state = D3D12_RESOURCE_STATE_RENDER_TARGET;
}
else if (dsv_format != DXGI_FORMAT_UNKNOWN)
{
optimized_clear_value.Format = dsv_format;
state = D3D12_RESOURCE_STATE_DEPTH_WRITE;
}
else
{
state = D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE;
}
D3D12MA::ALLOCATION_DESC allocationDesc = {};
allocationDesc.Flags = static_cast<D3D12MA::ALLOCATION_FLAGS>(alloc_flags) | D3D12MA::ALLOCATION_FLAG_WITHIN_BUDGET;
allocationDesc.HeapType = D3D12_HEAP_TYPE_DEFAULT;
ComPtr<ID3D12Resource> resource;
ComPtr<D3D12MA::Allocation> allocation;
HRESULT hr = g_d3d12_context->GetAllocator()->CreateResource(
&allocationDesc, &desc, state,
(rtv_format != DXGI_FORMAT_UNKNOWN || dsv_format != DXGI_FORMAT_UNKNOWN) ? &optimized_clear_value : nullptr,
allocation.put(), IID_PPV_ARGS(resource.put()));
if (FAILED(hr))
{
// OOM isn't fatal.
if (hr != E_OUTOFMEMORY)
Console.Error("Create texture failed: 0x%08X", hr);
return false;
}
DescriptorHandle srv_descriptor, rtv_descriptor;
bool is_depth_view = false;
if (srv_format != DXGI_FORMAT_UNKNOWN)
{
if (!CreateSRVDescriptor(resource.get(), levels, srv_format, &srv_descriptor))
return false;
}
if (rtv_format != DXGI_FORMAT_UNKNOWN)
{
pxAssert(dsv_format == DXGI_FORMAT_UNKNOWN);
if (!CreateRTVDescriptor(resource.get(), rtv_format, &rtv_descriptor))
{
g_d3d12_context->GetDescriptorHeapManager().Free(&srv_descriptor);
return false;
}
}
else if (dsv_format != DXGI_FORMAT_UNKNOWN)
{
if (!CreateDSVDescriptor(resource.get(), dsv_format, &rtv_descriptor))
{
g_d3d12_context->GetDescriptorHeapManager().Free(&srv_descriptor);
return false;
}
is_depth_view = true;
}
Destroy(true);
m_resource = std::move(resource);
m_allocation = std::move(allocation);
m_srv_descriptor = std::move(srv_descriptor);
m_rtv_or_dsv_descriptor = std::move(rtv_descriptor);
m_width = width;
m_height = height;
m_levels = levels;
m_format = format;
m_state = state;
m_is_depth_view = is_depth_view;
return true;
}
bool Texture::Adopt(ComPtr<ID3D12Resource> texture, DXGI_FORMAT srv_format, DXGI_FORMAT rtv_format,
DXGI_FORMAT dsv_format, D3D12_RESOURCE_STATES state)
{
const D3D12_RESOURCE_DESC desc(texture->GetDesc());
DescriptorHandle srv_descriptor, rtv_descriptor;
if (srv_format != DXGI_FORMAT_UNKNOWN)
{
if (!CreateSRVDescriptor(texture.get(), desc.MipLevels, srv_format, &srv_descriptor))
return false;
}
if (rtv_format != DXGI_FORMAT_UNKNOWN)
{
pxAssert(dsv_format == DXGI_FORMAT_UNKNOWN);
if (!CreateRTVDescriptor(texture.get(), rtv_format, &rtv_descriptor))
{
g_d3d12_context->GetDescriptorHeapManager().Free(&srv_descriptor);
return false;
}
}
else if (dsv_format != DXGI_FORMAT_UNKNOWN)
{
if (!CreateDSVDescriptor(texture.get(), dsv_format, &rtv_descriptor))
{
g_d3d12_context->GetDescriptorHeapManager().Free(&srv_descriptor);
return false;
}
}
m_resource = std::move(texture);
