qt5base-lts/tests/manual/rhi/multiwindow_threaded/multiwindow_threaded.cpp

760 lines
21 KiB
C++
Raw Normal View History

// Copyright (C) 2018 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR BSD-3-Clause
#include <QApplication>
#include <QWidget>
#include <QLabel>
#include <QPlainTextEdit>
#include <QPushButton>
#include <QCheckBox>
#include <QVBoxLayout>
#include <QThread>
#include <QMutex>
#include <QWaitCondition>
#include <QQueue>
#include <QEvent>
#include <QCommandLineParser>
#include <QElapsedTimer>
#include <QtGui/private/qshader_p.h>
#include <QFile>
#ifndef QT_NO_OPENGL
#include <QtGui/private/qrhigles2_p.h>
#include <QOffscreenSurface>
#endif
#if QT_CONFIG(vulkan)
#include <QLoggingCategory>
#include <QtGui/private/qrhivulkan_p.h>
#endif
#ifdef Q_OS_WIN
#include <QtGui/private/qrhid3d11_p.h>
#endif
#if defined(Q_OS_MACOS) || defined(Q_OS_IOS)
#include <QtGui/private/qrhimetal_p.h>
#endif
#ifdef Q_OS_DARWIN
#include <QtCore/private/qcore_mac_p.h>
#endif
#include "window.h"
#include "../shared/cube.h"
static QShader getShader(const QString &name)
{
QFile f(name);
if (f.open(QIODevice::ReadOnly))
return QShader::fromSerialized(f.readAll());
return QShader();
}
static GraphicsApi graphicsApi;
static QString graphicsApiName()
{
switch (graphicsApi) {
case OpenGL:
return QLatin1String("OpenGL 2.x");
case Vulkan:
return QLatin1String("Vulkan");
case D3D11:
return QLatin1String("Direct3D 11");
case Metal:
return QLatin1String("Metal");
default:
break;
}
return QString();
}
#if QT_CONFIG(vulkan)
QVulkanInstance *instance = nullptr;
#endif
// Window (main thread) emit signals -> Renderer::send* (main thread) -> event queue (add on main, process on render thread) -> Renderer::renderEvent (render thread)
// event queue is taken from the Qt Quick scenegraph as-is
// all this below is conceptually the same as the QSG threaded render loop
class RenderThreadEventQueue : public QQueue<QEvent *>
{
public:
RenderThreadEventQueue()
: waiting(false)
{
}
void addEvent(QEvent *e) {
mutex.lock();
enqueue(e);
if (waiting)
condition.wakeOne();
mutex.unlock();
}
QEvent *takeEvent(bool wait) {
mutex.lock();
if (isEmpty() && wait) {
waiting = true;
condition.wait(&mutex);
waiting = false;
}
QEvent *e = dequeue();
mutex.unlock();
return e;
}
bool hasMoreEvents() {
mutex.lock();
bool has = !isEmpty();
mutex.unlock();
return has;
}
private:
QMutex mutex;
QWaitCondition condition;
bool waiting;
};
struct Renderer;
struct Thread : public QThread
{
Thread(Renderer *renderer_)
: renderer(renderer_)
{
active = true;
start();
}
void run() override;
Renderer *renderer;
bool active;
RenderThreadEventQueue eventQueue;
bool sleeping = false;
bool stopEventProcessing = false;
bool pendingRender = false;
bool pendingRenderIsNewExpose = false;
// mutex and cond used to allow the main thread waiting until something completes on the render thread
QMutex mutex;
QWaitCondition cond;
};
class RenderThreadEvent : public QEvent
{
public:
RenderThreadEvent(QEvent::Type type) : QEvent(type) { }
};
class InitEvent : public RenderThreadEvent
{
public:
static const QEvent::Type TYPE = QEvent::Type(QEvent::User + 1);
InitEvent() : RenderThreadEvent(TYPE)
{ }
};
class RequestRenderEvent : public RenderThreadEvent
{
public:
static const QEvent::Type TYPE = QEvent::Type(QEvent::User + 2);
