mirror of
https://bitbucket.org/CPMADevs/cnq3
synced 2024-11-26 22:11:51 +00:00
1078 lines
25 KiB
C++
1078 lines
25 KiB
C++
/*
|
|
===========================================================================
|
|
Copyright (C) 2022-2023 Gian 'myT' Schellenbaum
|
|
|
|
This file is part of Challenge Quake 3 (CNQ3).
|
|
|
|
Challenge Quake 3 is free software; you can redistribute it
|
|
and/or modify it under the terms of the GNU General Public License as
|
|
published by the Free Software Foundation; either version 2 of the License,
|
|
or (at your option) any later version.
|
|
|
|
Challenge Quake 3 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 Challenge Quake 3. If not, see <https://www.gnu.org/licenses/>.
|
|
===========================================================================
|
|
*/
|
|
// Rendering Hardware Interface - private interface
|
|
|
|
|
|
#pragma once
|
|
|
|
|
|
#include "tr_local.h"
|
|
|
|
|
|
namespace RHI
|
|
{
|
|
// @TODO: turn into uint32_t constants too
|
|
#define RHI_MAX_TEXTURES_2D 4096 // real max: unlimited
|
|
#define RHI_MAX_RW_TEXTURES_2D 64 // real max: 64
|
|
#define RHI_MAX_SAMPLERS 128 // real max: 2048
|
|
|
|
// this has 2 meanings:
|
|
// 1. maximum number of frames queued
|
|
// 2. number of frames in the back buffer
|
|
const uint32_t FrameCount = 2;
|
|
const uint32_t MaxVertexBufferCount = 16;
|
|
const uint32_t MaxVertexAttributeCount = 32;
|
|
const uint32_t MaxRenderTargetCount = 8;
|
|
const uint32_t MaxDurationQueries = 64;
|
|
const uint32_t MaxTextureMips = 16;
|
|
const uint32_t InvalidDescriptorIndex = UINT16_MAX;
|
|
const uint32_t U32WholeRange = UINT32_MAX;
|
|
|
|
#define RHI_ENUM_OPERATORS(EnumType) \
|
|
inline EnumType operator|(EnumType a, EnumType b) { return (EnumType)((uint32_t)(a) | (uint32_t)(b)); } \
|
|
inline EnumType operator&(EnumType a, EnumType b) { return (EnumType)((uint32_t)(a) & (uint32_t)(b)); } \
|
|
inline EnumType operator|=(EnumType& a, EnumType b) { return a = (a | b); } \
|
|
inline EnumType operator&=(EnumType& a, EnumType b) { return a = (a & b); } \
|
|
inline EnumType operator~(EnumType a) { return (EnumType)(~(uint32_t)(a)); }
|
|
|
|
#define RHI_BIT(Bit) (1 << Bit)
|
|
#define RHI_BIT_MASK(BitCount) ((1 << BitCount) - 1)
|
|
|
|
#define RHI_GET_HANDLE_VALUE(Handle) (Handle.v)
|
|
#define RHI_MAKE_HANDLE(Value) { Value }
|
|
#define RHI_MAKE_NULL_HANDLE() { 0 }
|
|
|
|
struct IndexType
|
|
{
|
|
enum Id
|
|
{
|
|
UInt32,
|
|
UInt16,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct ResourceStates
|
|
{
|
|
enum Flags
|
|
{
|
|
Common = 0,
|
|
VertexBufferBit = RHI_BIT(0),
|
|
IndexBufferBit = RHI_BIT(1),
|
|
ConstantBufferBit = RHI_BIT(2),
|
|
RenderTargetBit = RHI_BIT(3),
|
|
VertexShaderAccessBit = RHI_BIT(4),
|
|
PixelShaderAccessBit = RHI_BIT(5),
|
|
ComputeShaderAccessBit = RHI_BIT(6),
|
|
CopySourceBit = RHI_BIT(7),
|
|
CopyDestinationBit = RHI_BIT(8),
|
|
DepthReadBit = RHI_BIT(9),
|
|
DepthWriteBit = RHI_BIT(10),
|
|
UnorderedAccessBit = RHI_BIT(11),
|
|
PresentBit = RHI_BIT(12),
|
|
ShaderAccessBits = VertexShaderAccessBit | PixelShaderAccessBit | ComputeShaderAccessBit,
|
|
DepthAccessBits = DepthReadBit | DepthWriteBit
|
|
};
|
|
};
|
|
RHI_ENUM_OPERATORS(ResourceStates::Flags);
|
|
|
|
struct MemoryUsage
|
|
{
|
|
enum Id
|
|
{
|
|
CPU, // Host
|
|
GPU, // DeviceLocal
|
|
Upload, // CPU -> GPU
|
|
Readback, // GPU -> CPU
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct ShaderStage
|
|
