Graphics: Add SpriteChainRenderer

This commit is contained in:
Jérôme Leclercq
2021-09-05 15:50:17 +02:00
parent a18d505ae2
commit 938d965e06
18 changed files with 705 additions and 14 deletions

View File

@@ -0,0 +1,258 @@
// Copyright (C) 2017 Jérôme Leclercq
// This file is part of the "Nazara Engine - Graphics module"
// For conditions of distribution and use, see copyright notice in Config.hpp
#include <Nazara/Graphics/SpriteChainRenderer.hpp>
#include <Nazara/Graphics/Graphics.hpp>
#include <Nazara/Graphics/RenderSpriteChain.hpp>
#include <Nazara/Renderer/CommandBufferBuilder.hpp>
#include <Nazara/Renderer/RenderFrame.hpp>
#include <Nazara/Renderer/UploadPool.hpp>
#include <utility>
#include <Nazara/Graphics/Debug.hpp>
namespace Nz
{
SpriteChainRenderer::SpriteChainRenderer(RenderDevice& device, std::size_t maxVertexBufferSize) :
m_device(device),
m_maxVertexBufferSize(maxVertexBufferSize),
m_maxVertexCount(m_maxVertexBufferSize / (2 * sizeof(float))) // Treat vec2 as the minimum declaration possible
{
std::size_t maxQuadCount = m_maxVertexCount / 4;
std::size_t indexCount = 6 * maxQuadCount;
m_indexBuffer = m_device.InstantiateBuffer(BufferType::Index);
if (!m_indexBuffer->Initialize(indexCount * sizeof(UInt16), BufferUsage::DeviceLocal))
throw std::runtime_error("failed to initialize index buffer");
// Generate indices for quad (0, 1, 2, 2, 1, 3, ...)
std::vector<UInt16> indices(indexCount);
UInt16* indexPtr = indices.data();
for (std::size_t i = 0; i < maxQuadCount; ++i)
{
*indexPtr++ = i * 4 + 0;
*indexPtr++ = i * 4 + 1;
*indexPtr++ = i * 4 + 2;
*indexPtr++ = i * 4 + 2;
*indexPtr++ = i * 4 + 1;
*indexPtr++ = i * 4 + 3;
}
m_indexBuffer->Fill(indices.data(), 0, indexCount * sizeof(UInt16));
}
std::unique_ptr<ElementRendererData> SpriteChainRenderer::InstanciateData()
{
return std::make_unique<SpriteChainRendererData>();
}
void SpriteChainRenderer::Prepare(ElementRendererData& rendererData, RenderFrame& currentFrame, const Pointer<const RenderElement>* elements, std::size_t elementCount)
{
auto& data = static_cast<SpriteChainRendererData&>(rendererData);
std::vector<std::pair<UploadPool::Allocation*, AbstractBuffer*>> pendingCopies;
std::size_t firstQuadIndex = 0;
SpriteChainRendererData::DrawCall* currentDrawCall = nullptr;
UploadPool::Allocation* currentAllocation = nullptr;
UInt8* currentAllocationMemPtr = nullptr;
const VertexDeclaration* currentVertexDeclaration = nullptr;
AbstractBuffer* currentVertexBuffer = nullptr;
const RenderPipeline* currentPipeline = nullptr;
const ShaderBinding* currentInstanceBinding = nullptr;
const ShaderBinding* currentMaterialBinding = nullptr;
auto FlushDrawCall = [&]()
{
currentDrawCall = nullptr;
};
auto Flush = [&]()
{
// changing vertex buffer always mean we have to switch draw calls
FlushDrawCall();
if (currentAllocation)
{
pendingCopies.emplace_back(currentAllocation, currentVertexBuffer);
firstQuadIndex = 0;
currentAllocation = nullptr;
currentVertexBuffer = nullptr;
}
};
std::size_t oldDrawCallCount = data.drawCalls.size();
for (std::size_t i = 0; i < elementCount; ++i)
{
assert(elements[i]->GetElementType() == UnderlyingCast(BasicRenderElement::SpriteChain));
const RenderSpriteChain& spriteChain = static_cast<const RenderSpriteChain&>(*elements[i]);
const VertexDeclaration* vertexDeclaration = spriteChain.GetVertexDeclaration();
std::size_t stride = vertexDeclaration->GetStride();
if (currentVertexDeclaration != vertexDeclaration)
{
// TODO: It's be possible to use another vertex declaration with the same vertex buffer but currently very complicated
// Wait until buffer rewrite
Flush();
currentVertexDeclaration = vertexDeclaration;
}
if (currentPipeline != spriteChain.GetRenderPipeline())
{
FlushDrawCall();
currentPipeline = spriteChain.GetRenderPipeline();
}
if (currentMaterialBinding != &spriteChain.GetMaterialBinding())
{
FlushDrawCall();
currentMaterialBinding = &spriteChain.GetMaterialBinding();
}
if (currentInstanceBinding != &spriteChain.GetInstanceBinding())
{
// TODO: Flushing draw calls on instance binding means we can have e.g. 1000 sprites rendered using a draw call for each one
// which is far from being efficient, using some bindless could help (or at least instancing?)