m_allocation.reset();
m_srv_descriptor = std::move(srv_descriptor);
m_rtv_or_dsv_descriptor = std::move(rtv_descriptor);
m_width = static_cast<u32>(desc.Width);
m_height = desc.Height;
m_levels = desc.MipLevels;
m_format = desc.Format;
m_state = state;
return true;
}
void Texture::Destroy(bool defer /* = true */)
{
if (defer)
{
g_d3d12_context->DeferDescriptorDestruction(g_d3d12_context->GetDescriptorHeapManager(), &m_srv_descriptor);
if (m_is_depth_view)
g_d3d12_context->DeferDescriptorDestruction(g_d3d12_context->GetDSVHeapManager(), &m_rtv_or_dsv_descriptor);
else
g_d3d12_context->DeferDescriptorDestruction(g_d3d12_context->GetRTVHeapManager(), &m_rtv_or_dsv_descriptor);
g_d3d12_context->DeferResourceDestruction(m_allocation.get(), m_resource.get());
m_resource.reset();
m_allocation.reset();
}
else
{
g_d3d12_context->GetDescriptorHeapManager().Free(&m_srv_descriptor);
if (m_is_depth_view)
g_d3d12_context->GetDSVHeapManager().Free(&m_rtv_or_dsv_descriptor);
else
g_d3d12_context->GetRTVHeapManager().Free(&m_rtv_or_dsv_descriptor);
m_resource.reset();
m_allocation.reset();
}
m_width = 0;
m_height = 0;
m_levels = 0;
m_format = DXGI_FORMAT_UNKNOWN;
m_is_depth_view = false;
}
void Texture::TransitionToState(ID3D12GraphicsCommandList* cmdlist, D3D12_RESOURCE_STATES state)
{
if (m_state == state)
return;
ResourceBarrier(cmdlist, m_resource.get(), m_state, state);
m_state = state;
}
void Texture::TransitionSubresourceToState(ID3D12GraphicsCommandList* cmdlist, u32 level,
D3D12_RESOURCE_STATES before_state, D3D12_RESOURCE_STATES after_state)
{
const D3D12_RESOURCE_BARRIER barrier = {D3D12_RESOURCE_BARRIER_TYPE_TRANSITION,
D3D12_RESOURCE_BARRIER_FLAG_NONE,
{{m_resource.get(), level, before_state, after_state}}};
cmdlist->ResourceBarrier(1, &barrier);
}
bool Texture::BeginStreamUpdate(u32 level, u32 x, u32 y, u32 width, u32 height, void** out_data, u32* out_data_pitch)
{
const u32 copy_pitch = Common::AlignUpPow2(width * GetTexelSize(m_format), D3D12_TEXTURE_DATA_PITCH_ALIGNMENT);
const u32 upload_size = copy_pitch * height;
if (!g_d3d12_context->GetTextureStreamBuffer().ReserveMemory(upload_size, D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT))
{
DevCon.WriteLn("Executing command buffer while waiting for %u bytes (%ux%u) in upload buffer", upload_size, width,
height);
g_d3d12_context->ExecuteCommandList(false);
if (!g_d3d12_context->GetTextureStreamBuffer().ReserveMemory(upload_size, D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT))
{
Console.Error("Failed to reserve %u bytes for %ux%u upload", upload_size, width, height);
return false;
}
}
*out_data = g_d3d12_context->GetTextureStreamBuffer().GetCurrentHostPointer();
*out_data_pitch = copy_pitch;
return true;
}
void Texture::EndStreamUpdate(u32 level, u32 x, u32 y, u32 width, u32 height)