{
|
|
enum Id
|
|
{
|
|
Vertex,
|
|
Pixel,
|
|
Compute,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct ShaderStages
|
|
{
|
|
enum Flags
|
|
{
|
|
None = 0,
|
|
VertexBit = 1 << ShaderStage::Vertex,
|
|
PixelBit = 1 << ShaderStage::Pixel,
|
|
ComputeBit = 1 << ShaderStage::Compute,
|
|
AllGraphicsBits = VertexBit | PixelBit
|
|
};
|
|
};
|
|
RHI_ENUM_OPERATORS(ShaderStages::Flags);
|
|
|
|
struct DataType
|
|
{
|
|
enum Id
|
|
{
|
|
Float32,
|
|
UNorm8,
|
|
UInt32,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct ShaderSemantic
|
|
{
|
|
enum Id
|
|
{
|
|
Position,
|
|
Normal,
|
|
TexCoord,
|
|
Color,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct TextureFormat
|
|
{
|
|
enum Id
|
|
{
|
|
RGBA32_UNorm,
|
|
RGBA64_Float,
|
|
Depth32_Float,
|
|
RG16_UNorm,
|
|
R8_UNorm,
|
|
Depth24_Stencil8,
|
|
R10G10B10A2_UNorm,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct ComparisonFunction
|
|
{
|
|
enum Id
|
|
{
|
|
Never,
|
|
Less,
|
|
Equal,
|
|
LessEqual,
|
|
Greater,
|
|
NotEqual,
|
|
GreaterEqual,
|
|
Always,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct DescriptorType
|
|
{
|
|
enum Id
|
|
{
|
|
Buffer, // CBV, HBuffer
|
|
RWBuffer, // UAV, HBuffer
|
|
Texture, // SRV, HTexture
|
|
RWTexture, // UAV, HTexture
|
|
Sampler,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct TextureFilter
|
|
{
|
|
enum Id
|
|
{
|
|
Point,
|
|
Linear,
|
|
Anisotropic,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct PipelineType
|
|
{
|
|
enum Id
|
|
{
|
|
Graphics,
|
|
Compute,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct StencilOp
|
|
{
|
|
enum Id
|
|
{
|
|
Keep,
|
|
Zero,
|
|
Replace,
|
|
SaturatedIncrement,
|
|
SaturatedDecrement,
|
|
Invert,
|
|
WrappedIncrement,
|
|
WrappedDecrement,
|
|
Count
|
|
};
|
|
};
|
|
|
|
struct RootSignatureDesc
|
|
{
|
|
RootSignatureDesc() = default;
|
|
RootSignatureDesc(const char* name_)
|
|
{
|
|
name = name_;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
bool usingVertexBuffers = false;
|
|
struct PerStageConstants
|
|
{
|
|
uint32_t byteCount = 0;
|
|
}
|
|
constants[ShaderStage::Count];
|
|
struct DescriptorRange
|
|
{
|
|
DescriptorType::Id type = DescriptorType::Count;
|
|
uint32_t firstIndex = 0;
|
|
uint32_t count = 0;
|
|
}
|
|
genericRanges[64];
|
|
uint32_t genericRangeCount = 0;
|
|
uint32_t samplerCount = 0;
|
|
ShaderStages::Flags genericVisibility = ShaderStages::None; // ignored by compute pipelines
|
|
ShaderStages::Flags samplerVisibility = ShaderStages::None; // ignored by compute pipelines
|
|
PipelineType::Id pipelineType = PipelineType::Graphics;
|
|
|
|
void AddRange(DescriptorType::Id type, uint32_t firstIndex, uint32_t count)
|
|
{
|
|
Q_assert(genericRangeCount < ARRAY_LEN(genericRanges));
|
|
DescriptorRange& r = genericRanges[genericRangeCount++];
|
|
r.type = type;
|
|
r.firstIndex = firstIndex;
|
|
r.count = count;
|
|
}
|
|
};
|
|
|
|
struct ShaderByteCode
|
|
{
|
|
ShaderByteCode() = default;
|
|
ShaderByteCode(const void* data_, uint32_t byteCount_)
|
|
{
|
|
data = data_;
|
|
byteCount = byteCount_;
|
|
}
|
|
|
|
template<uint32_t N>
|
|
ShaderByteCode(const uint8_t (&byteCode)[N])
|
|
{
|
|
data = byteCode;
|
|
byteCount = N;
|
|
}
|
|
|
|
const void* data = NULL;
|
|
uint32_t byteCount = 0;
|
|
};
|
|
|
|
struct ShaderMacro
|
|
{
|
|
ShaderMacro() = default;
|
|
ShaderMacro(const char* name_, const char* value_)
|
|
{
|
|
name = name_;
|
|
value = value_;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