RequestRenderEvent(bool newlyExposed_) : RenderThreadEvent(TYPE), newlyExposed(newlyExposed_)
{ }
bool newlyExposed;
};
class SurfaceCleanupEvent : public RenderThreadEvent
{
public:
static const QEvent::Type TYPE = QEvent::Type(QEvent::User + 3);
SurfaceCleanupEvent() : RenderThreadEvent(TYPE)
{ }
};
class CloseEvent : public RenderThreadEvent
{
public:
static const QEvent::Type TYPE = QEvent::Type(QEvent::User + 4);
CloseEvent() : RenderThreadEvent(TYPE)
{ }
};
class SyncSurfaceSizeEvent : public RenderThreadEvent
{
public:
static const QEvent::Type TYPE = QEvent::Type(QEvent::User + 5);
SyncSurfaceSizeEvent() : RenderThreadEvent(TYPE)
{ }
};
struct Renderer
{
// ctor and dtor and send* are called main thread, rest on the render thread
Renderer(QWindow *w, const QColor &bgColor, int rotationAxis);
~Renderer();
void sendInit();
void sendRender(bool newlyExposed);
void sendSurfaceGoingAway();
void sendSyncSurfaceSize();
QWindow *window;
Thread *thread;
QRhi *r = nullptr;
#ifndef QT_NO_OPENGL
QOffscreenSurface *fallbackSurface = nullptr;
#endif
void createRhi();
void destroyRhi();
void renderEvent(QEvent *e);
void init();
void releaseSwapChain();
void releaseResources();
void render(bool newlyExposed, bool wakeBeforePresent);
QColor m_bgColor;
int m_rotationAxis;
QList<QRhiResource *> m_releasePool;
bool m_hasSwapChain = false;
QRhiSwapChain *m_sc = nullptr;
QRhiRenderBuffer *m_ds = nullptr;
QRhiRenderPassDescriptor *m_rp = nullptr;
QRhiBuffer *m_vbuf = nullptr;
QRhiBuffer *m_ubuf = nullptr;
QRhiTexture *m_tex = nullptr;
QRhiSampler *m_sampler = nullptr;
QRhiShaderResourceBindings *m_srb = nullptr;
QRhiGraphicsPipeline *m_ps = nullptr;
QRhiResourceUpdateBatch *m_initialUpdates = nullptr;
QMatrix4x4 m_proj;
float m_rotation = 0;
int m_frameCount = 0;
};
void Thread::run()
{
while (active) {
#ifdef Q_OS_DARWIN
QMacAutoReleasePool autoReleasePool;
#endif
if (pendingRender) {
pendingRender = false;
renderer->render(pendingRenderIsNewExpose, false);
}
while (eventQueue.hasMoreEvents()) {
QEvent *e = eventQueue.takeEvent(false);
renderer->renderEvent(e);
delete e;
}
if (active && !pendingRender) {
sleeping = true;
stopEventProcessing = false;
while (!stopEventProcessing) {
QEvent *e = eventQueue.takeEvent(true);
renderer->renderEvent(e);
delete e;
}
sleeping = false;
}
}
}
Renderer::Renderer(QWindow *w, const QColor &bgColor, int rotationAxis)
: window(w),
m_bgColor(bgColor),
m_rotationAxis(rotationAxis)
{ // main thread
thread = new Thread(this);
#ifndef QT_NO_OPENGL
if (graphicsApi == OpenGL)
fallbackSurface = QRhiGles2InitParams::newFallbackSurface();
#endif
}
Renderer::~Renderer()
{ // main thread
thread->eventQueue.addEvent(new CloseEvent);
thread->wait();
delete thread;
#ifndef QT_NO_OPENGL
delete fallbackSurface;
#endif
}
void Renderer::createRhi()
{
if (r)
return;
qDebug() << "renderer" << this << "creating rhi";
QRhi::Flags rhiFlags;
#ifndef QT_NO_OPENGL
if (graphicsApi == OpenGL) {
QRhiGles2InitParams params;
params.fallbackSurface = fallbackSurface;
params.window = window;
r = QRhi::create(QRhi::OpenGLES2, &params, rhiFlags);
}
#endif
#if QT_CONFIG(vulkan)
if (graphicsApi == Vulkan) {
QRhiVulkanInitParams params;
params.inst = instance;
params.window = window;
r = QRhi::create(QRhi::Vulkan, &params, rhiFlags);
}
#endif
#ifdef Q_OS_WIN
if (graphicsApi == D3D11) {
QRhiD3D11InitParams params;
params.enableDebugLayer = true;