{
const u32 copy_pitch = Common::AlignUpPow2(width * GetTexelSize(m_format), D3D12_TEXTURE_DATA_PITCH_ALIGNMENT);
const u32 upload_size = copy_pitch * height;
StreamBuffer& sb = g_d3d12_context->GetTextureStreamBuffer();
const u32 sb_offset = sb.GetCurrentOffset();
sb.CommitMemory(upload_size);
CopyFromBuffer(level, x, y, width, height, copy_pitch, sb.GetBuffer(), sb_offset);
}
void Texture::CopyFromBuffer(u32 level, u32 x, u32 y, u32 width, u32 height, u32 pitch, ID3D12Resource* buffer, u32 buffer_offset)
{
D3D12_TEXTURE_COPY_LOCATION src;
src.pResource = buffer;
src.Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT;
src.PlacedFootprint.Offset = buffer_offset;
src.PlacedFootprint.Footprint.Width = width;
src.PlacedFootprint.Footprint.Height = height;
src.PlacedFootprint.Footprint.Depth = 1;
src.PlacedFootprint.Footprint.RowPitch = pitch;
src.PlacedFootprint.Footprint.Format = m_format;
D3D12_TEXTURE_COPY_LOCATION dst;
dst.pResource = m_resource.get();
dst.SubresourceIndex = level;
dst.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
const D3D12_BOX src_box{0u, 0u, 0u, width, height, 1u};
const D3D12_RESOURCE_STATES old_state = m_state;
ID3D12GraphicsCommandList* cmdlist = g_d3d12_context->GetCommandList();
TransitionToState(cmdlist, D3D12_RESOURCE_STATE_COPY_DEST);
cmdlist->CopyTextureRegion(&dst, x, y, 0, &src, &src_box);
TransitionToState(cmdlist, old_state);
}
static ID3D12Resource* CreateStagingBuffer(u32 height, const void* data, u32 pitch, u32 upload_pitch, u32 upload_size)
{
wil::com_ptr_nothrow<ID3D12Resource> resource;
wil::com_ptr_nothrow<D3D12MA::Allocation> allocation;
const D3D12MA::ALLOCATION_DESC allocation_desc = {D3D12MA::ALLOCATION_FLAG_NONE, D3D12_HEAP_TYPE_UPLOAD};
const D3D12_RESOURCE_DESC resource_desc = {
D3D12_RESOURCE_DIMENSION_BUFFER, 0, upload_size, 1, 1, 1, DXGI_FORMAT_UNKNOWN, {1, 0}, D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
D3D12_RESOURCE_FLAG_NONE};
HRESULT hr = g_d3d12_context->GetAllocator()->CreateResource(&allocation_desc, &resource_desc, D3D12_RESOURCE_STATE_COPY_SOURCE,
nullptr, allocation.put(), IID_PPV_ARGS(resource.put()));
if (FAILED(hr))
{
Console.Error("CreateResource() for upload staging buffer failed: %08X", hr);
return nullptr;
}
void* map;
const D3D12_RANGE read_range = {};
hr = resource->Map(0, &read_range, &map);
if (FAILED(hr))
{
Console.Error("Map() for upload staging buffer failed: %08X", hr);
return nullptr;
}
StringUtil::StrideMemCpy(map, upload_pitch, data, pitch, std::min(pitch, upload_pitch), height);
const D3D12_RANGE write_range = {0u, upload_size};
resource->Unmap(0, &write_range);
// queue them for destruction, since the upload happens in this cmdlist
g_d3d12_context->DeferResourceDestruction(allocation.get(), resource.get());
// AddRef()'ed by the defer above.