const char* value = NULL;
|
|
};
|
|
|
|
struct VertexAttribute
|
|
{
|
|
uint32_t vertexBufferIndex = 0; // also called "binding" or "input slot"
|
|
ShaderSemantic::Id semantic = ShaderSemantic::Count; // intended usage
|
|
DataType::Id dataType = DataType::Count; // for a single component of the vector
|
|
uint32_t vectorLength = 4; // number of components per vector
|
|
uint32_t structByteOffset = 0; // where in the struct to look when using interleaved data
|
|
};
|
|
|
|
struct DepthStencilOpDesc
|
|
{
|
|
ComparisonFunction::Id comparison = ComparisonFunction::Always;
|
|
StencilOp::Id passOp = StencilOp::Keep;
|
|
StencilOp::Id failOp = StencilOp::Keep;
|
|
StencilOp::Id depthFailOp = StencilOp::Keep;
|
|
};
|
|
|
|
struct GraphicsPipelineDesc
|
|
{
|
|
GraphicsPipelineDesc() = default;
|
|
GraphicsPipelineDesc(const char* name_, HRootSignature rootSignature_)
|
|
{
|
|
name = name_;
|
|
rootSignature = rootSignature_;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
HRootSignature rootSignature = RHI_MAKE_NULL_HANDLE();
|
|
ShaderByteCode vertexShader;
|
|
ShaderByteCode pixelShader;
|
|
struct VertexLayout
|
|
{
|
|
VertexAttribute attributes[MaxVertexAttributeCount];
|
|
uint32_t attributeCount = 0;
|
|
uint32_t bindingStrides[MaxVertexBufferCount] = { 0 }; // total byte size of a vertex for each buffer
|
|
|
|
void AddAttribute(
|
|
uint32_t vertexBufferIndex,
|
|
ShaderSemantic::Id semantic,
|
|
DataType::Id dataType,
|
|
uint32_t vectorLength,
|
|
uint32_t structByteOffset)
|
|
{
|
|
Q_assert(attributeCount < MaxVertexAttributeCount);
|
|
VertexAttribute& va = attributes[attributeCount++];
|
|
va.dataType = dataType;
|
|
va.semantic = semantic;
|
|
va.structByteOffset = structByteOffset;
|
|
va.vectorLength = vectorLength;
|
|
va.vertexBufferIndex = vertexBufferIndex;
|
|
}
|
|
}
|
|
vertexLayout;
|
|
struct DepthStencil
|
|
{
|
|
void DisableDepth()
|
|
{
|
|
enableDepthTest = false;
|
|
enableDepthWrites = false;
|
|
}
|
|
|
|
bool enableDepthTest = true;
|
|
bool enableDepthWrites = true;
|
|
ComparisonFunction::Id depthComparison = ComparisonFunction::GreaterEqual;
|
|
TextureFormat::Id depthStencilFormat = TextureFormat::Depth32_Float;
|
|
|
|
bool enableStencil = false;
|
|
uint8_t stencilReadMask = 0xFF;
|
|
uint8_t stencilWriteMask = 0xFF;
|
|
DepthStencilOpDesc frontFace;
|
|
DepthStencilOpDesc backFace;
|
|
}
|
|
depthStencil;
|
|
struct Rasterizer
|
|
{
|
|
cullType_t cullMode = CT_FRONT_SIDED;
|
|
bool polygonOffset = false;
|
|
bool clampDepth = false;
|
|
}
|
|
rasterizer;
|
|
struct RenderTarget
|
|
{
|
|
// @TODO:
|
|
//uint32_t q3BlendMode = GLS_SRCBLEND_SRC_ALPHA | GLS_DSTBLEND_ONE_MINUS_SRC_ALPHA;
|
|
uint32_t q3BlendMode = 0x00000005 | 0x00000060;
|
|
TextureFormat::Id format = TextureFormat::RGBA32_UNorm;
|
|
}
|
|
renderTargets[MaxRenderTargetCount];
|
|
uint32_t renderTargetCount = 0;
|
|
|
|
void AddRenderTarget(uint32_t q3BlendMode, TextureFormat::Id format)
|
|
{
|
|
Q_assert(renderTargetCount < MaxRenderTargetCount);
|
|
RenderTarget& rt = renderTargets[renderTargetCount++];
|
|
rt.q3BlendMode = q3BlendMode;
|
|
rt.format = format;
|
|
}
|
|
};
|
|
|
|
struct ComputePipelineDesc
|
|
{
|
|
ComputePipelineDesc() = default;
|
|
ComputePipelineDesc(const char* name_, HRootSignature rootSignature_)
|
|
{
|
|
name = name_;
|
|
rootSignature = rootSignature_;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
HRootSignature rootSignature = RHI_MAKE_NULL_HANDLE();
