Skip to content

Instantly share code, notes, and snippets.

@JSandusky

JSandusky/VEZRenderScript.cpp

Created February 6, 2020 01:57
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 1 You must be signed in to fork a gist
  • Save JSandusky/190cf1bd561e36ae9c41004414ea9563 to your computer and use it in GitHub Desktop.
Save JSandusky/190cf1bd561e36ae9c41004414ea9563 to your computer and use it in GitHub Desktop.
Download ZIP
VEZ Renderer
Raw
VEZRenderScript.cpp
#include "RenderScript.h"
#include "Buffer.h"
#include "Material.h"
#include "GraphicsDevice.h"
#include "Effect.h"
#include "LightShadow.h"
#include "Packing.h"
#include "Renderables.h"
#include "Renderer.h"
#include "ShaderConstants.h"
#include <algorithm>
#include <functional>
using namespace math;
using namespace std;
#define LOG_VULKAN(MSG, RESULT) device_->LogFormat(GLVU_ERROR, MSG ": %u", RESULT)
namespace GLVU
{
#define BINDING_QUAD_CUSTOM 0
#define BINDING_QUAD_VIEW_MATRIX 3
#define BINDING_QUAD_SIZE_DATA 4
VkCommandBuffer RS_StartCommandBuffer(GraphicsDevice* device)
{
auto buffer = device->GetGraphicsCmdBuffer();
vezBeginCommandBuffer(buffer, VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
return buffer;
}
void RS_SetupViewport(GraphicsDevice* device, const View& view)
{
VkViewport viewport = { static_cast<float>(view.viewport_[0]), static_cast<float>(view.viewport_[1]), static_cast<float>(view.viewport_[2]), static_cast<float>(view.viewport_[3]), 0.0f, 1.0f };
VkRect2D scissor = { { (int)view.viewport_[0], (int)view.viewport_[1] }, { view.viewport_.Width(), view.viewport_.Height() } };
vezCmdSetViewport(0, 1, &viewport);
vezCmdSetScissor(0, 1, &scissor);
vezCmdSetViewportState(1);
}
//****************************************************************************
//
// Function: RenderScript::BeginStage
//
// Purpose: For Vulkan there's a lot of boiler-plate even with VEZ and this
// function deals with that for binding a render-target setup to draw
// to and the render-targets that will be read.
//
//****************************************************************************
void RenderScript::BeginStage(GraphicsDevice* device, View view, RenderScriptStage* stage)
{
// setup our output target configuration
VezRenderPassBeginInfo passInfo = {};
vector<VezAttachmentInfo> attachmentInfo;
RenderTargetConfiguration* targetConfig = nullptr;
for (auto& targetID : stage->targets_)
{
VezAttachmentInfo attach = {};
attach.loadOp = VkAttachmentLoadOp::VK_ATTACHMENT_LOAD_OP_LOAD;
attach.storeOp = VkAttachmentStoreOp::VK_ATTACHMENT_STORE_OP_STORE;
attachmentInfo.push_back(attach);
}
if (stage && stage->targetConfig_.fbo_ != 0)
passInfo.framebuffer = stage->targetConfig_.fbo_->GetGPUObject();
else
{
passInfo.framebuffer = device->GetBackbuffer()->GetGPUObject();
VezAttachmentInfo attach = { };
attach.loadOp = VkAttachmentLoadOp::VK_ATTACHMENT_LOAD_OP_LOAD;
attach.storeOp = VkAttachmentStoreOp::VK_ATTACHMENT_STORE_OP_STORE;
attachmentInfo.push_back(attach);
attachmentInfo.push_back(attach);
}
passInfo.pAttachments = attachmentInfo.data();
passInfo.attachmentCount = attachmentInfo.size();
vezCmdBeginRenderPass(&passInfo);
// bind render target textures from inputs
for (auto& targetInputBinding : stage->targetBindings_)
{
bool texFound = false;
for (auto rec : targetTextures_)
{
if (rec->name_.nameHash_ == targetInputBinding.second.nameHash_)
{
if (auto found = GetTargetTexture(targetInputBinding.second.nameHash_))
{
if (found->texture_->IsTextureBuffer())
vezCmdBindBufferView(found->texture_->GetTextureBufferView(), 1, targetInputBinding.first, 0);
else
vezCmdBindImageView(rec->texture_->GetView(), device_->GetSampler(FILTER_LINEAR, false), 1, targetInputBinding.first, 0);
texFound = true;
}
else
{
if (auto system = device->GetSystemTexture(rec->name_.name_))
{
if (system->IsTextureBuffer())
vezCmdBindBufferView(system->GetTextureBufferView(), 1, targetInputBinding.first, 0);
else
vezCmdBindImageView(system->GetView(), device_->GetSampler(FILTER_LINEAR, false), 1, targetInputBinding.first, 0);
}
else // fallback on a default
vezCmdBindImageView(device->GetDefaultTexture()->GetView(), device_->GetSampler(FILTER_LINEAR, false), 1, targetInputBinding.first, 0);
}
break;
}
}
if (!texFound)
{
if (auto system = device->GetSystemTexture(targetInputBinding.second.name_))
{
if (system->IsTextureBuffer())
vezCmdBindBufferView(system->GetTextureBufferView(), 1, targetInputBinding.first, 0);
else
vezCmdBindImageView(system->GetView(), device_->GetSampler(FILTER_LINEAR, false), 1, targetInputBinding.first, 0);
}
}
}
}
//****************************************************************************
//
// Function: RenderScript::EndStage
//
// Purpose: Just closes a render-pass. Each stage consists of 1 VEZ renderpass.
//
//****************************************************************************
void RenderScript::EndStage(GraphicsDevice* device, RenderScriptStage* stage)
{
vezCmdEndRenderPass();
}
//****************************************************************************
//
// Function: RenderScript::Execute
//
// Purpose: Runs this script for the given view. An extensive function
// responsible for executing the stages and their commands.