return resource.get();
}
bool Texture::LoadData(u32 level, u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch)
{
const u32 texel_size = GetTexelSize(m_format);
const u32 upload_pitch = Common::AlignUpPow2(width * texel_size, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT);
const u32 upload_size = upload_pitch * height;
if (upload_size >= g_d3d12_context->GetTextureStreamBuffer().GetSize())
{
ID3D12Resource* staging_buffer = CreateStagingBuffer(height, data, pitch, upload_pitch, upload_size);
if (!staging_buffer)
return false;
CopyFromBuffer(level, x, y, width, height, upload_pitch, staging_buffer, 0);
return true;
}
void* write_ptr;
u32 write_pitch;
if (!BeginStreamUpdate(level, x, y, width, height, &write_ptr, &write_pitch))
return false;
StringUtil::StrideMemCpy(write_ptr, write_pitch, data, pitch, std::min(pitch, upload_pitch), height);
EndStreamUpdate(level, x, y, width, height);
return true;
}
bool Texture::CreateSRVDescriptor(ID3D12Resource* resource, u32 levels, DXGI_FORMAT format, DescriptorHandle* dh)
{
if (!g_d3d12_context->GetDescriptorHeapManager().Allocate(dh))
{
Console.Error("Failed to allocate SRV descriptor");
return false;
}
D3D12_SHADER_RESOURCE_VIEW_DESC desc = {format, D3D12_SRV_DIMENSION_TEXTURE2D, D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING};
desc.Texture2D.MipLevels = levels;
g_d3d12_context->GetDevice()->CreateShaderResourceView(resource, &desc, dh->cpu_handle);
return true;
}
bool Texture::CreateRTVDescriptor(ID3D12Resource* resource, DXGI_FORMAT format, DescriptorHandle* dh)
{
if (!g_d3d12_context->GetRTVHeapManager().Allocate(dh))
{
Console.Error("Failed to allocate SRV descriptor");
return false;
}
const D3D12_RENDER_TARGET_VIEW_DESC desc = {format, D3D12_RTV_DIMENSION_TEXTURE2D};
g_d3d12_context->GetDevice()->CreateRenderTargetView(resource, &desc, dh->cpu_handle);
return true;
}
bool Texture::CreateDSVDescriptor(ID3D12Resource* resource, DXGI_FORMAT format, DescriptorHandle* dh)
{
if (!g_d3d12_context->GetDSVHeapManager().Allocate(dh))
{
Console.Error("Failed to allocate SRV descriptor");
return false;
}
const D3D12_DEPTH_STENCIL_VIEW_DESC desc = {format, D3D12_DSV_DIMENSION_TEXTURE2D, D3D12_DSV_FLAG_NONE};
g_d3d12_context->GetDevice()->CreateDepthStencilView(resource, &desc, dh->cpu_handle);
return true;
}

100
common/D3D12/Texture.h Normal file
View File

@ -0,0 +1,100 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/RedtapeWindows.h"
#include "common/D3D12/DescriptorHeapManager.h"
#include <d3d12.h>
#include <wil/com.h>
namespace D3D12MA
{
class Allocation;
}
namespace D3D12
{
class StreamBuffer;
class Texture final
{
public:
template <typename T>
using ComPtr = wil::com_ptr_nothrow<T>;
Texture();
Texture(ID3D12Resource* resource, D3D12_RESOURCE_STATES state);
Texture(Texture&& texture);
Texture(const Texture&) = delete;
~Texture();
__fi ID3D12Resource* GetResource() const { return m_resource.get(); }
__fi D3D12MA::Allocation* GetAllocation() const { return m_allocation.get(); }
__fi const DescriptorHandle& GetSRVDescriptor() const { return m_srv_descriptor; }
__fi const DescriptorHandle& GetRTVOrDSVDescriptor() const { return m_rtv_or_dsv_descriptor; }
__fi D3D12_RESOURCE_STATES GetState() const { return m_state; }
__fi u32 GetWidth() const { return m_width; }
__fi u32 GetHeight() const { return m_height; }
__fi u32 GetLevels() const { return m_levels; }
__fi DXGI_FORMAT GetFormat() const { return m_format; }
__fi operator ID3D12Resource*() const { return m_resource.get(); }
__fi operator bool() const { return static_cast<bool>(m_resource); }
bool Create(u32 width, u32 height, u32 levels, DXGI_FORMAT format, DXGI_FORMAT srv_format,
DXGI_FORMAT rtv_format, DXGI_FORMAT dsv_format, D3D12_RESOURCE_FLAGS flags, u32 alloc_flags = 0);
bool Adopt(ComPtr<ID3D12Resource> texture, DXGI_FORMAT srv_format, DXGI_FORMAT rtv_format, DXGI_FORMAT dsv_format,
D3D12_RESOURCE_STATES state);
D3D12_RESOURCE_DESC GetDesc() const;
void Destroy(bool defer = true);
void TransitionToState(ID3D12GraphicsCommandList* cmdlist, D3D12_RESOURCE_STATES state);
void TransitionSubresourceToState(ID3D12GraphicsCommandList* cmdlist, u32 level,
D3D12_RESOURCE_STATES before_state, D3D12_RESOURCE_STATES after_state);
Texture& operator=(const Texture&) = delete;
Texture& operator=(Texture&& texture);
// NOTE: Does not handle compressed formats.