|
|
ShaderByteCode shader;
|
|
};
|
|
|
|
struct BufferDesc
|
|
{
|
|
BufferDesc() = default;
|
|
BufferDesc(const char* name_, uint32_t byteCount_, ResourceStates::Flags initialState_)
|
|
{
|
|
name = name_;
|
|
byteCount = byteCount_;
|
|
initialState = initialState_;
|
|
memoryUsage = MemoryUsage::GPU;
|
|
committedResource = false;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
uint32_t byteCount = 0;
|
|
ResourceStates::Flags initialState = ResourceStates::Common;
|
|
MemoryUsage::Id memoryUsage = MemoryUsage::GPU;
|
|
bool committedResource = false;
|
|
uint32_t structureByteCount = 0; // > 0 means structured buffer, == 0 means byte address buffer
|
|
};
|
|
|
|
struct TextureDesc
|
|
{
|
|
TextureDesc() = default;
|
|
TextureDesc(const char* name_, uint32_t width_, uint32_t height_, uint32_t mipCount_ = 1)
|
|
{
|
|
name = name_;
|
|
width = width_;
|
|
height = height_;
|
|
mipCount = mipCount_;
|
|
sampleCount = 1;
|
|
initialState = ResourceStates::PixelShaderAccessBit;
|
|
allowedState = ResourceStates::PixelShaderAccessBit;
|
|
format = TextureFormat::RGBA32_UNorm;
|
|
committedResource = false;
|
|
usePreferredClearValue = false;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
uint32_t width = 0;
|
|
uint32_t height = 0;
|
|
uint32_t mipCount = 1;
|
|
uint32_t sampleCount = 1;
|
|
ResourceStates::Flags initialState = ResourceStates::PixelShaderAccessBit;
|
|
ResourceStates::Flags allowedState = ResourceStates::PixelShaderAccessBit;
|
|
TextureFormat::Id format = TextureFormat::RGBA32_UNorm;
|
|
bool committedResource = false;
|
|
bool usePreferredClearValue = false; // for render targets and depth/stencil buffers
|
|
float clearColor[4] = { 0.0f, 0.0f, 0.0f, 1.0f };
|
|
float clearDepth = 1.0f;
|
|
byte clearStencil = 0;
|
|
void* nativeResource = NULL; // ID3D12Resource*, VkImage, etc.
|
|
|
|
void SetClearColor(const vec4_t rgba)
|
|
{
|
|
usePreferredClearValue = true;
|
|
Vector4Copy(rgba, clearColor);
|
|
}
|
|
|
|
void SetClearDepthStencil(float depth, byte stencil = 0)
|
|
{
|
|
usePreferredClearValue = true;
|
|
clearDepth = depth;
|
|
clearStencil = stencil;
|
|
}
|
|
};
|
|
|
|
struct SamplerDesc
|
|
{
|
|
SamplerDesc() = default;
|
|
SamplerDesc(textureWrap_t wrapMode_, TextureFilter::Id filterMode_, float minLOD_ = 0.0f, float mipLODBias_ = 0.0f)
|
|
{
|
|
wrapMode = wrapMode_;
|
|
filterMode = filterMode_;
|
|
minLOD = minLOD_;
|
|
mipLODBias = mipLODBias_;
|
|
}
|
|
|
|
// @TODO:
|
|
//textureWrap_t wrapMode = TW_REPEAT;
|
|
textureWrap_t wrapMode = (textureWrap_t)0;
|
|
TextureFilter::Id filterMode = TextureFilter::Linear;
|
|
float minLOD = 0.0f;
|
|
float mipLODBias = 0.0f;
|
|
};
|
|
|
|
struct DescriptorTableDesc
|
|
{
|
|
DescriptorTableDesc() = default;
|
|
DescriptorTableDesc(const char* name_, HRootSignature rootSignature_)
|
|
{
|
|
name = name_;
|
|
rootSignature = rootSignature_;
|
|
}
|
|
|
|
const char* name = NULL;
|
|
bool shortLifeTime = false;
|
|
HRootSignature rootSignature = RHI_MAKE_NULL_HANDLE();
|
|
};
|
|
|
|
struct BufferBarrier
|
|
{
|
|
BufferBarrier() = default;
|
|
BufferBarrier(HBuffer buffer_, ResourceStates::Flags newState_)
|
|
{
|
|
buffer = buffer_;
|
|
newState = newState_;
|
|
}
|
|
|
|
HBuffer buffer = RHI_MAKE_NULL_HANDLE();
|
|
ResourceStates::Flags newState = ResourceStates::Common;
|
|
};
|
|
|
|
struct TextureBarrier
|
|
{
|
|
TextureBarrier() = default;
|
|
TextureBarrier(HTexture texture_, ResourceStates::Flags newState_)