//
//****************************************************************************
void RenderScript::Execute(Renderer* renderer, View view, const vector<Batch>& batches, const vector< shared_ptr<Light> >& lights)
{
if (view.renderTarget_ == nullptr)
view.renderTarget_ = device_->GetBackbuffer();
view.camera_->SetViewport(view.viewport_);
if (width_ != view.renderTarget_->GetWidth() || height_ != view.renderTarget_->GetHeight())
OnBackbufferResize(device_, view.renderTarget_->GetWidth(), view.renderTarget_->GetHeight());
if (!viewUniformBuffer_)
{
viewUniformBuffer_ = device_->CreateUniformBuffer();
viewUniformBuffer_->SetTag(BufferTag_Dynamic);
viewUniformBuffer_->SetShadowed(true);
viewUniformBuffer_->SetSize(sizeof(viewBufferData_));
}
viewBufferData_.viewProj = view.camera_->GetViewMatrix();
viewBufferData_.invViewProj = view.camera_->GetInvViewProjection();
viewBufferData_.viewPos = { view.camera_->GetPosition().x, view.camera_->GetPosition().y, view.camera_->GetPosition().z, 0.0f };
auto camDir = view.camera_->GetDirection();
auto upDir = view.camera_->GetUp();
viewBufferData_.viewDir = { camDir.x, camDir.y, camDir.z, 1.0f };
viewBufferData_.viewUp = { upDir.x, upDir.y, upDir.z, 1.0f };
viewBufferData_.viewport = { view.viewport_[0], view.viewport_[1], view.viewport_[2], view.viewport_[3] };
viewBufferData_.nearFar = { view.camera_->GetNear(), view.camera_->GetFar(), 0, 0 };
viewUniformBuffer_->SetData(&viewBufferData_, sizeof(viewBufferData_));
// Verify that all of our shadow data is up to date
for (uint32_t b = 0; b < batches.size(); ++ b)
{
auto& batch = batches[b];
if (batch.material_)
batch.material_->CommitUniforms();
}
BatchQueue queue;
for (uint32_t b = 0; b < batches.size(); ++b)
{
auto& batch = batches[b];
queue.Add(batch);
}
queue.SortFrontToBack();
PrepareQueue(queue);
for (uint32_t stageIdx = 0; stageIdx < stages_.size(); ++stageIdx)
{
const auto& stage = stages_[stageIdx];
if (stage->active_)
{
if (!ShouldStageExecute(stage))
continue;
for (const auto& cmd : stage->commands_)
{
if (!cmd.enabled_)
continue;
if (cmd.commandType_ == ClearTargets)
{
auto cmdBuffer = RS_StartCommandBuffer(device_);
renderer->GetCommandBufferChain().push_back(cmdBuffer);
for (size_t i = 0; i < stage->targetConfig_.targets_.size(); ++i)
{
auto tgt = stage->targetConfig_.targets_[i];
bool clearDepth = cmd.cmdData_.clearData_.discardDepth_;
bool clearStencil = cmd.cmdData_.clearData_.discardStencil_;
bool clearColor = cmd.cmdData_.clearData_.discardColor_;
RS_SetupViewport(device_, view);
VezClearAttachment value;
value.colorAttachment = (uint32_t)i;
if (tgt->targetFormat_ == TEX_DEPTH && (clearDepth || clearStencil))
{
VkClearDepthStencilValue value = {};
value.depth = cmd.cmdData_.clearData_.depth_;
value.stencil = cmd.cmdData_.clearData_.stencilValue_;
VezImageSubresourceRange range = {};
range.layerCount = 1;
range.levelCount = 1;
vezCmdClearDepthStencilImage(tgt->texture_->GetImage(), &value, 1, &range);
}
else if (tgt->targetFormat_ != TEX_DEPTH && clearColor)
{
VezImageSubresourceRange range = {};
range.layerCount = 1;
range.levelCount = 1;
VkClearColorValue value = { cmd.cmdData_.clearData_.color_[0], cmd.cmdData_.clearData_.color_[1], cmd.cmdData_.clearData_.color_[2], cmd.cmdData_.clearData_.color_[3] };
vezCmdClearColorImage(tgt->texture_->GetImage(), &value, 1, &range);
}
}
if (stage->targetConfig_.targets_.empty())
{
// backbuffers
for (uint32_t i = 0; i < device_->GetBackbuffer()->GetTextureCount(); ++i)
{
auto tex = device_->GetBackbuffer()->GetTexture(i);
bool clearDepth = cmd.cmdData_.clearData_.discardDepth_;
bool clearStencil = cmd.cmdData_.clearData_.discardStencil_;
bool clearColor = cmd.cmdData_.clearData_.discardColor_;
VezClearAttachment value;
value.colorAttachment = (uint32_t)i;
if (tex->GetFormat() == TEX_DEPTH && (clearDepth || clearStencil))
{
VkClearDepthStencilValue value = {};
value.depth = cmd.cmdData_.clearData_.depth_;
value.stencil = cmd.cmdData_.clearData_.stencilValue_;
VezImageSubresourceRange range = {};
range.layerCount = 1;
range.levelCount = 1;
vezCmdClearDepthStencilImage(tex->GetImage(), &value, 1, &range);
}
else if (tex->GetFormat() != TEX_DEPTH && clearColor)
{
VezImageSubresourceRange range = {};
range.layerCount = 1;
range.levelCount = 1;
VkClearColorValue value = { cmd.cmdData_.clearData_.color_[0], cmd.cmdData_.clearData_.color_[1], cmd.cmdData_.clearData_.color_[2], cmd.cmdData_.clearData_.color_[3] };
vezCmdClearColorImage(tex->GetImage(), &value, 1, &range);
}
}
}
vezEndCommandBuffer();
}
}
for (const auto& cmd : stage->commands_)
{
if (!cmd.enabled_)
continue;
if (cmd.commandType_ == ClearTargets)
continue;
else if (cmd.commandType_ == GeometryPass)
{
if (device_->ThreadingIsSupported() && queue.ConsiderSplitting())
{
int threadCt = device_->GetNumThreads();
vector<BatchQueue> queues;
queue.Split(queues, threadCt);
vector<DrawBatchesTaskData> tasks(queues.size());
struct GeomPassThreadTaskData {
Renderer* renderer;
RenderScript* script;
RenderScriptStage* stage;
DrawBatchesTaskData* taskData;
BatchQueue* queue;
View view;
};
vector<GeomPassThreadTaskData> taskData;
size_t baseLoc = renderer->GetCommandBufferChain().size();
renderer->GetCommandBufferChain().resize(renderer->GetCommandBufferChain().size() + tasks.size());
for (uint32_t i = 0; i < tasks.size(); ++i)
{
tasks[i].uboBindings_.push_back({ 1, viewUniformBuffer_ });
tasks[i].contextNameHash_ = cmd.context_.nameHash_;
tasks[i].skinnedNameHash_ = Hash(string(cmd.context_.name_) + "_SKINNED");
tasks[i].instancedNameHash_ = Hash(string(cmd.context_.name_) + "_INST");
tasks[i].placeCmdBufferAt_ = baseLoc + i;
PrepareQueue(queues[i]);
taskData.push_back({ renderer, this, stage, &tasks[i], &queues[i], view });
}
//TODO
for (int i = 0; i < queues.size(); ++i)
{
device_->PushThreadJob(i, &taskData[i], [](uint32_t taskID, void* data) {
GeomPassThreadTaskData* taskData = (GeomPassThreadTaskData*)data;
taskData->script->DrawBatches(taskData->renderer, taskData->view, taskData->stage, *(taskData->queue), *(taskData->taskData));
});
}
device_->WaitForThreadJobs();
}
else
{
DrawBatchesTaskData task;
task.uboBindings_.push_back({ 1, viewUniformBuffer_ });
task.contextNameHash_ = cmd.context_.nameHash_;
task.skinnedNameHash_ = Hash(string(cmd.context_.name_) + "_SKINNED");
task.instancedNameHash_ = Hash(string(cmd.context_.name_) + "_INST");
DrawBatches(renderer, view, stage, queue, task);
}
}
else if (cmd.commandType_ == FullscreenQuad)
{
if (!cmd.effect_)
continue;
BufferPack pack(256);
struct ViewData {
float2 inputTexSize;
float2 invInputTexSize;
float2 outputTexSize;
float2 invOutputTexSize;
};
struct ParamData {
float params[32];
};
pack.Pack<float4x4>();
pack.Pack<ViewData>();
pack.Pack<ParamData>();
auto quadCmdUBO = device_->GetScratchUniformBuffer(pack.CalculateSize());
pack.AllocateUBO(quadCmdUBO);
// view matrix
pack.Get<float4x4>(0) = float4x4::OpenGLOrthoProjLH(0, 1, 2, 2);
// INPUT and OUTPUT size information sources
if (cmd.cmdData_.quadData_.inputSize_.IsValid())
{
if (auto info = GetTargetTexture(cmd.cmdData_.quadData_.inputSize_.nameHash_))
pack.Get<ViewData>(1).inputTexSize = { (float)info->width_, (float)info->height_ };
else
pack.Get<ViewData>(1).inputTexSize = { (float)view.viewport_[2], (float)view.viewport_[3] };
}
else
pack.Get<ViewData>(1).inputTexSize = { (float)view.viewport_[2], (float)view.viewport_[3] };
if (cmd.cmdData_.quadData_.outputSize_.IsValid())
{
if (auto info = GetTargetTexture(cmd.cmdData_.quadData_.outputSize_.nameHash_))
pack.Get<ViewData>(1).outputTexSize = { (float)info->width_, (float)info->height_ };
else
pack.Get<ViewData>(1).outputTexSize = { (float)view.viewport_[2], (float)view.viewport_[3] };
}
else
pack.Get<ViewData>(1).outputTexSize = { (float)view.viewport_[2], (float)view.viewport_[3] };
pack.Get<ViewData>(1).invInputTexSize = float2(1, 1) / pack.Get<ViewData>(1).inputTexSize;
pack.Get<ViewData>(1).invOutputTexSize = float2(1, 1) / pack.Get<ViewData>(1).outputTexSize;
// 128 bytes just isn't something to worry about, so copy the whole thing regardless how many are really used
memcpy(pack.Get<ParamData>(2).params, cmd.cmdData_.quadData_.params_, sizeof(float) * RENDER_SCRIPT_MAX_PARAMS);
pack.Transfer(quadCmdUBO, false);
auto cmdBuffer = RS_StartCommandBuffer(device_);
renderer->GetCommandBufferChain().push_back(cmdBuffer);
BeginStage(device_, view, stage);
{
auto pass = cmd.effect_->GetPasses()[0];
auto geo = device_->GetFSTriGeometry();
geo->layout_->Bind(geo.get(), { });
RS_SetupViewport(device_, view);
// Set rasterization state.