r = QRhi::create(QRhi::D3D11, &params, rhiFlags);
}
#endif
#if defined(Q_OS_MACOS) || defined(Q_OS_IOS)
if (graphicsApi == Metal) {
QRhiMetalInitParams params;
r = QRhi::create(QRhi::Metal, &params, rhiFlags);
}
#endif
if (!r)
qFatal("Failed to create RHI backend");
}
void Renderer::destroyRhi()
{
qDebug() << "renderer" << this << "destroying rhi";
delete r;
r = nullptr;
}
void Renderer::renderEvent(QEvent *e)
{
Q_ASSERT(QThread::currentThread() == thread);
if (thread->sleeping)
thread->stopEventProcessing = true;
switch (int(e->type())) {
case InitEvent::TYPE:
qDebug() << "renderer" << this << "for window" << window << "is initializing";
createRhi();
init();
break;
case RequestRenderEvent::TYPE:
thread->pendingRender = true;
thread->pendingRenderIsNewExpose = static_cast<RequestRenderEvent *>(e)->newlyExposed;
break;
case SurfaceCleanupEvent::TYPE: // when the QWindow is closed, before QPlatformWindow goes away
thread->mutex.lock();
qDebug() << "renderer" << this << "for window" << window << "is destroying swapchain";
thread->pendingRender = false;
releaseSwapChain();
thread->cond.wakeOne();
thread->mutex.unlock();
break;
case CloseEvent::TYPE: // when destroying the window+renderer (NB not the same as hitting X on the window, that's just QWindow close)
qDebug() << "renderer" << this << "for window" << window << "is shutting down";
thread->pendingRender = false;
thread->active = false;
thread->stopEventProcessing = true;
releaseResources();
destroyRhi();
break;
case SyncSurfaceSizeEvent::TYPE:
thread->mutex.lock();
thread->pendingRender = false;
render(false, true);
break;
default:
break;
}
}
void Renderer::init()
{
m_sc = r->newSwapChain();
m_ds = r->newRenderBuffer(QRhiRenderBuffer::DepthStencil,
QSize(),
1,
QRhiRenderBuffer::UsedWithSwapChainOnly);
m_releasePool << m_ds;
m_sc->setWindow(window);
m_sc->setDepthStencil(m_ds);
m_rp = m_sc->newCompatibleRenderPassDescriptor();
m_releasePool << m_rp;
m_sc->setRenderPassDescriptor(m_rp);
m_vbuf = r->newBuffer(QRhiBuffer::Immutable, QRhiBuffer::VertexBuffer, sizeof(cube));
m_releasePool << m_vbuf;
m_vbuf->create();
m_ubuf = r->newBuffer(QRhiBuffer::Dynamic, QRhiBuffer::UniformBuffer, 64 + 4);
m_releasePool << m_ubuf;
m_ubuf->create();
QImage image = QImage(QLatin1String(":/qt256.png")).convertToFormat(QImage::Format_RGBA8888);
m_tex = r->newTexture(QRhiTexture::RGBA8, image.size());
m_releasePool << m_tex;
m_tex->create();
m_sampler = r->newSampler(QRhiSampler::Linear, QRhiSampler::Linear, QRhiSampler::None,
QRhiSampler::ClampToEdge, QRhiSampler::ClampToEdge);
m_releasePool << m_sampler;
m_sampler->create();
m_srb = r->newShaderResourceBindings();
m_releasePool << m_srb;
m_srb->setBindings({
QRhiShaderResourceBinding::uniformBuffer(0, QRhiShaderResourceBinding::VertexStage | QRhiShaderResourceBinding::FragmentStage, m_ubuf),
QRhiShaderResourceBinding::sampledTexture(1, QRhiShaderResourceBinding::FragmentStage, m_tex, m_sampler)
});
m_srb->create();
m_ps = r->newGraphicsPipeline();
m_releasePool << m_ps;
m_ps->setDepthTest(true);
m_ps->setDepthWrite(true);
m_ps->setDepthOp(QRhiGraphicsPipeline::Less);
m_ps->setCullMode(QRhiGraphicsPipeline::Back);
m_ps->setFrontFace(QRhiGraphicsPipeline::CCW);
m_ps->setShaderStages({
{ QRhiShaderStage::Vertex, getShader(QLatin1String(":/texture.vert.qsb")) },
{ QRhiShaderStage::Fragment, getShader(QLatin1String(":/texture.frag.qsb")) }