|
|
{
|
|
texture = texture_;
|
|
newState = newState_;
|
|
}
|
|
|
|
HTexture texture = RHI_MAKE_NULL_HANDLE();
|
|
ResourceStates::Flags newState = ResourceStates::Common;
|
|
};
|
|
|
|
struct DescriptorTableUpdate
|
|
{
|
|
// note that for our texture UAVs,
|
|
// we only allow 2 options:
|
|
// 1) "bind all mips" (mip chain)
|
|
// 2) "bind this single mip" (slide)
|
|
|
|
uint32_t firstIndex = 0;
|
|
uint32_t resourceCount = 0;
|
|
DescriptorType::Id type = DescriptorType::Count;
|
|
union // based on type
|
|
{
|
|
const HTexture* textures = NULL;
|
|
const HBuffer* buffers;
|
|
const HSampler* samplers;
|
|
};
|
|
uint32_t uavMipSlice = 0; // UAV textures: bind this specific mip
|
|
bool uavMipChain = false; // UAV textures: bind all mips if true, the specific mip slice otherwise
|
|
|
|
void SetSamplers(uint32_t count, const HSampler* samplers_, uint32_t tableIndex = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::Sampler;
|
|
samplers = samplers_;
|
|
}
|
|
|
|
void SetBuffers(uint32_t count, const HBuffer* buffers_, uint32_t tableIndex = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::Buffer;
|
|
buffers = buffers_;
|
|
}
|
|
|
|
void SetRWBuffers(uint32_t count, const HBuffer* buffers_, uint32_t tableIndex = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::RWBuffer;
|
|
buffers = buffers_;
|
|
}
|
|
|
|
void SetTextures(uint32_t count, const HTexture* textures_, uint32_t tableIndex = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::Texture;
|
|
textures = textures_;
|
|
}
|
|
|
|
void SetRWTexturesSlice(uint32_t count, const HTexture* textures_, uint32_t tableIndex = 0, uint32_t slice = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::RWTexture;
|
|
textures = textures_;
|
|
uavMipChain = false;
|
|
uavMipSlice = slice;
|
|
}
|
|
|
|
void SetRWTexturesChain(uint32_t count, const HTexture* textures_, uint32_t tableIndex = 0)
|
|
{
|
|
firstIndex = tableIndex;
|
|
resourceCount = count;
|
|
type = DescriptorType::RWTexture;
|
|
textures = textures_;
|
|
uavMipChain = true;
|
|
uavMipSlice = 0;
|
|
}
|
|
};
|
|
|
|
struct Rect
|
|
{
|
|
Rect() = default;
|
|
Rect(uint32_t x_, uint32_t y_, uint32_t w_, uint32_t h_)
|
|
{
|
|
x = x_;
|
|
y = y_;
|
|
w = w_;
|
|
h = h_;
|
|
}
|
|
|
|
uint32_t x = 0;
|
|
uint32_t y = 0;
|
|
uint32_t w = 0;
|
|
uint32_t h = 0;
|
|
};
|
|
|
|
struct ShaderDesc
|
|
{
|
|
ShaderDesc() = default;
|
|
ShaderDesc(ShaderStage::Id stage_, uint32_t sourceLength_, const void* source_, const char* entryPoint_ = "main", uint32_t macroCount_ = 0, const ShaderMacro* macros_ = NULL)
|
|
{
|
|
stage = stage_;
|
|
source = source_;
|
|
sourceLength = sourceLength_;
|
|
entryPoint = entryPoint_;
|
|
macroCount = macroCount_;
|
|
macros = macros_;
|
|
}
|
|
|
|
ShaderStage::Id stage = ShaderStage::Count;
|
|
uint32_t sourceLength = 0;
|
|
const void* source = NULL;
|
|
const char* entryPoint = "main";
|
|
uint32_t macroCount = 0;
|
|
const ShaderMacro* macros = NULL;
|
|
};
|
|
|
|
void Init();
|
|
void ShutDown(bool destroyWindow);
|
|
|
|
void BeginFrame();
|
|
void EndFrame();
|
|
|
|
uint32_t GetFrameIndex();
|
|
HTexture GetSwapChainTexture();
|
|
|
|
HBuffer CreateBuffer(const BufferDesc& desc);
|
|
void DestroyBuffer(HBuffer buffer);
|
|
uint8_t* MapBuffer(HBuffer buffer);
|
|
void UnmapBuffer(HBuffer buffer);
|
|
|
|
HTexture CreateTexture(const TextureDesc& desc);
|
|
void DestroyTexture(HTexture texture);
|
|
|
|
HSampler CreateSampler(const SamplerDesc& sampler);