VezRasterizationState rasterizationState = { };
rasterizationState.polygonMode = VK_POLYGON_MODE_FILL;
rasterizationState.cullMode = VK_CULL_MODE_NONE;
rasterizationState.frontFace = VK_FRONT_FACE_CLOCKWISE;
vezCmdSetRasterizationState(&rasterizationState);
// Set depth stencil state.
VezDepthStencilState depthStencilState = {};
depthStencilState.depthTestEnable = VK_FALSE;
depthStencilState.depthWriteEnable = VK_FALSE;
depthStencilState.depthCompareOp = VK_COMPARE_OP_ALWAYS;
depthStencilState.stencilTestEnable = VK_FALSE;
depthStencilState.depthBoundsTestEnable = VK_FALSE;
vezCmdSetDepthStencilState(&depthStencilState);
for (auto tex : cmd.textures_)
{
if (tex.second->IsTextureBuffer())
vezCmdBindBufferView(tex.second->GetTextureBufferView(), 1, tex.first, 0);
else
vezCmdBindImageView(tex.second->GetView(), device_->GetSampler(FILTER_POINT, true), 1, tex.first, 0);
}
VezInputAssemblyState state = { };
state.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
vezCmdSetInputAssemblyState(&state);
vezCmdBindPipeline(pass->GetPipeline());
auto vbo = geo->vertexBuffers_[0]->GetGPUObject();
VkDeviceSize junk = 0;
vezCmdBindVertexBuffers(0, 1, &vbo, &junk);
vezCmdBindIndexBuffer(device_->GetSequentialIndexBuffer()->GetGPUObject(), 0, VK_INDEX_TYPE_UINT16);
vezCmdBindBuffer(quadCmdUBO->GetGPUObject(), pack.OffsetOf(0), pack.SizeOf(0), 0, BINDING_QUAD_VIEW_MATRIX, 0);
vezCmdBindBuffer(quadCmdUBO->GetGPUObject(), pack.OffsetOf(1), pack.SizeOf(1), 0, BINDING_QUAD_SIZE_DATA, 0);
vezCmdBindBuffer(quadCmdUBO->GetGPUObject(), pack.OffsetOf(2), pack.SizeOf(2), 0, 5, 0);
vezCmdDrawIndexed(3, 1, 0, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, 1);
}
EndStage(device_, stage);
vezEndCommandBuffer();
}
else if (cmd.commandType_ == LightVolumes)
{
auto lightFX = device_->GetDeferredLightEffect();
BatchQueue lightQueue;
vector<float4x4> transforms;
transforms.resize(lights.size());
shared_ptr<Geometry> geometry;
auto transformBuffer = device_->GetScratchUniformBuffer(sizeof(float4x4));
LightData lightData;
auto cmdBuffer = RS_StartCommandBuffer(device_);
renderer->GetCommandBufferChain().push_back(cmdBuffer);
VezRasterizationState frontFaces = {};
frontFaces.polygonMode = VK_POLYGON_MODE_FILL;
frontFaces.cullMode = VK_CULL_MODE_BACK_BIT;
frontFaces.frontFace = VK_FRONT_FACE_CLOCKWISE;
frontFaces.depthBiasEnable = VK_FALSE;
VezRasterizationState backFaces = {};
backFaces.polygonMode = VK_POLYGON_MODE_FILL;
backFaces.cullMode = VK_CULL_MODE_FRONT_BIT;
backFaces.frontFace = VK_FRONT_FACE_CLOCKWISE;
backFaces.depthBiasEnable = VK_FALSE;
VezRasterizationState noCullFaces = {};
noCullFaces.polygonMode = VK_POLYGON_MODE_FILL;
noCullFaces.cullMode = VK_CULL_MODE_NONE;
noCullFaces.frontFace = VK_FRONT_FACE_CLOCKWISE;
noCullFaces.depthBiasEnable = VK_FALSE;
noCullFaces.depthClampEnable = VK_TRUE;
VezDepthStencilState depthStencil = {};
depthStencil.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL;
depthStencil.depthTestEnable = VK_TRUE;
depthStencil.depthWriteEnable = VK_FALSE;
depthStencil.stencilTestEnable = VK_FALSE;
shared_ptr<ShaderPass> lastPass = nullptr;
BeginStage(device_, view, stage);
if (lights.size() > 0)
RS_SetupViewport(device_, view);
RollingBufferAllocator lightBuffers(device_);
for (uint32_t i = 0; i < lights.size(); ++i)
{
auto light = lights[i];
lightData.lightMat = light->GetShadowMatrix(0);
lightData.lightPos = float4(lights[i]->GetPosition(), (int)light->GetKind());
lightData.lightDir = float4(lights[i]->GetDirection(), light->GetRadius());
lightData.color = light->GetColor();
lightData.extraParams.x = light->GetFOV();
lightData.extraParams.y = light->IsShadowCasting() ? 1.0f : 0.0f;
lightData.shadowMapCoords[0] = light->IsShadowCasting() ? light->GetShadowDomain(0) : float4::zero;
lightData.shadowMapCoords[1] = light->IsShadowCasting() ? light->GetShadowDomain(1) : float4::zero;
auto alloc = lightBuffers.Allocate(sizeof(lightData));
memcpy(alloc.first, &lightData, sizeof(lightData));
//auto lightDataBuffer = device_->GetScratchUniformBuffer(sizeof(LightData));
//lightDataBuffer->SetData(&lightData, sizeof(LightData));
}
lightBuffers.Finish();
for (uint32_t i = 0; i < lights.size(); ++i)
{
auto light = lights[i];
if (light->GetKind() == Light::POINT)
{
auto pass = lightFX->GetPass("POINT_LIGHT", PRIM_UNKNOWN);
geometry = device_->GetPointLightGeometry();
auto transform = float4x4::FromTRS(light->GetPosition(), Quat::identity, float3(light->GetRadius()));
auto dataT = lightBuffers.GetAllocation(i);
vezCmdBindBuffer(dataT.buffer_->GetGPUObject(), dataT.allocationOffset_, dataT.allocationSize_, 0, 0, 0);
//vezCmdBindBuffer(lightDataBuffer->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 0, 0);
vezCmdBindBuffer(viewUniformBuffer_->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 1, 0);
ApplyPass(pass, view);
ApplyGeometry(pass, geometry.get(), false);
ApplyBlendMode(light->GetColor().w < 0 ? Blend_Subtract : Blend_Add, view, stage);
vezCmdPushConstants(0, sizeof(float4x4), &transform);
if (light->Contains(view.camera_))
{
vezCmdSetRasterizationState(&backFaces);
depthStencil.depthCompareOp = VK_COMPARE_OP_GREATER;
}
else
{
vezCmdSetRasterizationState(&frontFaces);
depthStencil.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL;
}
if (light->GetShadowMapTexture())
vezCmdBindImageView(light->GetShadowMapTexture()->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
else
vezCmdBindImageView(device_->GetSystemTexture(PIPELINE_RESOURCE_SHADOWMAP)->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
vezCmdSetDepthStencilState(&depthStencil);
vezCmdDrawIndexed(geometry->indexCount_, 1, 0, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, geometry->primCount_);
//EndStage(device, stage);
}
else if (light->GetKind() == Light::SPOT)
{
auto pass = lightFX->GetPass("SPOT_LIGHT", PRIM_UNKNOWN);
geometry = device_->GetSpotLightGeometry();
auto dataT = lightBuffers.GetAllocation(i);