});
QRhiVertexInputLayout inputLayout;
inputLayout.setBindings({
{ 3 * sizeof(float) },
{ 2 * sizeof(float) }
});
inputLayout.setAttributes({
{ 0, 0, QRhiVertexInputAttribute::Float3, 0 },
{ 1, 1, QRhiVertexInputAttribute::Float2, 0 }
});
m_ps->setVertexInputLayout(inputLayout);
m_ps->setShaderResourceBindings(m_srb);
m_ps->setRenderPassDescriptor(m_rp);
m_ps->create();
m_initialUpdates = r->nextResourceUpdateBatch();
m_initialUpdates->uploadStaticBuffer(m_vbuf, cube);
qint32 flip = 0;
m_initialUpdates->updateDynamicBuffer(m_ubuf, 64, 4, &flip);
m_initialUpdates->uploadTexture(m_tex, image);
}
void Renderer::releaseSwapChain()
{
if (m_hasSwapChain) {
m_hasSwapChain = false;
m_sc->destroy();
}
}
void Renderer::releaseResources()
{
qDeleteAll(m_releasePool);
m_releasePool.clear();
delete m_sc;
m_sc = nullptr;
}
void Renderer::render(bool newlyExposed, bool wakeBeforePresent)
{
// This function handles both resizing and rendering. Resizes have some
// complications due to the threaded model (check exposeEvent and
// sendSyncSurfaceSize) but don't have to worry about that in here.
auto buildOrResizeSwapChain = [this] {
qDebug() << "renderer" << this << "build or resize swapchain for window" << window;
m_hasSwapChain = m_sc->createOrResize();
const QSize outputSize = m_sc->currentPixelSize();
qDebug() << " size is" << outputSize;
m_proj = r->clipSpaceCorrMatrix();
m_proj.perspective(45.0f, outputSize.width() / (float) outputSize.height(), 0.01f, 100.0f);
m_proj.translate(0, 0, -4);
};
auto wakeUpIfNeeded = [wakeBeforePresent, this] {
// make sure the main/gui thread is not blocked when issuing the Present (or equivalent)
if (wakeBeforePresent) {
thread->cond.wakeOne();
thread->mutex.unlock();
}
};
const QSize surfaceSize = m_sc->surfacePixelSize();
if (surfaceSize.isEmpty()) {
wakeUpIfNeeded();
return;
}
if (newlyExposed || m_sc->currentPixelSize() != surfaceSize)
buildOrResizeSwapChain();
if (!m_hasSwapChain) {
wakeUpIfNeeded();
return;
}
QRhi::FrameOpResult result = r->beginFrame(m_sc);
if (result == QRhi::FrameOpSwapChainOutOfDate) {
buildOrResizeSwapChain();
if (!m_hasSwapChain) {
wakeUpIfNeeded();
return;
}
result = r->beginFrame(m_sc);
}
if (result != QRhi::FrameOpSuccess) {
wakeUpIfNeeded();
return;
}
QRhiCommandBuffer *cb = m_sc->currentFrameCommandBuffer();
const QSize outputSize = m_sc->currentPixelSize();
QRhiResourceUpdateBatch *u = r->nextResourceUpdateBatch();
if (m_initialUpdates) {
u->merge(m_initialUpdates);
m_initialUpdates->release();
m_initialUpdates = nullptr;
}
m_rotation += 1.0f;
QMatrix4x4 mvp = m_proj;
mvp.scale(0.5f);
mvp.rotate(m_rotation, m_rotationAxis == 0 ? 1 : 0, m_rotationAxis == 1 ? 1 : 0, m_rotationAxis == 2 ? 1 : 0);
u->updateDynamicBuffer(m_ubuf, 0, 64, mvp.constData());
cb->beginPass(m_sc->currentFrameRenderTarget(),
QColor::fromRgbF(float(m_bgColor.redF()), float(m_bgColor.greenF()), float(m_bgColor.blueF()), 1.0f),
{ 1.0f, 0 },
u);
cb->setGraphicsPipeline(m_ps);
cb->setViewport(QRhiViewport(0, 0, outputSize.width(), outputSize.height()));
cb->setShaderResources();
const QRhiCommandBuffer::VertexInput vbufBindings[] = {
{ m_vbuf, 0 },
{ m_vbuf, quint32(36 * 3 * sizeof(float)) }
};
cb->setVertexInput(0, 2, vbufBindings);
cb->draw(36);
cb->endPass();
wakeUpIfNeeded();
r->endFrame(m_sc);
m_frameCount += 1;
}
void Renderer::sendInit()
{ // main thread