|
|
void DestroySampler(HSampler sampler);
|
|
|
|
HRootSignature CreateRootSignature(const RootSignatureDesc& desc);
|
|
void DestroyRootSignature(HRootSignature signature);
|
|
|
|
HDescriptorTable CreateDescriptorTable(const DescriptorTableDesc& desc);
|
|
void UpdateDescriptorTable(HDescriptorTable table, const DescriptorTableUpdate& update);
|
|
void DestroyDescriptorTable(HDescriptorTable table);
|
|
|
|
HPipeline CreateGraphicsPipeline(const GraphicsPipelineDesc& desc);
|
|
HPipeline CreateComputePipeline(const ComputePipelineDesc& desc);
|
|
void DestroyPipeline(HPipeline pipeline);
|
|
|
|
HShader CreateShader(const ShaderDesc& desc);
|
|
ShaderByteCode GetShaderByteCode(HShader shader);
|
|
void DestroyShader(HShader shader);
|
|
|
|
void CmdBindRenderTargets(uint32_t colorCount, const HTexture* colorTargets, const HTexture* depthStencilTarget);
|
|
void CmdBindRootSignature(HRootSignature rootSignature);
|
|
void CmdBindDescriptorTable(HRootSignature sigHandle, HDescriptorTable table);
|
|
void CmdBindPipeline(HPipeline pipeline);
|
|
void CmdBindVertexBuffers(uint32_t count, const HBuffer* vertexBuffers, const uint32_t* byteStrides, const uint32_t* startByteOffsets);
|
|
void CmdBindIndexBuffer(HBuffer indexBuffer, IndexType::Id type, uint32_t startByteOffset);
|
|
void CmdSetViewport(uint32_t x, uint32_t y, uint32_t w, uint32_t h, float minDepth = 0.0f, float maxDepth = 1.0f);
|
|
void CmdSetScissor(uint32_t x, uint32_t y, uint32_t w, uint32_t h);
|
|
void CmdSetRootConstants(HRootSignature rootSignature, ShaderStage::Id shaderType, const void* constants);
|
|
void CmdDraw(uint32_t vertexCount, uint32_t firstVertex);
|
|
void CmdDrawIndexed(uint32_t indexCount, uint32_t firstIndex, uint32_t firstVertex);
|
|
void CmdDispatch(uint32_t groupCountX, uint32_t groupCountY, uint32_t groupCountZ);
|
|
uint32_t CmdBeginDurationQuery();
|
|
void CmdEndDurationQuery(uint32_t index);
|
|
void CmdBarrier(uint32_t texCount, const TextureBarrier* textures, uint32_t buffCount = 0, const BufferBarrier* buffers = NULL);
|
|
void CmdClearColorTarget(HTexture texture, const vec4_t clearColor, const Rect* rect = NULL);
|
|
void CmdClearDepthStencilTarget(HTexture texture, bool clearDepth, float depth, bool clearStencil = false, uint8_t stencil = 0, const Rect* rect = NULL);
|
|
void CmdInsertDebugLabel(const char* name, float r = 1.0f, float g = 1.0f, float b = 1.0f);
|
|
void CmdBeginDebugLabel(const char* name, float r = 1.0f, float g = 1.0f, float b = 1.0f);
|
|
void CmdEndDebugLabel();
|
|
void CmdSetStencilReference(uint8_t stencilRef);
|
|
void CmdCopyBuffer(HBuffer dest, HBuffer source);
|
|
|
|
#if 0
|
|
void CmdClearUAV(HTexture htexture, uint32_t mip);
|
|
void CmdNullUAVBarrier();
|
|
#endif
|
|
|
|
// the duration at index 0 is for the entire frame
|
|
uint32_t GetDurationCount();
|
|
void GetDurations(uint32_t* gpuMicroSeconds);
|
|
|
|
uint8_t* BeginBufferUpload(HBuffer buffer);
|
|
void EndBufferUpload(HBuffer buffer);
|
|
|
|
void BeginTextureUpload(MappedTexture& mappedTexture, HTexture texture);
|
|
void EndTextureUpload();
|
|
|
|
// the temporary command list is guaranteed to be done executing before the next BeginFrame call ends
|
|
void BeginTempCommandList();
|
|
void EndTempCommandList();
|
|
|
|
void BeginTextureReadback(MappedTexture& mappedTexture, HTexture texture);
|
|
void EndTextureReadback();
|
|
|
|
void WaitUntilDeviceIsIdle();
|
|
|
|
const Handle HandleIndexBitCount = 16;