vezCmdBindBuffer(dataT.buffer_->GetGPUObject(), dataT.allocationOffset_, dataT.allocationSize_, 0, 0, 0);
//vezCmdBindBuffer(lightDataBuffer->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 0, 0);
vezCmdBindBuffer(viewUniformBuffer_->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 1, 0);
ApplyPass(pass, view);
ApplyGeometry(pass, geometry.get(), false);
ApplyBlendMode(light->GetColor().w < 0 ? Blend_Subtract : Blend_Add, view, stage);
float yScale = tanf(math::DegToRad(light->GetFOV()) * 0.5f) * light->GetRadius() * 0.53f;
float xScale = yScale;
auto transform = float4x4::FromTRS(light->GetPosition(), light->GetRotation(), float3(xScale, yScale, light->GetRadius()));
vezCmdPushConstants(0, sizeof(float4x4), &transform);
if (light->Contains(view.camera_))
{
vezCmdSetRasterizationState(&backFaces);
depthStencil.depthCompareOp = VK_COMPARE_OP_GREATER;
}
else
{
vezCmdSetRasterizationState(&frontFaces);
depthStencil.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL;
}
if (light->GetShadowMapTexture())
vezCmdBindImageView(light->GetShadowMapTexture()->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
else
vezCmdBindImageView(device_->GetSystemTexture(PIPELINE_RESOURCE_SHADOWMAP)->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
vezCmdSetDepthStencilState(&depthStencil);
vezCmdDrawIndexed(geometry->indexCount_, 1, 0, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, geometry->primCount_);
}
else if (light->GetKind() == Light::DIRECTIONAL)
{
auto pass = lightFX->GetPass("DIRECTIONAL_LIGHT", PRIM_UNKNOWN);
if (pass != lastPass)
{
ApplyPass(pass, view);
lastPass = pass;
}
geometry = device_->GetFSTriGeometry();
ApplyGeometry(pass, geometry.get(), false);
auto viewMatrix = float4x4::OpenGLOrthoProjLH(0, 1, 2, 2);
ViewBufferData dirLightOverride = viewBufferData_;
dirLightOverride.invViewProj = viewMatrix;
dirLightOverride.viewProj = viewMatrix;
auto viewDataBuffer = device_->GetScratchUniformBuffer(sizeof(ViewBufferData));
viewDataBuffer->SetData(&dirLightOverride, sizeof(ViewBufferData));
ApplyBlendMode(Blend_Add, view, stage);
auto dataT = lightBuffers.GetAllocation(i);
vezCmdBindBuffer(dataT.buffer_->GetGPUObject(), dataT.allocationOffset_, dataT.allocationSize_, 0, 0, 0);
//vezCmdBindBuffer(lightDataBuffer->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 0, 0);
vezCmdBindBuffer(viewDataBuffer->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 1, 0);
vezCmdSetRasterizationState(&noCullFaces);
if (light->GetShadowMapTexture())
vezCmdBindImageView(light->GetShadowMapTexture()->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
else
vezCmdBindImageView(device_->GetSystemTexture(PIPELINE_RESOURCE_SHADOWMAP)->GetView(), device_->GetSampler(FILTER_SHADOW, false), 1, SHADER_TEX_SHADOWMAP, 0);
auto transform = float4x4::identity;
vezCmdPushConstants(0, sizeof(float4x4), &transform);
VezDepthStencilState depthStencilState = {};
depthStencilState.depthTestEnable = VK_FALSE;
depthStencilState.depthWriteEnable = VK_FALSE;
depthStencilState.depthCompareOp = VK_COMPARE_OP_ALWAYS;
depthStencilState.stencilTestEnable = VK_FALSE;
depthStencilState.depthBoundsTestEnable = VK_FALSE;
vezCmdSetDepthStencilState(&depthStencilState);
vezCmdDraw(3, 1, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, 1);
}
}
EndStage(device_, stage);
vezEndCommandBuffer();
}
else if (cmd.commandType_ == DeferredTiledLights)
{
// get deferred tiled pass
VkBuffer fsQuad = device_->GetFullscreenTriVertices()->GetGPUObject();
device_->GetLayout_PosUV()->Bind(nullptr, { });
vezCmdBindVertexBuffers(0, 1, &fsQuad, 0);
vezCmdDraw(3, 1, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, 1);
}
else if (cmd.commandType_ == ForwardTiledLights)
{
ShaderPass* activePass = nullptr;
shared_ptr<Geometry> activeGeometry;
shared_ptr<Material> activeMaterial;
static uint32_t tiledLightingPassID = 0;
// render the geometries now that we've bound the tiled lights
shared_ptr<ShaderPass> lastPass = nullptr;
Geometry* lastGeometry = nullptr;
Material* lastMaterial = nullptr;
for (const auto& batch : batches)
{
if (auto pass = batch.material_->GetEffect()->GetPass(tiledLightingPassID, batch.geometry_->primType_))
{
if (pass != lastPass)
{
lastPass = pass;
vezCmdBindPipeline(lastPass->GetPipeline());
}
if (batch.material_ != lastMaterial)
{
lastMaterial = batch.material_;
}
if (batch.geometry_ != lastGeometry)
{
lastGeometry = batch.geometry_;
vector<VkBuffer> vbos;
for (auto b : lastGeometry->vertexBuffers_)
vbos.push_back(b->GetGPUObject());
VkDeviceSize junk = 0;
vezCmdBindVertexBuffers(0, (uint32_t)vbos.size(), vbos.data(), &junk);
if (lastGeometry->indexBuffer_)
vezCmdBindIndexBuffer(lastGeometry->indexBuffer_->GetGPUObject(), 0, lastGeometry->indexBuffer_->HasTag(BufferTag_32Bit) ? VK_INDEX_TYPE_UINT32 : VK_INDEX_TYPE_UINT16);
}
if (lastGeometry->indexBuffer_)
{
vezCmdDrawIndexed(lastGeometry->GetIndexCount(), batch.numTransforms_, lastGeometry->indexStart_, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, lastGeometry->primCount_ * batch.numTransforms_);
}
else
{
vezCmdDraw(lastGeometry->vertexCount_, batch.numTransforms_, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, lastGeometry->primCount_ * batch.numTransforms_);
}
}
}
}
else if (cmd.commandType_ == ForwardLights)
{
for (uint32_t i = 0; i < lights.size(); ++i)
{
auto light = lights[i];
AABB dumbBnds;
dumbBnds.SetNegativeInfinity();
auto batches = view.scene_->GetBatches(light.get(), UINT_MAX, dumbBnds);
if (batches.empty())
continue;
LightData lightData;
lightData.lightMat = light->GetShadowMatrix(0);
lightData.lightPos = float4(lights[i]->GetPosition(), (int)light->GetKind());
lightData.lightDir = float4(lights[i]->GetDirection(), light->GetRadius());
lightData.color = light->GetColor();
lightData.extraParams.x = light->GetFOV();
lightData.extraParams.y = light->IsShadowCasting() ? 1.0f : 0.0f;
lightData.shadowMapCoords[0] = light->IsShadowCasting() ? light->GetShadowDomain(0) : float4::zero;