FlushDrawCall();
currentInstanceBinding = &spriteChain.GetInstanceBinding();
}
std::size_t remainingQuads = spriteChain.GetSpriteCount();
while (remainingQuads > 0)
{
if (!currentAllocation)
{
currentAllocation = &currentFrame.GetUploadPool().Allocate(m_maxVertexBufferSize);
currentAllocationMemPtr = static_cast<UInt8*>(currentAllocation->mappedPtr);
std::shared_ptr<AbstractBuffer> vertexBuffer = m_device.InstantiateBuffer(BufferType::Vertex);
vertexBuffer->Initialize(m_maxVertexBufferSize, BufferUsage::DeviceLocal);
currentVertexBuffer = vertexBuffer.get();
data.vertexBuffers.emplace_back(std::move(vertexBuffer));
}
if (!currentDrawCall)
{
data.drawCalls.push_back(SpriteChainRendererData::DrawCall{
currentVertexBuffer,
currentPipeline,
currentInstanceBinding,
currentMaterialBinding,
6 * firstQuadIndex,
0,
});
currentDrawCall = &data.drawCalls.back();
}
std::size_t remainingSpace = m_maxVertexBufferSize - (currentAllocationMemPtr - currentAllocation->mappedPtr);
std::size_t maxQuads = remainingSpace / (4 * stride);
if (maxQuads == 0)
{
Flush();
continue;
}
std::size_t copiedQuadCount = std::min(maxQuads, remainingQuads);
std::size_t copiedSize = 4 * copiedQuadCount * stride;
std::memcpy(currentAllocationMemPtr, spriteChain.GetSpriteData(), copiedSize);
currentAllocationMemPtr += copiedSize;
firstQuadIndex += copiedQuadCount;
currentDrawCall->quadCount += copiedQuadCount;
remainingQuads -= copiedQuadCount;
// If there's still data to copy, it means buffer is full, flush it
if (remainingQuads > 0)
Flush();
}
}
//TODO: Add Finish()/PrepareEnd() call to allow to reuse buffers/draw calls for multiple Prepare calls
Flush();
const RenderSpriteChain* firstSpriteChain = static_cast<const RenderSpriteChain*>(elements[0]);
std::size_t drawCallCount = data.drawCalls.size() - oldDrawCallCount;
data.drawCallPerElement[firstSpriteChain] = SpriteChainRendererData::DrawCallIndices{ oldDrawCallCount, drawCallCount };
if (!pendingCopies.empty())
{
currentFrame.Execute([&](CommandBufferBuilder& builder)
{
for (auto&& [allocation, buffer] : pendingCopies)
builder.CopyBuffer(*allocation, buffer);
builder.PostTransferBarrier();
}, Nz::QueueType::Transfer);
}
}
void SpriteChainRenderer::Render(ElementRendererData& rendererData, CommandBufferBuilder& commandBuffer, const Pointer<const RenderElement>* elements, std::size_t elementCount)
{
auto& data = static_cast<SpriteChainRendererData&>(rendererData);
commandBuffer.BindIndexBuffer(*m_indexBuffer);
const AbstractBuffer* currentVertexBuffer = nullptr;
const RenderPipeline* currentPipeline = nullptr;
const ShaderBinding* currentInstanceBinding = nullptr;
const ShaderBinding* currentMaterialBinding = nullptr;
const RenderSpriteChain* firstSpriteChain = static_cast<const RenderSpriteChain*>(elements[0]);
auto it = data.drawCallPerElement.find(firstSpriteChain);
assert(it != data.drawCallPerElement.end());
const auto& indices = it->second;
for (std::size_t i = 0; i < indices.count; ++i)
{
const auto& drawCall = data.drawCalls[indices.start + i];
if (currentVertexBuffer != drawCall.vertexBuffer)
{
commandBuffer.BindVertexBuffer(0, *drawCall.vertexBuffer);
currentVertexBuffer = drawCall.vertexBuffer;
}
if (currentPipeline != drawCall.renderPipeline)
{
commandBuffer.BindPipeline(*drawCall.renderPipeline);
currentPipeline = drawCall.renderPipeline;
}
if (currentMaterialBinding != drawCall.materialBinding)
{
commandBuffer.BindShaderBinding(Graphics::MaterialBindingSet, *drawCall.materialBinding);
currentMaterialBinding = drawCall.materialBinding;
}
if (currentInstanceBinding != drawCall.instanceBinding)
{
commandBuffer.BindShaderBinding(Graphics::WorldBindingSet, *drawCall.instanceBinding);
currentInstanceBinding = drawCall.instanceBinding;
}
commandBuffer.DrawIndexed(drawCall.quadCount * 6, 1U, drawCall.firstIndex);
}
}
void SpriteChainRenderer::Reset(ElementRendererData& rendererData, RenderFrame& currentFrame)
{
auto& data = static_cast<SpriteChainRendererData&>(rendererData);
// TODO: Reuse vertex buffers
for (auto& vertexBufferPtr : data.vertexBuffers)
currentFrame.PushForRelease(std::move(vertexBufferPtr));
data.drawCalls.clear();
data.vertexBuffers.clear();
}
}