|
|
const Handle HandleIndexBitOffset = 0;
|
|
const Handle HandleGenBitCount = 10;
|
|
const Handle HandleGenBitOffset = 16;
|
|
const Handle HandleTypeBitCount = 6;
|
|
const Handle HandleTypeBitOffset = 26;
|
|
|
|
inline Handle CreateHandle(Handle type, Handle index, Handle generation)
|
|
{
|
|
return
|
|
(type << HandleTypeBitOffset) |
|
|
(index << HandleIndexBitOffset) |
|
|
(generation << HandleGenBitOffset);
|
|
}
|
|
|
|
inline void DecomposeHandle(Handle* type, Handle* index, Handle* generation, Handle handle)
|
|
{
|
|
*type = (handle >> HandleTypeBitOffset) & RHI_BIT_MASK(HandleTypeBitCount);
|
|
*index = (handle >> HandleIndexBitOffset) & RHI_BIT_MASK(HandleIndexBitCount);
|
|
*generation = (handle >> HandleGenBitOffset) & RHI_BIT_MASK(HandleGenBitCount);
|
|
}
|
|
|
|
template<typename T>
|
|
bool IsNullHandle(T handle)
|
|
{
|
|
return RHI_GET_HANDLE_VALUE(handle) == 0;
|
|
}
|
|
|
|
template<typename T, typename HT, RHI::Handle RT, int N>
|
|
struct StaticPool
|
|
{
|
|
private:
|
|
struct Item
|
|
{
|
|
T item;
|
|
uint16_t generation;
|
|
uint16_t next : 15;
|
|
uint16_t used : 1;
|
|
};
|
|
|
|
public:
|
|
StaticPool()
|
|
{
|
|
Clear();
|
|
}
|
|
|
|
void Clear()
|
|
{
|
|
freeList = 0;
|
|
for(int i = 0; i < N; ++i)
|
|
{
|
|
At(i).generation = 0;
|
|
At(i).used = 0;
|
|
At(i).next = i + 1;
|
|
}
|
|
At(N - 1).next = RHI_BIT_MASK(15);
|
|
}
|
|
|
|
HT Add(const T& item)
|
|
{
|
|
ASSERT_OR_DIE(freeList < N, "The pool is full");
|
|
At(freeList).item = item;
|
|
At(freeList).used = qtrue;
|
|
const Handle handle = CreateHandle(RT, freeList, At(freeList).generation);
|
|
freeList = At(freeList).next;
|
|
return RHI_MAKE_HANDLE(handle);
|
|
}
|
|
|
|
void Remove(HT handle)
|
|
{
|
|
ASSERT_OR_DIE(!IsNullHandle(handle), "Null pool handle");
|
|
Item& item = GetItemRef(handle);
|
|
ASSERT_OR_DIE(item.used, "Memory pool item was already freed");
|
|
item.generation = (item.generation + 1) & RHI_BIT_MASK(HandleGenBitCount);
|
|
item.used = 0;
|
|
item.next = freeList;
|
|
freeList = (uint16_t)(&item - &At(0));
|
|
}
|
|
|
|
T& Get(HT handle)
|
|
{
|
|
ASSERT_OR_DIE(!IsNullHandle(handle), "Null pool handle");
|
|
return GetItemRef(handle).item;
|
|
}
|
|
|
|
T* TryGet(HT handle)
|
|
{
|
|
if(handle == 0)
|
|
{
|
|
return NULL;
|
|
}
|
|
|
|
return &GetItemRef(handle).item;
|
|
}
|
|
|
|
bool FindNext(T** object, HT* handle, int* index)
|
|
{
|
|
Q_assert(object);
|
|
Q_assert(handle);
|
|
Q_assert(index);
|
|
|
|
for(int i = *index; i < N; ++i)
|
|
{
|
|
if(At(i).used)
|
|
{
|
|
*object = &At(i).item;
|
|
*handle = RHI_MAKE_HANDLE(CreateHandle(RT, i, At(i).generation));
|
|
*index = i + 1;
|
|
return true;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
int CountUsedSlots() const
|
|
{
|
|
int used = 0;
|
|
for(int i = 0; i < N; ++i)
|
|
{
|
|
if(At(i).used)
|
|
{
|
|
used++;
|
|
}
|
|
}
|
|
|
|
return used;
|
|
}
|
|
|
|
private:
|
|
StaticPool(const StaticPool<T, HT, RT, N>&);
|
|
void operator=(const StaticPool<T, HT, RT, N>&);
|
|
|
|
Item& GetItemRef(HT handle)
|
|
{
|
|
ASSERT_OR_DIE(!IsNullHandle(handle), "Null pool handle");
|
|
|
|
Handle type, index, gen;
|
|
DecomposeHandle(&type, &index, &gen, RHI_GET_HANDLE_VALUE(handle));
|
|
ASSERT_OR_DIE(type == RT, "Invalid pool handle (wrong resource type)");
|
|
ASSERT_OR_DIE(index <= (Handle)N, "Invalid pool handle (bad index)");
|
|
|
|
Item& item = At(index);
|
|
ASSERT_OR_DIE(item.used, "Invalid pool handle (unused slot)");