InitEvent *e = new InitEvent;
thread->eventQueue.addEvent(e);
}
void Renderer::sendRender(bool newlyExposed)
{ // main thread
RequestRenderEvent *e = new RequestRenderEvent(newlyExposed);
thread->eventQueue.addEvent(e);
}
void Renderer::sendSurfaceGoingAway()
{ // main thread
SurfaceCleanupEvent *e = new SurfaceCleanupEvent;
// cannot let this thread to proceed with tearing down the native window
// before the render thread completes the swapchain release
thread->mutex.lock();
thread->eventQueue.addEvent(e);
thread->cond.wait(&thread->mutex);
thread->mutex.unlock();
}
void Renderer::sendSyncSurfaceSize()
{ // main thread
SyncSurfaceSizeEvent *e = new SyncSurfaceSizeEvent;
// must block to prevent surface size mess. the render thread will do a
// full rendering round before it unlocks which is good since it can thus
// pick up and the surface (window) size atomically.
thread->mutex.lock();
thread->eventQueue.addEvent(e);
thread->cond.wait(&thread->mutex);
thread->mutex.unlock();
}
struct WindowAndRenderer
{
QWindow *window;
Renderer *renderer;
};
QList<WindowAndRenderer> windows;
void createWindow()
{
static QColor colors[] = { Qt::red, Qt::green, Qt::blue, Qt::yellow, Qt::cyan, Qt::gray };
const int n = windows.count();
Window *w = new Window(QString::asprintf("Window+Thread #%d (%s)", n, qPrintable(graphicsApiName())), graphicsApi);
Renderer *renderer = new Renderer(w, colors[n % 6], n % 3);;
QObject::connect(w, &Window::initRequested, w, [renderer] {
renderer->sendInit();
});
QObject::connect(w, &Window::renderRequested, w, [w, renderer](bool newlyExposed) {
renderer->sendRender(newlyExposed);
w->requestUpdate();
});
QObject::connect(w, &Window::surfaceGoingAway, w, [renderer] {
renderer->sendSurfaceGoingAway();
});
QObject::connect(w, &Window::syncSurfaceSizeRequested, w, [renderer] {
renderer->sendSyncSurfaceSize();
});
windows.append({ w, renderer });
w->show();
}
void closeWindow()
{
WindowAndRenderer wr = windows.takeLast();
delete wr.renderer;
delete wr.window;
}
int main(int argc, char **argv)
{
QApplication app(argc, argv);
#if defined(Q_OS_WIN)
graphicsApi = D3D11;
#elif defined(Q_OS_MACOS) || defined(Q_OS_IOS)
graphicsApi = Metal;
#elif QT_CONFIG(vulkan)
graphicsApi = Vulkan;
#else
graphicsApi = OpenGL;
#endif
QCommandLineParser cmdLineParser;
cmdLineParser.addHelpOption();
QCommandLineOption glOption({ "g", "opengl" }, QLatin1String("OpenGL (2.x)"));
cmdLineParser.addOption(glOption);
QCommandLineOption vkOption({ "v", "vulkan" }, QLatin1String("Vulkan"));
cmdLineParser.addOption(vkOption);
QCommandLineOption d3dOption({ "d", "d3d11" }, QLatin1String("Direct3D 11"));
cmdLineParser.addOption(d3dOption);
QCommandLineOption mtlOption({ "m", "metal" }, QLatin1String("Metal"));
cmdLineParser.addOption(mtlOption);
cmdLineParser.process(app);
if (cmdLineParser.isSet(glOption))
graphicsApi = OpenGL;
if (cmdLineParser.isSet(vkOption))
graphicsApi = Vulkan;
if (cmdLineParser.isSet(d3dOption))
graphicsApi = D3D11;
if (cmdLineParser.isSet(mtlOption))
graphicsApi = Metal;
qDebug("Selected graphics API is %s", qPrintable(graphicsApiName()));
qDebug("This is a multi-api example, use command line arguments to override:\n%s", qPrintable(cmdLineParser.helpText()));