lightData.shadowMapCoords[1] = light->IsShadowCasting() ? light->GetShadowDomain(1) : float4::zero;
auto lightDataBuffer = device_->GetScratchUniformBuffer(sizeof(LightData));
lightDataBuffer->SetData(&lightData, sizeof(lightData));
BatchQueue lightQueue;
for (uint32_t i = 0; i < batches.size(); ++i)
lightQueue.Add(batches[i]);
DrawBatchesTaskData task;
task.uboBindings_.push_back({ SHADER_BUFFER_LIGHT_DATA, lightDataBuffer });
task.clearTargets_ = false;
task.textureBindings_.push_back({ SHADER_TEX_SHADOWMAP, light->GetShadowMapTexture() });
if (light->GetKind() == Light::POINT)
{
task.contextNameHash_ = Hash(string(cmd.context_.name_) + "_LITPOINT");
task.skinnedNameHash_ = Hash(string(cmd.context_.name_) + "_LITPOINT_SKINNED");
task.instancedNameHash_ = Hash(string(cmd.context_.name_) + "_LITPOINT_INST");
}
else if (light->GetKind() == Light::SPOT)
{
task.contextNameHash_ = Hash(string(cmd.context_.name_) + "_LITSPOT");
task.skinnedNameHash_ = Hash(string(cmd.context_.name_) + "_LITSPOT_SKINNED");
task.instancedNameHash_ = Hash(string(cmd.context_.name_) + "_LITSPOT_INST");
}
else if (light->GetKind() == Light::DIRECTIONAL)
{
task.contextNameHash_ = Hash(string(cmd.context_.name_) + "_LITDIR");
task.skinnedNameHash_ = Hash(string(cmd.context_.name_) + "_LITDIR_SKINNED");
task.instancedNameHash_ = Hash(string(cmd.context_.name_) + "_LITDIR_INST");
}
DrawBatches(renderer, view, stage, lightQueue, task);
}
}
else if (cmd.commandType_ == Blit)
{
auto src = GetTargetTexture(cmd.cmdData_.blitData_.source_.nameHash_);
auto dest = GetTargetTexture(cmd.cmdData_.blitData_.dest_.nameHash_);
if (src && dest)
{
VezImageBlit blitInfo = { };
blitInfo.srcSubresource.mipLevel = 0;
blitInfo.srcSubresource.baseArrayLayer = 0;
blitInfo.srcSubresource.layerCount = 1;
blitInfo.srcOffsets[1].x = src->width_;
blitInfo.srcOffsets[1].y = src->height_;
blitInfo.srcOffsets[1].z = 1;
blitInfo.dstSubresource.mipLevel = 0;
blitInfo.dstSubresource.baseArrayLayer = 0;
blitInfo.dstSubresource.layerCount = 1;
blitInfo.dstOffsets[1].x = dest->width_;
blitInfo.dstOffsets[1].y = dest->height_;
blitInfo.dstOffsets[1].z = 1;
vezCmdBlitImage(src->texture_->GetImage(), dest->texture_->GetImage(), 1, &blitInfo, VK_FILTER_LINEAR);
}
}
else if (cmd.commandType_ == GenMips)
{
if (auto tgt = GetTargetTexture(cmd.cmdData_.genMipsData_.texture_.nameHash_))
{
auto width = tgt->width_;
auto height = tgt->height_;
const uint32_t layerCt = cmd.cmdData_.genMipsData_.layer_ == -1 ? tgt->texture_->GetLayers() : 1;
const uint32_t layerBase = cmd.cmdData_.genMipsData_.layer_ == -1 ? 0 : cmd.cmdData_.genMipsData_.layer_;
for (auto level = 1; level < tgt->texture_->GetMips(); ++level)
{
VezImageBlit blitInfo = {};
blitInfo.srcSubresource.mipLevel = level - 1;
blitInfo.srcSubresource.baseArrayLayer = layerBase;
blitInfo.srcSubresource.layerCount = layerCt;
blitInfo.srcOffsets[1].x = tgt->width_ >> (level - 1);
blitInfo.srcOffsets[1].y = tgt->height_ >> (level - 1);
blitInfo.srcOffsets[1].z = 1;
blitInfo.dstSubresource.mipLevel = level;
blitInfo.dstSubresource.baseArrayLayer = layerBase;
blitInfo.dstSubresource.layerCount = layerCt;
blitInfo.dstOffsets[1].x = tgt->width_ >> level;
blitInfo.dstOffsets[1].y = tgt->height_ >> level;
blitInfo.dstOffsets[1].z = 1;
vezCmdBlitImage(tgt->texture_->GetImage(), tgt->texture_->GetImage(), 1, &blitInfo, VK_FILTER_LINEAR);
}
}
}
else if (cmd.commandType_ == BufferCopy)
{
auto srcBuffer = FindBuffer(cmd.cmdData_.buffCopyData_.source_);
auto destBuffer = FindBuffer(cmd.cmdData_.buffCopyData_.dest_);
if (srcBuffer && destBuffer)
{
VezBufferCopy reg = { };
reg.size = srcBuffer->GetSize();
reg.dstOffset = 0;
reg.srcOffset = 0;
vezCmdCopyBuffer(srcBuffer->GetGPUObject(), destBuffer->GetGPUObject(), 0, &reg);
}
}
else if (cmd.commandType_ == ComputePass)
{
if (auto pass = cmd.effect_->GetPass(cmd.context_.nameHash_, PRIM_UNKNOWN))
{
ApplyPass(pass, view);
for (auto& uboRecord : cmd.buffers_)
{
if (auto found = GetDataBuffer(uboRecord.second.nameHash_))
vezCmdBindBuffer(found->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, uboRecord.first, 0);
}
for (auto& texRecord : cmd.textures_)
cmd.effect_->BindTexture(texRecord.second, texRecord.first);
if (cmd.cmdData_.computeData_.numParams_ > 0)
{
auto scratch = device_->GetScratchUniformBuffer(sizeof(float) * RENDER_SCRIPT_MAX_PARAMS);
scratch->SetData((void*)cmd.cmdData_.computeData_.params_, sizeof(float) * RENDER_SCRIPT_MAX_PARAMS);
vezCmdBindBuffer(scratch->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 0, 0);
}
vezCmdDispatch(cmd.cmdData_.computeData_.groupsX_, cmd.cmdData_.computeData_.groupsY_, cmd.cmdData_.computeData_.groupsZ_);
}
}
else if (cmd.commandType_ == RenderCallback)
{
device_->InvokeCallback(cmd.cmdData_.callData_.callID_.name_, this, &view, (float*)cmd.cmdData_.callData_.params_, cmd.cmdData_.callData_.numParams_);
}
}
}
}
}
//****************************************************************************
//
// Function: RenderScript::DrawBatches
//
// Purpose: Executes the bind + draw loop for a queue of batches that are
// to be drawn to a specific view. Task data is used to encapsulate
// a lot of the detail configuration such as which passes to look
// for and UBOs/textures that should be inherited.
//
//****************************************************************************
void RenderScript::DrawBatches(Renderer* renderer, View view, RenderScriptStage* stage, const BatchQueue& batchQueue, DrawBatchesTaskData& taskData)
{
shared_ptr<ShaderPass> activePass = nullptr;
Geometry* activeGeometry = nullptr;
Material* activeMaterial = nullptr;
static auto applyPass = [](const shared_ptr<ShaderPass> pass, View view) {
vezCmdBindPipeline(pass->GetPipeline());
VkViewport v = {};
v.x = (float)view.viewport_[0];
v.y = (float)view.viewport_[1];
v.width = view.viewport_.Width();
v.height = view.viewport_.Height();
v.minDepth = 0.0f;
v.maxDepth = 1.0f;
vezCmdSetViewport(0, 1, &v);
if (pass->GetDrawState().stencilTest_)
{
//??