|
|
if(gen > (Handle)item.generation)
|
|
{
|
|
DIE("Invalid pool handle (allocation from the future)");
|
|
}
|
|
else if(gen < (Handle)item.generation)
|
|
{
|
|
DIE("Invalid pool handle (the object has been freed)");
|
|
}
|
|
|
|
return item;
|
|
}
|
|
|
|
Item& At(uint32_t index)
|
|
{
|
|
ASSERT_OR_DIE(index < N, "Invalid pool index");
|
|
return *(Item*)&items[index * sizeof(Item)];
|
|
}
|
|
|
|
const Item& At(uint32_t index) const
|
|
{
|
|
ASSERT_OR_DIE(index < N, "Invalid pool index");
|
|
return *(Item*)&items[index * sizeof(Item)];
|
|
}
|
|
|
|
byte items[N * sizeof(Item)];
|
|
uint16_t freeList;
|
|
|
|
public:
|
|
const int size = N;
|
|
};
|
|
|
|
template<typename T, uint32_t N>
|
|
struct StaticUnorderedArray
|
|
{
|
|
StaticUnorderedArray()
|
|
{
|
|
Clear();
|
|
}
|
|
|
|
void Add(const T& value)
|
|
{
|
|
Q_assert(count < N);
|
|
if(count >= N)
|
|
{
|
|
return;
|
|
}
|
|
|
|
items[count++] = value;
|
|
}
|
|
|
|
void Remove(uint32_t index)
|
|
{
|
|
Q_assert(index < N);
|
|
if(count >= N)
|
|
{
|
|
return;
|
|
}
|
|
|
|
if(index < count - 1)
|
|
{
|
|
items[index] = items[count - 1];
|
|
}
|
|
count--;
|
|
}
|
|
|
|
void Clear()
|
|
{
|
|
count = 0;
|
|
}
|
|
|
|
T& operator[](uint32_t index)
|
|
{
|
|
Q_assert(index < N);
|
|
|
|
return items[index];
|
|
}
|
|
|
|
const T& operator[](uint32_t index) const
|
|
{
|
|
Q_assert(index < N);
|
|
|
|
return items[index];
|
|
}
|
|
|
|
private:
|
|
StaticUnorderedArray(const StaticUnorderedArray<T, N>&);
|
|
void operator=(const StaticUnorderedArray<T, N>&);
|
|
|
|
public:
|
|
T items[N];
|
|
uint32_t count;
|
|
};
|
|
|
|
template<typename T, uint32_t Invalid>
|
|
struct StaticFreeList
|
|
{
|
|
void Init(T* items_, uint32_t size_)
|
|
{
|
|
items = items_;
|
|
size = size_;
|
|
Clear();
|
|
}
|
|
|
|
uint32_t Allocate()
|
|
{
|
|
ASSERT_OR_DIE(firstFree != Invalid, "Free list out of memory");
|
|
|
|
const T index = firstFree;
|
|
firstFree = items[index];
|
|
items[index] = Invalid;
|
|
allocatedItemCount++;
|
|
|
|
return index;
|
|
}
|
|
|
|
void Free(uint32_t index)
|
|
{
|
|
ASSERT_OR_DIE(index < size, "Invalid free list slot");
|
|
|
|
const T oldList = firstFree;
|
|
firstFree = index;
|
|
items[index] = oldList;
|
|
allocatedItemCount--;
|
|
}
|
|
|
|
void Clear()
|
|
{
|
|
for(uint32_t i = 0; i < size - 1; ++i)
|
|
{
|
|
items[i] = i + 1;
|
|
}
|
|
items[size - 1] = Invalid;
|
|
firstFree = 0;
|
|
allocatedItemCount = 0;
|
|
}
|
|
|
|
T* items;
|
|
T firstFree;
|
|
uint32_t allocatedItemCount;
|
|
uint32_t size;
|
|
};
|
|
|
|
struct LinearAllocator
|
|
{
|
|
void Init(byte* data_, uint32_t size_)
|
|
{
|
|
data = data_;
|
|
size = size_;
|
|
offset = 0;
|
|
}
|
|
|
|
void Clear()
|
|
{
|
|
offset = 0;
|
|
}
|
|
|
|
const char* Allocate(const char* string)
|
|
{
|
|
const uint32_t l = strlen(string);
|
|
if(offset + l + 1 > size)
|
|
{
|
|
Q_assert(!"StringAllocator ran out of memory");
|
|
return "out of memory";
|
|
}
|
|
|
|
char* newString = (char*)data + offset;
|
|
memcpy(newString, string, l);
|
|
newString[l] = '\0';
|
|
offset += l + 1;
|
|
|
|
return newString;
|
|
}
|
|
|
|
byte* Allocate(uint32_t byteCount)
|
|
{
|
|
if(offset + byteCount > size)
|
|
{
|
|
Q_assert(!"StringAllocator ran out of memory");
|
|
return data;
|
|
}
|
|
|
|
byte* newData = data + offset;
|
|
offset += byteCount;
|
|
|
|
return newData;
|
|
}
|
|
|
|
byte* data = NULL;
|
|
uint32_t size = 0;
|
|
uint32_t offset = 0;
|
|
};
|
|
}
|