rhi: gl: Avoid magic adjustments to the context/window format ...by removing the entire adjustedFormat() helper. Qt Quick has never used this, which indicates it is not that useful. Same goes for Qt Multimedia or Qt 3D. Ensuring depth and stencil is requested is already solved by using QSurfaceFormat::setDefaultFormat() or by adjusting the formats everywhere as appropriate. The helper function's usages are in the manual tests that use it as a shortcut, and in the GL backend itself. Remove it and leave it up the client to set the depth or stencil buffer size, typically in the global default surface format. (which in fact many of the mentioned manual tests already did, so some of calls to window->setFormat(adjustedFormat()) were completely unnecessary) By not having the built-in magic that tries to always force depth and stencil, we avoid problems that arise then the helper cannot be easily invoked (thinking of widgets and backingstores), and so one ends up with unexpected stencil (or depth) in the context (where the GL backend auto-adjusts), but not in the window (which is not under QRhi's control). It was in practice possible to trigger EGL_BAD_MATCH failures with the new rhi-based widget composition on EGL-based systems. For example, if an application with a QOpenGLWidget did not set both depth and stencil (but only one, or none), it ended up failing due to the context - surface EGLConfig mismatches. On other platforms this matters less due to less strict config/pixelformat management. Pick-to: 6.4 Change-Id: I28ae2de163de63ee91bee3ceae08b58e106e1380 Fixes: QTBUG-104951 Reviewed-by: Andy Nichols <andy.nichols@qt.io>
2022-07-15 12:16:23 +00:00
QSurfaceFormat fmt;
fmt.setDepthBufferSize(24);
QSurfaceFormat::setDefaultFormat(fmt);
#if QT_CONFIG(vulkan)
instance = new QVulkanInstance;
if (graphicsApi == Vulkan) {
instance->setLayers({ "VK_LAYER_KHRONOS_validation" });
if (!instance->create()) {
qWarning("Failed to create Vulkan instance, switching to OpenGL");
graphicsApi = OpenGL;
}
}
#endif
int winCount = 0;
QWidget w;
w.resize(800, 600);
w.setWindowTitle(QCoreApplication::applicationName() + QLatin1String(" - ") + graphicsApiName());
QVBoxLayout *layout = new QVBoxLayout(&w);
QPlainTextEdit *info = new QPlainTextEdit(
QLatin1String("This application tests rendering on a separate thread per window, with dedicated QRhi instances and resources. "
"\n\nThis is the same concept as the Qt Quick Scenegraph's threaded render loop. This should allow rendering to the different windows "
"without unintentionally throttling each other's threads."
"\n\nUsing API: ") + graphicsApiName());
info->setReadOnly(true);
layout->addWidget(info);
QLabel *label = new QLabel(QLatin1String("Window and thread count: 0"));
layout->addWidget(label);
QPushButton *btn = new QPushButton(QLatin1String("New window"));
QObject::connect(btn, &QPushButton::clicked, btn, [label, &winCount] {
winCount += 1;
label->setText(QString::asprintf("Window count: %d", winCount));
createWindow();
});
layout->addWidget(btn);
btn = new QPushButton(QLatin1String("Close window"));
QObject::connect(btn, &QPushButton::clicked, btn, [label, &winCount] {
if (winCount > 0) {
winCount -= 1;
label->setText(QString::asprintf("Window count: %d", winCount));
closeWindow();
}
});
layout->addWidget(btn);
w.show();
int result = app.exec();
for (const WindowAndRenderer &wr : windows) {
delete wr.renderer;
delete wr.window;
}
#if QT_CONFIG(vulkan)
delete instance;
#endif
return result;
}