}
};
static auto applyMaterial = [](GraphicsDevice* device, View view, RenderScriptStage* stage, shared_ptr<ShaderPass> pass, Material* mat, Geometry* geom, DrawBatchesTaskData& taskData) {
for (auto& uboRecord : mat->GetUBOs())
{
bool ignore = false;
for (auto& rec : taskData.uboBindings_)
{
if (rec.first == uboRecord.first)
{
ignore = true;
break;
}
}
if (!ignore)
vezCmdBindBuffer(uboRecord.second->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, uboRecord.first, 0);
}
for (auto& rec : taskData.uboBindings_)
vezCmdBindBuffer(rec.second->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, rec.first, 0);
for (auto& tex : taskData.textureBindings_)
{
auto traits = mat->GetEffect()->GetSamplerTraits(tex.first);
if (tex.second->IsTextureBuffer())
vezCmdBindBufferView(tex.second->GetTextureBufferView(), 1, tex.first, 0);
else
vezCmdBindImageView(tex.second->GetView(), device->GetSampler(traits.filter_, traits.wrap_), 1, tex.first, 0);
}
for (auto& texRecord : mat->GetTextures())
{
if (texRecord.second)
{
auto traits = mat->GetEffect()->GetSamplerTraits(texRecord.first);
if (texRecord.second->IsTextureBuffer())
vezCmdBindBufferView(texRecord.second->GetTextureBufferView(), 1, texRecord.first, 0);
else
vezCmdBindImageView(texRecord.second->GetView(), device->GetSampler(traits.filter_, traits.wrap_), 1, texRecord.first, 0);
}
}
static VkPolygonMode polyModes[] = {
VK_POLYGON_MODE_FILL,
VK_POLYGON_MODE_POINT,
VK_POLYGON_MODE_LINE
};
static VkCullModeFlags cullModes[] = {
VK_CULL_MODE_NONE,
VK_CULL_MODE_FRONT_BIT,
VK_CULL_MODE_BACK_BIT
};
VezRasterizationState raster = {};
raster.polygonMode = polyModes[0];
raster.cullMode = cullModes[pass->GetDrawState().culling_];
raster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
raster.depthBiasEnable = pass->GetDrawState().depthBias_ != 0.0f || pass->GetDrawState().slopeBias_ != 0.0f;
vezCmdSetRasterizationState(&raster);
if (raster.depthBiasEnable)
vezCmdSetDepthBias(pass->GetDrawState().depthBias_, 0.0f, pass->GetDrawState().slopeBias_);
static const VkCompareOp compareTable[] = {
VK_COMPARE_OP_EQUAL,
VK_COMPARE_OP_LESS_OR_EQUAL,
VK_COMPARE_OP_GREATER_OR_EQUAL,
VK_COMPARE_OP_LESS,
VK_COMPARE_OP_GREATER,
VK_COMPARE_OP_NOT_EQUAL,
VK_COMPARE_OP_ALWAYS,
VK_COMPARE_OP_NEVER
};
VezDepthStencilState depthStencil = {};
depthStencil.depthCompareOp = compareTable[pass->GetDrawState().depthCompare_];
depthStencil.depthTestEnable = pass->GetDrawState().depthTest_ ? VK_TRUE : VK_FALSE;
depthStencil.depthWriteEnable = pass->GetDrawState().depthWrite_ ? VK_TRUE : VK_FALSE;
depthStencil.stencilTestEnable = pass->GetDrawState().stencilTest_ ? VK_TRUE : VK_FALSE;
if (pass->GetDrawState().stencilMask_)
{
depthStencil.front.passOp = VK_STENCIL_OP_REPLACE;
depthStencil.back.passOp = VK_STENCIL_OP_REPLACE;
vezCmdSetStencilReference(VkStencilFaceFlagBits::VK_STENCIL_FRONT_AND_BACK, pass->GetDrawState().stencilMask_);
}
vezCmdSetDepthStencilState(&depthStencil);
VkColorComponentFlags vkColorCompAll = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;
static VezColorBlendAttachmentState blendStates[] = {
// Blend none
{ false, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
// Alpha
{ true, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
// Add
{ true, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_ADD, vkColorCompAll },
// Subtract
{ true, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_REVERSE_SUBTRACT, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_REVERSE_SUBTRACT, vkColorCompAll },
// Multiply
{ true, VK_BLEND_FACTOR_DST_COLOR, VK_BLEND_FACTOR_ZERO, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_DST_COLOR, VK_BLEND_FACTOR_ZERO, VK_BLEND_OP_ADD, vkColorCompAll },
// Premultiplied
{ true, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
};
VezColorBlendState blendState = { };
vector<VezColorBlendAttachmentState> attachStates;
auto fbo = stage && stage->targetConfig_.fbo_ ? stage->targetConfig_.fbo_ : view.renderTarget_;
for (uint32_t i = 0; i < fbo->GetTextureCount(); ++i)
{
auto tgt = fbo->GetTexture(i);
if (tgt->GetFormat() != TEX_DEPTH )
attachStates.push_back(blendStates[pass->GetDrawState().blendMode_]);
}
blendState.attachmentCount = (uint32_t)attachStates.size();
blendState.pAttachments = attachStates.data();
blendState.logicOpEnable = false;
vezCmdSetColorBlendState(&blendState);
};
if (taskData.cmdBuffer_ == 0)
{
auto cmdBuffer = RS_StartCommandBuffer(device_);
if (taskData.placeCmdBufferAt_ != -1)
renderer->GetCommandBufferChain()[taskData.placeCmdBufferAt_] = cmdBuffer;
else
renderer->GetCommandBufferChain().push_back(cmdBuffer);
}
else
vezBeginCommandBuffer(taskData.cmdBuffer_, VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
RS_SetupViewport(device_, view);
if (stage != nullptr)
BeginStage(device_, view, stage);
else
{
VezRenderPassBeginInfo info = { };
info.attachmentCount = view.renderTarget_->GetTextureCount();
info.framebuffer = view.renderTarget_->GetGPUObject();
vector<VezAttachmentInfo> attach;
for (unsigned i = 0; i < view.renderTarget_->GetTextureCount(); ++i)
{
VezAttachmentInfo at = { };
at.loadOp = taskData.clearTargets_ ? VK_ATTACHMENT_LOAD_OP_CLEAR : VK_ATTACHMENT_LOAD_OP_LOAD;
at.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
at.clearValue.color = { 0, 0, 0, 1.0f };
at.clearValue.depthStencil.depth = 1.0f;
at.clearValue.depthStencil.stencil = 0;
attach.push_back(at);
}
info.pAttachments = attach.data();
vezCmdBeginRenderPass(&info);
}
for (const auto& grp : batchQueue.groups_)
{
assert(grp.second.geometry_ && grp.second.material_);
if (grp.second.material_ == nullptr || grp.second.geometry_ == nullptr)
continue;
if (auto pass = grp.second.material_->GetEffect()->GetPass(taskData.instancedNameHash_, grp.second.geometry_->primType_))
{
if (pass != activePass)
{
applyPass(pass, view);
activePass = pass;
}
if (activeMaterial != grp.second.material_ || activeGeometry != grp.second.geometry_)
{
applyMaterial(device_, view, stage, pass, grp.second.material_, grp.second.geometry_, taskData);
activeMaterial = grp.second.material_;
}
if (activeGeometry != grp.second.geometry_)
{
ApplyGeometry(pass, grp.second.geometry_, true);
activeGeometry = grp.second.geometry_;
}
uint32_t drawCt = (uint32_t)grp.second.transforms_.size();
for (size_t i = 0; i < grp.second.instanceTransformBuffers_.size(); ++i)
{
//uint32_t drawCt = grp.transforms_.size();
auto instBuff = grp.second.instanceTransformBuffers_[i].second;
uint32_t drawCt = grp.second.instanceTransformBuffers_[i].first;
//ApplyGeometry(pass, activeGeometry, true, { grp.second.instanceTransformBuffers_[i].second });
VkDeviceSize inOfs = 0;
VkBuffer buff = grp.second.instanceTransformBuffers_[i].second->GetGPUObject();
vezCmdBindVertexBuffers(activeGeometry->vertexBuffers_.size(), 1, &buff, &inOfs);
if (activeGeometry->indexBuffer_)
{
vezCmdDrawIndexed(activeGeometry->indexCount_, drawCt, activeGeometry->indexStart_, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_ * drawCt);
device_->AddStat(STAT_INSTANCES, drawCt);
}
else
{
vezCmdDraw(activeGeometry->vertexCount_, drawCt, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_ * drawCt);
device_->AddStat(STAT_INSTANCES, drawCt);
}
}
}
}
// batches that aren't instanced
for (const auto& batch : batchQueue.batches_)
{
assert(batch.geometry_ && batch.material_);
if (batch.geometry_ == nullptr || batch.material_ == nullptr)
continue;
if (auto pass = batch.material_->GetEffect()->GetPass(taskData.contextNameHash_, batch.geometry_->primType_))
{
if (pass != activePass)
{
applyPass(pass, view);
activePass = pass;
}
if (activeMaterial != batch.material_ || activeGeometry != batch.geometry_)
{
activeMaterial = batch.material_;
applyMaterial(device_, view, stage, activePass, activeMaterial, batch.geometry_, taskData);
}
if (batch.geometry_ != activeGeometry)
{
activeGeometry = batch.geometry_;
ApplyGeometry(activePass, activeGeometry, false);
}
if (batch.isSkinned_)
{
vezCmdBindBuffer(batch.bonesBuffer_->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 1, 0);
vezCmdPushConstants(0, sizeof(float4x4), batch.transforms_); // we still have this one guy
if (activeGeometry->indexBuffer_ != nullptr)
{
vezCmdDrawIndexed(activeGeometry->indexCount_, 1, activeGeometry->indexStart_, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_);
}
else
{
vezCmdDraw(activeGeometry->vertexCount_, 1, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_);
}
}
else
{
// we'll iterate, fake instancing - though this is *effectively* what instancing does
for (uint32_t i = 0; i < batch.numTransforms_; ++i)
{
vezCmdPushConstants(0, sizeof(float4x4), batch.transforms_ + sizeof(float4x4) * i);
if (activeGeometry->indexBuffer_ != nullptr)
{
vezCmdDrawIndexed(activeGeometry->indexCount_, 1, activeGeometry->indexStart_, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_);
}
else
{
vezCmdDraw(activeGeometry->vertexCount_, 1, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, activeGeometry->primCount_);
}
}
}
}
}
if (stage != nullptr)
EndStage(device_, stage);
else
vezCmdEndRenderPass();
vezEndCommandBuffer();
}
//****************************************************************************
//
// Function: RenderScript::ApplyPass
//
// Purpose: Not much for vulkan to do.