bool BeginStreamUpdate(u32 level, u32 x, u32 y, u32 width, u32 height, void** out_data, u32* out_data_pitch);
void EndStreamUpdate(u32 level, u32 x, u32 y, u32 width, u32 height);
bool LoadData(u32 level, u32 x, u32 y, u32 width, u32 height, const void* data, u32 pitch);
void CopyFromBuffer(u32 level, u32 x, u32 y, u32 width, u32 height, u32 pitch, ID3D12Resource* buffer, u32 buffer_offset);
private:
static bool CreateSRVDescriptor(ID3D12Resource* resource, u32 levels, DXGI_FORMAT format, DescriptorHandle* dh);
static bool CreateRTVDescriptor(ID3D12Resource* resource, DXGI_FORMAT format, DescriptorHandle* dh);
static bool CreateDSVDescriptor(ID3D12Resource* resource, DXGI_FORMAT format, DescriptorHandle* dh);
ComPtr<ID3D12Resource> m_resource;
ComPtr<D3D12MA::Allocation> m_allocation;
DescriptorHandle m_srv_descriptor = {};
DescriptorHandle m_rtv_or_dsv_descriptor = {};
u32 m_width = 0;
u32 m_height = 0;
u32 m_levels = 0;
DXGI_FORMAT m_format = DXGI_FORMAT_UNKNOWN;
D3D12_RESOURCE_STATES m_state = D3D12_RESOURCE_STATE_COMMON;
bool m_is_depth_view = false;
};
} // namespace D3D12

93
common/D3D12/Util.cpp Normal file
View File

@ -0,0 +1,93 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#include "common/PrecompiledHeader.h"
#include "common/D3D12/Util.h"
#include "common/Assertions.h"
#include "common/StringUtil.h"
u32 D3D12::GetTexelSize(DXGI_FORMAT format)
{
switch (format)
{
case DXGI_FORMAT_R32G32B32A32_FLOAT:
case DXGI_FORMAT_BC1_UNORM:
case DXGI_FORMAT_BC2_UNORM:
case DXGI_FORMAT_BC3_UNORM:
case DXGI_FORMAT_BC7_UNORM:
return 16;
case DXGI_FORMAT_D32_FLOAT_S8X24_UINT:
return 4;
case DXGI_FORMAT_R8G8B8A8_UNORM:
case DXGI_FORMAT_R8G8B8A8_SNORM:
case DXGI_FORMAT_R8G8B8A8_TYPELESS:
case DXGI_FORMAT_B8G8R8A8_UNORM:
case DXGI_FORMAT_B8G8R8A8_TYPELESS:
case DXGI_FORMAT_R32_UINT:
case DXGI_FORMAT_R32_SINT:
return 4;
case DXGI_FORMAT_B5G5R5A1_UNORM:
case DXGI_FORMAT_B5G6R5_UNORM:
case DXGI_FORMAT_R16_UINT:
case DXGI_FORMAT_R16_SINT:
return 2;
case DXGI_FORMAT_A8_UNORM:
case DXGI_FORMAT_R8_UNORM:
return 1;
default:
pxFailRel("Unknown format");
return 1;
}
}
void D3D12::SetDefaultSampler(D3D12_SAMPLER_DESC* desc)
{
desc->Filter = D3D12_FILTER_MIN_MAG_MIP_LINEAR;
desc->AddressU = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
desc->AddressV = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
desc->AddressW = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
desc->MipLODBias = 0;
desc->MaxAnisotropy = 1;
desc->ComparisonFunc = D3D12_COMPARISON_FUNC_NEVER;
desc->BorderColor[0] = 1.0f;
desc->BorderColor[1] = 1.0f;
desc->BorderColor[2] = 1.0f;
desc->BorderColor[3] = 1.0f;
desc->MinLOD = -3.402823466e+38F; // -FLT_MAX
desc->MaxLOD = 3.402823466e+38F; // FLT_MAX
}
#ifdef _DEBUG
void D3D12::SetObjectName(ID3D12Object* object, const char* name)
{
object->SetName(StringUtil::UTF8StringToWideString(name).c_str());
}
void D3D12::SetObjectNameFormatted(ID3D12Object* object, const char* format, ...)
{
std::va_list ap;
va_start(ap, format);
SetObjectName(object, StringUtil::StdStringFromFormatV(format, ap).c_str());
va_end(ap);
}
#endif

77
common/D3D12/Util.h Normal file
View File

@ -0,0 +1,77 @@
/* PCSX2 - PS2 Emulator for PCs
* Copyright (C) 2002-2022 PCSX2 Dev Team
*
* PCSX2 is free software: you can redistribute it and/or modify it under the terms
* of the GNU Lesser General Public License as published by the Free Software Found-
* ation, either version 3 of the License, or (at your option) any later version.
*
* PCSX2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
* without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
* PURPOSE. See the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along with PCSX2.
* If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "common/Pcsx2Defs.h"
#include "common/RedtapeWindows.h"
#include <array>
#include <d3d12.h>
namespace D3D12
{
static inline void ResourceBarrier(ID3D12GraphicsCommandList* cmdlist, ID3D12Resource* resource,
D3D12_RESOURCE_STATES from_state, D3D12_RESOURCE_STATES to_state)
{
const D3D12_RESOURCE_BARRIER barrier = {D3D12_RESOURCE_BARRIER_TYPE_TRANSITION,
D3D12_RESOURCE_BARRIER_FLAG_NONE,
{{resource, D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES, from_state, to_state}}};
cmdlist->ResourceBarrier(1, &barrier);
}
static inline void SetViewport(ID3D12GraphicsCommandList* cmdlist, int x, int y, int width, int height,
float min_depth = 0.0f, float max_depth = 1.0f)
{
const D3D12_VIEWPORT vp{static_cast<float>(x),
static_cast<float>(y),
static_cast<float>(width),
static_cast<float>(height),
min_depth,
max_depth};
cmdlist->RSSetViewports(1, &vp);
}
static inline void SetScissor(ID3D12GraphicsCommandList* cmdlist, int x, int y, int width, int height)
{
const D3D12_RECT r{x, y, x + width, y + height};
cmdlist->RSSetScissorRects(1, &r);
}
static inline void SetViewportAndScissor(ID3D12GraphicsCommandList* cmdlist, int x, int y, int width, int height,
float min_depth = 0.0f, float max_depth = 1.0f)
{
SetViewport(cmdlist, x, y, width, height, min_depth, max_depth);
SetScissor(cmdlist, x, y, width, height);
}
u32 GetTexelSize(DXGI_FORMAT format);
void SetDefaultSampler(D3D12_SAMPLER_DESC* desc);
#ifdef _DEBUG
void SetObjectName(ID3D12Object* object, const char* name);
void SetObjectNameFormatted(ID3D12Object* object, const char* format, ...);
#else
static inline void SetObjectName(ID3D12Object* object, const char* name)
{
}
static inline void SetObjectNameFormatted(ID3D12Object* object, const char* format, ...) {}
#endif
} // namespace D3D12

View File

@ -33,7 +33,7 @@
</PropertyGroup>
<ItemDefinitionGroup>
<ClCompile>
<AdditionalIncludeDirectories>$(SolutionDir)3rdparty\glad\include;$(SolutionDir)3rdparty\glslang\glslang;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
<AdditionalIncludeDirectories>$(SolutionDir)3rdparty\d3d12memalloc\include;$(SolutionDir)3rdparty\glad\include;$(SolutionDir)3rdparty\glslang\glslang;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
<ExceptionHandling>Async</ExceptionHandling>
<PrecompiledHeader>Use</PrecompiledHeader>
<ForcedIncludeFiles>PrecompiledHeader.h</ForcedIncludeFiles>
@ -47,6 +47,13 @@
<ClCompile Include="Console.cpp" />
<ClCompile Include="D3D11\ShaderCache.cpp" />
<ClCompile Include="D3D11\ShaderCompiler.cpp" />
<ClCompile Include="D3D12\Builders.cpp" />
<ClCompile Include="D3D12\Context.cpp" />
<ClCompile Include="D3D12\DescriptorHeapManager.cpp" />
<ClCompile Include="D3D12\ShaderCache.cpp" />
<ClCompile Include="D3D12\StreamBuffer.cpp" />
<ClCompile Include="D3D12\Texture.cpp" />
<ClCompile Include="D3D12\Util.cpp" />
<ClCompile Include="Exceptions.cpp" />
<ClCompile Include="FastFormatString.cpp" />
<ClCompile Include="FastJmp.cpp">
@ -117,6 +124,13 @@
<ClInclude Include="BitCast.h" />
<ClInclude Include="D3D11\ShaderCache.h" />
<ClInclude Include="D3D11\ShaderCompiler.h" />
<ClInclude Include="D3D12\Builders.h" />
<ClInclude Include="D3D12\Context.h" />
<ClInclude Include="D3D12\DescriptorHeapManager.h" />
<ClInclude Include="D3D12\ShaderCache.h" />
<ClInclude Include="D3D12\StreamBuffer.h" />
<ClInclude Include="D3D12\Texture.h" />
<ClInclude Include="D3D12\Util.h" />
<ClInclude Include="EmbeddedImage.h" />
<ClInclude Include="boost_spsc_queue.hpp" />
<ClInclude Include="FastJmp.h" />
@ -190,6 +204,9 @@
<ClInclude Include="ZipHelpers.h" />
</ItemGroup>
<ItemGroup>
<ProjectReference Include="..\3rdparty\d3d12memalloc\d3d12memalloc.vcxproj">
<Project>{d45cec7a-3171-40dd-975d-e1544cf16139}</Project>
</ProjectReference>
<ProjectReference Include="..\3rdparty\glad\glad.vcxproj">
<Project>{c0293b32-5acf-40f0-aa6c-e6da6f3bf33a}</Project>
</ProjectReference>

View File

@ -178,6 +178,27 @@
<ClCompile Include="D3D11\ShaderCompiler.cpp">
<Filter>Source Files\D3D11</Filter>
</ClCompile>
<ClCompile Include="D3D12\Context.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\DescriptorHeapManager.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\ShaderCache.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\StreamBuffer.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\Texture.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\Util.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
<ClCompile Include="D3D12\Builders.cpp">
<Filter>Source Files\D3D12</Filter>
</ClCompile>
</ItemGroup>
<ItemGroup>
<ClInclude Include="AlignedMalloc.h">
@ -414,6 +435,27 @@
<ClInclude Include="ZipHelpers.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="D3D12\Context.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\DescriptorHeapManager.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\ShaderCache.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\StreamBuffer.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\Texture.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\Util.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
<ClInclude Include="D3D12\Builders.h">
<Filter>Header Files\D3D12</Filter>
</ClInclude>
</ItemGroup>
<ItemGroup>
<Filter Include="Source Files">
@ -440,6 +482,12 @@
<Filter Include="Header Files\D3D11">
<UniqueIdentifier>{764ddf71-77a6-41b8-bc65-1eef94b6997b}</UniqueIdentifier>
</Filter>
<Filter Include="Source Files\D3D12">
<UniqueIdentifier>{96f78eb9-089b-4166-a23e-78c4e7d90b64}</UniqueIdentifier>
</Filter>
<Filter Include="Header Files\D3D12">
<UniqueIdentifier>{cf37623d-bb05-4c54-8c72-1c20b5331c69}</UniqueIdentifier>
</Filter>
</ItemGroup>
<ItemGroup>
<None Include="EventSource.inl">
@ -454,4 +502,4 @@
<Filter>Source Files</Filter>
</MASM>
</ItemGroup>
</Project>
</Project>