//
//****************************************************************************
void RenderScript::ApplyPass(const shared_ptr<ShaderPass> pass, View view)
{
vezCmdBindPipeline(pass->GetPipeline());
VkViewport v = {};
v.x = (float)view.viewport_[0];
v.y = (float)view.viewport_[1];
v.width = view.viewport_.Width();
v.height = view.viewport_.Height();
v.minDepth = 0.0f;
v.maxDepth = 1.0f;
vezCmdSetViewport(0, 1, &v);
VkRect2D r;
r.offset.x = (int)v.x;
r.offset.y = (int)v.y;
r.extent.width = (uint32_t)v.width;
r.extent.height = (uint32_t)v.height;
vezCmdSetScissor(0, 1, &r);
}
//****************************************************************************
//
// Function: RenderScript::ApplyBlendMode
//
// Purpose: Utility for reliably setting blend-modes, previously this was done poorly.
// Unused parameters are Vulkan specific.
//
//****************************************************************************
void RenderScript::ApplyBlendMode(BlendMode mode, View view, RenderScriptStage* stage)
{
VkColorComponentFlags vkColorCompAll = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;
static VezColorBlendAttachmentState blendStates[] = {
// Blend none
{ VK_FALSE, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
// Alpha
{ VK_TRUE, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_SRC_ALPHA, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
// Add
{ VK_TRUE, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_ADD, vkColorCompAll },
// Subtract
{ VK_TRUE, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_REVERSE_SUBTRACT, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE, VK_BLEND_OP_REVERSE_SUBTRACT, vkColorCompAll },
// Multiply
{ VK_TRUE, VK_BLEND_FACTOR_DST_COLOR, VK_BLEND_FACTOR_ZERO, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_DST_COLOR, VK_BLEND_FACTOR_ZERO, VK_BLEND_OP_ADD, vkColorCompAll },
// Premultiplied
{ VK_TRUE, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, VK_BLEND_FACTOR_ONE, VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA, VK_BLEND_OP_ADD, vkColorCompAll },
};
VezColorBlendState blendState = { };
vector<VezColorBlendAttachmentState> attachStates;
auto fbo = stage && stage->targetConfig_.fbo_ ? stage->targetConfig_.fbo_ : view.renderTarget_;
for (uint32_t i = 0; i < fbo->GetTextureCount(); ++i)
{
auto tgt = fbo->GetTexture(i);
if (tgt->GetFormat() != TEX_DEPTH)
attachStates.push_back(blendStates[mode]);
}
blendState.attachmentCount = (uint32_t)attachStates.size();
blendState.pAttachments = attachStates.data();
blendState.logicOpEnable = false;
vezCmdSetColorBlendState(&blendState);
}
//****************************************************************************
//
// Function: RenderScript::ApplyGeometry
//
// Purpose: Does the repeated task of setting of the primtive assembly and bindings buffers
// Watch out for the bit with instancing, that's about the automatic instancing.
//
//****************************************************************************
void RenderScript::ApplyGeometry(shared_ptr<ShaderPass> pass, Geometry* geometry, bool isInstanced, const std::vector<std::shared_ptr<Buffer>>& extraBuffers, bool laterInstanced)
{
static VkPrimitiveTopology topoModes[] = {
VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST,
VK_PRIMITIVE_TOPOLOGY_POINT_LIST,
VK_PRIMITIVE_TOPOLOGY_LINE_LIST
};
VezInputAssemblyState inputAssemblyState = {};
inputAssemblyState.topology = pass->IsTessellating() ? VK_PRIMITIVE_TOPOLOGY_PATCH_LIST : topoModes[geometry->primType_];
inputAssemblyState.primitiveRestartEnable = false;
vezCmdSetInputAssemblyState(&inputAssemblyState);
auto layout = !isInstanced ? geometry->layout_ : geometry->layout_->GetInstancedVariant();
layout->Bind(geometry, { });
size_t numBuffers = geometry->vertexBuffers_.size();
vector<VkDeviceSize> vtxOffsets;
vector<VkBuffer> buffers;
buffers.reserve(numBuffers);
vtxOffsets.reserve(numBuffers);
for (const auto& buf : geometry->vertexBuffers_)
{
buffers.push_back(buf->GetGPUObject());
vtxOffsets.push_back(0);
}
for (const auto& extraBuf : extraBuffers)
{
buffers.push_back(extraBuf->GetGPUObject());
vtxOffsets.push_back(0);
}
vezCmdBindVertexBuffers(0, (uint32_t)buffers.size(), buffers.data(), vtxOffsets.data());
if (geometry->indexBuffer_ != nullptr)
vezCmdBindIndexBuffer(geometry->indexBuffer_->GetGPUObject(), geometry->indexStart_, geometry->indexBuffer_->HasTag(BufferTag_32Bit) ? VK_INDEX_TYPE_UINT32 : VK_INDEX_TYPE_UINT16);
}
//****************************************************************************
//
// Function: RenderScript::RenderFullscreen
//
// Purpose: Uses the given effect to do a generic full-screen render.
// This function is makes rocks look smart.
//
//****************************************************************************
void RenderScript::RenderFullscreen(std::shared_ptr<Effect> effect, View view, const Material::UBOBindingList& extraBuffers, const Material::TextureBindingList& extraTextures)
{
auto pass = effect->GetPasses()[0];
ApplyPass(pass, view);
auto geo = device_->GetFSTriGeometry();
ApplyGeometry(pass, geo.get(), false, { });
// Set rasterization state.
VezRasterizationState rasterizationState = { };
rasterizationState.polygonMode = VK_POLYGON_MODE_FILL;
rasterizationState.cullMode = VK_CULL_MODE_NONE;
rasterizationState.frontFace = VK_FRONT_FACE_CLOCKWISE;
vezCmdSetRasterizationState(&rasterizationState);
// Set depth stencil state which is special for a fullscreen pass
VezDepthStencilState depthStencilState = {};
depthStencilState.depthTestEnable = VK_FALSE;
depthStencilState.depthWriteEnable = VK_FALSE;
depthStencilState.depthCompareOp = VK_COMPARE_OP_ALWAYS;
depthStencilState.stencilTestEnable = VK_FALSE;
depthStencilState.depthBoundsTestEnable = VK_FALSE;
vezCmdSetDepthStencilState(&depthStencilState);
auto vbo = geo->vertexBuffers_[0]->GetGPUObject();
VkDeviceSize junk = 0;
vezCmdBindVertexBuffers(0, 1, &vbo, &junk);
vezCmdBindIndexBuffer(device_->GetSequentialIndexBuffer()->GetGPUObject(), 0, VK_INDEX_TYPE_UINT16);
vezCmdDraw(3, 1, 0, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, 1);
}
//****************************************************************************
//
// Function: Renderer::Draw2DBatches
//
// Purpose: Performs generic 2D rendering, intended for use with naive
// GUIs like dear-imgui.
//
//****************************************************************************
void Renderer::Draw2DBatches(const std::vector<Draw2D>& calls, View forView, RenderScript* script)
{
if (forView.renderTarget_ == nullptr)
forView.renderTarget_ = device_->GetBackbuffer();
if (calls.size() > 0)
{
if (uiGeometry_ == nullptr)
{
uiGeometry_.reset(new Geometry());
uiGeometry_->vertexBuffers_.push_back(device_->CreateVertexBuffer());
uiGeometry_->vertexBuffers_[0]->SetTag(BufferTag_Dynamic);
uiGeometry_->layout_ = device_->GetLayout_2D();
uiUBO_ = device_->CreateUniformBuffer();
uiUBO_->SetTag(BufferTag_Dynamic);
}
float4 sz = calls[0].domain_;
float L = sz.x;
float R = sz.x + sz.z;
float T = sz.y;
float B = sz.y + sz.w;
float scale[2];
scale[0] = 2.0f / (R - L);
scale[1] = 2.0f / (B - T);
float translate[2];
translate[0] = -1.0f - sz.x * scale[0];
translate[1] = -1.0f - sz.y * scale[1];
auto mat = float4x4(
translate[0], translate[1], 0,0,
scale[0], scale[1], 0,0,
0,0,0,0,
0,0,0,0
);
//mat.Transpose();
uiUBO_->SetData(&mat, sizeof(float4x4));
Vertex2D* dataStart = &((*calls[0].vertices_)[0]);
uiGeometry_->vertexBuffers_[0]->SetData(dataStart, calls[0].vertices_->size() * sizeof(Vertex2D));
uiGeometry_->InferValuesFromData();
auto cmdBuffer = device_->GetGraphicsCmdBuffer();
GetCommandBufferChain().push_back(cmdBuffer);
vezBeginCommandBuffer(cmdBuffer, VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT);
RS_SetupViewport(device_, forView);
VezRenderPassBeginInfo pass = { };
pass.framebuffer = forView.renderTarget_->GetGPUObject();
std::vector<VezAttachmentInfo> attachments;
for (uint32_t i = 0; i < forView.renderTarget_->GetTextureCount(); ++i)
{
VezAttachmentInfo attach = {};
attach.loadOp = VkAttachmentLoadOp::VK_ATTACHMENT_LOAD_OP_LOAD;
attach.storeOp = VkAttachmentStoreOp::VK_ATTACHMENT_STORE_OP_STORE;
attachments.push_back(attach);
}
pass.pAttachments = attachments.data();
pass.attachmentCount = attachments.size();
vezCmdBeginRenderPass(&pass);
vezCmdBindBuffer(uiUBO_->GetGPUObject(), 0, VK_WHOLE_SIZE, 0, 0, 0);
auto fxUI = device_->GetGUIEffect();
script->ApplyPass(fxUI->GetPasses()[0], forView);
int4 vpt = calls[0].viewport_;
VkViewport v;
v.x = vpt.x;
v.y = vpt.y;
v.width = vpt.Width();
v.height = vpt.Height();
v.minDepth = 0.0f;
v.maxDepth = 1.0f;
vezCmdSetViewport(0, 1, &v);
int4 clip = calls[0].clipRect_;
VkRect2D scissor = { calls[0].clipRect_.x, calls[0].clipRect_.y, calls[0].clipRect_.Width(), calls[0].clipRect_.Height() };
vezCmdSetScissor(0, 1, &scissor);
for (uint32_t i = 0; i < calls.size(); ++i)
{
const auto& call = calls[i];
if (memcmp(&calls[i].viewport_, &vpt, sizeof(int4)) != 0)
{
vpt = calls[i].viewport_;
VkViewport v;
v.x = vpt.x;
v.y = vpt.y;
v.width = vpt.Width();
v.height = vpt.Height();
v.minDepth = 0.0f;
v.maxDepth = 1.0f;
vezCmdSetViewport(0, 1, &v);
}
if (memcmp(&calls[i].clipRect_, &clip, sizeof(int4)) != 0)
{
clip = calls[i].clipRect_;
VkRect2D scissor = { call.clipRect_.x, call.clipRect_.y, call.clipRect_.Width(), call.clipRect_.Height() };
vezCmdSetScissor(0, 1, &scissor);
}
script->ApplyBlendMode(call.blendMode_, forView, nullptr);
script->ApplyGeometry(fxUI->GetPasses()[0], uiGeometry_.get(), false);
if (call.texture_)
vezCmdBindImageView(call.texture_->GetView(), device_->GetSampler(FILTER_POINT, true), 1, 0, 0);
vezCmdDraw(call.vertexCount_, 1, call.vertexStart_, 0);
device_->AddStat(STAT_BATCHES, 1);
device_->AddStat(STAT_PRIMITIVES, call.vertexCount_ / 3);
}
vezCmdEndRenderPass();
vezEndCommandBuffer();
VkQueue graphicsQueue = 0;
vezGetDeviceGraphicsQueue(device_->GetVKDevice(), 0, &graphicsQueue);
array<VkSemaphore, 3> semaphores;
VezSubmitInfo submission = {};
submission.commandBufferCount = 1;
submission.pCommandBuffers = &cmdBuffer;
submission.signalSemaphoreCount = 1;
submission.pSignalSemaphores = &semaphores[0];
VkResult result = vezQueueSubmit(graphicsQueue, 1, &submission, nullptr);
if (result != VK_SUCCESS)
LOG_VULKAN("Failed to submit to queue: %u", result);
}
}
//****************************************************************************
//
// Function: Renderer::ResetCommandBufferChain
//
// Purpose: Wipes our local list of buffers to execute, GraphicsDevice
// is responsible for the actual pool management.
//
//****************************************************************************
void Renderer::ResetCommandBufferChain()
{
commandBufferChain_.clear();
}
//****************************************************************************
//
// Function: Renderer::GetCommandBuffer
//
// Purpose: Just wraps the call the GraphicsDevice.
//
// TODO: Consider removal? When we have renderer we have the device too?
// Will there be no special management concerns at this point?
//
//****************************************************************************
VkCommandBuffer Renderer::GetCommandBuffer()
{
auto buffer = device_->GetGraphicsCmdBuffer();
return buffer;
}
//****************************************************************************
//
// Function: Renderer::FinishRendering
//
// Purpose: Submits our constructed VkCommandBuffer lists.
//
// TODO: Incremental submissions? Submit-after-N? Avoid waiting.
// Or do resource uploads/set-data have it covered there?
//
// This function needs changes when GraphicsDeviceHead is implemented.
// Right now we're assuming a singular swap-chain, in the future
// there will be a swapchain per-head, this will make semaphore management
// *interesting*.
//
//****************************************************************************
void Renderer::FinishRendering()
{
VkQueue graphicsQueue = 0;
vezGetDeviceGraphicsQueue(device_->GetVKDevice(), 0, &graphicsQueue);
array<VkSemaphore, 3> semaphores;
VezSubmitInfo submission = {};
submission.commandBufferCount = (uint32_t)commandBufferChain_.size();
submission.pCommandBuffers = commandBufferChain_.data();
submission.signalSemaphoreCount = 1;
submission.pSignalSemaphores = &semaphores[0];
VkResult result = vezQueueSubmit(graphicsQueue, 1, &submission, nullptr);
if (result != VK_SUCCESS)
LOG_VULKAN("Failed to submit command-buffers", result);
VkPipelineStageFlags waitDstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
auto swapchain = device_->GetSwapchain();
auto srcImage = device_->GetBackbuffer()->GetTexture(0)->GetImage();
VezPresentInfo presentInfo = {};
presentInfo.waitSemaphoreCount = 1;
presentInfo.pWaitSemaphores = &semaphores[0];
presentInfo.pWaitDstStageMask = &waitDstStageMask;
presentInfo.swapchainCount = 1;
presentInfo.pSwapchains = &swapchain;
presentInfo.pImages = &srcImage;
result = vezQueuePresent(graphicsQueue, &presentInfo);
if (result != VK_SUCCESS)
LOG_VULKAN("Failed to present swap-chain", result);
ResetCommandBufferChain();
}
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment