1	// Copyright (C) 2019 The Qt Company Ltd.
2	// Copyright (C) 2016 Jolla Ltd, author: <gunnar.sletta@jollamobile.com>
3	// Copyright (C) 2016 Robin Burchell <robin.burchell@viroteck.net>
4	// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only
5
6	#include "qsgbatchrenderer_p.h"
7
8	#include <qmath.h>
9
10	#include <QtCore/QElapsedTimer>
11	#include <QtCore/QtNumeric>
12
13	#include <QtGui/QGuiApplication>
14
15	#include <private/qnumeric_p.h>
16	#include "qsgmaterialshader_p.h"
17
18	#include "qsgrhivisualizer_p.h"
19
20	#include <algorithm>
21
22	QT_BEGIN_NAMESPACE
23
24	#ifndef QT_NO_DEBUG
25	Q_QUICK_EXPORT bool qsg_test_and_clear_material_failure();
26	#endif
27
28	int qt_sg_envInt(const char *name, int defaultValue);
29
30	namespace QSGBatchRenderer
31	{
32
33	#define DECLARE_DEBUG_VAR(variable) \
34	static bool debug_ ## variable() \
35	{ static bool value = qgetenv("QSG_RENDERER_DEBUG").contains(QT_STRINGIFY(variable)); return value; }
36	DECLARE_DEBUG_VAR(render)
37	DECLARE_DEBUG_VAR(build)
38	DECLARE_DEBUG_VAR(change)
39	DECLARE_DEBUG_VAR(upload)
40	DECLARE_DEBUG_VAR(roots)
41	DECLARE_DEBUG_VAR(dump)
42	DECLARE_DEBUG_VAR(pools)
43	DECLARE_DEBUG_VAR(noalpha)
44	DECLARE_DEBUG_VAR(noopaque)
45	DECLARE_DEBUG_VAR(noclip)
46	#undef DECLARE_DEBUG_VAR
47
48	#define QSGNODE_TRAVERSE(NODE) for (QSGNode *child = NODE->firstChild(); child; child = child->nextSibling())
49	#define SHADOWNODE_TRAVERSE(NODE) for (Node *child = NODE->firstChild(); child; child = child->sibling())
50
51	static inline int size_of_type(int type)
52	{
53	static int sizes[] = {
54	sizeof(char),
55	sizeof(unsigned char),
56	sizeof(short),
57	sizeof(unsigned short),
58	sizeof(int),
59	sizeof(unsigned int),
60	sizeof(float),
61	2,
62	3,
63	4,
64	sizeof(double)
65	};
66	Q_ASSERT(type >= QSGGeometry::ByteType && type <= QSGGeometry::DoubleType);
67	return sizes[type - QSGGeometry::ByteType];
68	}
69
70	bool qsg_sort_element_increasing_order(Element *a, Element *b) { return a->order < b->order; }
71	bool qsg_sort_element_decreasing_order(Element *a, Element *b) { return a->order > b->order; }
72	bool qsg_sort_batch_is_valid(Batch *a, Batch *b) { return a->first && !b->first; }
73	bool qsg_sort_batch_increasing_order(Batch *a, Batch *b) { return a->first->order < b->first->order; }
74	bool qsg_sort_batch_decreasing_order(Batch *a, Batch *b) { return a->first->order > b->first->order; }
75
76	QSGMaterial::Flag QSGMaterial_FullMatrix = (QSGMaterial::Flag) (QSGMaterial::RequiresFullMatrix & ~QSGMaterial::RequiresFullMatrixExceptTranslate);
77
78	static bool isTranslate(const QMatrix4x4 &m) { return m.flags() <= QMatrix4x4::Translation; }
79	static bool isScale(const QMatrix4x4 &m) { return m.flags() <= QMatrix4x4::Scale; }
80	static bool is2DSafe(const QMatrix4x4 &m) { return m.flags() < QMatrix4x4::Rotation; }
81
82	const float OPAQUE_LIMIT = 0.999f;
83
84	const uint DYNAMIC_VERTEX_INDEX_BUFFER_THRESHOLD = 4;
85	const int VERTEX_BUFFER_BINDING = 0;
86	const int ZORDER_BUFFER_BINDING = VERTEX_BUFFER_BINDING + 1;
87
88	const float VIEWPORT_MIN_DEPTH = 0.0f;
89	const float VIEWPORT_MAX_DEPTH = 1.0f;
90
91	const quint32 DEFAULT_BUFFER_POOL_SIZE_LIMIT = 2 * 1024 * 1024; // 2 MB for m_vboPool and m_iboPool each
92
93	template <class Int>
94	inline Int aligned(Int v, Int byteAlign)
95	{
96	return (v + byteAlign - 1) & ~(byteAlign - 1);
97	}
98
99	QRhiVertexInputAttribute::Format qsg_vertexInputFormat(const QSGGeometry::Attribute &a)
100	{
101	switch (a.type) {
102	case QSGGeometry::FloatType:
103	if (a.tupleSize == 4)
104	return QRhiVertexInputAttribute::Float4;
105	if (a.tupleSize == 3)
106	return QRhiVertexInputAttribute::Float3;
107	if (a.tupleSize == 2)
108	return QRhiVertexInputAttribute::Float2;
109	if (a.tupleSize == 1)
110	return QRhiVertexInputAttribute::Float;
111	break;
112	case QSGGeometry::UnsignedByteType:
113	if (a.tupleSize == 4)
114	return QRhiVertexInputAttribute::UNormByte4;
115	if (a.tupleSize == 2)
116	return QRhiVertexInputAttribute::UNormByte2;
117	if (a.tupleSize == 1)
118	return QRhiVertexInputAttribute::UNormByte;
119	break;
120	default:
121	break;
122	}
123	qWarning(msg: "Unsupported attribute type 0x%x with %d components", a.type, a.tupleSize);
124	Q_UNREACHABLE_RETURN(QRhiVertexInputAttribute::Float);
125	}
126
127	static QRhiVertexInputLayout calculateVertexInputLayout(const QSGMaterialShader *s, const QSGGeometry *geometry, bool batchable)
128	{
129	Q_ASSERT(geometry);
130	const QSGMaterialShaderPrivate *sd = QSGMaterialShaderPrivate::get(s);
131	if (!sd->vertexShader) {
132	qWarning(msg: "No vertex shader in QSGMaterialShader %p", s);
133	return QRhiVertexInputLayout();
134	}
135
136	const int attrCount = geometry->attributeCount();
137	QVarLengthArray<QRhiVertexInputAttribute, 8> inputAttributes;
138	inputAttributes.reserve(sz: attrCount + 1);
139	quint32 offset = 0;
140	for (int i = 0; i < attrCount; ++i) {
141	const QSGGeometry::Attribute &a = geometry->attributes()[i];
142	if (!sd->vertexShader->vertexInputLocations.contains(t: a.position)) {
143	qWarning(msg: "Vertex input %d is present in material but not in shader. This is wrong.",
144	a.position);
145	}
146	inputAttributes.append(t: QRhiVertexInputAttribute(VERTEX_BUFFER_BINDING, a.position, qsg_vertexInputFormat(a), offset));
147	offset += a.tupleSize * size_of_type(type: a.type);
148	}
149	if (batchable) {
150	inputAttributes.append(t: QRhiVertexInputAttribute(ZORDER_BUFFER_BINDING, sd->vertexShader->qt_order_attrib_location,
151	QRhiVertexInputAttribute::Float, 0));
152	}
153
154	Q_ASSERT(VERTEX_BUFFER_BINDING == 0 && ZORDER_BUFFER_BINDING == 1); // not very flexible
155	QVarLengthArray<QRhiVertexInputBinding, 2> inputBindings;
156	inputBindings.append(t: QRhiVertexInputBinding(geometry->sizeOfVertex()));
157	if (batchable)
158	inputBindings.append(t: QRhiVertexInputBinding(sizeof(float)));
159
160	QRhiVertexInputLayout inputLayout;
161	inputLayout.setBindings(first: inputBindings.cbegin(), last: inputBindings.cend());
162	inputLayout.setAttributes(first: inputAttributes.cbegin(), last: inputAttributes.cend());
163
164	return inputLayout;
165	}
166
167	QRhiCommandBuffer::IndexFormat qsg_indexFormat(const QSGGeometry *geometry)
168	{
169	switch (geometry->indexType()) {
170	case QSGGeometry::UnsignedShortType:
171	return QRhiCommandBuffer::IndexUInt16;
172	break;
173	case QSGGeometry::UnsignedIntType:
174	return QRhiCommandBuffer::IndexUInt32;
175	break;
176	default:
177	Q_UNREACHABLE_RETURN(QRhiCommandBuffer::IndexUInt16);
178	}
179	}
180
181	QRhiGraphicsPipeline::Topology qsg_topology(int geomDrawMode)
182	{
183	QRhiGraphicsPipeline::Topology topology = QRhiGraphicsPipeline::Triangles;
184	switch (geomDrawMode) {
185	case QSGGeometry::DrawPoints:
186	topology = QRhiGraphicsPipeline::Points;
187	break;
188	case QSGGeometry::DrawLines:
189	topology = QRhiGraphicsPipeline::Lines;
190	break;
191	case QSGGeometry::DrawLineStrip:
192	topology = QRhiGraphicsPipeline::LineStrip;
193	break;
194	case QSGGeometry::DrawTriangles:
195	topology = QRhiGraphicsPipeline::Triangles;
196	break;
197	case QSGGeometry::DrawTriangleStrip:
198	topology = QRhiGraphicsPipeline::TriangleStrip;
199	break;
200	default:
201	qWarning(msg: "Primitive topology 0x%x not supported", geomDrawMode);
202	break;
203	}
204	return topology;
205	}
206
207	void qsg_setMultiViewFlagsOnMaterial(QSGMaterial *material, int multiViewCount)
208	{
209	material->setFlag(flags: QSGMaterial::MultiView2, on: multiViewCount == 2);
210	material->setFlag(flags: QSGMaterial::MultiView3, on: multiViewCount == 3);
211	material->setFlag(flags: QSGMaterial::MultiView4, on: multiViewCount == 4);
212	}
213
214	ShaderManager::Shader *ShaderManager::prepareMaterial(QSGMaterial *material,
215	const QSGGeometry *geometry,
216	QSGRendererInterface::RenderMode renderMode,
217	int multiViewCount)
218	{
219	qsg_setMultiViewFlagsOnMaterial(material, multiViewCount);
220
221	QSGMaterialType *type = material->type();
222	ShaderKey key = { .type: type, .renderMode: renderMode, .multiViewCount: multiViewCount };
223	Shader *shader = rewrittenShaders.value(key, defaultValue: nullptr);
224	if (shader)
225	return shader;
226
227	shader = new Shader;
228	QSGMaterialShader *s = static_cast<QSGMaterialShader *>(material->createShader(renderMode));
229	context->initializeRhiShader(shader: s, shaderVariant: QShader::BatchableVertexShader);
230	shader->materialShader = s;
231	shader->inputLayout = calculateVertexInputLayout(s, geometry, batchable: true);
232	QSGMaterialShaderPrivate *sD = QSGMaterialShaderPrivate::get(s);
233	shader->stages = {
234	{ QRhiShaderStage::Vertex, sD->shader(stage: QShader::VertexStage), QShader::BatchableVertexShader },
235	{ QRhiShaderStage::Fragment, sD->shader(stage: QShader::FragmentStage) }
236	};
237
238	shader->lastOpacity = 0;
239
240	rewrittenShaders[key] = shader;
241	return shader;
242	}
243
244	ShaderManager::Shader *ShaderManager::prepareMaterialNoRewrite(QSGMaterial *material,
245	const QSGGeometry *geometry,
246	QSGRendererInterface::RenderMode renderMode,
247	int multiViewCount)
248	{
249	qsg_setMultiViewFlagsOnMaterial(material, multiViewCount);
250
251	QSGMaterialType *type = material->type();
252	ShaderKey key = { .type: type, .renderMode: renderMode, .multiViewCount: multiViewCount };
253	Shader *shader = stockShaders.value(key, defaultValue: nullptr);
254	if (shader)
255	return shader;
256
257	shader = new Shader;
258	QSGMaterialShader *s = static_cast<QSGMaterialShader *>(material->createShader(renderMode));
259	context->initializeRhiShader(shader: s, shaderVariant: QShader::StandardShader);
260	shader->materialShader = s;
261	shader->inputLayout = calculateVertexInputLayout(s, geometry, batchable: false);
262	QSGMaterialShaderPrivate *sD = QSGMaterialShaderPrivate::get(s);
263	shader->stages = {
264	{ QRhiShaderStage::Vertex, sD->shader(stage: QShader::VertexStage) },
265	{ QRhiShaderStage::Fragment, sD->shader(stage: QShader::FragmentStage) }
266	};
267
268	shader->lastOpacity = 0;
269
270	stockShaders[key] = shader;
271
272	return shader;
273	}
274
275	void ShaderManager::invalidated()
276	{
277	qDeleteAll(c: stockShaders);
278	stockShaders.clear();
279	qDeleteAll(c: rewrittenShaders);
280	rewrittenShaders.clear();
281
282	qDeleteAll(c: pipelineCache);
283	pipelineCache.clear();
284
285	qDeleteAll(c: srbPool);
286	srbPool.clear();
287	}
288
289	void ShaderManager::clearCachedRendererData()
290	{
291	for (ShaderManager::Shader *sms : std::as_const(t&: stockShaders)) {
292	QSGMaterialShader *s = sms->materialShader;
293	if (s) {
294	QSGMaterialShaderPrivate *sd = QSGMaterialShaderPrivate::get(s);
295	sd->clearCachedRendererData();
296	}
297	}
298	for (ShaderManager::Shader *sms : std::as_const(t&: rewrittenShaders)) {
299	QSGMaterialShader *s = sms->materialShader;
300	if (s) {
301	QSGMaterialShaderPrivate *sd = QSGMaterialShaderPrivate::get(s);
302	sd->clearCachedRendererData();
303	}
304	}
305	}
306
307	void qsg_dumpShadowRoots(BatchRootInfo *i, int indent)
308	{
309	static int extraIndent = 0;
310	++extraIndent;
311
312	QByteArray ind(indent + extraIndent + 10, ' ');
313
314	if (!i) {
315	qDebug(msg: "%s - no info", ind.constData());
316	} else {
317	qDebug() << ind.constData() << "- parent:" << i->parentRoot << "orders" << i->firstOrder << "->" << i->lastOrder << ", avail:" << i->availableOrders;
318	for (QSet<Node *>::const_iterator it = i->subRoots.constBegin();
319	it != i->subRoots.constEnd(); ++it) {
320	qDebug() << ind.constData() << "-" << *it;
321	qsg_dumpShadowRoots(i: (*it)->rootInfo(), indent);
322	}
323	}
324
325	--extraIndent;
326	}
327
328	void qsg_dumpShadowRoots(Node *n)
329	{
330	#ifndef QT_NO_DEBUG_OUTPUT
331	static int indent = 0;
332	++indent;
333
334	QByteArray ind(indent, ' ');
335
336	if (n->type() == QSGNode::ClipNodeType || n->isBatchRoot) {
337	qDebug() << ind.constData() << "[X]" << n->sgNode << Qt::hex << uint(n->sgNode->flags());
338	qsg_dumpShadowRoots(i: n->rootInfo(), indent);
339	} else {
340	QDebug d = qDebug();
341	d << ind.constData() << "[ ]" << n->sgNode << Qt::hex << uint(n->sgNode->flags());
342	if (n->type() == QSGNode::GeometryNodeType)
343	d << "order" << Qt::dec << n->element()->order;
344	}
345
346	SHADOWNODE_TRAVERSE(n)
347	qsg_dumpShadowRoots(n: child);
348
349	--indent;
350	#else
351	Q_UNUSED(n);
352	#endif
353	}
354
355	Updater::Updater(Renderer *r)
356	: renderer(r)
357	, m_roots(32)
358	, m_rootMatrices(8)
359	{
360	m_roots.add(t: 0);
361	m_combined_matrix_stack.add(t: &m_identityMatrix);
362	m_rootMatrices.add(t: m_identityMatrix);
363	}
364
365	void Updater::updateStates(QSGNode *n)
366	{
367	m_current_clip = nullptr;
368
369	m_added = 0;
370	m_transformChange = 0;
371	m_opacityChange = 0;
372
373	Node *sn = renderer->m_nodes.value(key: n, defaultValue: 0);
374	Q_ASSERT(sn);
375
376	if (Q_UNLIKELY(debug_roots()))
377	qsg_dumpShadowRoots(n: sn);
378
379	if (Q_UNLIKELY(debug_build())) {
380	qDebug(msg: "Updater::updateStates()");
381	if (sn->dirtyState & (QSGNode::DirtyNodeAdded << 16))
382	qDebug(msg: " - nodes have been added");
383	if (sn->dirtyState & (QSGNode::DirtyMatrix << 16))
384	qDebug(msg: " - transforms have changed");
385	if (sn->dirtyState & (QSGNode::DirtyOpacity << 16))
386	qDebug(msg: " - opacity has changed");
387	if (uint(sn->dirtyState) & uint(QSGNode::DirtyForceUpdate << 16))
388	qDebug(msg: " - forceupdate");
389	}
390
391	if (Q_UNLIKELY(renderer->m_visualizer->mode() == Visualizer::VisualizeChanges))
392	renderer->m_visualizer->visualizeChangesPrepare(n: sn);
393
394	visitNode(n: sn);
395	}
396
397	void Updater::visitNode(Node *n)
398	{
399	if (m_added == 0 && n->dirtyState == 0 && m_force_update == 0 && m_transformChange == 0 && m_opacityChange == 0)
400	return;
401
402	int count = m_added;
403	if (n->dirtyState & QSGNode::DirtyNodeAdded)
404	++m_added;
405
406	int force = m_force_update;
407	if (n->dirtyState & QSGNode::DirtyForceUpdate)
408	++m_force_update;
409
410	switch (n->type()) {
411	case QSGNode::OpacityNodeType:
412	visitOpacityNode(n);
413	break;
414	case QSGNode::TransformNodeType:
415	visitTransformNode(n);
416	break;
417	case QSGNode::GeometryNodeType:
418	visitGeometryNode(n);
419	break;
420	case QSGNode::ClipNodeType:
421	visitClipNode(n);
422	break;
423	case QSGNode::RenderNodeType:
424	if (m_added)
425	n->renderNodeElement()->root = m_roots.last();
426	Q_FALLTHROUGH(); // to visit children
427	default:
428	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
429	break;
430	}
431
432	m_added = count;
433	m_force_update = force;
434	n->dirtyState = {};
435	}
436
437	void Updater::visitClipNode(Node *n)
438	{
439	ClipBatchRootInfo *extra = n->clipInfo();
440
441	QSGClipNode *cn = static_cast<QSGClipNode *>(n->sgNode);
442
443	if (m_roots.last() && m_added > 0)
444	renderer->registerBatchRoot(childRoot: n, parentRoot: m_roots.last());
445
446	cn->setRendererClipList(m_current_clip);
447	m_current_clip = cn;
448	m_roots << n;
449	m_rootMatrices.add(t: m_rootMatrices.last() * *m_combined_matrix_stack.last());
450	extra->matrix = m_rootMatrices.last();
451	cn->setRendererMatrix(&extra->matrix);
452	m_combined_matrix_stack << &m_identityMatrix;
453
454	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
455
456	m_current_clip = cn->clipList();
457	m_rootMatrices.pop_back();
458	m_combined_matrix_stack.pop_back();
459	m_roots.pop_back();
460	}
461
462	void Updater::visitOpacityNode(Node *n)
463	{
464	QSGOpacityNode *on = static_cast<QSGOpacityNode *>(n->sgNode);
465
466	qreal combined = m_opacity_stack.last() * on->opacity();
467	on->setCombinedOpacity(combined);
468	m_opacity_stack.add(t: combined);
469
470	if (m_added == 0 && n->dirtyState & QSGNode::DirtyOpacity) {
471	bool was = n->isOpaque;
472	bool is = on->opacity() > OPAQUE_LIMIT;
473	if (was != is) {
474	renderer->m_rebuild = Renderer::FullRebuild;
475	n->isOpaque = is;
476	}
477	++m_opacityChange;
478	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
479	--m_opacityChange;
480	} else {
481	if (m_added > 0)
482	n->isOpaque = on->opacity() > OPAQUE_LIMIT;
483	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
484	}
485
486	m_opacity_stack.pop_back();
487	}
488
489	void Updater::visitTransformNode(Node *n)
490	{
491	bool popMatrixStack = false;
492	bool popRootStack = false;
493	bool dirty = n->dirtyState & QSGNode::DirtyMatrix;
494
495	QSGTransformNode *tn = static_cast<QSGTransformNode *>(n->sgNode);
496
497	if (n->isBatchRoot) {
498	if (m_added > 0 && m_roots.last())
499	renderer->registerBatchRoot(childRoot: n, parentRoot: m_roots.last());
500	tn->setCombinedMatrix(m_rootMatrices.last() * *m_combined_matrix_stack.last() * tn->matrix());
501
502	// The only change in this subtree is ourselves and we are a batch root, so
503	// only update subroots and return, saving tons of child-processing (flickable-panning)
504
505	if (!n->becameBatchRoot && m_added == 0 && m_force_update == 0 && m_opacityChange == 0 && dirty && (n->dirtyState & ~QSGNode::DirtyMatrix) == 0) {
506	BatchRootInfo *info = renderer->batchRootInfo(node: n);
507	for (QSet<Node *>::const_iterator it = info->subRoots.constBegin();
508	it != info->subRoots.constEnd(); ++it) {
509	updateRootTransforms(n: *it, root: n, combined: tn->combinedMatrix());
510	}
511	return;
512	}
513
514	n->becameBatchRoot = false;
515
516	m_combined_matrix_stack.add(t: &m_identityMatrix);
517	m_roots.add(t: n);
518	m_rootMatrices.add(t: tn->combinedMatrix());
519
520	popMatrixStack = true;
521	popRootStack = true;
522	} else if (!tn->matrix().isIdentity()) {
523	tn->setCombinedMatrix(*m_combined_matrix_stack.last() * tn->matrix());
524	m_combined_matrix_stack.add(t: &tn->combinedMatrix());
525	popMatrixStack = true;
526	} else {
527	tn->setCombinedMatrix(*m_combined_matrix_stack.last());
528	}
529
530	if (dirty)
531	++m_transformChange;
532
533	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
534
535	if (dirty)
536	--m_transformChange;
537	if (popMatrixStack)
538	m_combined_matrix_stack.pop_back();
539	if (popRootStack) {
540	m_roots.pop_back();
541	m_rootMatrices.pop_back();
542	}
543	}
544
545	void Updater::visitGeometryNode(Node *n)
546	{
547	QSGGeometryNode *gn = static_cast<QSGGeometryNode *>(n->sgNode);
548
549	gn->setRendererMatrix(m_combined_matrix_stack.last());
550	gn->setRendererClipList(m_current_clip);
551	gn->setInheritedOpacity(m_opacity_stack.last());
552
553	if (m_added) {
554	Element *e = n->element();
555	e->root = m_roots.last();
556	e->translateOnlyToRoot = isTranslate(m: *gn->matrix());
557
558	if (e->root) {
559	BatchRootInfo *info = renderer->batchRootInfo(node: e->root);
560	while (info != nullptr) {
561	info->availableOrders--;
562	if (info->availableOrders < 0) {
563	renderer->m_rebuild |= Renderer::BuildRenderLists;
564	} else {
565	renderer->m_rebuild |= Renderer::BuildRenderListsForTaggedRoots;
566	renderer->m_taggedRoots << e->root;
567	}
568	if (info->parentRoot != nullptr)
569	info = renderer->batchRootInfo(node: info->parentRoot);
570	else
571	info = nullptr;
572	}
573	} else {
574	renderer->m_rebuild |= Renderer::FullRebuild;
575	}
576	} else {
577	if (m_transformChange) {
578	Element *e = n->element();
579	e->translateOnlyToRoot = isTranslate(m: *gn->matrix());
580	}
581	if (m_opacityChange) {
582	Element *e = n->element();
583	if (e->batch)
584	renderer->invalidateBatchAndOverlappingRenderOrders(batch: e->batch);
585	}
586	}
587
588	SHADOWNODE_TRAVERSE(n) visitNode(n: child);
589	}
590
591	void Updater::updateRootTransforms(Node *node, Node *root, const QMatrix4x4 &combined)
592	{
593	BatchRootInfo *info = renderer->batchRootInfo(node);
594	QMatrix4x4 m;
595	Node *n = node;
596
597	while (n != root) {
598	if (n->type() == QSGNode::TransformNodeType)
599	m = static_cast<QSGTransformNode *>(n->sgNode)->matrix() * m;
600	n = n->parent();
601	}
602
603	m = combined * m;
604
605	if (node->type() == QSGNode::ClipNodeType) {
606	static_cast<ClipBatchRootInfo *>(info)->matrix = m;
607	} else {
608	Q_ASSERT(node->type() == QSGNode::TransformNodeType);
609	static_cast<QSGTransformNode *>(node->sgNode)->setCombinedMatrix(m);
610	}
611
612	for (QSet<Node *>::const_iterator it = info->subRoots.constBegin();
613	it != info->subRoots.constEnd(); ++it) {
614	updateRootTransforms(node: *it, root: node, combined: m);
615	}
616	}
617
618	int qsg_positionAttribute(QSGGeometry *g)
619	{
620	int vaOffset = 0;
621	for (int a=0; a<g->attributeCount(); ++a) {
622	const QSGGeometry::Attribute &attr = g->attributes()[a];
623	if (attr.isVertexCoordinate && attr.tupleSize == 2 && attr.type == QSGGeometry::FloatType) {
624	return vaOffset;
625	}
626	vaOffset += attr.tupleSize * size_of_type(type: attr.type);
627	}
628	return -1;
629	}
630
631
632	void Rect::map(const QMatrix4x4 &matrix)
633	{
634	const float *m = matrix.constData();
635	if (isScale(m: matrix)) {
636	tl.x = tl.x * m[0] + m[12];
637	tl.y = tl.y * m[5] + m[13];
638	br.x = br.x * m[0] + m[12];
639	br.y = br.y * m[5] + m[13];
640	if (tl.x > br.x)
641	qSwap(value1&: tl.x, value2&: br.x);
642	if (tl.y > br.y)
643	qSwap(value1&: tl.y, value2&: br.y);
644	} else {
645	Pt mtl = tl;
646	Pt mtr = { .x: br.x, .y: tl.y };
647	Pt mbl = { .x: tl.x, .y: br.y };
648	Pt mbr = br;
649
650	mtl.map(mat: matrix);
651	mtr.map(mat: matrix);
652	mbl.map(mat: matrix);
653	mbr.map(mat: matrix);
654
655	set(FLT_MAX, FLT_MAX, right: -FLT_MAX, bottom: -FLT_MAX);
656	(*this) |= mtl;
657	(*this) |= mtr;
658	(*this) |= mbl;
659	(*this) |= mbr;
660	}
661	}
662
663	void Element::computeBounds()
664	{
665	Q_ASSERT(!boundsComputed);
666	boundsComputed = true;
667
668	QSGGeometry *g = node->geometry();
669	int offset = qsg_positionAttribute(g);
670	if (offset == -1) {
671	// No position attribute means overlaps with everything..
672	bounds.set(left: -FLT_MAX, top: -FLT_MAX, FLT_MAX, FLT_MAX);
673	return;
674	}
675
676	bounds.set(FLT_MAX, FLT_MAX, right: -FLT_MAX, bottom: -FLT_MAX);
677	char *vd = (char *) g->vertexData() + offset;
678	for (int i=0; i<g->vertexCount(); ++i) {
679	bounds |= *(Pt *) vd;
680	vd += g->sizeOfVertex();
681	}
682	bounds.map(matrix: *node->matrix());
683
684	if (!qt_is_finite(f: bounds.tl.x) || bounds.tl.x == FLT_MAX)
685	bounds.tl.x = -FLT_MAX;
686	if (!qt_is_finite(f: bounds.tl.y) || bounds.tl.y == FLT_MAX)
687	bounds.tl.y = -FLT_MAX;
688	if (!qt_is_finite(f: bounds.br.x) || bounds.br.x == -FLT_MAX)
689	bounds.br.x = FLT_MAX;
690	if (!qt_is_finite(f: bounds.br.y) || bounds.br.y == -FLT_MAX)
691	bounds.br.y = FLT_MAX;
692
693	Q_ASSERT(bounds.tl.x <= bounds.br.x);
694	Q_ASSERT(bounds.tl.y <= bounds.br.y);
695
696	boundsOutsideFloatRange = bounds.isOutsideFloatRange();
697	}
698
699	BatchCompatibility Batch::isMaterialCompatible(Element *e) const
700	{
701	Element *n = first;
702	// Skip to the first node other than e which has not been removed
703	while (n && (n == e || n->removed))
704	n = n->nextInBatch;
705
706	// Only 'e' in this batch, so a material change doesn't change anything as long as
707	// its blending is still in sync with this batch...
708	if (!n)
709	return BatchIsCompatible;
710
711	QSGMaterial *m = e->node->activeMaterial();
712	QSGMaterial *nm = n->node->activeMaterial();
713	return (nm->type() == m->type() && nm->viewCount() == m->viewCount() && nm->compare(other: m) == 0)
714	? BatchIsCompatible
715	: BatchBreaksOnCompare;
716	}
717
718	/*
719	* Marks this batch as dirty or in the case where the geometry node has
720	* changed to be incompatible with this batch, return false so that
721	* the caller can mark the entire sg for a full rebuild...
722	*/
723	bool Batch::geometryWasChanged(QSGGeometryNode *gn)
724	{
725	Element *e = first;
726	Q_ASSERT_X(e, "Batch::geometryWasChanged", "Batch is expected to 'valid' at this time");
727	// 'gn' is the first node in the batch, compare against the next one.
728	while (e && (e->node == gn || e->removed))
729	e = e->nextInBatch;
730	if (!e || e->node->geometry()->attributes() == gn->geometry()->attributes()) {
731	needsUpload = true;
732	return true;
733	} else {
734	return false;
735	}
736	}
737
738	void Batch::cleanupRemovedElements()
739	{
740	if (!needsPurge)
741	return;
742
743	// remove from front of batch..
744	while (first && first->removed) {
745	first = first->nextInBatch;
746	}
747
748	// Then continue and remove other nodes further out in the batch..
749	if (first) {
750	Element *e = first;
751	while (e->nextInBatch) {
752	if (e->nextInBatch->removed)
753	e->nextInBatch = e->nextInBatch->nextInBatch;
754	else
755	e = e->nextInBatch;
756
757	}
758	}
759
760	needsPurge = false;
761	}
762
763	/*
764	* Iterates through all geometry nodes in this batch and unsets their batch,
765	* thus forcing them to be rebuilt
766	*/
767	void Batch::invalidate()
768	{
769	cleanupRemovedElements();
770	Element *e = first;
771	first = nullptr;
772	root = nullptr;
773	while (e) {
774	e->batch = nullptr;
775	Element *n = e->nextInBatch;
776	e->nextInBatch = nullptr;
777	e = n;
778	}
779	}
780
781	bool Batch::isTranslateOnlyToRoot() const {
782	bool only = true;
783	Element *e = first;
784	while (e && only) {
785	only &= e->translateOnlyToRoot;
786	e = e->nextInBatch;
787	}
788	return only;
789	}
790
791	/*
792	* Iterates through all the nodes in the batch and returns true if the
793	* nodes are all safe to batch. There are two separate criteria:
794	*
795	* - The matrix is such that the z component of the result is of no
796	* consequence.
797	*
798	* - The bounds are inside the stable floating point range. This applies
799	* to desktop only where we in this case can trigger a fallback to
800	* unmerged in which case we pass the geometry straight through and
801	* just apply the matrix.
802	*
803	* NOTE: This also means a slight performance impact for geometries which
804	* are defined to be outside the stable floating point range and still
805	* use single precision float, but given that this implicitly fixes
806	* huge lists and tables, it is worth it.
807	*/
808	bool Batch::isSafeToBatch() const {
809	Element *e = first;
810	while (e) {
811	if (e->boundsOutsideFloatRange)
812	return false;
813	if (!is2DSafe(m: *e->node->matrix()))
814	return false;
815	e = e->nextInBatch;
816	}
817	return true;
818	}
819
820	static int qsg_countNodesInBatch(const Batch *batch)
821	{
822	int sum = 0;
823	Element *e = batch->first;
824	while (e) {
825	++sum;
826	e = e->nextInBatch;
827	}
828	return sum;
829	}
830
831	static int qsg_countNodesInBatches(const QDataBuffer<Batch *> &batches)
832	{
833	int sum = 0;
834	for (int i=0; i<batches.size(); ++i) {
835	sum += qsg_countNodesInBatch(batch: batches.at(i));
836	}
837	return sum;
838	}
839
840	Renderer::Renderer(QSGDefaultRenderContext *ctx, QSGRendererInterface::RenderMode renderMode)
841	: QSGRenderer(ctx)
842	, m_context(ctx)
843	, m_renderMode(renderMode)
844	, m_opaqueRenderList(64)
845	, m_alphaRenderList(64)
846	, m_nextRenderOrder(0)
847	, m_partialRebuild(false)
848	, m_partialRebuildRoot(nullptr)
849	, m_forceNoDepthBuffer(false)
850	, m_opaqueBatches(16)
851	, m_alphaBatches(16)
852	, m_batchPool(16)
853	, m_elementsToDelete(64)
854	, m_tmpAlphaElements(16)
855	, m_tmpOpaqueElements(16)
856	, m_vboPool(16)
857	, m_iboPool(16)
858	, m_vboPoolCost(0)
859	, m_iboPoolCost(0)
860	, m_rebuild(FullRebuild)
861	, m_zRange(0)
862	#if defined(QSGBATCHRENDERER_INVALIDATE_WEDGED_NODES)
863	, m_renderOrderRebuildLower(-1)
864	, m_renderOrderRebuildUpper(-1)
865	#endif
866	, m_currentMaterial(nullptr)
867	, m_currentShader(nullptr)
868	, m_vertexUploadPool(256)
869	, m_indexUploadPool(64)
870	{
871	m_rhi = m_context->rhi();
872	Q_ASSERT(m_rhi); // no more direct OpenGL code path in Qt 6
873
874	m_ubufAlignment = m_rhi->ubufAlignment();
875
876	m_uint32IndexForRhi = !m_rhi->isFeatureSupported(feature: QRhi::NonFourAlignedEffectiveIndexBufferOffset);
877	if (qEnvironmentVariableIntValue(varName: "QSG_RHI_UINT32_INDEX"))
878	m_uint32IndexForRhi = true;
879
880	m_visualizer = new RhiVisualizer(this);
881
882	setNodeUpdater(new Updater(this));
883
884	// The shader manager is shared between renderers (think for example Item
885	// layers that create a new Renderer each) with the same rendercontext (and
886	// so same QRhi).
887	m_shaderManager = ctx->findChild<ShaderManager *>(aName: QString(), options: Qt::FindDirectChildrenOnly);
888	if (!m_shaderManager) {
889	m_shaderManager = new ShaderManager(ctx);
890	m_shaderManager->setObjectName(QStringLiteral("__qt_ShaderManager"));
891	m_shaderManager->setParent(ctx);
892	QObject::connect(sender: ctx, SIGNAL(invalidated()), receiver: m_shaderManager, SLOT(invalidated()), Qt::DirectConnection);
893	}
894
895	m_batchNodeThreshold = qt_sg_envInt(name: "QSG_RENDERER_BATCH_NODE_THRESHOLD", defaultValue: 64);
896	m_batchVertexThreshold = qt_sg_envInt(name: "QSG_RENDERER_BATCH_VERTEX_THRESHOLD", defaultValue: 1024);
897	m_srbPoolThreshold = qt_sg_envInt(name: "QSG_RENDERER_SRB_POOL_THRESHOLD", defaultValue: 1024);
898	m_bufferPoolSizeLimit = qt_sg_envInt(name: "QSG_RENDERER_BUFFER_POOL_LIMIT", defaultValue: DEFAULT_BUFFER_POOL_SIZE_LIMIT);
899
900	if (Q_UNLIKELY(debug_build() || debug_render() || debug_pools())) {
901	qDebug(msg: "Batch thresholds: nodes: %d vertices: %d srb pool: %d buffer pool: %d",
902	m_batchNodeThreshold, m_batchVertexThreshold, m_srbPoolThreshold, m_bufferPoolSizeLimit);
903	}
904	}
905
906	static void qsg_wipeBuffer(Buffer *buffer)
907	{
908	delete buffer->buf;
909
910	// The free here is ok because we're in one of two situations.
911	// 1. We're using the upload pool in which case unmap will have set the
912	// data pointer to 0 and calling free on 0 is ok.
913	// 2. We're using dedicated buffers because of visualization or IBO workaround
914	// and the data something we malloced and must be freed.
915	free(ptr: buffer->data);
916	}
917
918	static void qsg_wipeBatch(Batch *batch)
919	{
920	qsg_wipeBuffer(buffer: &batch->vbo);
921	qsg_wipeBuffer(buffer: &batch->ibo);
922	delete batch->ubuf;
923	batch->stencilClipState.reset();
924	delete batch;
925	}
926
927	Renderer::~Renderer()
928	{
929	if (m_rhi) {
930	// Clean up batches and buffers
931	for (int i = 0; i < m_opaqueBatches.size(); ++i)
932	qsg_wipeBatch(batch: m_opaqueBatches.at(i));
933	for (int i = 0; i < m_alphaBatches.size(); ++i)
934	qsg_wipeBatch(batch: m_alphaBatches.at(i));
935	for (int i = 0; i < m_batchPool.size(); ++i)
936	qsg_wipeBatch(batch: m_batchPool.at(i));
937	for (int i = 0; i < m_vboPool.size(); ++i)
938	delete m_vboPool.at(i);
939	for (int i = 0; i < m_iboPool.size(); ++i)
940	delete m_iboPool.at(i);
941	}
942
943	for (Node *n : std::as_const(t&: m_nodes)) {
944	if (n->type() == QSGNode::GeometryNodeType) {
945	Element *e = n->element();
946	if (!e->removed)
947	m_elementsToDelete.add(t: e);
948	} else if (n->type() == QSGNode::ClipNodeType) {
949	delete n->clipInfo();
950	} else if (n->type() == QSGNode::RenderNodeType) {
951	RenderNodeElement *e = n->renderNodeElement();
952	if (!e->removed)
953	m_elementsToDelete.add(t: e);
954	}
955
956	m_nodeAllocator.release(t: n);
957	}
958
959	// Remaining elements...
960	for (int i=0; i<m_elementsToDelete.size(); ++i)
961	releaseElement(e: m_elementsToDelete.at(i), inDestructor: true);
962
963	destroyGraphicsResources();
964
965	delete m_visualizer;
966	}
967
968	void Renderer::destroyGraphicsResources()
969	{
970	// If this is from the dtor, then the shader manager and its already
971	// prepared shaders will stay around for other renderers -> the cached data
972	// in the rhi shaders have to be purged as it may refer to samplers we
973	// are going to destroy.
974	m_shaderManager->clearCachedRendererData();
975
976	qDeleteAll(c: m_samplers);
977	m_stencilClipCommon.reset();
978	delete m_dummyTexture;
979	m_visualizer->releaseResources();
980	}
981
982	void Renderer::releaseCachedResources()
983	{
984	m_shaderManager->invalidated();
985
986	destroyGraphicsResources();
987
988	m_samplers.clear();
989	m_dummyTexture = nullptr;
990
991	m_rhi->releaseCachedResources();
992
993	m_vertexUploadPool.shrink(size: 0);
994	m_vertexUploadPool.reset();
995	m_indexUploadPool.shrink(size: 0);
996	m_indexUploadPool.reset();
997
998	for (int i = 0; i < m_vboPool.size(); ++i)
999	delete m_vboPool.at(i);
1000	m_vboPool.reset();
1001	m_vboPoolCost = 0;
1002
1003	for (int i = 0; i < m_iboPool.size(); ++i)
1004	delete m_iboPool.at(i);
1005	m_iboPool.reset();
1006	m_iboPoolCost = 0;
1007	}
1008
1009	void Renderer::invalidateAndRecycleBatch(Batch *b)
1010	{
1011	if (b->vbo.buf != nullptr && m_vboPoolCost + b->vbo.buf->size() <= quint32(m_bufferPoolSizeLimit)) {
1012	m_vboPool.add(t: b->vbo.buf);
1013	m_vboPoolCost += b->vbo.buf->size();
1014	} else {
1015	delete b->vbo.buf;
1016	}
1017	if (b->ibo.buf != nullptr && m_iboPoolCost + b->ibo.buf->size() <= quint32(m_bufferPoolSizeLimit)) {
1018	m_iboPool.add(t: b->ibo.buf);
1019	m_iboPoolCost += b->ibo.buf->size();
1020	} else {
1021	delete b->ibo.buf;
1022	}
1023	b->vbo.buf = nullptr;
1024	b->ibo.buf = nullptr;
1025	b->invalidate();
1026	for (int i=0; i<m_batchPool.size(); ++i)
1027	if (b == m_batchPool.at(i))
1028	return;
1029	m_batchPool.add(t: b);
1030	}
1031
1032	void Renderer::map(Buffer *buffer, quint32 byteSize, bool isIndexBuf)
1033	{
1034	if (m_visualizer->mode() == Visualizer::VisualizeNothing) {
1035	// Common case, use a shared memory pool for uploading vertex data to avoid
1036	// excessive reevaluation
1037	QDataBuffer<char> &pool = isIndexBuf ? m_indexUploadPool : m_vertexUploadPool;
1038	if (byteSize > quint32(pool.size()))
1039	pool.resize(size: byteSize);
1040	buffer->data = pool.data();
1041	} else if (buffer->size != byteSize) {
1042	free(ptr: buffer->data);
1043	buffer->data = (char *) malloc(size: byteSize);
1044	Q_CHECK_PTR(buffer->data);
1045	}
1046	buffer->size = byteSize;
1047	}
1048
1049	void Renderer::unmap(Buffer *buffer, bool isIndexBuf)
1050	{
1051	// Batches are pooled and reused which means the QRhiBuffer will be
1052	// still valid in a recycled Batch. We only hit the newBuffer() path
1053	// when there are no buffers to recycle.
1054	QDataBuffer<QRhiBuffer *> *bufferPool = isIndexBuf ? &m_iboPool : &m_vboPool;
1055	if (!buffer->buf && bufferPool->isEmpty()) {
1056	buffer->buf = m_rhi->newBuffer(type: QRhiBuffer::Immutable,
1057	usage: isIndexBuf ? QRhiBuffer::IndexBuffer : QRhiBuffer::VertexBuffer,
1058	size: buffer->size);
1059	if (!buffer->buf->create()) {
1060	qWarning(msg: "Failed to build vertex/index buffer of size %u", buffer->size);
1061	delete buffer->buf;
1062	buffer->buf = nullptr;
1063	}
1064	} else {
1065	if (!buffer->buf) {
1066	const quint32 expectedSize = buffer->size;
1067	qsizetype foundBufferIndex = 0;
1068	for (qsizetype i = 0; i < bufferPool->size(); ++i) {
1069	QRhiBuffer *testBuffer = bufferPool->at(i);
1070	if (!buffer->buf
1071	|| (testBuffer->size() >= expectedSize && testBuffer->size() < buffer->buf->size())
1072	|| (testBuffer->size() < expectedSize && testBuffer->size() > buffer->buf->size())) {
1073	foundBufferIndex = i;
1074	buffer->buf = testBuffer;
1075	if (buffer->buf->size() == expectedSize)
1076	break;
1077	}
1078	}
1079
1080	const qsizetype lastBufferIndex = bufferPool->size() - 1;
1081	if (foundBufferIndex < lastBufferIndex) {
1082	qSwap(value1&: bufferPool->data()[foundBufferIndex],
1083	value2&: bufferPool->data()[lastBufferIndex]);
1084	}
1085	if (isIndexBuf)
1086	m_iboPoolCost -= bufferPool->data()[lastBufferIndex]->size();
1087	else
1088	m_vboPoolCost -= bufferPool->data()[lastBufferIndex]->size();
1089	bufferPool->pop_back();
1090	}
1091
1092	bool needsRebuild = false;
1093	if (buffer->buf->size() < buffer->size) {
1094	buffer->buf->setSize(buffer->size);
1095	needsRebuild = true;
1096	}
1097	if (buffer->buf->type() != QRhiBuffer::Dynamic
1098	&& buffer->nonDynamicChangeCount > DYNAMIC_VERTEX_INDEX_BUFFER_THRESHOLD)
1099	{
1100	buffer->buf->setType(QRhiBuffer::Dynamic);
1101	buffer->nonDynamicChangeCount = 0;
1102	needsRebuild = true;
1103	}
1104	if (needsRebuild) {
1105	if (!buffer->buf->create()) {
1106	qWarning(msg: "Failed to (re)build vertex/index buffer of size %u", buffer->size);
1107	delete buffer->buf;
1108	buffer->buf = nullptr;
1109	}
1110	}
1111	}
1112	if (buffer->buf) {
1113	if (buffer->buf->type() != QRhiBuffer::Dynamic) {
1114	m_resourceUpdates->uploadStaticBuffer(buf: buffer->buf, offset: 0, size: buffer->size, data: buffer->data);
1115	buffer->nonDynamicChangeCount += 1;
1116	} else {
1117	if (m_rhi->resourceLimit(limit: QRhi::FramesInFlight) == 1)
1118	buffer->buf->fullDynamicBufferUpdateForCurrentFrame(data: buffer->data, size: buffer->size);
1119	else
1120	m_resourceUpdates->updateDynamicBuffer(buf: buffer->buf, offset: 0, size: buffer->size, data: buffer->data);
1121	}
1122	}
1123	if (m_visualizer->mode() == Visualizer::VisualizeNothing)
1124	buffer->data = nullptr;
1125	}
1126
1127	BatchRootInfo *Renderer::batchRootInfo(Node *node)
1128	{
1129	BatchRootInfo *info = node->rootInfo();
1130	if (!info) {
1131	if (node->type() == QSGNode::ClipNodeType)
1132	info = new ClipBatchRootInfo;
1133	else {
1134	Q_ASSERT(node->type() == QSGNode::TransformNodeType);
1135	info = new BatchRootInfo;
1136	}
1137	node->data = info;
1138	}
1139	return info;
1140	}
1141
1142	void Renderer::removeBatchRootFromParent(Node *childRoot)
1143	{
1144	BatchRootInfo *childInfo = batchRootInfo(node: childRoot);
1145	if (!childInfo->parentRoot)
1146	return;
1147	BatchRootInfo *parentInfo = batchRootInfo(node: childInfo->parentRoot);
1148
1149	Q_ASSERT(parentInfo->subRoots.contains(childRoot));
1150	parentInfo->subRoots.remove(value: childRoot);
1151	childInfo->parentRoot = nullptr;
1152	}
1153
1154	void Renderer::registerBatchRoot(Node *subRoot, Node *parentRoot)
1155	{
1156	BatchRootInfo *subInfo = batchRootInfo(node: subRoot);
1157	BatchRootInfo *parentInfo = batchRootInfo(node: parentRoot);
1158	subInfo->parentRoot = parentRoot;
1159	parentInfo->subRoots << subRoot;
1160	}
1161
1162	bool Renderer::changeBatchRoot(Node *node, Node *root)
1163	{
1164	BatchRootInfo *subInfo = batchRootInfo(node);
1165	if (subInfo->parentRoot == root)
1166	return false;
1167	if (subInfo->parentRoot) {
1168	BatchRootInfo *oldRootInfo = batchRootInfo(node: subInfo->parentRoot);
1169	oldRootInfo->subRoots.remove(value: node);
1170	}
1171	BatchRootInfo *newRootInfo = batchRootInfo(node: root);
1172	newRootInfo->subRoots << node;
1173	subInfo->parentRoot = root;
1174	return true;
1175	}
1176
1177	void Renderer::nodeChangedBatchRoot(Node *node, Node *root)
1178	{
1179	if (node->type() == QSGNode::ClipNodeType || node->isBatchRoot) {
1180	// When we reach a batchroot, we only need to update it. Its subtree
1181	// is relative to that root, so no need to recurse further.
1182	changeBatchRoot(node, root);
1183	return;
1184	} else if (node->type() == QSGNode::GeometryNodeType) {
1185	// Only need to change the root as nodeChanged anyway flags a full update.
1186	Element *e = node->element();
1187	if (e) {
1188	e->root = root;
1189	e->boundsComputed = false;
1190	}
1191	} else if (node->type() == QSGNode::RenderNodeType) {
1192	RenderNodeElement *e = node->renderNodeElement();
1193	if (e)
1194	e->root = root;
1195	}
1196
1197	SHADOWNODE_TRAVERSE(node)
1198	nodeChangedBatchRoot(node: child, root);
1199	}
1200
1201	void Renderer::nodeWasTransformed(Node *node, int *vertexCount)
1202	{
1203	if (node->type() == QSGNode::GeometryNodeType) {
1204	QSGGeometryNode *gn = static_cast<QSGGeometryNode *>(node->sgNode);
1205	*vertexCount += gn->geometry()->vertexCount();
1206	Element *e = node->element();
1207	if (e) {
1208	e->boundsComputed = false;
1209	if (e->batch) {
1210	if (!e->batch->isOpaque) {
1211	invalidateBatchAndOverlappingRenderOrders(batch: e->batch);
1212	} else if (e->batch->merged) {
1213	e->batch->needsUpload = true;
1214	}
1215	}
1216	}
1217	}
1218
1219	SHADOWNODE_TRAVERSE(node)
1220	nodeWasTransformed(node: child, vertexCount);
1221	}
1222
1223	void Renderer::nodeWasAdded(QSGNode *node, Node *shadowParent)
1224	{
1225	Q_ASSERT(!m_nodes.contains(node));
1226	if (node->isSubtreeBlocked())
1227	return;
1228
1229	Node *snode = m_nodeAllocator.allocate();
1230	snode->sgNode = node;
1231	m_nodes.insert(key: node, value: snode);
1232	if (shadowParent)
1233	shadowParent->append(child: snode);
1234
1235	if (node->type() == QSGNode::GeometryNodeType) {
1236	snode->data = m_elementAllocator.allocate();
1237	snode->element()->setNode(static_cast<QSGGeometryNode *>(node));
1238
1239	} else if (node->type() == QSGNode::ClipNodeType) {
1240	snode->data = new ClipBatchRootInfo;
1241	m_rebuild |= FullRebuild;
1242
1243	} else if (node->type() == QSGNode::RenderNodeType) {
1244	QSGRenderNode *rn = static_cast<QSGRenderNode *>(node);
1245	RenderNodeElement *e = new RenderNodeElement(rn);
1246	snode->data = e;
1247	Q_ASSERT(!m_renderNodeElements.contains(rn));
1248	m_renderNodeElements.insert(key: e->renderNode, value: e);
1249	if (!rn->flags().testFlag(flag: QSGRenderNode::DepthAwareRendering))
1250	m_forceNoDepthBuffer = true;
1251	m_rebuild |= FullRebuild;
1252	}
1253
1254	QSGNODE_TRAVERSE(node)
1255	nodeWasAdded(node: child, shadowParent: snode);
1256	}
1257
1258	void Renderer::nodeWasRemoved(Node *node)
1259	{
1260	// Prefix traversal as removeBatchRootFromParent below removes nodes
1261	// in a bottom-up manner. Note that we *cannot* use SHADOWNODE_TRAVERSE
1262	// here, because we delete 'child' (when recursed, down below), so we'd
1263	// have a use-after-free.
1264	{
1265	Node *child = node->firstChild();
1266	while (child) {
1267	// Remove (and delete) child
1268	node->remove(child);
1269	nodeWasRemoved(node: child);
1270	child = node->firstChild();
1271	}
1272	}
1273
1274	if (node->type() == QSGNode::GeometryNodeType) {
1275	Element *e = node->element();
1276	if (e) {
1277	e->removed = true;
1278	m_elementsToDelete.add(t: e);
1279	e->node = nullptr;
1280	if (e->root) {
1281	BatchRootInfo *info = batchRootInfo(node: e->root);
1282	info->availableOrders++;
1283	}
1284	if (e->batch) {
1285	e->batch->needsUpload = true;
1286	e->batch->needsPurge = true;
1287	}
1288
1289	}
1290
1291	} else if (node->type() == QSGNode::ClipNodeType) {
1292	removeBatchRootFromParent(childRoot: node);
1293	delete node->clipInfo();
1294	m_rebuild |= FullRebuild;
1295	m_taggedRoots.remove(value: node);
1296
1297	} else if (node->isBatchRoot) {
1298	removeBatchRootFromParent(childRoot: node);
1299	delete node->rootInfo();
1300	m_rebuild |= FullRebuild;
1301	m_taggedRoots.remove(value: node);
1302
1303	} else if (node->type() == QSGNode::RenderNodeType) {
1304	RenderNodeElement *e = m_renderNodeElements.take(key: static_cast<QSGRenderNode *>(node->sgNode));
1305	if (e) {
1306	e->removed = true;
1307	m_elementsToDelete.add(t: e);
1308	if (m_renderNodeElements.isEmpty()) {
1309	m_forceNoDepthBuffer = false;
1310	// Must have a full rebuild given useDepthBuffer() now returns
1311	// a different value than before, meaning there can once again
1312	// be an opaque pass.
1313	m_rebuild |= FullRebuild;
1314	}
1315
1316	if (e->batch != nullptr)
1317	e->batch->needsPurge = true;
1318	}
1319	}
1320
1321	Q_ASSERT(m_nodes.contains(node->sgNode));
1322
1323	m_nodeAllocator.release(t: m_nodes.take(key: node->sgNode));
1324	}
1325
1326	void Renderer::turnNodeIntoBatchRoot(Node *node)
1327	{
1328	if (Q_UNLIKELY(debug_change())) qDebug(msg: " - new batch root");
1329	m_rebuild |= FullRebuild;
1330	node->isBatchRoot = true;
1331	node->becameBatchRoot = true;
1332
1333	Node *p = node->parent();
1334	while (p) {
1335	if (p->type() == QSGNode::ClipNodeType || p->isBatchRoot) {
1336	registerBatchRoot(subRoot: node, parentRoot: p);
1337	break;
1338	}
1339	p = p->parent();
1340	}
1341
1342	SHADOWNODE_TRAVERSE(node)
1343	nodeChangedBatchRoot(node: child, root: node);
1344	}
1345
1346
1347	void Renderer::nodeChanged(QSGNode *node, QSGNode::DirtyState state)
1348	{
1349	#ifndef QT_NO_DEBUG_OUTPUT
1350	if (Q_UNLIKELY(debug_change())) {
1351	QDebug debug = qDebug();
1352	debug << "dirty:";
1353	if (state & QSGNode::DirtyGeometry)
1354	debug << "Geometry";
1355	if (state & QSGNode::DirtyMaterial)
1356	debug << "Material";
1357	if (state & QSGNode::DirtyMatrix)
1358	debug << "Matrix";
1359	if (state & QSGNode::DirtyNodeAdded)
1360	debug << "Added";
1361	if (state & QSGNode::DirtyNodeRemoved)
1362	debug << "Removed";
1363	if (state & QSGNode::DirtyOpacity)
1364	debug << "Opacity";
1365	if (state & QSGNode::DirtySubtreeBlocked)
1366	debug << "SubtreeBlocked";
1367	if (state & QSGNode::DirtyForceUpdate)
1368	debug << "ForceUpdate";
1369
1370	// when removed, some parts of the node could already have been destroyed
1371	// so don't debug it out.
1372	if (state & QSGNode::DirtyNodeRemoved)
1373	debug << (void *) node << node->type();
1374	else
1375	debug << node;
1376	}
1377	#endif
1378	// As this function calls nodeChanged recursively, we do it at the top
1379	// to avoid that any of the others are processed twice.
1380	if (state & QSGNode::DirtySubtreeBlocked) {
1381	Node *sn = m_nodes.value(key: node);
1382
1383	// Force a batch rebuild if this includes an opacity change
1384	if (state & QSGNode::DirtyOpacity)
1385	m_rebuild |= FullRebuild;
1386
1387	bool blocked = node->isSubtreeBlocked();
1388	if (blocked && sn) {
1389	nodeChanged(node, state: QSGNode::DirtyNodeRemoved);
1390	Q_ASSERT(m_nodes.value(node) == 0);
1391	} else if (!blocked && !sn) {
1392	nodeChanged(node, state: QSGNode::DirtyNodeAdded);
1393	}
1394	return;
1395	}
1396
1397	if (state & QSGNode::DirtyNodeAdded) {
1398	if (nodeUpdater()->isNodeBlocked(n: node, root: rootNode())) {
1399	QSGRenderer::nodeChanged(node, state);
1400	return;
1401	}
1402	if (node == rootNode())
1403	nodeWasAdded(node, shadowParent: nullptr);
1404	else
1405	nodeWasAdded(node, shadowParent: m_nodes.value(key: node->parent()));
1406	}
1407
1408	// Mark this node dirty in the shadow tree.
1409	Node *shadowNode = m_nodes.value(key: node);
1410
1411	// Blocked subtrees won't have shadow nodes, so we can safely abort
1412	// here..
1413	if (!shadowNode) {
1414	QSGRenderer::nodeChanged(node, state);
1415	return;
1416	}
1417
1418	shadowNode->dirtyState |= state;
1419
1420	if (state & QSGNode::DirtyMatrix && !shadowNode->isBatchRoot) {
1421	Q_ASSERT(node->type() == QSGNode::TransformNodeType);
1422	if (node->m_subtreeRenderableCount > m_batchNodeThreshold) {
1423	turnNodeIntoBatchRoot(node: shadowNode);
1424	} else {
1425	int vertices = 0;
1426	nodeWasTransformed(node: shadowNode, vertexCount: &vertices);
1427	if (vertices > m_batchVertexThreshold) {
1428	turnNodeIntoBatchRoot(node: shadowNode);
1429	}
1430	}
1431	}
1432
1433	if (state & QSGNode::DirtyGeometry && node->type() == QSGNode::GeometryNodeType) {
1434	QSGGeometryNode *gn = static_cast<QSGGeometryNode *>(node);
1435	Element *e = shadowNode->element();
1436	if (e) {
1437	e->boundsComputed = false;
1438	Batch *b = e->batch;
1439	if (b) {
1440	if (!e->batch->geometryWasChanged(gn) || !e->batch->isOpaque) {
1441	invalidateBatchAndOverlappingRenderOrders(batch: e->batch);
1442	} else {
1443	b->needsUpload = true;
1444	}
1445	}
1446	}
1447	}
1448
1449	if (state & QSGNode::DirtyMaterial && node->type() == QSGNode::GeometryNodeType) {
1450	Element *e = shadowNode->element();
1451	if (e) {
1452	bool blended = hasMaterialWithBlending(n: static_cast<QSGGeometryNode *>(node));
1453	if (e->isMaterialBlended != blended) {
1454	m_rebuild |= Renderer::FullRebuild;
1455	e->isMaterialBlended = blended;
1456	} else if (e->batch) {
1457	if (e->batch->isMaterialCompatible(e) == BatchBreaksOnCompare)
1458	invalidateBatchAndOverlappingRenderOrders(batch: e->batch);
1459	} else {
1460	m_rebuild |= Renderer::BuildBatches;
1461	}
1462	}
1463	}
1464
1465	// Mark the shadow tree dirty all the way back to the root...
1466	QSGNode::DirtyState dirtyChain = state & (QSGNode::DirtyNodeAdded
1467	| QSGNode::DirtyOpacity
1468	| QSGNode::DirtyMatrix
1469	| QSGNode::DirtySubtreeBlocked
1470	| QSGNode::DirtyForceUpdate);
1471	if (dirtyChain != 0) {
1472	dirtyChain = QSGNode::DirtyState(dirtyChain << 16);
1473	Node *sn = shadowNode->parent();
1474	while (sn) {
1475	sn->dirtyState |= dirtyChain;
1476	sn = sn->parent();
1477	}
1478	}
1479
1480	// Delete happens at the very end because it deletes the shadownode.
1481	if (state & QSGNode::DirtyNodeRemoved) {
1482	Node *parent = shadowNode->parent();
1483	if (parent)
1484	parent->remove(child: shadowNode);
1485	nodeWasRemoved(node: shadowNode);
1486	Q_ASSERT(m_nodes.value(node) == 0);
1487	}
1488
1489	QSGRenderer::nodeChanged(node, state);
1490	}
1491
1492	/*
1493	* Traverses the tree and builds two list of geometry nodes. One for
1494	* the opaque and one for the translucent. These are populated
1495	* in the order they should visually appear in, meaning first
1496	* to the back and last to the front.
1497	*
1498	* We split opaque and translucent as we can perform different
1499	* types of reordering / batching strategies on them, depending
1500	*
1501	* Note: It would be tempting to use the shadow nodes instead of the QSGNodes
1502	* for traversal to avoid hash lookups, but the order of the children
1503	* is important and they are not preserved in the shadow tree, so we must
1504	* use the actual QSGNode tree.
1505	*/
1506	void Renderer::buildRenderLists(QSGNode *node)
1507	{
1508	if (node->isSubtreeBlocked())
1509	return;
1510
1511	Node *shadowNode = m_nodes.value(key: node);
1512	Q_ASSERT(shadowNode);
1513
1514	if (node->type() == QSGNode::GeometryNodeType) {
1515	QSGGeometryNode *gn = static_cast<QSGGeometryNode *>(node);
1516
1517	Element *e = shadowNode->element();
1518	Q_ASSERT(e);
1519
1520	bool opaque = gn->inheritedOpacity() > OPAQUE_LIMIT && !(gn->activeMaterial()->flags() & QSGMaterial::Blending);
1521	if (opaque && useDepthBuffer())
1522	m_opaqueRenderList << e;
1523	else
1524	m_alphaRenderList << e;
1525
1526	e->order = ++m_nextRenderOrder;
1527	// Used while rebuilding partial roots.
1528	if (m_partialRebuild)
1529	e->orphaned = false;
1530
1531	} else if (node->type() == QSGNode::ClipNodeType || shadowNode->isBatchRoot) {
1532	Q_ASSERT(m_nodes.contains(node));
1533	BatchRootInfo *info = batchRootInfo(node: shadowNode);
1534	if (node == m_partialRebuildRoot) {
1535	m_nextRenderOrder = info->firstOrder;
1536	QSGNODE_TRAVERSE(node)
1537	buildRenderLists(node: child);
1538	m_nextRenderOrder = info->lastOrder + 1;
1539	} else {
1540	int currentOrder = m_nextRenderOrder;
1541	QSGNODE_TRAVERSE(node)
1542	buildRenderLists(node: child);
1543	int padding = (m_nextRenderOrder - currentOrder) >> 2;
1544	info->firstOrder = currentOrder;
1545	info->availableOrders = padding;
1546	info->lastOrder = m_nextRenderOrder + padding;
1547	m_nextRenderOrder = info->lastOrder;
1548	}
1549	return;
1550	} else if (node->type() == QSGNode::RenderNodeType) {
1551	RenderNodeElement *e = shadowNode->renderNodeElement();
1552	m_alphaRenderList << e;
1553	e->order = ++m_nextRenderOrder;
1554	Q_ASSERT(e);
1555	}
1556
1557	QSGNODE_TRAVERSE(node)
1558	buildRenderLists(node: child);
1559	}
1560
1561	void Renderer::tagSubRoots(Node *node)
1562	{
1563	BatchRootInfo *i = batchRootInfo(node);
1564	m_taggedRoots << node;
1565	for (QSet<Node *>::const_iterator it = i->subRoots.constBegin();
1566	it != i->subRoots.constEnd(); ++it) {
1567	tagSubRoots(node: *it);
1568	}
1569	}
1570
1571	static void qsg_addOrphanedElements(QDataBuffer<Element *> &orphans, const QDataBuffer<Element *> &renderList)
1572	{
1573	orphans.reset();
1574	for (int i=0; i<renderList.size(); ++i) {
1575	Element *e = renderList.at(i);
1576	if (e && !e->removed) {
1577	e->orphaned = true;
1578	orphans.add(t: e);
1579	}
1580	}
1581	}
1582
1583	static void qsg_addBackOrphanedElements(QDataBuffer<Element *> &orphans, QDataBuffer<Element *> &renderList)
1584	{
1585	for (int i=0; i<orphans.size(); ++i) {
1586	Element *e = orphans.at(i);
1587	if (e->orphaned)
1588	renderList.add(t: e);
1589	}
1590	orphans.reset();
1591	}
1592
1593	/*
1594	* To rebuild the tagged roots, we start by putting all subroots of tagged
1595	* roots into the list of tagged roots. This is to make the rest of the
1596	* algorithm simpler.
1597	*
1598	* Second, we invalidate all batches which belong to tagged roots, which now
1599	* includes the entire subtree under a given root
1600	*
1601	* Then we call buildRenderLists for all tagged subroots which do not have
1602	* parents which are tagged, aka, we traverse only the topmosts roots.
1603	*
1604	* Then we sort the render lists based on their render order, to restore the
1605	* right order for rendering.
1606	*/
1607	void Renderer::buildRenderListsForTaggedRoots()
1608	{
1609	// Flag any element that is currently in the render lists, but which
1610	// is not in a batch. This happens when we have a partial rebuild
1611	// in one sub tree while we have a BuildBatches change in another
1612	// isolated subtree. So that batch-building takes into account
1613	// these "orphaned" nodes, we flag them now. The ones under tagged
1614	// roots will be cleared again. The remaining ones are added into the
1615	// render lists so that they contain all visual nodes after the
1616	// function completes.
1617	qsg_addOrphanedElements(orphans&: m_tmpOpaqueElements, renderList: m_opaqueRenderList);
1618	qsg_addOrphanedElements(orphans&: m_tmpAlphaElements, renderList: m_alphaRenderList);
1619
1620	// Take a copy now, as we will be adding to this while traversing..
1621	QSet<Node *> roots = m_taggedRoots;
1622	for (QSet<Node *>::const_iterator it = roots.constBegin();
1623	it != roots.constEnd(); ++it) {
1624	tagSubRoots(node: *it);
1625	}
1626
1627	for (int i=0; i<m_opaqueBatches.size(); ++i) {
1628	Batch *b = m_opaqueBatches.at(i);
1629	if (m_taggedRoots.contains(value: b->root))
1630	invalidateAndRecycleBatch(b);
1631
1632	}
1633	for (int i=0; i<m_alphaBatches.size(); ++i) {
1634	Batch *b = m_alphaBatches.at(i);
1635	if (m_taggedRoots.contains(value: b->root))
1636	invalidateAndRecycleBatch(b);
1637	}
1638
1639	m_opaqueRenderList.reset();
1640	m_alphaRenderList.reset();
1641	int maxRenderOrder = m_nextRenderOrder;
1642	m_partialRebuild = true;
1643	// Traverse each root, assigning it
1644	for (QSet<Node *>::const_iterator it = m_taggedRoots.constBegin();
1645	it != m_taggedRoots.constEnd(); ++it) {
1646	Node *root = *it;
1647	BatchRootInfo *i = batchRootInfo(node: root);
1648	if ((!i->parentRoot || !m_taggedRoots.contains(value: i->parentRoot))
1649	&& !nodeUpdater()->isNodeBlocked(n: root->sgNode, root: rootNode())) {
1650	m_nextRenderOrder = i->firstOrder;
1651	m_partialRebuildRoot = root->sgNode;
1652	buildRenderLists(node: root->sgNode);
1653	}
1654	}
1655	m_partialRebuild = false;
1656	m_partialRebuildRoot = nullptr;
1657	m_taggedRoots.clear();
1658	m_nextRenderOrder = qMax(a: m_nextRenderOrder, b: maxRenderOrder);
1659
1660	// Add orphaned elements back into the list and then sort it..
1661	qsg_addBackOrphanedElements(orphans&: m_tmpOpaqueElements, renderList&: m_opaqueRenderList);
1662	qsg_addBackOrphanedElements(orphans&: m_tmpAlphaElements, renderList&: m_alphaRenderList);
1663
1664	if (m_opaqueRenderList.size())
1665	std::sort(first: &m_opaqueRenderList.first(), last: &m_opaqueRenderList.last() + 1, comp: qsg_sort_element_decreasing_order);
1666	if (m_alphaRenderList.size())
1667	std::sort(first: &m_alphaRenderList.first(), last: &m_alphaRenderList.last() + 1, comp: qsg_sort_element_increasing_order);
1668
1669	}
1670
1671	void Renderer::buildRenderListsFromScratch()
1672	{
1673	m_opaqueRenderList.reset();
1674	m_alphaRenderList.reset();
1675
1676	for (int i=0; i<m_opaqueBatches.size(); ++i)
1677	invalidateAndRecycleBatch(b: m_opaqueBatches.at(i));
1678	for (int i=0; i<m_alphaBatches.size(); ++i)
1679	invalidateAndRecycleBatch(b: m_alphaBatches.at(i));
1680	m_opaqueBatches.reset();
1681	m_alphaBatches.reset();
1682
1683	m_nextRenderOrder = 0;
1684
1685	buildRenderLists(node: rootNode());
1686	}
1687
1688	void Renderer::invalidateBatchAndOverlappingRenderOrders(Batch *batch)
1689	{
1690	Q_ASSERT(batch);
1691	Q_ASSERT(batch->first);
1692
1693	#if defined(QSGBATCHRENDERER_INVALIDATE_WEDGED_NODES)
1694	if (m_renderOrderRebuildLower < 0 || batch->first->order < m_renderOrderRebuildLower)
1695	m_renderOrderRebuildLower = batch->first->order;
1696	if (m_renderOrderRebuildUpper < 0 || batch->lastOrderInBatch > m_renderOrderRebuildUpper)
1697	m_renderOrderRebuildUpper = batch->lastOrderInBatch;
1698
1699	int first = m_renderOrderRebuildLower;
1700	int last = m_renderOrderRebuildUpper;
1701	#else
1702	int first = batch->first->order;
1703	int last = batch->lastOrderInBatch;
1704	#endif
1705
1706	batch->invalidate();
1707
1708	for (int i=0; i<m_alphaBatches.size(); ++i) {
1709	Batch *b = m_alphaBatches.at(i);
1710	if (b->first) {
1711	int bf = b->first->order;
1712	int bl = b->lastOrderInBatch;
1713	if (bl > first && bf < last)
1714	b->invalidate();
1715	}
1716	}
1717
1718	m_rebuild |= BuildBatches;
1719	}
1720
1721	/* Clean up batches by making it a consecutive list of "valid"
1722	* batches and moving all invalidated batches to the batches pool.
1723	*/
1724	void Renderer::cleanupBatches(QDataBuffer<Batch *> *batches) {
1725	if (batches->size()) {
1726	std::stable_sort(first: &batches->first(), last: &batches->last() + 1, comp: qsg_sort_batch_is_valid);
1727	int count = 0;
1728	while (count < batches->size() && batches->at(i: count)->first)
1729	++count;
1730	for (int i=count; i<batches->size(); ++i)
1731	invalidateAndRecycleBatch(b: batches->at(i));
1732	batches->resize(size: count);
1733	}
1734	}
1735
1736	void Renderer::prepareOpaqueBatches()
1737	{
1738	for (int i=m_opaqueRenderList.size() - 1; i >= 0; --i) {
1739	Element *ei = m_opaqueRenderList.at(i);
1740	if (!ei || ei->batch || ei->node->geometry()->vertexCount() == 0)
1741	continue;
1742	Batch *batch = newBatch();
1743	batch->first = ei;
1744	batch->root = ei->root;
1745	batch->isOpaque = true;
1746	batch->needsUpload = true;
1747	batch->positionAttribute = qsg_positionAttribute(g: ei->node->geometry());
1748
1749	m_opaqueBatches.add(t: batch);
1750
1751	ei->batch = batch;
1752	Element *next = ei;
1753
1754	QSGGeometryNode *gni = ei->node;
1755
1756	for (int j = i - 1; j >= 0; --j) {
1757	Element *ej = m_opaqueRenderList.at(i: j);
1758	if (!ej)
1759	continue;
1760	if (ej->root != ei->root)
1761	break;
1762	if (ej->batch || ej->node->geometry()->vertexCount() == 0)
1763	continue;
1764
1765	QSGGeometryNode *gnj = ej->node;
1766
1767	const QSGGeometry *gniGeometry = gni->geometry();
1768	const QSGMaterial *gniMaterial = gni->activeMaterial();
1769	const QSGGeometry *gnjGeometry = gnj->geometry();
1770	const QSGMaterial *gnjMaterial = gnj->activeMaterial();
1771	if (gni->clipList() == gnj->clipList()
1772	&& gniGeometry->drawingMode() == gnjGeometry->drawingMode()
1773	&& (gniGeometry->drawingMode() != QSGGeometry::DrawLines || gniGeometry->lineWidth() == gnjGeometry->lineWidth())
1774	&& gniGeometry->attributes() == gnjGeometry->attributes()
1775	&& gniGeometry->indexType() == gnjGeometry->indexType()
1776	&& gni->inheritedOpacity() == gnj->inheritedOpacity()
1777	&& gniMaterial->type() == gnjMaterial->type()
1778	&& gniMaterial->viewCount() == gnjMaterial->viewCount()
1779	&& gniMaterial->compare(other: gnjMaterial) == 0)
1780	{
1781	ej->batch = batch;
1782	next->nextInBatch = ej;
1783	next = ej;
1784	}
1785	}
1786
1787	batch->lastOrderInBatch = next->order;
1788	}
1789	}
1790
1791	bool Renderer::checkOverlap(int first, int last, const Rect &bounds)
1792	{
1793	for (int i=first; i<=last; ++i) {
1794	Element *e = m_alphaRenderList.at(i);
1795	#if defined(QSGBATCHRENDERER_INVALIDATE_WEDGED_NODES)
1796	if (!e || e->batch)
1797	#else
1798	if (!e)
1799	#endif
1800	continue;
1801	Q_ASSERT(e->boundsComputed);
1802	if (e->bounds.intersects(r: bounds))
1803	return true;
1804	}
1805	return false;
1806	}
1807
1808	/*
1809	*
1810	* To avoid the O(n^2) checkOverlap check in most cases, we have the
1811	* overlapBounds which is the union of all bounding rects to check overlap
1812	* for. We know that if it does not overlap, then none of the individual
1813	* ones will either. For the typical list case, this results in no calls
1814	* to checkOverlap what-so-ever. This also ensures that when all consecutive
1815	* items are matching (such as a table of text), we don't build up an
1816	* overlap bounds and thus do not require full overlap checks.
1817	*/
1818
1819	void Renderer::prepareAlphaBatches()
1820	{
1821	for (int i=0; i<m_alphaRenderList.size(); ++i) {
1822	Element *e = m_alphaRenderList.at(i);
1823	if (!e || e->isRenderNode)
1824	continue;
1825	Q_ASSERT(!e->removed);
1826	e->ensureBoundsValid();
1827	}
1828
1829	for (int i=0; i<m_alphaRenderList.size(); ++i) {
1830	Element *ei = m_alphaRenderList.at(i);
1831	if (!ei || ei->batch)
1832	continue;
1833
1834	if (ei->isRenderNode) {
1835	Batch *rnb = newBatch();
1836	rnb->first = ei;
1837	rnb->root = ei->root;
1838	rnb->isOpaque = false;
1839	rnb->isRenderNode = true;
1840	ei->batch = rnb;
1841	m_alphaBatches.add(t: rnb);
1842	continue;
1843	}
1844
1845	if (ei->node->geometry()->vertexCount() == 0)
1846	continue;
1847
1848	Batch *batch = newBatch();
1849	batch->first = ei;
1850	batch->root = ei->root;
1851	batch->isOpaque = false;
1852	batch->needsUpload = true;
1853	m_alphaBatches.add(t: batch);
1854	ei->batch = batch;
1855
1856	QSGGeometryNode *gni = ei->node;
1857	batch->positionAttribute = qsg_positionAttribute(g: gni->geometry());
1858
1859	Rect overlapBounds;
1860	overlapBounds.set(FLT_MAX, FLT_MAX, right: -FLT_MAX, bottom: -FLT_MAX);
1861
1862	Element *next = ei;
1863
1864	for (int j = i + 1; j < m_alphaRenderList.size(); ++j) {
1865	Element *ej = m_alphaRenderList.at(i: j);
1866	if (!ej)
1867	continue;
1868	if (ej->root != ei->root || ej->isRenderNode)
1869	break;
1870	if (ej->batch) {
1871	#if !defined(QSGBATCHRENDERER_INVALIDATE_WEDGED_NODES)
1872	overlapBounds |= ej->bounds;
1873	#endif
1874	continue;
1875	}
1876
1877	QSGGeometryNode *gnj = ej->node;
1878	if (gnj->geometry()->vertexCount() == 0)
1879	continue;
1880
1881	const QSGGeometry *gniGeometry = gni->geometry();
1882	const QSGMaterial *gniMaterial = gni->activeMaterial();
1883	const QSGGeometry *gnjGeometry = gnj->geometry();
1884	const QSGMaterial *gnjMaterial = gnj->activeMaterial();
1885	if (gni->clipList() == gnj->clipList()
1886	&& gniGeometry->drawingMode() == gnjGeometry->drawingMode()
1887	&& (gniGeometry->drawingMode() != QSGGeometry::DrawLines
1888	|| (gniGeometry->lineWidth() == gnjGeometry->lineWidth()
1889	// Must not do overlap checks when the line width is not 1,
1890	// we have no knowledge how such lines are rasterized.
1891	&& gniGeometry->lineWidth() == 1.0f))
1892	&& gniGeometry->attributes() == gnjGeometry->attributes()
1893	&& gniGeometry->indexType() == gnjGeometry->indexType()
1894	&& gni->inheritedOpacity() == gnj->inheritedOpacity()
1895	&& gniMaterial->type() == gnjMaterial->type()
1896	&& gniMaterial->viewCount() == gnjMaterial->viewCount()
1897	&& gniMaterial->compare(other: gnjMaterial) == 0)
1898	{
1899	if (!overlapBounds.intersects(r: ej->bounds) || !checkOverlap(first: i+1, last: j - 1, bounds: ej->bounds)) {
1900	ej->batch = batch;
1901	next->nextInBatch = ej;
1902	next = ej;
1903	} else {
1904	/* When we come across a compatible element which hits an overlap, we
1905	* need to stop the batch right away. We cannot add more elements
1906	* to the current batch as they will be rendered before the batch that the
1907	* current 'ej' will be added to.
1908	*/
1909	break;
1910	}
1911	} else {
1912	overlapBounds |= ej->bounds;
1913	}
1914	}
1915
1916	batch->lastOrderInBatch = next->order;
1917	}
1918
1919
1920	}
1921
1922	static inline int qsg_fixIndexCount(int iCount, int drawMode)
1923	{
1924	switch (drawMode) {
1925	case QSGGeometry::DrawTriangleStrip:
1926	// Merged triangle strips need to contain degenerate triangles at the beginning and end.
1927	// One could save 2 uploaded ushorts here by ditching the padding for the front of the
1928	// first and the end of the last, but for simplicity, we simply don't care.
1929	// Those extra triangles will be skipped while drawing to preserve the strip's parity
1930	// anyhow.
1931	return iCount + 2;
1932	case QSGGeometry::DrawLines:
1933	// For lines we drop the last vertex if the number of vertices is uneven.
1934	return iCount - (iCount % 2);
1935	case QSGGeometry::DrawTriangles:
1936	// For triangles we drop trailing vertices until the result is divisible by 3.
1937	return iCount - (iCount % 3);
1938	default:
1939	return iCount;
1940	}
1941	}
1942
1943	static inline float calculateElementZOrder(const Element *e, qreal zRange)
1944	{
1945	// Clamp the zOrder to within the min and max depth of the viewport.
1946	return std::clamp(val: 1.0f - float(e->order * zRange), lo: VIEWPORT_MIN_DEPTH, hi: VIEWPORT_MAX_DEPTH);
1947	}
1948
1949	/* These parameters warrant some explanation...
1950	*
1951	* vaOffset: The byte offset into the vertex data to the location of the
1952	* 2D float point vertex attributes.
1953	*
1954	* vertexData: destination where the geometry's vertex data should go
1955	*
1956	* zData: destination of geometries injected Z positioning
1957	*
1958	* indexData: destination of the indices for this element
1959	*
1960	* iBase: The starting index for this element in the batch
1961	*/
1962
1963	void Renderer::uploadMergedElement(Element *e, int vaOffset, char **vertexData, char **zData, char **indexData, void *iBasePtr, int *indexCount)
1964	{
1965	if (Q_UNLIKELY(debug_upload())) qDebug() << " - uploading element:" << e << e->node << (void *) *vertexData << (qintptr) (*zData - *vertexData) << (qintptr) (*indexData - *vertexData);
1966	QSGGeometry *g = e->node->geometry();
1967
1968	const QMatrix4x4 &localx = *e->node->matrix();
1969	const float *localxdata = localx.constData();
1970
1971	const int vCount = g->vertexCount();
1972	const int vSize = g->sizeOfVertex();
1973	memcpy(dest: *vertexData, src: g->vertexData(), n: vSize * vCount);
1974
1975	// apply vertex transform..
1976	char *vdata = *vertexData + vaOffset;
1977	if (localx.flags() == QMatrix4x4::Translation) {
1978	for (int i=0; i<vCount; ++i) {
1979	Pt *p = (Pt *) vdata;
1980	p->x += localxdata[12];
1981	p->y += localxdata[13];
1982	vdata += vSize;
1983	}
1984	} else if (localx.flags() > QMatrix4x4::Translation) {
1985	for (int i=0; i<vCount; ++i) {
1986	((Pt *) vdata)->map(mat: localx);
1987	vdata += vSize;
1988	}
1989	}
1990
1991	if (useDepthBuffer()) {
1992	float *vzorder = (float *) *zData;
1993	float zorder = calculateElementZOrder(e, zRange: m_zRange);
1994	for (int i=0; i<vCount; ++i)
1995	vzorder[i] = zorder;
1996	*zData += vCount * sizeof(float);
1997	}
1998
1999	int iCount = g->indexCount();
2000	if (m_uint32IndexForRhi) {
2001	// can only happen when using the rhi
2002	quint32 *iBase = (quint32 *) iBasePtr;
2003	quint32 *indices = (quint32 *) *indexData;
2004	if (iCount == 0) {
2005	iCount = vCount;
2006	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip)
2007	*indices++ = *iBase;
2008	else
2009	iCount = qsg_fixIndexCount(iCount, drawMode: g->drawingMode());
2010
2011	for (int i=0; i<iCount; ++i)
2012	indices[i] = *iBase + i;
2013	} else {
2014	// source index data in QSGGeometry is always ushort (we would not merge otherwise)
2015	const quint16 *srcIndices = g->indexDataAsUShort();
2016	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip)
2017	*indices++ = *iBase + srcIndices[0];
2018	else
2019	iCount = qsg_fixIndexCount(iCount, drawMode: g->drawingMode());
2020
2021	for (int i=0; i<iCount; ++i)
2022	indices[i] = *iBase + srcIndices[i];
2023	}
2024	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip) {
2025	indices[iCount] = indices[iCount - 1];
2026	iCount += 2;
2027	}
2028	*iBase += vCount;
2029	} else {
2030	// normally batching is only done for ushort index data
2031	quint16 *iBase = (quint16 *) iBasePtr;
2032	quint16 *indices = (quint16 *) *indexData;
2033	if (iCount == 0) {
2034	iCount = vCount;
2035	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip)
2036	*indices++ = *iBase;
2037	else
2038	iCount = qsg_fixIndexCount(iCount, drawMode: g->drawingMode());
2039
2040	for (int i=0; i<iCount; ++i)
2041	indices[i] = *iBase + i;
2042	} else {
2043	const quint16 *srcIndices = g->indexDataAsUShort();
2044	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip)
2045	*indices++ = *iBase + srcIndices[0];
2046	else
2047	iCount = qsg_fixIndexCount(iCount, drawMode: g->drawingMode());
2048
2049	for (int i=0; i<iCount; ++i)
2050	indices[i] = *iBase + srcIndices[i];
2051	}
2052	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip) {
2053	indices[iCount] = indices[iCount - 1];
2054	iCount += 2;
2055	}
2056	*iBase += vCount;
2057	}
2058
2059	*vertexData += vCount * vSize;
2060	*indexData += iCount * mergedIndexElemSize();
2061	*indexCount += iCount;
2062	}
2063
2064	QMatrix4x4 qsg_matrixForRoot(Node *node)
2065	{
2066	if (node->type() == QSGNode::TransformNodeType)
2067	return static_cast<QSGTransformNode *>(node->sgNode)->combinedMatrix();
2068	Q_ASSERT(node->type() == QSGNode::ClipNodeType);
2069	QSGClipNode *c = static_cast<QSGClipNode *>(node->sgNode);
2070	return *c->matrix();
2071	}
2072
2073	void Renderer::uploadBatch(Batch *b)
2074	{
2075	// Early out if nothing has changed in this batch..
2076	if (!b->needsUpload) {
2077	if (Q_UNLIKELY(debug_upload())) qDebug() << " Batch:" << b << "already uploaded...";
2078	return;
2079	}
2080
2081	if (!b->first) {
2082	if (Q_UNLIKELY(debug_upload())) qDebug() << " Batch:" << b << "is invalid...";
2083	return;
2084	}
2085
2086	if (b->isRenderNode) {
2087	if (Q_UNLIKELY(debug_upload())) qDebug() << " Batch: " << b << "is a render node...";
2088	return;
2089	}
2090
2091	// Figure out if we can merge or not, if not, then just render the batch as is..
2092	Q_ASSERT(b->first);
2093	Q_ASSERT(b->first->node);
2094
2095	QSGGeometryNode *gn = b->first->node;
2096	QSGGeometry *g = gn->geometry();
2097	QSGMaterial::Flags flags = gn->activeMaterial()->flags();
2098	bool canMerge = (g->drawingMode() == QSGGeometry::DrawTriangles || g->drawingMode() == QSGGeometry::DrawTriangleStrip ||
2099	g->drawingMode() == QSGGeometry::DrawLines || g->drawingMode() == QSGGeometry::DrawPoints)
2100	&& b->positionAttribute >= 0
2101	&& g->indexType() == QSGGeometry::UnsignedShortType
2102	&& (flags & (QSGMaterial::NoBatching | QSGMaterial_FullMatrix)) == 0
2103	&& ((flags & QSGMaterial::RequiresFullMatrixExceptTranslate) == 0 || b->isTranslateOnlyToRoot())
2104	&& b->isSafeToBatch();
2105
2106	b->merged = canMerge;
2107
2108	// Figure out how much memory we need...
2109	b->vertexCount = 0;
2110	b->indexCount = 0;
2111	int unmergedIndexSize = 0;
2112	Element *e = b->first;
2113
2114	// Merged batches always do indexed draw calls. Non-indexed geometry gets
2115	// indices generated automatically, when merged.
2116	while (e) {
2117	QSGGeometry *eg = e->node->geometry();
2118	b->vertexCount += eg->vertexCount();
2119	int iCount = eg->indexCount();
2120	if (b->merged) {
2121	if (iCount == 0)
2122	iCount = eg->vertexCount();
2123	iCount = qsg_fixIndexCount(iCount, drawMode: g->drawingMode());
2124	} else {
2125	const int effectiveIndexSize = m_uint32IndexForRhi ? sizeof(quint32) : eg->sizeOfIndex();
2126	unmergedIndexSize += iCount * effectiveIndexSize;
2127	}
2128	b->indexCount += iCount;
2129	e = e->nextInBatch;
2130	}
2131
2132	// Abort if there are no vertices in this batch.. We abort this late as
2133	// this is a broken usecase which we do not care to optimize for...
2134	if (b->vertexCount == 0 || (b->merged && b->indexCount == 0))
2135	return;
2136
2137	/* Allocate memory for this batch. Merged batches are divided into three separate blocks
2138	1. Vertex data for all elements, as they were in the QSGGeometry object, but
2139	with the tranform relative to this batch's root applied. The vertex data
2140	is otherwise unmodified.
2141	2. Z data for all elements, derived from each elements "render order".
2142	This is present for merged data only.
2143	3. Indices for all elements, as they were in the QSGGeometry object, but
2144	adjusted so that each index matches its.
2145	And for TRIANGLE_STRIPs, we need to insert degenerate between each
2146	primitive. These are unsigned shorts for merged and arbitrary for
2147	non-merged.
2148	*/
2149	int bufferSize = b->vertexCount * g->sizeOfVertex();
2150	int ibufferSize = 0;
2151	if (b->merged) {
2152	ibufferSize = b->indexCount * mergedIndexElemSize();
2153	if (useDepthBuffer())
2154	bufferSize += b->vertexCount * sizeof(float);
2155	} else {
2156	ibufferSize = unmergedIndexSize;
2157	}
2158
2159	map(buffer: &b->ibo, byteSize: ibufferSize, isIndexBuf: true);
2160	map(buffer: &b->vbo, byteSize: bufferSize);
2161
2162	if (Q_UNLIKELY(debug_upload())) qDebug() << " - batch" << b << " first:" << b->first << " root:"
2163	<< b->root << " merged:" << b->merged << " positionAttribute" << b->positionAttribute
2164	<< " vbo:" << b->vbo.buf << ":" << b->vbo.size;
2165
2166	if (b->merged) {
2167	char *vertexData = b->vbo.data;
2168	char *zData = vertexData + b->vertexCount * g->sizeOfVertex();
2169	char *indexData = b->ibo.data;
2170
2171	quint16 iOffset16 = 0;
2172	quint32 iOffset32 = 0;
2173	e = b->first;
2174	uint verticesInSet = 0;
2175	// Start a new set already after 65534 vertices because 0xFFFF may be
2176	// used for an always-on primitive restart with some apis (adapt for
2177	// uint32 indices as appropriate).
2178	const uint verticesInSetLimit = m_uint32IndexForRhi ? 0xfffffffe : 0xfffe;
2179	int indicesInSet = 0;
2180	b->drawSets.reset();
2181	int drawSetIndices = 0;
2182	const char *indexBase = b->ibo.data;
2183	b->drawSets << DrawSet(0, zData - vertexData, drawSetIndices);
2184	while (e) {
2185	verticesInSet += e->node->geometry()->vertexCount();
2186	if (verticesInSet > verticesInSetLimit) {
2187	b->drawSets.last().indexCount = indicesInSet;
2188	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip) {
2189	b->drawSets.last().indices += 1 * mergedIndexElemSize();
2190	b->drawSets.last().indexCount -= 2;
2191	}
2192	drawSetIndices = indexData - indexBase;
2193	b->drawSets << DrawSet(vertexData - b->vbo.data,
2194	zData - b->vbo.data,
2195	drawSetIndices);
2196	iOffset16 = 0;
2197	iOffset32 = 0;
2198	verticesInSet = e->node->geometry()->vertexCount();
2199	indicesInSet = 0;
2200	}
2201	void *iBasePtr = &iOffset16;
2202	if (m_uint32IndexForRhi)
2203	iBasePtr = &iOffset32;
2204	uploadMergedElement(e, vaOffset: b->positionAttribute, vertexData: &vertexData, zData: &zData, indexData: &indexData, iBasePtr, indexCount: &indicesInSet);
2205	e = e->nextInBatch;
2206	}
2207	b->drawSets.last().indexCount = indicesInSet;
2208	// We skip the very first and very last degenerate triangles since they aren't needed
2209	// and the first one would reverse the vertex ordering of the merged strips.
2210	if (g->drawingMode() == QSGGeometry::DrawTriangleStrip) {
2211	b->drawSets.last().indices += 1 * mergedIndexElemSize();
2212	b->drawSets.last().indexCount -= 2;
2213	}
2214	} else {
2215	char *vboData = b->vbo.data;
2216	char *iboData = b->ibo.data;
2217	Element *e = b->first;
2218	while (e) {
2219	QSGGeometry *g = e->node->geometry();
2220	int vbs = g->vertexCount() * g->sizeOfVertex();
2221	memcpy(dest: vboData, src: g->vertexData(), n: vbs);
2222	vboData = vboData + vbs;
2223	const int indexCount = g->indexCount();
2224	if (indexCount) {
2225	const int effectiveIndexSize = m_uint32IndexForRhi ? sizeof(quint32) : g->sizeOfIndex();
2226	const int ibs = indexCount * effectiveIndexSize;
2227	if (g->sizeOfIndex() == effectiveIndexSize) {
2228	memcpy(dest: iboData, src: g->indexData(), n: ibs);
2229	} else {
2230	if (g->sizeOfIndex() == sizeof(quint16) && effectiveIndexSize == sizeof(quint32)) {
2231	quint16 *src = g->indexDataAsUShort();
2232	quint32 *dst = (quint32 *) iboData;
2233	for (int i = 0; i < indexCount; ++i)
2234	dst[i] = src[i];
2235	} else {
2236	Q_ASSERT_X(false, "uploadBatch (unmerged)", "uint index with ushort effective index - cannot happen");
2237	}
2238	}
2239	iboData += ibs;
2240	}
2241	e = e->nextInBatch;
2242	}
2243	}
2244	#ifndef QT_NO_DEBUG_OUTPUT
2245	if (Q_UNLIKELY(debug_upload())) {
2246	const char *vd = b->vbo.data;
2247	qDebug() << " -- Vertex Data, count:" << b->vertexCount << " - " << g->sizeOfVertex() << "bytes/vertex";
2248	for (int i=0; i<b->vertexCount; ++i) {
2249	QDebug dump = qDebug().nospace();
2250	dump << " --- " << i << ": ";
2251	int offset = 0;
2252	for (int a=0; a<g->attributeCount(); ++a) {
2253	const QSGGeometry::Attribute &attr = g->attributes()[a];
2254	dump << attr.position << ":(" << attr.tupleSize << ",";
2255	if (attr.type == QSGGeometry::FloatType) {
2256	dump << "float ";
2257	if (attr.isVertexCoordinate)
2258	dump << "* ";
2259	for (int t=0; t<attr.tupleSize; ++t)
2260	dump << *(const float *)(vd + offset + t * sizeof(float)) << " ";
2261	} else if (attr.type == QSGGeometry::UnsignedByteType) {
2262	dump << "ubyte ";
2263	for (int t=0; t<attr.tupleSize; ++t)
2264	dump << *(const unsigned char *)(vd + offset + t * sizeof(unsigned char)) << " ";
2265	}
2266	dump << ") ";
2267	offset += attr.tupleSize * size_of_type(type: attr.type);
2268	}
2269	if (b->merged && useDepthBuffer()) {
2270	float zorder = ((float*)(b->vbo.data + b->vertexCount * g->sizeOfVertex()))[i];
2271	dump << " Z:(" << zorder << ")";
2272	}
2273	vd += g->sizeOfVertex();
2274	}
2275
2276	if (!b->drawSets.isEmpty()) {
2277	if (m_uint32IndexForRhi) {
2278	const quint32 *id = (const quint32 *) b->ibo.data;
2279	{
2280	QDebug iDump = qDebug();
2281	iDump << " -- Index Data, count:" << b->indexCount;
2282	for (int i=0; i<b->indexCount; ++i) {
2283	if ((i % 24) == 0)
2284	iDump << Qt::endl << " --- ";
2285	iDump << id[i];
2286	}
2287	}
2288	} else {
2289	const quint16 *id = (const quint16 *) b->ibo.data;
2290	{
2291	QDebug iDump = qDebug();
2292	iDump << " -- Index Data, count:" << b->indexCount;
2293	for (int i=0; i<b->indexCount; ++i) {
2294	if ((i % 24) == 0)
2295	iDump << Qt::endl << " --- ";
2296	iDump << id[i];
2297	}
2298	}
2299	}
2300
2301	for (int i=0; i<b->drawSets.size(); ++i) {
2302	const DrawSet &s = b->drawSets.at(i);
2303	qDebug() << " -- DrawSet: indexCount:" << s.indexCount << " vertices:" << s.vertices << " z:" << s.zorders << " indices:" << s.indices;
2304	}
2305	}
2306	}
2307	#endif // QT_NO_DEBUG_OUTPUT
2308
2309	unmap(buffer: &b->vbo);
2310	unmap(buffer: &b->ibo, isIndexBuf: true);
2311
2312	if (Q_UNLIKELY(debug_upload() || debug_pools()))
2313	qDebug() << " --- vertex/index buffers unmapped, batch upload completed... vbo pool size" << m_vboPoolCost << "ibo pool size" << m_iboPoolCost;
2314
2315	b->needsUpload = false;
2316
2317	if (Q_UNLIKELY(debug_render()))
2318	b->uploadedThisFrame = true;
2319	}
2320
2321	void Renderer::applyClipStateToGraphicsState()
2322	{
2323	m_gstate.usesScissor = (m_currentClipState.type & ClipState::ScissorClip);
2324	m_gstate.stencilTest = (m_currentClipState.type & ClipState::StencilClip);
2325	}
2326
2327	QRhiGraphicsPipeline *Renderer::buildStencilPipeline(const Batch *batch, bool firstStencilClipInBatch)
2328	{
2329	QRhiGraphicsPipeline *ps = m_rhi->newGraphicsPipeline();
2330	ps->setFlags(QRhiGraphicsPipeline::UsesStencilRef);
2331	QRhiGraphicsPipeline::TargetBlend blend;
2332	blend.colorWrite = {};
2333	ps->setTargetBlends({ blend });
2334	ps->setSampleCount(renderTarget().rt->sampleCount());
2335	ps->setStencilTest(true);
2336	QRhiGraphicsPipeline::StencilOpState stencilOp;
2337	if (firstStencilClipInBatch) {
2338	stencilOp.compareOp = QRhiGraphicsPipeline::Always;
2339	stencilOp.failOp = QRhiGraphicsPipeline::Keep;
2340	stencilOp.depthFailOp = QRhiGraphicsPipeline::Keep;
2341	stencilOp.passOp = QRhiGraphicsPipeline::Replace;
2342	} else {
2343	stencilOp.compareOp = QRhiGraphicsPipeline::Equal;
2344	stencilOp.failOp = QRhiGraphicsPipeline::Keep;
2345	stencilOp.depthFailOp = QRhiGraphicsPipeline::Keep;
2346	stencilOp.passOp = QRhiGraphicsPipeline::IncrementAndClamp;
2347	}
2348	ps->setStencilFront(stencilOp);
2349	ps->setStencilBack(stencilOp);
2350
2351	ps->setTopology(m_stencilClipCommon.topology);
2352
2353	ps->setMultiViewCount(renderTarget().multiViewCount);
2354
2355	ps->setShaderStages({ QRhiShaderStage(QRhiShaderStage::Vertex, m_stencilClipCommon.vs),
2356	QRhiShaderStage(QRhiShaderStage::Fragment, m_stencilClipCommon.fs) });
2357	ps->setVertexInputLayout(m_stencilClipCommon.inputLayout);
2358	ps->setShaderResourceBindings(batch->stencilClipState.srb); // use something, it just needs to be layout-compatible
2359	ps->setRenderPassDescriptor(renderTarget().rpDesc);
2360
2361	if (!ps->create()) {
2362	qWarning(msg: "Failed to build stencil clip pipeline");
2363	delete ps;
2364	return nullptr;
2365	}
2366
2367	return ps;
2368	}
2369
2370	void Renderer::updateClipState(const QSGClipNode *clipList, Batch *batch)
2371	{
2372	// Note: No use of the clip-related speparate m_current* vars is allowed
2373	// here. All stored in batch->clipState instead. To collect state during
2374	// the prepare steps, m_currentClipState is used. It should not be used in
2375	// the render steps afterwards.
2376
2377	// The stenciling logic is slightly different from Qt 5's direct OpenGL version
2378	// as we cannot just randomly clear the stencil buffer. We now put all clip
2379	// shapes into the stencil buffer for all batches in the frame. This means
2380	// that the number of total clips in a scene is reduced (since the stencil
2381	// value cannot exceed 255) but we do not need any clears inbetween.
2382
2383	Q_ASSERT(m_rhi);
2384	batch->stencilClipState.updateStencilBuffer = false;
2385	if (clipList == m_currentClipState.clipList || Q_UNLIKELY(debug_noclip())) {
2386	applyClipStateToGraphicsState();
2387	batch->clipState = m_currentClipState;
2388	return;
2389	}
2390
2391	ClipState::ClipType clipType = ClipState::NoClip;
2392	QRect scissorRect;
2393	QVarLengthArray<const QSGClipNode *, 4> stencilClipNodes;
2394	const QSGClipNode *clip = clipList;
2395
2396	batch->stencilClipState.drawCalls.reset();
2397	quint32 totalVSize = 0;
2398	quint32 totalISize = 0;
2399	quint32 totalUSize = 0;
2400	const quint32 StencilClipUbufSize = 64;
2401
2402	while (clip) {
2403	QMatrix4x4 m = m_current_projection_matrix_native_ndc[0]; // never hit for 3D and so multiview
2404	if (clip->matrix())
2405	m *= *clip->matrix();
2406
2407	bool isRectangleWithNoPerspective = clip->isRectangular()
2408	&& qFuzzyIsNull(f: m(3, 0)) && qFuzzyIsNull(f: m(3, 1));
2409	bool noRotate = qFuzzyIsNull(f: m(0, 1)) && qFuzzyIsNull(f: m(1, 0));
2410	bool isRotate90 = qFuzzyIsNull(f: m(0, 0)) && qFuzzyIsNull(f: m(1, 1));
2411
2412	if (isRectangleWithNoPerspective && (noRotate || isRotate90)) {
2413	QRectF bbox = clip->clipRect();
2414	qreal invW = 1 / m(3, 3);
2415	qreal fx1, fy1, fx2, fy2;
2416	if (noRotate) {
2417	fx1 = (bbox.left() * m(0, 0) + m(0, 3)) * invW;
2418	fy1 = (bbox.bottom() * m(1, 1) + m(1, 3)) * invW;
2419	fx2 = (bbox.right() * m(0, 0) + m(0, 3)) * invW;
2420	fy2 = (bbox.top() * m(1, 1) + m(1, 3)) * invW;
2421	} else {
2422	Q_ASSERT(isRotate90);
2423	fx1 = (bbox.bottom() * m(0, 1) + m(0, 3)) * invW;
2424	fy1 = (bbox.left() * m(1, 0) + m(1, 3)) * invW;
2425	fx2 = (bbox.top() * m(0, 1) + m(0, 3)) * invW;
2426	fy2 = (bbox.right() * m(1, 0) + m(1, 3)) * invW;
2427	}
2428
2429	if (fx1 > fx2)
2430	qSwap(value1&: fx1, value2&: fx2);
2431	if (fy1 > fy2)
2432	qSwap(value1&: fy1, value2&: fy2);
2433
2434	QRect deviceRect = this->deviceRect();
2435
2436	qint32 ix1 = qRound(d: (fx1 + 1) * deviceRect.width() * qreal(0.5));
2437	qint32 iy1 = qRound(d: (fy1 + 1) * deviceRect.height() * qreal(0.5));
2438	qint32 ix2 = qRound(d: (fx2 + 1) * deviceRect.width() * qreal(0.5));
2439	qint32 iy2 = qRound(d: (fy2 + 1) * deviceRect.height() * qreal(0.5));
2440
2441	if (!(clipType & ClipState::ScissorClip)) {
2442	clipType |= ClipState::ScissorClip;
2443	scissorRect = QRect(ix1, iy1, ix2 - ix1, iy2 - iy1);
2444	} else {
2445	scissorRect &= QRect(ix1, iy1, ix2 - ix1, iy2 - iy1);
2446	}
2447	} else {
2448	clipType |= ClipState::StencilClip;
2449
2450	const QSGGeometry *g = clip->geometry();
2451	Q_ASSERT(g->attributeCount() > 0);
2452
2453	const int vertexByteSize = g->sizeOfVertex() * g->vertexCount();
2454	// the 4 byte alignment may not actually be needed here
2455	totalVSize = aligned(v: totalVSize, byteAlign: 4u) + vertexByteSize;
2456	if (g->indexCount()) {
2457	const int indexByteSize = g->sizeOfIndex() * g->indexCount();
2458	// so no need to worry about NonFourAlignedEffectiveIndexBufferOffset
2459	totalISize = aligned(v: totalISize, byteAlign: 4u) + indexByteSize;
2460	}
2461	// ubuf start offsets must be aligned (typically to 256 bytes)
2462	totalUSize = aligned(v: totalUSize, byteAlign: m_ubufAlignment) + StencilClipUbufSize;
2463
2464	stencilClipNodes.append(t: clip);
2465	}
2466
2467	clip = clip->clipList();
2468	}
2469
2470	if (clipType & ClipState::StencilClip) {
2471	bool rebuildVBuf = false;
2472	if (!batch->stencilClipState.vbuf) {
2473	batch->stencilClipState.vbuf = m_rhi->newBuffer(type: QRhiBuffer::Dynamic, usage: QRhiBuffer::VertexBuffer, size: totalVSize);
2474	rebuildVBuf = true;
2475	} else if (batch->stencilClipState.vbuf->size() < totalVSize) {
2476	batch->stencilClipState.vbuf->setSize(totalVSize);
2477	rebuildVBuf = true;
2478	}
2479	if (rebuildVBuf) {
2480	if (!batch->stencilClipState.vbuf->create()) {
2481	qWarning(msg: "Failed to build stencil clip vertex buffer");
2482	delete batch->stencilClipState.vbuf;
2483	batch->stencilClipState.vbuf = nullptr;
2484	return;
2485	}
2486	}
2487
2488	if (totalISize) {
2489	bool rebuildIBuf = false;
2490	if (!batch->stencilClipState.ibuf) {
2491	batch->stencilClipState.ibuf = m_rhi->newBuffer(type: QRhiBuffer::Dynamic, usage: QRhiBuffer::IndexBuffer, size: totalISize);
2492	rebuildIBuf = true;
2493	} else if (batch->stencilClipState.ibuf->size() < totalISize) {
2494	batch->stencilClipState.ibuf->setSize(totalISize);
2495	rebuildIBuf = true;
2496	}
2497	if (rebuildIBuf) {
2498	if (!batch->stencilClipState.ibuf->create()) {
2499	qWarning(msg: "Failed to build stencil clip index buffer");
2500	delete batch->stencilClipState.ibuf;
2501	batch->stencilClipState.ibuf = nullptr;
2502	return;
2503	}
2504	}
2505	}
2506
2507	bool rebuildUBuf = false;
2508	if (!batch->stencilClipState.ubuf) {
2509	batch->stencilClipState.ubuf = m_rhi->newBuffer(type: QRhiBuffer::Dynamic, usage: QRhiBuffer::UniformBuffer, size: totalUSize);
2510	rebuildUBuf = true;
2511	} else if (batch->stencilClipState.ubuf->size() < totalUSize) {
2512	batch->stencilClipState.ubuf->setSize(totalUSize);
2513	rebuildUBuf = true;
2514	}
2515	if (rebuildUBuf) {
2516	if (!batch->stencilClipState.ubuf->create()) {
2517	qWarning(msg: "Failed to build stencil clip uniform buffer");
2518	delete batch->stencilClipState.ubuf;
2519	batch->stencilClipState.ubuf = nullptr;
2520	return;
2521	}
2522	}
2523
2524	if (!batch->stencilClipState.srb) {
2525	batch->stencilClipState.srb = m_rhi->newShaderResourceBindings();
2526	const QRhiShaderResourceBinding ubufBinding = QRhiShaderResourceBinding::uniformBufferWithDynamicOffset(
2527	binding: 0, stage: QRhiShaderResourceBinding::VertexStage, buf: batch->stencilClipState.ubuf, size: StencilClipUbufSize);
2528	batch->stencilClipState.srb->setBindings({ ubufBinding });
2529	if (!batch->stencilClipState.srb->create()) {
2530	qWarning(msg: "Failed to build stencil clip srb");
2531	delete batch->stencilClipState.srb;
2532	batch->stencilClipState.srb = nullptr;
2533	return;
2534	}
2535	}
2536
2537	quint32 vOffset = 0;
2538	quint32 iOffset = 0;
2539	quint32 uOffset = 0;
2540	for (const QSGClipNode *clip : stencilClipNodes) {
2541	const QSGGeometry *g = clip->geometry();
2542	const QSGGeometry::Attribute *a = g->attributes();
2543	StencilClipState::StencilDrawCall drawCall;
2544	const bool firstStencilClipInBatch = batch->stencilClipState.drawCalls.isEmpty();
2545
2546	if (firstStencilClipInBatch) {
2547	m_stencilClipCommon.inputLayout.setBindings({ QRhiVertexInputBinding(g->sizeOfVertex()) });
2548	m_stencilClipCommon.inputLayout.setAttributes({ QRhiVertexInputAttribute(0, 0, qsg_vertexInputFormat(a: *a), 0) });
2549	m_stencilClipCommon.topology = qsg_topology(geomDrawMode: g->drawingMode());
2550	}
2551	#ifndef QT_NO_DEBUG
2552	else {
2553	if (qsg_topology(geomDrawMode: g->drawingMode()) != m_stencilClipCommon.topology)
2554	qWarning(msg: "updateClipState: Clip list entries have different primitive topologies, this is not currently supported.");
2555	if (qsg_vertexInputFormat(a: *a) != m_stencilClipCommon.inputLayout.cbeginAttributes()->format())
2556	qWarning(msg: "updateClipState: Clip list entries have different vertex input layouts, this is must not happen.");
2557	}
2558	#endif
2559
2560	drawCall.vbufOffset = aligned(v: vOffset, byteAlign: 4u);
2561	const int vertexByteSize = g->sizeOfVertex() * g->vertexCount();
2562	vOffset = drawCall.vbufOffset + vertexByteSize;
2563
2564	int indexByteSize = 0;
2565	if (g->indexCount()) {
2566	drawCall.ibufOffset = aligned(v: iOffset, byteAlign: 4u);
2567	indexByteSize = g->sizeOfIndex() * g->indexCount();
2568	iOffset = drawCall.ibufOffset + indexByteSize;
2569	}
2570
2571	drawCall.ubufOffset = aligned(v: uOffset, byteAlign: m_ubufAlignment);
2572	uOffset = drawCall.ubufOffset + StencilClipUbufSize;
2573
2574	QMatrix4x4 matrixYUpNDC = m_current_projection_matrix[0];
2575	if (clip->matrix())
2576	matrixYUpNDC *= *clip->matrix();
2577
2578	m_resourceUpdates->updateDynamicBuffer(buf: batch->stencilClipState.ubuf, offset: drawCall.ubufOffset, size: 64, data: matrixYUpNDC.constData());
2579	m_resourceUpdates->updateDynamicBuffer(buf: batch->stencilClipState.vbuf, offset: drawCall.vbufOffset, size: vertexByteSize, data: g->vertexData());
2580	if (indexByteSize)
2581	m_resourceUpdates->updateDynamicBuffer(buf: batch->stencilClipState.ibuf, offset: drawCall.ibufOffset, size: indexByteSize, data: g->indexData());
2582
2583	// stencil ref goes 1, 1, 2, 3, 4, ..., N for the clips in the first batch,
2584	// then N+1, N+1, N+2, N+3, ... for the next batch,
2585	// and so on.
2586	// Note the different stencilOp for the first and the subsequent clips.
2587	drawCall.stencilRef = firstStencilClipInBatch ? m_currentClipState.stencilRef + 1 : m_currentClipState.stencilRef;
2588	m_currentClipState.stencilRef += 1;
2589
2590	drawCall.vertexCount = g->vertexCount();
2591	drawCall.indexCount = g->indexCount();
2592	drawCall.indexFormat = qsg_indexFormat(geometry: g);
2593	batch->stencilClipState.drawCalls.add(t: drawCall);
2594	}
2595
2596	if (!m_stencilClipCommon.vs.isValid())
2597	m_stencilClipCommon.vs = QSGMaterialShaderPrivate::loadShader(filename: QLatin1String(":/qt-project.org/scenegraph/shaders_ng/stencilclip.vert.qsb"));
2598
2599	if (!m_stencilClipCommon.fs.isValid())
2600	m_stencilClipCommon.fs = QSGMaterialShaderPrivate::loadShader(filename: QLatin1String(":/qt-project.org/scenegraph/shaders_ng/stencilclip.frag.qsb"));
2601
2602	if (!m_stencilClipCommon.replacePs)
2603	m_stencilClipCommon.replacePs = buildStencilPipeline(batch, firstStencilClipInBatch: true);
2604
2605	if (!m_stencilClipCommon.incrPs)
2606	m_stencilClipCommon.incrPs = buildStencilPipeline(batch, firstStencilClipInBatch: false);
2607
2608	batch->stencilClipState.updateStencilBuffer = true;
2609	}
2610
2611	m_currentClipState.clipList = clipList;
2612	m_currentClipState.type = clipType;
2613	m_currentClipState.scissor = QRhiScissor(scissorRect.x(), scissorRect.y(),
2614	scissorRect.width(), scissorRect.height());
2615
2616	applyClipStateToGraphicsState();
2617	batch->clipState = m_currentClipState;
2618	}
2619
2620	void Renderer::enqueueStencilDraw(const Batch *batch)
2621	{
2622	// cliptype stencil + updateStencilBuffer==false means the batch uses
2623	// stenciling but relies on the stencil data generated by a previous batch
2624	// (due to the having the same clip node). Do not enqueue draw calls for
2625	// stencil in this case as the stencil buffer is already up-to-date.
2626	if (!batch->stencilClipState.updateStencilBuffer)
2627	return;
2628
2629	QRhiCommandBuffer *cb = renderTarget().cb;
2630	const int count = batch->stencilClipState.drawCalls.size();
2631	for (int i = 0; i < count; ++i) {
2632	const StencilClipState::StencilDrawCall &drawCall(batch->stencilClipState.drawCalls.at(i));
2633	QRhiShaderResourceBindings *srb = batch->stencilClipState.srb;
2634	QRhiCommandBuffer::DynamicOffset ubufOffset(0, drawCall.ubufOffset);
2635	if (i == 0) {
2636	cb->setGraphicsPipeline(m_stencilClipCommon.replacePs);
2637	cb->setViewport(m_pstate.viewport);
2638	} else if (i == 1) {
2639	cb->setGraphicsPipeline(m_stencilClipCommon.incrPs);
2640	cb->setViewport(m_pstate.viewport);
2641	}
2642	// else incrPs is already bound
2643	cb->setShaderResources(srb, dynamicOffsetCount: 1, dynamicOffsets: &ubufOffset);
2644	cb->setStencilRef(drawCall.stencilRef);
2645	const QRhiCommandBuffer::VertexInput vbufBinding(batch->stencilClipState.vbuf, drawCall.vbufOffset);
2646	if (drawCall.indexCount) {
2647	cb->setVertexInput(startBinding: 0, bindingCount: 1, bindings: &vbufBinding,
2648	indexBuf: batch->stencilClipState.ibuf, indexOffset: drawCall.ibufOffset, indexFormat: drawCall.indexFormat);
2649	cb->drawIndexed(indexCount: drawCall.indexCount);
2650	} else {
2651	cb->setVertexInput(startBinding: 0, bindingCount: 1, bindings: &vbufBinding);
2652	cb->draw(vertexCount: drawCall.vertexCount);
2653	}
2654	}
2655	}
2656
2657	void Renderer::setActiveRhiShader(QSGMaterialShader *program, ShaderManager::Shader *shader)
2658	{
2659	Q_ASSERT(m_rhi);
2660	m_currentProgram = program;
2661	m_currentShader = shader;
2662	m_currentMaterial = nullptr;
2663	}
2664
2665	static inline bool needsBlendConstant(QRhiGraphicsPipeline::BlendFactor f)
2666	{
2667	return f == QRhiGraphicsPipeline::ConstantColor
2668	|| f == QRhiGraphicsPipeline::OneMinusConstantColor
2669	|| f == QRhiGraphicsPipeline::ConstantAlpha
2670	|| f == QRhiGraphicsPipeline::OneMinusConstantAlpha;
2671	}
2672
2673	// With QRhi renderBatches() is split to two steps: prepare and render.
2674	//
2675	// Prepare goes through the batches and elements, and set up a graphics
2676	// pipeline, srb, uniform buffer, calculates clipping, based on m_gstate, the
2677	// material (shaders), and the batches. This step does not touch the command
2678	// buffer or renderpass-related state (m_pstate).
2679	//
2680	// The render step then starts a renderpass, and goes through all
2681	// batches/elements again and records setGraphicsPipeline, drawIndexed, etc. on
2682	// the command buffer. The prepare step's accumulated global state like
2683	// m_gstate must not be used here. Rather, all data needed for rendering is
2684	// available from Batch/Element at this stage. Bookkeeping of state in the
2685	// renderpass is done via m_pstate.
2686
2687	bool Renderer::ensurePipelineState(Element *e, const ShaderManager::Shader *sms, bool depthPostPass)
2688	{
2689	// Note the key's == and qHash implementations: the renderpass descriptor
2690	// and srb are tested for compatibility, not pointer equality.
2691	//
2692	// We do not store the srb pointer itself because the ownership stays with
2693	// the Element and that can go away more often that we would like it
2694	// to. (think scrolling a list view, constantly dropping and creating new
2695	// nodes) Rather, use an opaque blob of a few uints and store and compare
2696	// that. This works because once the pipeline is built, we will always call
2697	// setShaderResources with an explicitly specified srb which is fine even if
2698	// e->srb we used here to bake the pipeline is already gone by that point.
2699	//
2700	// A typical QSGMaterial's serialized srb layout is 8 uints. (uniform buffer
2701	// + texture, 4 fields each) Regardless, using an implicitly shared
2702	// container is essential here. (won't detach so no more allocs and copies
2703	// are done, unless the Element decides to rebake the srb with a different
2704	// layout - but then the detach is exactly what we need)
2705	//
2706	// Same story for the renderpass descriptor: the object can go away but
2707	// that's fine because that has no effect on an already built pipeline, and
2708	// for comparison we only rely on the serialized blob in order decide if the
2709	// render target is compatible with the pipeline.
2710
2711	const GraphicsPipelineStateKey k = GraphicsPipelineStateKey::create(state: m_gstate, sms, rpDesc: renderTarget().rpDesc, srb: e->srb);
2712
2713	// Note: dynamic state (viewport rect, scissor rect, stencil ref, blend
2714	// constant) is never a part of GraphicsState/QRhiGraphicsPipeline.
2715
2716	// See if there is an existing, matching pipeline state object.
2717	auto it = m_shaderManager->pipelineCache.constFind(key: k);
2718	if (it != m_shaderManager->pipelineCache.constEnd()) {
2719	if (depthPostPass)
2720	e->depthPostPassPs = *it;
2721	else
2722	e->ps = *it;
2723	return true;
2724	}
2725
2726	// Build a new one. This is potentially expensive.
2727	QRhiGraphicsPipeline *ps = m_rhi->newGraphicsPipeline();
2728	ps->setShaderStages(first: sms->stages.cbegin(), last: sms->stages.cend());
2729	ps->setVertexInputLayout(sms->inputLayout);
2730	ps->setShaderResourceBindings(e->srb);
2731	ps->setRenderPassDescriptor(renderTarget().rpDesc);
2732
2733	QRhiGraphicsPipeline::Flags flags;
2734	if (needsBlendConstant(f: m_gstate.srcColor) || needsBlendConstant(f: m_gstate.dstColor)
2735	|| needsBlendConstant(f: m_gstate.srcAlpha) || needsBlendConstant(f: m_gstate.dstAlpha))
2736	{
2737	flags |= QRhiGraphicsPipeline::UsesBlendConstants;
2738	}
2739	if (m_gstate.usesScissor)
2740	flags |= QRhiGraphicsPipeline::UsesScissor;
2741	if (m_gstate.stencilTest)
2742	flags |= QRhiGraphicsPipeline::UsesStencilRef;
2743
2744	ps->setFlags(flags);
2745	ps->setTopology(qsg_topology(geomDrawMode: m_gstate.drawMode));
2746	ps->setCullMode(m_gstate.cullMode);
2747	ps->setPolygonMode(m_gstate.polygonMode);
2748	ps->setMultiViewCount(m_gstate.multiViewCount);
2749
2750	QRhiGraphicsPipeline::TargetBlend blend;
2751	blend.colorWrite = m_gstate.colorWrite;
2752	blend.enable = m_gstate.blending;
2753	blend.srcColor = m_gstate.srcColor;
2754	blend.dstColor = m_gstate.dstColor;
2755	blend.srcAlpha = m_gstate.srcAlpha;
2756	blend.dstAlpha = m_gstate.dstAlpha;
2757	blend.opColor = m_gstate.opColor;
2758	blend.opAlpha = m_gstate.opAlpha;
2759	ps->setTargetBlends({ blend });
2760
2761	ps->setDepthTest(m_gstate.depthTest);
2762	ps->setDepthWrite(m_gstate.depthWrite);
2763	ps->setDepthOp(m_gstate.depthFunc);
2764
2765	if (m_gstate.stencilTest) {
2766	ps->setStencilTest(true);
2767	QRhiGraphicsPipeline::StencilOpState stencilOp;
2768	stencilOp.compareOp = QRhiGraphicsPipeline::Equal;
2769	stencilOp.failOp = QRhiGraphicsPipeline::Keep;
2770	stencilOp.depthFailOp = QRhiGraphicsPipeline::Keep;
2771	stencilOp.passOp = QRhiGraphicsPipeline::Keep;
2772	ps->setStencilFront(stencilOp);
2773	ps->setStencilBack(stencilOp);
2774	}
2775
2776	ps->setSampleCount(m_gstate.sampleCount);
2777
2778	ps->setLineWidth(m_gstate.lineWidth);
2779
2780	if (!ps->create()) {
2781	qWarning(msg: "Failed to build graphics pipeline state");
2782	delete ps;
2783	return false;
2784	}
2785
2786	m_shaderManager->pipelineCache.insert(key: k, value: ps);
2787	if (depthPostPass)
2788	e->depthPostPassPs = ps;
2789	else
2790	e->ps = ps;
2791	return true;
2792	}
2793
2794	static QRhiSampler *newSampler(QRhi *rhi, const QSGSamplerDescription &desc)
2795	{
2796	QRhiSampler::Filter magFilter;
2797	QRhiSampler::Filter minFilter;
2798	QRhiSampler::Filter mipmapMode;
2799	QRhiSampler::AddressMode u;
2800	QRhiSampler::AddressMode v;
2801
2802	switch (desc.filtering) {
2803	case QSGTexture::None:
2804	Q_FALLTHROUGH();
2805	case QSGTexture::Nearest:
2806	magFilter = minFilter = QRhiSampler::Nearest;
2807	break;
2808	case QSGTexture::Linear:
2809	magFilter = minFilter = QRhiSampler::Linear;
2810	break;
2811	default:
2812	Q_UNREACHABLE();
2813	magFilter = minFilter = QRhiSampler::Nearest;
2814	break;
2815	}
2816
2817	switch (desc.mipmapFiltering) {
2818	case QSGTexture::None:
2819	mipmapMode = QRhiSampler::None;
2820	break;
2821	case QSGTexture::Nearest:
2822	mipmapMode = QRhiSampler::Nearest;
2823	break;
2824	case QSGTexture::Linear:
2825	mipmapMode = QRhiSampler::Linear;
2826	break;
2827	default:
2828	Q_UNREACHABLE();
2829	mipmapMode = QRhiSampler::None;
2830	break;
2831	}
2832
2833	switch (desc.horizontalWrap) {
2834	case QSGTexture::Repeat:
2835	u = QRhiSampler::Repeat;
2836	break;
2837	case QSGTexture::ClampToEdge:
2838	u = QRhiSampler::ClampToEdge;
2839	break;
2840	case QSGTexture::MirroredRepeat:
2841	u = QRhiSampler::Mirror;
2842	break;
2843	default:
2844	Q_UNREACHABLE();
2845	u = QRhiSampler::ClampToEdge;
2846	break;
2847	}
2848
2849	switch (desc.verticalWrap) {
2850	case QSGTexture::Repeat:
2851	v = QRhiSampler::Repeat;
2852	break;
2853	case QSGTexture::ClampToEdge:
2854	v = QRhiSampler::ClampToEdge;
2855	break;
2856	case QSGTexture::MirroredRepeat:
2857	v = QRhiSampler::Mirror;
2858	break;
2859	default:
2860	Q_UNREACHABLE();
2861	v = QRhiSampler::ClampToEdge;
2862	break;
2863	}
2864
2865	return rhi->newSampler(magFilter, minFilter, mipmapMode, addressU: u, addressV: v);
2866	}
2867
2868	QRhiTexture *Renderer::dummyTexture()
2869	{
2870	if (!m_dummyTexture) {
2871	m_dummyTexture = m_rhi->newTexture(format: QRhiTexture::RGBA8, pixelSize: QSize(64, 64));
2872	if (m_dummyTexture->create()) {
2873	if (m_resourceUpdates) {
2874	QImage img(m_dummyTexture->pixelSize(), QImage::Format_RGBA8888_Premultiplied);
2875	img.fill(pixel: 0);
2876	m_resourceUpdates->uploadTexture(tex: m_dummyTexture, image: img);
2877	}
2878	}
2879	}
2880	return m_dummyTexture;
2881	}
2882
2883	static void rendererToMaterialGraphicsState(QSGMaterialShader::GraphicsPipelineState *dst,
2884	GraphicsState *src)
2885	{
2886	dst->blendEnable = src->blending;
2887
2888	// the enum values should match, sanity check it
2889	Q_ASSERT(int(QSGMaterialShader::GraphicsPipelineState::OneMinusSrc1Alpha) == int(QRhiGraphicsPipeline::OneMinusSrc1Alpha));
2890	Q_ASSERT(int(QSGMaterialShader::GraphicsPipelineState::BlendOp::Max) == int(QRhiGraphicsPipeline::Max));
2891	Q_ASSERT(int(QSGMaterialShader::GraphicsPipelineState::A) == int(QRhiGraphicsPipeline::A));
2892	Q_ASSERT(int(QSGMaterialShader::GraphicsPipelineState::CullBack) == int(QRhiGraphicsPipeline::Back));
2893	Q_ASSERT(int(QSGMaterialShader::GraphicsPipelineState::Line) == int(QRhiGraphicsPipeline::Line));
2894	dst->srcColor = QSGMaterialShader::GraphicsPipelineState::BlendFactor(src->srcColor);
2895	dst->dstColor = QSGMaterialShader::GraphicsPipelineState::BlendFactor(src->dstColor);
2896
2897	// For compatibility with any existing code, separateBlendFactors defaults
2898	// to _false_ which means that materials that do not touch srcAlpha and
2899	// dstAlpha will continue to use srcColor and dstColor as the alpha
2900	// blending factors. New code that needs different values for color/alpha,
2901	// can explicitly set separateBlendFactors to true and then set srcAlpha
2902	// and dstAlpha as well.
2903	dst->separateBlendFactors = false;
2904
2905	dst->srcAlpha = QSGMaterialShader::GraphicsPipelineState::BlendFactor(src->srcAlpha);
2906	dst->dstAlpha = QSGMaterialShader::GraphicsPipelineState::BlendFactor(src->dstAlpha);
2907
2908	dst->opColor = QSGMaterialShader::GraphicsPipelineState::BlendOp(src->opColor);
2909	dst->opAlpha = QSGMaterialShader::GraphicsPipelineState::BlendOp(src->opAlpha);
2910
2911	dst->colorWrite = QSGMaterialShader::GraphicsPipelineState::ColorMask(int(src->colorWrite));
2912
2913	dst->cullMode = QSGMaterialShader::GraphicsPipelineState::CullMode(src->cullMode);
2914	dst->polygonMode = QSGMaterialShader::GraphicsPipelineState::PolygonMode(src->polygonMode);
2915	}
2916
2917	static void materialToRendererGraphicsState(GraphicsState *dst,
2918	QSGMaterialShader::GraphicsPipelineState *src)
2919	{
2920	dst->blending = src->blendEnable;
2921	dst->srcColor = QRhiGraphicsPipeline::BlendFactor(src->srcColor);
2922	dst->dstColor = QRhiGraphicsPipeline::BlendFactor(src->dstColor);
2923	if (src->separateBlendFactors) {
2924	dst->srcAlpha = QRhiGraphicsPipeline::BlendFactor(src->srcAlpha);
2925	dst->dstAlpha = QRhiGraphicsPipeline::BlendFactor(src->dstAlpha);
2926	} else {
2927	dst->srcAlpha = dst->srcColor;
2928	dst->dstAlpha = dst->dstColor;
2929	}
2930	dst->opColor = QRhiGraphicsPipeline::BlendOp(src->opColor);
2931	dst->opAlpha = QRhiGraphicsPipeline::BlendOp(src->opAlpha);
2932	dst->colorWrite = QRhiGraphicsPipeline::ColorMask(int(src->colorWrite));
2933	dst->cullMode = QRhiGraphicsPipeline::CullMode(src->cullMode);
2934	dst->polygonMode = QRhiGraphicsPipeline::PolygonMode(src->polygonMode);
2935	}
2936
2937	void Renderer::updateMaterialDynamicData(ShaderManager::Shader *sms,
2938	QSGMaterialShader::RenderState &renderState,
2939	QSGMaterial *material,
2940	const Batch *batch,
2941	Element *e,
2942	int ubufOffset,
2943	int ubufRegionSize,
2944	char *directUpdatePtr)
2945	{
2946	m_current_resource_update_batch = m_resourceUpdates;
2947
2948	QSGMaterialShader *shader = sms->materialShader;
2949	QSGMaterialShaderPrivate *pd = QSGMaterialShaderPrivate::get(s: shader);
2950	QVarLengthArray<QRhiShaderResourceBinding, 8> bindings;
2951
2952	if (pd->ubufBinding >= 0) {
2953	m_current_uniform_data = &pd->masterUniformData;
2954	const bool changed = shader->updateUniformData(state&: renderState, newMaterial: material, oldMaterial: m_currentMaterial);
2955	m_current_uniform_data = nullptr;
2956
2957	if (changed || !batch->ubufDataValid) {
2958	if (directUpdatePtr)
2959	memcpy(dest: directUpdatePtr + ubufOffset, src: pd->masterUniformData.constData(), n: ubufRegionSize);
2960	else
2961	m_resourceUpdates->updateDynamicBuffer(buf: batch->ubuf, offset: ubufOffset, size: ubufRegionSize, data: pd->masterUniformData.constData());
2962	}
2963
2964	bindings.append(t: QRhiShaderResourceBinding::uniformBuffer(binding: pd->ubufBinding,
2965	stage: pd->ubufStages,
2966	buf: batch->ubuf,
2967	offset: ubufOffset,
2968	size: ubufRegionSize));
2969	}
2970
2971	for (int binding = 0; binding < QSGMaterialShaderPrivate::MAX_SHADER_RESOURCE_BINDINGS; ++binding) {
2972	const QRhiShaderResourceBinding::StageFlags stages = pd->combinedImageSamplerBindings[binding];
2973	if (!stages)
2974	continue;
2975
2976	const QVarLengthArray<QSGTexture *, 4> &prevTex(pd->textureBindingTable[binding]);
2977	QVarLengthArray<QSGTexture *, 4> nextTex = prevTex;
2978
2979	const int count = pd->combinedImageSamplerCount[binding];
2980	nextTex.resize(sz: count);
2981
2982	shader->updateSampledImage(state&: renderState, binding, texture: nextTex.data(), newMaterial: material,
2983	oldMaterial: m_currentMaterial);
2984
2985	if (nextTex.contains(t: nullptr)) {
2986	qWarning(msg: "No QSGTexture provided from updateSampledImage(). This is wrong.");
2987	continue;
2988	}
2989
2990	bool hasDirtySamplerOptions = false;
2991	bool isAnisotropic = false;
2992	for (QSGTexture *t : nextTex) {
2993	QSGTexturePrivate *td = QSGTexturePrivate::get(t);
2994	hasDirtySamplerOptions |= td->hasDirtySamplerOptions();
2995	isAnisotropic |= t->anisotropyLevel() != QSGTexture::AnisotropyNone;
2996	td->resetDirtySamplerOptions();
2997	}
2998
2999	// prevTex may be invalid at this point, avoid dereferencing it
3000	if (nextTex != prevTex || hasDirtySamplerOptions) {
3001
3002	// The QSGTexture, and so the sampler parameters, may have changed.
3003	// The rhiTexture is not relevant here.
3004	pd->textureBindingTable[binding] = nextTex; // does not own
3005	pd->samplerBindingTable[binding].clear();
3006
3007	if (isAnisotropic) // ###
3008	qWarning(msg: "QSGTexture anisotropy levels are not currently supported");
3009
3010	QVarLengthArray<QRhiSampler *, 4> samplers;
3011
3012	for (QSGTexture *t : nextTex) {
3013	const QSGSamplerDescription samplerDesc = QSGSamplerDescription::fromTexture(t);
3014
3015	QRhiSampler *sampler = m_samplers[samplerDesc];
3016
3017	if (!sampler) {
3018	sampler = newSampler(rhi: m_rhi, desc: samplerDesc);
3019	if (!sampler->create()) {
3020	qWarning(msg: "Failed to build sampler");
3021	delete sampler;
3022	continue;
3023	}
3024	m_samplers[samplerDesc] = sampler;
3025	}
3026	samplers.append(t: sampler);
3027	}
3028
3029	pd->samplerBindingTable[binding] = samplers; // does not own
3030	}
3031
3032	if (pd->textureBindingTable[binding].size() == pd->samplerBindingTable[binding].size()) {
3033
3034	QVarLengthArray<QRhiShaderResourceBinding::TextureAndSampler, 4> textureSamplers;
3035
3036	for (int i = 0; i < pd->textureBindingTable[binding].size(); ++i) {
3037
3038	QRhiTexture *texture = pd->textureBindingTable[binding].at(idx: i)->rhiTexture();
3039
3040	// texture may be null if the update above failed for any reason,
3041	// or if the QSGTexture chose to return null intentionally. This is
3042	// valid and we still need to provide something to the shader.
3043	if (!texture)
3044	texture = dummyTexture();
3045
3046	QRhiSampler *sampler = pd->samplerBindingTable[binding].at(idx: i);
3047
3048	textureSamplers.append(
3049	t: QRhiShaderResourceBinding::TextureAndSampler { .tex: texture, .sampler: sampler });
3050	}
3051
3052	if (!textureSamplers.isEmpty())
3053	bindings.append(t: QRhiShaderResourceBinding::sampledTextures(
3054	binding, stage: stages, count, texSamplers: textureSamplers.constData()));
3055	}
3056	}
3057
3058	#ifndef QT_NO_DEBUG
3059	if (bindings.isEmpty())
3060	qWarning(msg: "No shader resources for material %p, this is odd.", material);
3061	#endif
3062
3063	enum class SrbAction {
3064	Unknown,
3065	DoNothing,
3066	UpdateResources,
3067	Rebake
3068	} srbAction = SrbAction::Unknown;
3069
3070	// First, if the Element has no srb created at all, then try to find an existing,
3071	// currently unused srb that is layout-compatible with our binding list.
3072	if (!e->srb) {
3073	// reuse a QVector as our work area, thus possibly reusing the underlying allocation too
3074	QVector<quint32> &layoutDesc(m_shaderManager->srbLayoutDescSerializeWorkspace);
3075	layoutDesc.clear();
3076	QRhiShaderResourceBinding::serializeLayoutDescription(first: bindings.cbegin(), last: bindings.cend(), dst: std::back_inserter(x&: layoutDesc));
3077	e->srb = m_shaderManager->srbPool.take(key: layoutDesc);
3078	if (e->srb) {
3079	// Here we know layout compatibility is satisfied, but do not spend time on full
3080	// comparison. The chance of getting an srb that refers to the same resources
3081	// (buffer, textures) is low in practice. So reuse, but write new resources.
3082	srbAction = SrbAction::UpdateResources;
3083	}
3084	}
3085
3086	// If the Element had an existing srb, investigate:
3087	// - It may be used as-is (when nothing changed in the scene regarding this node compared to the previous frame).
3088	// - Otherwise it may be able to go with a lightweight update (replace resources, binding list layout is the same).
3089	// - If all else fails rebake the full thing, meaning we reuse the memory allocation but will recreate everything underneath.
3090	if (srbAction == SrbAction::Unknown && e->srb) {
3091	if (std::equal(first1: e->srb->cbeginBindings(), last1: e->srb->cendBindings(), first2: bindings.cbegin(), last2: bindings.cend())) {
3092	srbAction = SrbAction::DoNothing;
3093	} else if (std::equal(first1: e->srb->cbeginBindings(), last1: e->srb->cendBindings(), first2: bindings.cbegin(), last2: bindings.cend(),
3094	binary_pred: [](const auto &a, const auto &b) { return a.isLayoutCompatible(b); }))
3095	{
3096	srbAction = SrbAction::UpdateResources;
3097	} else {
3098	srbAction = SrbAction::Rebake;
3099	}
3100	}
3101
3102	// If the Element had no srb associated at all and could not find a layout-compatible
3103	// one from the pool, then create a whole new object.
3104	if (!e->srb) {
3105	e->srb = m_rhi->newShaderResourceBindings();
3106	srbAction = SrbAction::Rebake;
3107	}
3108
3109	Q_ASSERT(srbAction != SrbAction::Unknown && e->srb);
3110
3111	switch (srbAction) {
3112	case SrbAction::DoNothing:
3113	break;
3114	case SrbAction::UpdateResources:
3115	{
3116	e->srb->setBindings(first: bindings.cbegin(), last: bindings.cend());
3117	QRhiShaderResourceBindings::UpdateFlags flags;
3118	// Due to the way the binding list is built up above, if we have a uniform buffer
3119	// at binding point 0 (or none at all) then the sampledTexture bindings are added
3120	// with increasing binding points afterwards, so the list is already sorted based
3121	// on the binding points, thus we can save some time by telling the QRhi backend
3122	// not to sort again.
3123	if (pd->ubufBinding <= 0 || bindings.size() <= 1)
3124	flags |= QRhiShaderResourceBindings::BindingsAreSorted;
3125
3126	e->srb->updateResources(flags);
3127	}
3128	break;
3129	case SrbAction::Rebake:
3130	e->srb->setBindings(first: bindings.cbegin(), last: bindings.cend());
3131	if (!e->srb->create())
3132	qWarning(msg: "Failed to build srb");
3133	break;
3134	default:
3135	Q_ASSERT_X(false, "updateMaterialDynamicData", "No srb action set, this cannot happen");
3136	}
3137	}
3138
3139	void Renderer::updateMaterialStaticData(ShaderManager::Shader *sms,
3140	QSGMaterialShader::RenderState &renderState,
3141	QSGMaterial *material,
3142	Batch *batch,
3143	bool *gstateChanged)
3144	{
3145	QSGMaterialShader *shader = sms->materialShader;
3146	*gstateChanged = false;
3147	if (shader->flags().testFlag(flag: QSGMaterialShader::UpdatesGraphicsPipelineState)) {
3148	// generate the public mini-state from m_gstate, invoke the material,
3149	// write the changes, if any, back to m_gstate, together with a way to
3150	// roll those back.
3151	QSGMaterialShader::GraphicsPipelineState shaderPs;
3152	rendererToMaterialGraphicsState(dst: &shaderPs, src: &m_gstate);
3153	const bool changed = shader->updateGraphicsPipelineState(state&: renderState, ps: &shaderPs, newMaterial: material, oldMaterial: m_currentMaterial);
3154	if (changed) {
3155	m_gstateStack.push(t: m_gstate);
3156	materialToRendererGraphicsState(dst: &m_gstate, src: &shaderPs);
3157	if (needsBlendConstant(f: m_gstate.srcColor) || needsBlendConstant(f: m_gstate.dstColor)
3158	|| needsBlendConstant(f: m_gstate.srcAlpha) || needsBlendConstant(f: m_gstate.dstAlpha))
3159	{
3160	batch->blendConstant = shaderPs.blendConstant;
3161	}
3162	*gstateChanged = true;
3163	}
3164	}
3165	}
3166
3167	bool Renderer::prepareRenderMergedBatch(Batch *batch, PreparedRenderBatch *renderBatch)
3168	{
3169	if (batch->vertexCount == 0 || batch->indexCount == 0)
3170	return false;
3171
3172	Element *e = batch->first;
3173	Q_ASSERT(e);
3174
3175	#ifndef QT_NO_DEBUG_OUTPUT
3176	if (Q_UNLIKELY(debug_render())) {
3177	QDebug debug = qDebug();
3178	debug << " -"
3179	<< batch
3180	<< (batch->uploadedThisFrame ? "[ upload]" : "[retained]")
3181	<< (e->node->clipList() ? "[ clip]" : "[noclip]")
3182	<< (batch->isOpaque ? "[opaque]" : "[ alpha]")
3183	<< "[ merged]"
3184	<< " Nodes:" << QString::fromLatin1(ba: "%1").arg(a: qsg_countNodesInBatch(batch), fieldWidth: 4).toLatin1().constData()
3185	<< " Vertices:" << QString::fromLatin1(ba: "%1").arg(a: batch->vertexCount, fieldWidth: 5).toLatin1().constData()
3186	<< " Indices:" << QString::fromLatin1(ba: "%1").arg(a: batch->indexCount, fieldWidth: 5).toLatin1().constData()
3187	<< " root:" << batch->root;
3188	if (batch->drawSets.size() > 1)
3189	debug << "sets:" << batch->drawSets.size();
3190	if (!batch->isOpaque)
3191	debug << "opacity:" << e->node->inheritedOpacity();
3192	batch->uploadedThisFrame = false;
3193	}
3194	#endif
3195
3196	QSGGeometryNode *gn = e->node;
3197
3198	// We always have dirty matrix as all batches are at a unique z range.
3199	QSGMaterialShader::RenderState::DirtyStates dirty = QSGMaterialShader::RenderState::DirtyMatrix;
3200	if (batch->root)
3201	m_current_model_view_matrix = qsg_matrixForRoot(node: batch->root);
3202	else
3203	m_current_model_view_matrix.setToIdentity();
3204	m_current_determinant = m_current_model_view_matrix.determinant();
3205
3206	const int viewCount = projectionMatrixCount();
3207	m_current_projection_matrix.resize(sz: viewCount);
3208	for (int viewIndex = 0; viewIndex < viewCount; ++viewIndex)
3209	m_current_projection_matrix[viewIndex] = projectionMatrix(index: viewIndex);
3210
3211	m_current_projection_matrix_native_ndc.resize(sz: projectionMatrixWithNativeNDCCount());
3212	for (int viewIndex = 0; viewIndex < projectionMatrixWithNativeNDCCount(); ++viewIndex)
3213	m_current_projection_matrix_native_ndc[viewIndex] = projectionMatrixWithNativeNDC(index: viewIndex);
3214
3215	QSGMaterial *material = gn->activeMaterial();
3216	if (m_renderMode != QSGRendererInterface::RenderMode3D)
3217	updateClipState(clipList: gn->clipList(), batch);
3218
3219	const QSGGeometry *g = gn->geometry();
3220	const int multiViewCount = renderTarget().multiViewCount;
3221	ShaderManager::Shader *sms = useDepthBuffer() ? m_shaderManager->prepareMaterial(material, geometry: g, renderMode: m_renderMode, multiViewCount)
3222	: m_shaderManager->prepareMaterialNoRewrite(material, geometry: g, renderMode: m_renderMode, multiViewCount);
3223	if (!sms)
3224	return false;
3225
3226	Q_ASSERT(sms->materialShader);
3227	if (m_currentShader != sms)
3228	setActiveRhiShader(program: sms->materialShader, shader: sms);
3229
3230	m_current_opacity = gn->inheritedOpacity();
3231	if (!qFuzzyCompare(p1: sms->lastOpacity, p2: float(m_current_opacity))) {
3232	dirty |= QSGMaterialShader::RenderState::DirtyOpacity;
3233	sms->lastOpacity = m_current_opacity;
3234	}
3235
3236	QSGMaterialShaderPrivate *pd = QSGMaterialShaderPrivate::get(s: sms->materialShader);
3237	const quint32 ubufSize = quint32(pd->masterUniformData.size());
3238	if (pd->ubufBinding >= 0) {
3239	bool ubufRebuild = false;
3240	if (!batch->ubuf) {
3241	batch->ubuf = m_rhi->newBuffer(type: QRhiBuffer::Dynamic, usage: QRhiBuffer::UniformBuffer, size: ubufSize);
3242	ubufRebuild = true;
3243	} else {
3244	if (batch->ubuf->size() < ubufSize) {
3245	batch->ubuf->setSize(ubufSize);
3246	ubufRebuild = true;
3247	}
3248	}
3249	if (ubufRebuild) {
3250	batch->ubufDataValid = false;
3251	if (!batch->ubuf->create()) {
3252	qWarning(msg: "Failed to build uniform buffer of size %u bytes", ubufSize);
3253	delete batch->ubuf;
3254	batch->ubuf = nullptr;
3255	return false;
3256	}
3257	}
3258	}
3259
3260	QSGMaterialShader::RenderState renderState = state(dirty: QSGMaterialShader::RenderState::DirtyStates(int(dirty)));
3261
3262	bool pendingGStatePop = false;
3263	updateMaterialStaticData(sms, renderState, material, batch, gstateChanged: &pendingGStatePop);
3264
3265	char *directUpdatePtr = nullptr;
3266	if (batch->ubuf->nativeBuffer().slotCount == 0)
3267	directUpdatePtr = batch->ubuf->beginFullDynamicBufferUpdateForCurrentFrame();
3268
3269	updateMaterialDynamicData(sms, renderState, material, batch, e, ubufOffset: 0, ubufRegionSize: ubufSize, directUpdatePtr);
3270
3271	if (directUpdatePtr)
3272	batch->ubuf->endFullDynamicBufferUpdateForCurrentFrame();
3273
3274	#ifndef QT_NO_DEBUG
3275	if (qsg_test_and_clear_material_failure()) {
3276	qDebug(msg: "QSGMaterial::updateState triggered an error (merged), batch will be skipped:");
3277	Element *ee = e;
3278	while (ee) {
3279	qDebug() << " -" << ee->node;
3280	ee = ee->nextInBatch;
3281	}
3282	QSGNodeDumper::dump(n: rootNode());
3283	qFatal(msg: "Aborting: scene graph is invalid...");
3284	}
3285	#endif
3286
3287	m_gstate.drawMode = QSGGeometry::DrawingMode(g->drawingMode());
3288	m_gstate.lineWidth = g->lineWidth();
3289
3290	const bool hasPipeline = ensurePipelineState(e, sms);
3291
3292	if (pendingGStatePop)
3293	m_gstate = m_gstateStack.pop();
3294
3295	if (!hasPipeline)
3296	return false;
3297
3298	if (m_renderMode == QSGRendererInterface::RenderMode3D) {
3299	m_gstateStack.push(t: m_gstate);
3300	setStateForDepthPostPass();
3301	ensurePipelineState(e, sms, depthPostPass: true);
3302	m_gstate = m_gstateStack.pop();
3303	}
3304
3305	batch->ubufDataValid = true;
3306
3307	m_currentMaterial = material;
3308
3309	renderBatch->batch = batch;
3310	renderBatch->sms = sms;
3311
3312	return true;
3313	}
3314
3315	void Renderer::checkLineWidth(QSGGeometry *g)
3316	{
3317	if (g->drawingMode() == QSGGeometry::DrawLines || g->drawingMode() == QSGGeometry::DrawLineLoop
3318	|| g->drawingMode() == QSGGeometry::DrawLineStrip)
3319	{
3320	if (g->lineWidth() != 1.0f) {
3321	static bool checkedWideLineSupport = false;
3322	if (!checkedWideLineSupport) {
3323	checkedWideLineSupport = true;
3324	if (!m_rhi->isFeatureSupported(feature: QRhi::WideLines))
3325	qWarning(msg: "Line widths other than 1 are not supported by the graphics API");
3326	}
3327	}
3328	} else if (g->drawingMode() == QSGGeometry::DrawPoints) {
3329	if (g->lineWidth() != 1.0f) {
3330	static bool warnedPointSize = false;
3331	if (!warnedPointSize) {
3332	warnedPointSize = true;
3333	qWarning(msg: "Point size is not controllable by QSGGeometry. "
3334	"Set gl_PointSize from the vertex shader instead.");
3335	}
3336	}
3337	}
3338	}
3339
3340	void Renderer::renderMergedBatch(PreparedRenderBatch *renderBatch, bool depthPostPass)
3341	{
3342	const Batch *batch = renderBatch->batch;
3343	if (!batch->vbo.buf || !batch->ibo.buf)
3344	return;
3345
3346	Element *e = batch->first;
3347	QSGGeometryNode *gn = e->node;
3348	QSGGeometry *g = gn->geometry();
3349	checkLineWidth(g);
3350
3351	if (batch->clipState.type & ClipState::StencilClip)
3352	enqueueStencilDraw(batch);
3353
3354	QRhiCommandBuffer *cb = renderTarget().cb;
3355	setGraphicsPipeline(cb, batch, e, depthPostPass);
3356
3357	for (int i = 0, ie = batch->drawSets.size(); i != ie; ++i) {
3358	const DrawSet &draw = batch->drawSets.at(i);
3359	const QRhiCommandBuffer::VertexInput vbufBindings[] = {
3360	{ batch->vbo.buf, quint32(draw.vertices) },
3361	{ batch->vbo.buf, quint32(draw.zorders) }
3362	};
3363	cb->setVertexInput(startBinding: VERTEX_BUFFER_BINDING, bindingCount: useDepthBuffer() ? 2 : 1, bindings: vbufBindings,
3364	indexBuf: batch->ibo.buf, indexOffset: draw.indices,
3365	indexFormat: m_uint32IndexForRhi ? QRhiCommandBuffer::IndexUInt32 : QRhiCommandBuffer::IndexUInt16);
3366	cb->drawIndexed(indexCount: draw.indexCount);
3367	}
3368	}
3369
3370	bool Renderer::prepareRenderUnmergedBatch(Batch *batch, PreparedRenderBatch *renderBatch)
3371	{
3372	if (batch->vertexCount == 0)
3373	return false;
3374
3375	Element *e = batch->first;
3376	Q_ASSERT(e);
3377
3378	if (Q_UNLIKELY(debug_render())) {
3379	qDebug() << " -"
3380	<< batch
3381	<< (batch->uploadedThisFrame ? "[ upload]" : "[retained]")
3382	<< (e->node->clipList() ? "[ clip]" : "[noclip]")
3383	<< (batch->isOpaque ? "[opaque]" : "[ alpha]")
3384	<< "[unmerged]"
3385	<< " Nodes:" << QString::fromLatin1(ba: "%1").arg(a: qsg_countNodesInBatch(batch), fieldWidth: 4).toLatin1().constData()
3386	<< " Vertices:" << QString::fromLatin1(ba: "%1").arg(a: batch->vertexCount, fieldWidth: 5).toLatin1().constData()
3387	<< " Indices:" << QString::fromLatin1(ba: "%1").arg(a: batch->indexCount, fieldWidth: 5).toLatin1().constData()
3388	<< " root:" << batch->root;
3389
3390	batch->uploadedThisFrame = false;
3391	}
3392
3393	const int viewCount = projectionMatrixCount();
3394	m_current_projection_matrix.resize(sz: viewCount);
3395	for (int viewIndex = 0; viewIndex < viewCount; ++viewIndex)
3396	m_current_projection_matrix[viewIndex] = projectionMatrix(index: viewIndex);
3397
3398	m_current_projection_matrix_native_ndc.resize(sz: projectionMatrixWithNativeNDCCount());
3399	for (int viewIndex = 0; viewIndex < projectionMatrixWithNativeNDCCount(); ++viewIndex)
3400	m_current_projection_matrix_native_ndc[viewIndex] = projectionMatrixWithNativeNDC(index: viewIndex);
3401
3402	QSGGeometryNode *gn = e->node;
3403	if (m_renderMode != QSGRendererInterface::RenderMode3D)
3404	updateClipState(clipList: gn->clipList(), batch);
3405
3406	// We always have dirty matrix as all batches are at a unique z range.
3407	QSGMaterialShader::RenderState::DirtyStates dirty = QSGMaterialShader::RenderState::DirtyMatrix;
3408
3409	// The vertex attributes are assumed to be the same for all elements in the
3410	// unmerged batch since the material (and so the shaders) is the same.
3411	QSGGeometry *g = gn->geometry();
3412	QSGMaterial *material = gn->activeMaterial();
3413	ShaderManager::Shader *sms = m_shaderManager->prepareMaterialNoRewrite(material, geometry: g, renderMode: m_renderMode, multiViewCount: renderTarget().multiViewCount);
3414	if (!sms)
3415	return false;
3416
3417	Q_ASSERT(sms->materialShader);
3418	if (m_currentShader != sms)
3419	setActiveRhiShader(program: sms->materialShader, shader: sms);
3420
3421	m_current_opacity = gn->inheritedOpacity();
3422	if (sms->lastOpacity != m_current_opacity) {
3423	dirty |= QSGMaterialShader::RenderState::DirtyOpacity;
3424	sms->lastOpacity = m_current_opacity;
3425	}
3426
3427	QMatrix4x4 rootMatrix = batch->root ? qsg_matrixForRoot(node: batch->root) : QMatrix4x4();
3428
3429	QSGMaterialShaderPrivate *pd = QSGMaterialShaderPrivate::get(s: sms->materialShader);
3430	const quint32 ubufSize = quint32(pd->masterUniformData.size());
3431	if (pd->ubufBinding >= 0) {
3432	quint32 totalUBufSize = 0;
3433	while (e) {
3434	totalUBufSize += aligned(v: ubufSize, byteAlign: m_ubufAlignment);
3435	e = e->nextInBatch;
3436	}
3437	bool ubufRebuild = false;
3438	if (!batch->ubuf) {
3439	batch->ubuf = m_rhi->newBuffer(type: QRhiBuffer::Dynamic, usage: QRhiBuffer::UniformBuffer, size: totalUBufSize);
3440	ubufRebuild = true;
3441	} else {
3442	if (batch->ubuf->size() < totalUBufSize) {
3443	batch->ubuf->setSize(totalUBufSize);
3444	ubufRebuild = true;
3445	}
3446	}
3447	if (ubufRebuild) {
3448	batch->ubufDataValid = false;
3449	if (!batch->ubuf->create()) {
3450	qWarning(msg: "Failed to build uniform buffer of size %u bytes", totalUBufSize);
3451	delete batch->ubuf;
3452	batch->ubuf = nullptr;
3453	return false;
3454	}
3455	}
3456	}
3457
3458	QSGMaterialShader::RenderState renderState = state(dirty: QSGMaterialShader::RenderState::DirtyStates(int(dirty)));
3459	bool pendingGStatePop = false;
3460	updateMaterialStaticData(sms, renderState,
3461	material, batch, gstateChanged: &pendingGStatePop);
3462
3463	int ubufOffset = 0;
3464	QRhiGraphicsPipeline *ps = nullptr;
3465	QRhiGraphicsPipeline *depthPostPassPs = nullptr;
3466	e = batch->first;
3467
3468	char *directUpdatePtr = nullptr;
3469	if (batch->ubuf->nativeBuffer().slotCount == 0)
3470	directUpdatePtr = batch->ubuf->beginFullDynamicBufferUpdateForCurrentFrame();
3471
3472	while (e) {
3473	gn = e->node;
3474
3475	m_current_model_view_matrix = rootMatrix * *gn->matrix();
3476	m_current_determinant = m_current_model_view_matrix.determinant();
3477
3478	const int viewCount = projectionMatrixCount();
3479	m_current_projection_matrix.resize(sz: viewCount);
3480	for (int viewIndex = 0; viewIndex < viewCount; ++viewIndex)
3481	m_current_projection_matrix[viewIndex] = projectionMatrix(index: viewIndex);
3482
3483	m_current_projection_matrix_native_ndc.resize(sz: projectionMatrixWithNativeNDCCount());
3484	for (int viewIndex = 0; viewIndex < projectionMatrixWithNativeNDCCount(); ++viewIndex)
3485	m_current_projection_matrix_native_ndc[viewIndex] = projectionMatrixWithNativeNDC(index: viewIndex);
3486
3487	if (useDepthBuffer()) {
3488	// this cannot be multiview
3489	m_current_projection_matrix[0](2, 2) = m_zRange;
3490	m_current_projection_matrix[0](2, 3) = calculateElementZOrder(e, zRange: m_zRange);
3491	}
3492
3493	QSGMaterialShader::RenderState renderState = state(dirty: QSGMaterialShader::RenderState::DirtyStates(int(dirty)));
3494	updateMaterialDynamicData(sms, renderState, material, batch, e, ubufOffset, ubufRegionSize: ubufSize, directUpdatePtr);
3495
3496	#ifndef QT_NO_DEBUG
3497	if (qsg_test_and_clear_material_failure()) {
3498	qDebug(msg: "QSGMaterial::updateState() triggered an error (unmerged), batch will be skipped:");
3499	qDebug() << " - offending node is" << e->node;
3500	QSGNodeDumper::dump(n: rootNode());
3501	qFatal(msg: "Aborting: scene graph is invalid...");
3502	return false;
3503	}
3504	#endif
3505
3506	ubufOffset += aligned(v: ubufSize, byteAlign: m_ubufAlignment);
3507
3508	const QSGGeometry::DrawingMode prevDrawMode = m_gstate.drawMode;
3509	const float prevLineWidth = m_gstate.lineWidth;
3510	m_gstate.drawMode = QSGGeometry::DrawingMode(g->drawingMode());
3511	m_gstate.lineWidth = g->lineWidth();
3512
3513	// Do not bother even looking up the ps if the topology has not changed
3514	// since everything else is the same for all elements in the batch.
3515	// (except if the material modified blend state)
3516	if (!ps || m_gstate.drawMode != prevDrawMode || m_gstate.lineWidth != prevLineWidth || pendingGStatePop) {
3517	if (!ensurePipelineState(e, sms)) {
3518	if (pendingGStatePop)
3519	m_gstate = m_gstateStack.pop();
3520	return false;
3521	}
3522	ps = e->ps;
3523	if (m_renderMode == QSGRendererInterface::RenderMode3D) {
3524	m_gstateStack.push(t: m_gstate);
3525	setStateForDepthPostPass();
3526	ensurePipelineState(e, sms, depthPostPass: true);
3527	m_gstate = m_gstateStack.pop();
3528	depthPostPassPs = e->depthPostPassPs;
3529	}
3530	} else {
3531	e->ps = ps;
3532	if (m_renderMode == QSGRendererInterface::RenderMode3D)
3533	e->depthPostPassPs = depthPostPassPs;
3534	}
3535
3536	// We don't need to bother with asking each node for its material as they
3537	// are all identical (compare==0) since they are in the same batch.
3538	m_currentMaterial = material;
3539
3540	// We only need to push this on the very first iteration...
3541	dirty &= ~QSGMaterialShader::RenderState::DirtyOpacity;
3542
3543	e = e->nextInBatch;
3544	}
3545
3546	if (directUpdatePtr)
3547	batch->ubuf->endFullDynamicBufferUpdateForCurrentFrame();
3548
3549	if (pendingGStatePop)
3550	m_gstate = m_gstateStack.pop();
3551
3552	batch->ubufDataValid = true;
3553
3554	renderBatch->batch = batch;
3555	renderBatch->sms = sms;
3556
3557	return true;
3558	}
3559
3560	void Renderer::renderUnmergedBatch(PreparedRenderBatch *renderBatch, bool depthPostPass)
3561	{
3562	const Batch *batch = renderBatch->batch;
3563	if (!batch->vbo.buf)
3564	return;
3565
3566	Element *e = batch->first;
3567
3568	if (batch->clipState.type & ClipState::StencilClip)
3569	enqueueStencilDraw(batch);
3570
3571	quint32 vOffset = 0;
3572	quint32 iOffset = 0;
3573	QRhiCommandBuffer *cb = renderTarget().cb;
3574
3575	while (e) {
3576	QSGGeometry *g = e->node->geometry();
3577	checkLineWidth(g);
3578	const int effectiveIndexSize = m_uint32IndexForRhi ? sizeof(quint32) : g->sizeOfIndex();
3579
3580	setGraphicsPipeline(cb, batch, e, depthPostPass);
3581
3582	const QRhiCommandBuffer::VertexInput vbufBinding(batch->vbo.buf, vOffset);
3583	if (g->indexCount()) {
3584	if (batch->ibo.buf) {
3585	cb->setVertexInput(startBinding: VERTEX_BUFFER_BINDING, bindingCount: 1, bindings: &vbufBinding,
3586	indexBuf: batch->ibo.buf, indexOffset: iOffset,
3587	indexFormat: effectiveIndexSize == sizeof(quint32) ? QRhiCommandBuffer::IndexUInt32
3588	: QRhiCommandBuffer::IndexUInt16);
3589	cb->drawIndexed(indexCount: g->indexCount());
3590	}
3591	} else {
3592	cb->setVertexInput(startBinding: VERTEX_BUFFER_BINDING, bindingCount: 1, bindings: &vbufBinding);
3593	cb->draw(vertexCount: g->vertexCount());
3594	}
3595
3596	vOffset += g->sizeOfVertex() * g->vertexCount();
3597	iOffset += g->indexCount() * effectiveIndexSize;
3598
3599	e = e->nextInBatch;
3600	}
3601	}
3602
3603	void Renderer::setGraphicsPipeline(QRhiCommandBuffer *cb, const Batch *batch, Element *e, bool depthPostPass)
3604	{
3605	cb->setGraphicsPipeline(depthPostPass ? e->depthPostPassPs : e->ps);
3606
3607	if (!m_pstate.viewportSet) {
3608	m_pstate.viewportSet = true;
3609	cb->setViewport(m_pstate.viewport);
3610	}
3611	if (batch->clipState.type & ClipState::ScissorClip) {
3612	Q_ASSERT(e->ps->flags().testFlag(QRhiGraphicsPipeline::UsesScissor));
3613	m_pstate.scissorSet = true;
3614	cb->setScissor(batch->clipState.scissor);
3615	} else {
3616	Q_ASSERT(!e->ps->flags().testFlag(QRhiGraphicsPipeline::UsesScissor));
3617	// Regardless of the ps not using scissor, the scissor may need to be
3618	// reset, depending on the backend. So set the viewport again, which in
3619	// turn also sets the scissor on backends where a scissor rect is
3620	// always-on (Vulkan).
3621	if (m_pstate.scissorSet) {
3622	m_pstate.scissorSet = false;
3623	cb->setViewport(m_pstate.viewport);
3624	}
3625	}
3626	if (batch->clipState.type & ClipState::StencilClip) {
3627	Q_ASSERT(e->ps->flags().testFlag(QRhiGraphicsPipeline::UsesStencilRef));
3628	cb->setStencilRef(batch->clipState.stencilRef);
3629	}
3630	if (!depthPostPass && e->ps->flags().testFlag(flag: QRhiGraphicsPipeline::UsesBlendConstants))
3631	cb->setBlendConstants(batch->blendConstant);
3632
3633	cb->setShaderResources(srb: e->srb);
3634	}
3635
3636	void Renderer::releaseElement(Element *e, bool inDestructor)
3637	{
3638	if (e->isRenderNode) {
3639	delete static_cast<RenderNodeElement *>(e);
3640	} else {
3641	if (e->srb) {
3642	if (!inDestructor) {
3643	if (m_shaderManager->srbPool.size() < m_srbPoolThreshold)
3644	m_shaderManager->srbPool.insert(key: e->srb->serializedLayoutDescription(), value: e->srb);
3645	else
3646	delete e->srb;
3647	} else {
3648	delete e->srb;
3649	}
3650	e->srb = nullptr;
3651	}
3652	m_elementAllocator.release(t: e);
3653	}
3654	}
3655
3656	void Renderer::deleteRemovedElements()
3657	{
3658	if (!m_elementsToDelete.size())
3659	return;
3660
3661	for (int i=0; i<m_opaqueRenderList.size(); ++i) {
3662	Element **e = m_opaqueRenderList.data() + i;
3663	if (*e && (*e)->removed)
3664	*e = nullptr;
3665	}
3666	for (int i=0; i<m_alphaRenderList.size(); ++i) {
3667	Element **e = m_alphaRenderList.data() + i;
3668	if (*e && (*e)->removed)
3669	*e = nullptr;
3670	}
3671
3672	for (int i=0; i<m_elementsToDelete.size(); ++i)
3673	releaseElement(e: m_elementsToDelete.at(i));
3674
3675	m_elementsToDelete.reset();
3676	}
3677
3678	void Renderer::render()
3679	{
3680	// Gracefully handle the lack of a render target - some autotests may rely
3681	// on this in odd cases.
3682	if (!renderTarget().rt)
3683	return;
3684
3685	prepareRenderPass(ctx: &m_mainRenderPassContext);
3686	beginRenderPass(ctx: &m_mainRenderPassContext);
3687	recordRenderPass(ctx: &m_mainRenderPassContext);
3688	endRenderPass(ctx: &m_mainRenderPassContext);
3689	}
3690
3691	// An alternative to render() is to call prepareInline() and renderInline() at
3692	// the appropriate times (i.e. outside of a QRhi::beginPass() and then inside,
3693	// respectively) These allow rendering within a render pass that is started by
3694	// another component. In contrast, render() records a full render pass on its
3695	// own.
3696
3697	void Renderer::prepareInline()
3698	{
3699	prepareRenderPass(ctx: &m_mainRenderPassContext);
3700	}
3701
3702	void Renderer::renderInline()
3703	{
3704	recordRenderPass(ctx: &m_mainRenderPassContext);
3705	}
3706
3707	void Renderer::prepareRenderPass(RenderPassContext *ctx)
3708	{
3709	if (ctx->valid)
3710	qWarning(msg: "prepareRenderPass() called with an already prepared render pass context");
3711
3712	ctx->valid = true;
3713
3714	if (Q_UNLIKELY(debug_dump())) {
3715	qDebug(msg: "\n");
3716	QSGNodeDumper::dump(n: rootNode());
3717	}
3718
3719	ctx->timeRenderLists = 0;
3720	ctx->timePrepareOpaque = 0;
3721	ctx->timePrepareAlpha = 0;
3722	ctx->timeSorting = 0;
3723	ctx->timeUploadOpaque = 0;
3724	ctx->timeUploadAlpha = 0;
3725
3726	if (Q_UNLIKELY(debug_render() || debug_build())) {
3727	QByteArray type("rebuild:");
3728	if (m_rebuild == 0)
3729	type += " none";
3730	if (m_rebuild == FullRebuild)
3731	type += " full";
3732	else {
3733	if (m_rebuild & BuildRenderLists)
3734	type += " renderlists";
3735	else if (m_rebuild & BuildRenderListsForTaggedRoots)
3736	type += " partial";
3737	else if (m_rebuild & BuildBatches)
3738	type += " batches";
3739	}
3740
3741	qDebug() << "Renderer::render()" << this << type;
3742	ctx->timer.start();
3743	}
3744
3745	m_resourceUpdates = m_rhi->nextResourceUpdateBatch();
3746
3747	if (m_rebuild & (BuildRenderLists | BuildRenderListsForTaggedRoots)) {
3748	bool complete = (m_rebuild & BuildRenderLists) != 0;
3749	if (complete)
3750	buildRenderListsFromScratch();
3751	else
3752	buildRenderListsForTaggedRoots();
3753	m_rebuild |= BuildBatches;
3754
3755	if (Q_UNLIKELY(debug_build())) {
3756	qDebug(msg: "Opaque render lists %s:", (complete ? "(complete)" : "(partial)"));
3757	for (int i=0; i<m_opaqueRenderList.size(); ++i) {
3758	Element *e = m_opaqueRenderList.at(i);
3759	qDebug() << " - element:" << e << " batch:" << e->batch << " node:" << e->node << " order:" << e->order;
3760	}
3761	qDebug(msg: "Alpha render list %s:", complete ? "(complete)" : "(partial)");
3762	for (int i=0; i<m_alphaRenderList.size(); ++i) {
3763	Element *e = m_alphaRenderList.at(i);
3764	qDebug() << " - element:" << e << " batch:" << e->batch << " node:" << e->node << " order:" << e->order;
3765	}
3766	}
3767	}
3768	if (Q_UNLIKELY(debug_render())) ctx->timeRenderLists = ctx->timer.restart();
3769
3770	for (int i=0; i<m_opaqueBatches.size(); ++i)
3771	m_opaqueBatches.at(i)->cleanupRemovedElements();
3772	for (int i=0; i<m_alphaBatches.size(); ++i)
3773	m_alphaBatches.at(i)->cleanupRemovedElements();
3774	deleteRemovedElements();
3775
3776	cleanupBatches(batches: &m_opaqueBatches);
3777	cleanupBatches(batches: &m_alphaBatches);
3778
3779	if (m_rebuild & BuildBatches) {
3780	prepareOpaqueBatches();
3781	if (Q_UNLIKELY(debug_render())) ctx->timePrepareOpaque = ctx->timer.restart();
3782	prepareAlphaBatches();
3783	if (Q_UNLIKELY(debug_render())) ctx->timePrepareAlpha = ctx->timer.restart();
3784
3785	if (Q_UNLIKELY(debug_build())) {
3786	qDebug(msg: "Opaque Batches:");
3787	for (int i=0; i<m_opaqueBatches.size(); ++i) {
3788	Batch *b = m_opaqueBatches.at(i);
3789	qDebug() << " - Batch " << i << b << (b->needsUpload ? "upload" : "") << " root:" << b->root;
3790	for (Element *e = b->first; e; e = e->nextInBatch) {
3791	qDebug() << " - element:" << e << " node:" << e->node << e->order;
3792	}
3793	}
3794	qDebug(msg: "Alpha Batches:");
3795	for (int i=0; i<m_alphaBatches.size(); ++i) {
3796	Batch *b = m_alphaBatches.at(i);
3797	qDebug() << " - Batch " << i << b << (b->needsUpload ? "upload" : "") << " root:" << b->root;
3798	for (Element *e = b->first; e; e = e->nextInBatch) {
3799	qDebug() << " - element:" << e << e->bounds << " node:" << e->node << " order:" << e->order;
3800	}
3801	}
3802	}
3803	} else {
3804	if (Q_UNLIKELY(debug_render())) ctx->timePrepareOpaque = ctx->timePrepareAlpha = ctx->timer.restart();
3805	}
3806
3807
3808	deleteRemovedElements();
3809
3810	if (m_rebuild != 0) {
3811	// Then sort opaque batches so that we're drawing the batches with the highest
3812	// order first, maximizing the benefit of front-to-back z-ordering.
3813	if (m_opaqueBatches.size())
3814	std::sort(first: &m_opaqueBatches.first(), last: &m_opaqueBatches.last() + 1, comp: qsg_sort_batch_decreasing_order);
3815
3816	// Sort alpha batches back to front so that they render correctly.
3817	if (m_alphaBatches.size())
3818	std::sort(first: &m_alphaBatches.first(), last: &m_alphaBatches.last() + 1, comp: qsg_sort_batch_increasing_order);
3819
3820	m_zRange = m_nextRenderOrder != 0
3821	? 1.0 / (m_nextRenderOrder)
3822	: 0;
3823	}
3824
3825	if (Q_UNLIKELY(debug_render())) ctx->timeSorting = ctx->timer.restart();
3826
3827	// Set size to 0, nothing is deallocated, they will "grow" again
3828	// as part of uploadBatch.
3829	m_vertexUploadPool.reset();
3830	m_indexUploadPool.reset();
3831
3832	if (Q_UNLIKELY(debug_upload())) qDebug(msg: "Uploading Opaque Batches:");
3833	for (int i=0; i<m_opaqueBatches.size(); ++i) {
3834	Batch *b = m_opaqueBatches.at(i);
3835	uploadBatch(b);
3836	}
3837	if (Q_UNLIKELY(debug_render())) ctx->timeUploadOpaque = ctx->timer.restart();
3838
3839	if (Q_UNLIKELY(debug_upload())) qDebug(msg: "Uploading Alpha Batches:");
3840	for (int i=0; i<m_alphaBatches.size(); ++i) {
3841	Batch *b = m_alphaBatches.at(i);
3842	uploadBatch(b);
3843	}
3844	if (Q_UNLIKELY(debug_render())) ctx->timeUploadAlpha = ctx->timer.restart();
3845
3846	if (Q_UNLIKELY(debug_render())) {
3847	qDebug().nospace() << "Rendering:" << Qt::endl
3848	<< " -> Opaque: " << qsg_countNodesInBatches(batches: m_opaqueBatches) << " nodes in " << m_opaqueBatches.size() << " batches..." << Qt::endl
3849	<< " -> Alpha: " << qsg_countNodesInBatches(batches: m_alphaBatches) << " nodes in " << m_alphaBatches.size() << " batches...";
3850	}
3851
3852	m_current_opacity = 1;
3853	m_currentMaterial = nullptr;
3854	m_currentShader = nullptr;
3855	m_currentProgram = nullptr;
3856	m_currentClipState.reset();
3857
3858	const QRect viewport = viewportRect();
3859
3860	bool renderOpaque = !debug_noopaque();
3861	bool renderAlpha = !debug_noalpha();
3862
3863	m_pstate.viewport =
3864	QRhiViewport(viewport.x(), deviceRect().bottom() - viewport.bottom(), viewport.width(),
3865	viewport.height(), VIEWPORT_MIN_DEPTH, VIEWPORT_MAX_DEPTH);
3866	m_pstate.clearColor = clearColor();
3867	m_pstate.dsClear = QRhiDepthStencilClearValue(1.0f, 0);
3868	m_pstate.viewportSet = false;
3869	m_pstate.scissorSet = false;
3870
3871	m_gstate.depthTest = useDepthBuffer();
3872	m_gstate.depthWrite = useDepthBuffer();
3873	m_gstate.depthFunc = QRhiGraphicsPipeline::Less;
3874	m_gstate.blending = false;
3875
3876	m_gstate.cullMode = QRhiGraphicsPipeline::None;
3877	m_gstate.polygonMode = QRhiGraphicsPipeline::Fill;
3878	m_gstate.colorWrite = QRhiGraphicsPipeline::R
3879	| QRhiGraphicsPipeline::G
3880	| QRhiGraphicsPipeline::B
3881	| QRhiGraphicsPipeline::A;
3882	m_gstate.usesScissor = false;
3883	m_gstate.stencilTest = false;
3884
3885	m_gstate.sampleCount = renderTarget().rt->sampleCount();
3886	m_gstate.multiViewCount = renderTarget().multiViewCount;
3887
3888	ctx->opaqueRenderBatches.clear();
3889	if (Q_LIKELY(renderOpaque)) {
3890	for (int i = 0, ie = m_opaqueBatches.size(); i != ie; ++i) {
3891	Batch *b = m_opaqueBatches.at(i);
3892	PreparedRenderBatch renderBatch;
3893	bool ok;
3894	if (b->merged)
3895	ok = prepareRenderMergedBatch(batch: b, renderBatch: &renderBatch);
3896	else
3897	ok = prepareRenderUnmergedBatch(batch: b, renderBatch: &renderBatch);
3898	if (ok)
3899	ctx->opaqueRenderBatches.append(t: renderBatch);
3900	}
3901	}
3902
3903	m_gstate.blending = true;
3904	// factors never change, always set for premultiplied alpha based blending
3905
3906	// depth test stays enabled (if useDepthBuffer(), that is) but no need
3907	// to write out depth from the transparent (back-to-front) pass
3908	m_gstate.depthWrite = false;
3909
3910	// special case: the 3D plane mode tests against the depth buffer, but does
3911	// not write (and all batches are alpha because this render mode evaluates
3912	// to useDepthBuffer()==false)
3913	if (m_renderMode == QSGRendererInterface::RenderMode3D) {
3914	Q_ASSERT(m_opaqueBatches.isEmpty());
3915	m_gstate.depthTest = true;
3916	}
3917
3918	ctx->alphaRenderBatches.clear();
3919	if (Q_LIKELY(renderAlpha)) {
3920	for (int i = 0, ie = m_alphaBatches.size(); i != ie; ++i) {
3921	Batch *b = m_alphaBatches.at(i);
3922	PreparedRenderBatch renderBatch;
3923	bool ok;
3924	if (b->merged)
3925	ok = prepareRenderMergedBatch(batch: b, renderBatch: &renderBatch);
3926	else if (b->isRenderNode)
3927	ok = prepareRhiRenderNode(batch: b, renderBatch: &renderBatch);
3928	else
3929	ok = prepareRenderUnmergedBatch(batch: b, renderBatch: &renderBatch);
3930	if (ok)
3931	ctx->alphaRenderBatches.append(t: renderBatch);
3932	}
3933	}
3934
3935	m_rebuild = 0;
3936
3937	#if defined(QSGBATCHRENDERER_INVALIDATE_WEDGED_NODES)
3938	m_renderOrderRebuildLower = -1;
3939	m_renderOrderRebuildUpper = -1;
3940	#endif
3941
3942	if (m_visualizer->mode() != Visualizer::VisualizeNothing)
3943	m_visualizer->prepareVisualize();
3944
3945	renderTarget().cb->resourceUpdate(resourceUpdates: m_resourceUpdates);
3946	m_resourceUpdates = nullptr;
3947	}
3948
3949	void Renderer::beginRenderPass(RenderPassContext *)
3950	{
3951	const QSGRenderTarget &rt(renderTarget());
3952	rt.cb->beginPass(rt: rt.rt, colorClearValue: m_pstate.clearColor, depthStencilClearValue: m_pstate.dsClear, resourceUpdates: nullptr,
3953	// we cannot tell if the application will have
3954	// native rendering thrown in to this pass
3955	// (QQuickWindow::beginExternalCommands()), so
3956	// we have no choice but to set the flag always
3957	// (thus triggering using secondary command
3958	// buffers with Vulkan)
3959	flags: QRhiCommandBuffer::ExternalContent
3960	// We do not use GPU compute at all at the moment, this means we can
3961	// get a small performance gain with OpenGL by declaring this.
3962	| QRhiCommandBuffer::DoNotTrackResourcesForCompute);
3963
3964	if (m_renderPassRecordingCallbacks.start)
3965	m_renderPassRecordingCallbacks.start(m_renderPassRecordingCallbacks.userData);
3966	}
3967
3968	void Renderer::recordRenderPass(RenderPassContext *ctx)
3969	{
3970	// prepareRenderPass and recordRenderPass must always be called together.
3971	// They are separate because beginRenderPass and endRenderPass are optional.
3972	//
3973	// The valid call sequence are therefore:
3974	// prepare, begin, record, end
3975	// or
3976	// prepare, record
3977
3978	if (!ctx->valid)
3979	qWarning(msg: "recordRenderPass() called without a prepared render pass context");
3980
3981	ctx->valid = false;
3982
3983	QRhiCommandBuffer *cb = renderTarget().cb;
3984	cb->debugMarkBegin(QByteArrayLiteral("Qt Quick scene render"));
3985
3986	for (int i = 0, ie = ctx->opaqueRenderBatches.size(); i != ie; ++i) {
3987	if (i == 0)
3988	cb->debugMarkMsg(QByteArrayLiteral("Qt Quick opaque batches"));
3989	PreparedRenderBatch *renderBatch = &ctx->opaqueRenderBatches[i];
3990	if (renderBatch->batch->merged)
3991	renderMergedBatch(renderBatch);
3992	else
3993	renderUnmergedBatch(renderBatch);
3994	}
3995
3996	for (int i = 0, ie = ctx->alphaRenderBatches.size(); i != ie; ++i) {
3997	if (i == 0) {
3998	if (m_renderMode == QSGRendererInterface::RenderMode3D)
3999	cb->debugMarkMsg(QByteArrayLiteral("Qt Quick 2D-in-3D batches"));
4000	else
4001	cb->debugMarkMsg(QByteArrayLiteral("Qt Quick alpha batches"));
4002	}
4003	PreparedRenderBatch *renderBatch = &ctx->alphaRenderBatches[i];
4004	if (renderBatch->batch->merged)
4005	renderMergedBatch(renderBatch);
4006	else if (renderBatch->batch->isRenderNode)
4007	renderRhiRenderNode(batch: renderBatch->batch);
4008	else
4009	renderUnmergedBatch(renderBatch);
4010	}
4011
4012	if (m_renderMode == QSGRendererInterface::RenderMode3D) {
4013	// Depth post-pass to fill up the depth buffer in a way that it
4014	// corresponds to what got rendered to the color buffer in the previous
4015	// (alpha) pass. The previous pass cannot enable depth write due to Z
4016	// fighting. Rather, do it separately in a dedicated color-write-off,
4017	// depth-write-on pass. This enables the 3D content drawn afterwards to
4018	// depth test against the 2D items' rendering.
4019	for (int i = 0, ie = ctx->alphaRenderBatches.size(); i != ie; ++i) {
4020	if (i == 0)
4021	cb->debugMarkMsg(QByteArrayLiteral("Qt Quick 2D-in-3D depth post-pass"));
4022	PreparedRenderBatch *renderBatch = &ctx->alphaRenderBatches[i];
4023	if (renderBatch->batch->merged)
4024	renderMergedBatch(renderBatch, depthPostPass: true);
4025	else if (!renderBatch->batch->isRenderNode) // rendernodes are skipped here for now
4026	renderUnmergedBatch(renderBatch, depthPostPass: true);
4027	}
4028	}
4029
4030	if (m_currentShader)
4031	setActiveRhiShader(program: nullptr, shader: nullptr);
4032
4033	cb->debugMarkEnd();
4034
4035	if (Q_UNLIKELY(debug_render())) {
4036	qDebug(msg: " -> times: build: %d, prepare(opaque/alpha): %d/%d, sorting: %d, upload(opaque/alpha): %d/%d, record rendering: %d",
4037	(int) ctx->timeRenderLists,
4038	(int) ctx->timePrepareOpaque, (int) ctx->timePrepareAlpha,
4039	(int) ctx->timeSorting,
4040	(int) ctx->timeUploadOpaque, (int) ctx->timeUploadAlpha,
4041	(int) ctx->timer.elapsed());
4042	}
4043	}
4044
4045	void Renderer::endRenderPass(RenderPassContext *)
4046	{
4047	if (m_renderPassRecordingCallbacks.end)
4048	m_renderPassRecordingCallbacks.end(m_renderPassRecordingCallbacks.userData);
4049
4050	if (m_visualizer->mode() != Visualizer::VisualizeNothing)
4051	m_visualizer->visualize();
4052
4053	renderTarget().cb->endPass();
4054	}
4055
4056	struct RenderNodeState : public QSGRenderNode::RenderState
4057	{
4058	const QMatrix4x4 *projectionMatrix() const override { return m_projectionMatrix; }
4059	QRect scissorRect() const override { return m_scissorRect; }
4060	bool scissorEnabled() const override { return m_scissorEnabled; }
4061	int stencilValue() const override { return m_stencilValue; }
4062	bool stencilEnabled() const override { return m_stencilEnabled; }
4063	const QRegion *clipRegion() const override { return nullptr; }
4064
4065	const QMatrix4x4 *m_projectionMatrix;
4066	QRect m_scissorRect;
4067	int m_stencilValue;
4068	bool m_scissorEnabled;
4069	bool m_stencilEnabled;
4070	};
4071
4072	bool Renderer::prepareRhiRenderNode(Batch *batch, PreparedRenderBatch *renderBatch)
4073	{
4074	if (Q_UNLIKELY(debug_render()))
4075	qDebug() << " -" << batch << "rendernode";
4076
4077	Q_ASSERT(batch->first->isRenderNode);
4078	RenderNodeElement *e = static_cast<RenderNodeElement *>(batch->first);
4079
4080	setActiveRhiShader(program: nullptr, shader: nullptr);
4081
4082	QSGRenderNodePrivate *rd = QSGRenderNodePrivate::get(node: e->renderNode);
4083	rd->m_clip_list = nullptr;
4084	if (m_renderMode != QSGRendererInterface::RenderMode3D) {
4085	QSGNode *clip = e->renderNode->parent();
4086	while (clip != rootNode()) {
4087	if (clip->type() == QSGNode::ClipNodeType) {
4088	rd->m_clip_list = static_cast<QSGClipNode *>(clip);
4089	break;
4090	}
4091	clip = clip->parent();
4092	}
4093	updateClipState(clipList: rd->m_clip_list, batch);
4094	}
4095
4096	QSGNode *xform = e->renderNode->parent();
4097	QMatrix4x4 matrix;
4098	QSGNode *root = rootNode();
4099	if (e->root) {
4100	matrix = qsg_matrixForRoot(node: e->root);
4101	root = e->root->sgNode;
4102	}
4103	while (xform != root) {
4104	if (xform->type() == QSGNode::TransformNodeType) {
4105	matrix = matrix * static_cast<QSGTransformNode *>(xform)->combinedMatrix();
4106	break;
4107	}
4108	xform = xform->parent();
4109	}
4110	rd->m_localMatrix = matrix;
4111	rd->m_matrix = &rd->m_localMatrix;
4112
4113	QSGNode *opacity = e->renderNode->parent();
4114	rd->m_opacity = 1.0;
4115	while (opacity != rootNode()) {
4116	if (opacity->type() == QSGNode::OpacityNodeType) {
4117	rd->m_opacity = static_cast<QSGOpacityNode *>(opacity)->combinedOpacity();
4118	break;
4119	}
4120	opacity = opacity->parent();
4121	}
4122
4123	rd->m_rt = renderTarget();
4124
4125	const int viewCount = projectionMatrixCount();
4126	rd->m_projectionMatrix.resize(sz: viewCount);
4127	for (int viewIndex = 0; viewIndex < viewCount; ++viewIndex)
4128	rd->m_projectionMatrix[viewIndex] = projectionMatrix(index: viewIndex);
4129
4130	if (useDepthBuffer()) {
4131	// this cannot be multiview
4132	rd->m_projectionMatrix[0](2, 2) = m_zRange;
4133	rd->m_projectionMatrix[0](2, 3) = calculateElementZOrder(e, zRange: m_zRange);
4134	}
4135
4136	e->renderNode->prepare();
4137
4138	renderBatch->batch = batch;
4139	renderBatch->sms = nullptr;
4140
4141	return true;
4142	}
4143
4144	void Renderer::renderRhiRenderNode(const Batch *batch)
4145	{
4146	if (batch->clipState.type & ClipState::StencilClip)
4147	enqueueStencilDraw(batch);
4148
4149	RenderNodeElement *e = static_cast<RenderNodeElement *>(batch->first);
4150	QSGRenderNodePrivate *rd = QSGRenderNodePrivate::get(node: e->renderNode);
4151
4152	RenderNodeState state;
4153	// Expose only the first matrix through the state object, the rest are
4154	// queriable through the QSGRenderNode getters anyway.
4155	state.m_projectionMatrix = &rd->m_projectionMatrix[0];
4156	const std::array<int, 4> scissor = batch->clipState.scissor.scissor();
4157	state.m_scissorRect = QRect(scissor[0], scissor[1], scissor[2], scissor[3]);
4158	state.m_stencilValue = batch->clipState.stencilRef;
4159	state.m_scissorEnabled = batch->clipState.type & ClipState::ScissorClip;
4160	state.m_stencilEnabled = batch->clipState.type & ClipState::StencilClip;
4161
4162	const QSGRenderNode::StateFlags changes = e->renderNode->changedStates();
4163
4164	QRhiCommandBuffer *cb = renderTarget().cb;
4165	const bool needsExternal = !e->renderNode->flags().testFlag(flag: QSGRenderNode::NoExternalRendering);
4166	if (needsExternal)
4167	cb->beginExternal();
4168	e->renderNode->render(state: &state);
4169	if (needsExternal)
4170	cb->endExternal();
4171
4172	rd->m_matrix = nullptr;
4173	rd->m_clip_list = nullptr;
4174
4175	if ((changes & QSGRenderNode::ViewportState)
4176	|| (changes & QSGRenderNode::ScissorState))
4177	{
4178	// Reset both flags if either is reported as changed, since with the rhi
4179	// it could be setViewport() that will record the resetting of the scissor.
4180	m_pstate.viewportSet = false;
4181	m_pstate.scissorSet = false;
4182	}
4183
4184	// Do not bother with RenderTargetState. Where applicable, endExternal()
4185	// ensures the correct target is rebound. For others (like Vulkan) it makes
4186	// no sense since render() could not possibly do that on our command buffer
4187	// which is in renderpass recording state.
4188	}
4189
4190	void Renderer::setVisualizationMode(const QByteArray &mode)
4191	{
4192	if (mode.isEmpty())
4193	m_visualizer->setMode(Visualizer::VisualizeNothing);
4194	else if (mode == "clip")
4195	m_visualizer->setMode(Visualizer::VisualizeClipping);
4196	else if (mode == "overdraw")
4197	m_visualizer->setMode(Visualizer::VisualizeOverdraw);
4198	else if (mode == "batches")
4199	m_visualizer->setMode(Visualizer::VisualizeBatches);
4200	else if (mode == "changes")
4201	m_visualizer->setMode(Visualizer::VisualizeChanges);
4202	}
4203
4204	bool Renderer::hasVisualizationModeWithContinuousUpdate() const
4205	{
4206	return m_visualizer->mode() == Visualizer::VisualizeOverdraw;
4207	}
4208
4209	bool operator==(const GraphicsState &a, const GraphicsState &b) noexcept
4210	{
4211	return a.depthTest == b.depthTest
4212	&& a.depthWrite == b.depthWrite
4213	&& a.depthFunc == b.depthFunc
4214	&& a.blending == b.blending
4215	&& a.srcColor == b.srcColor
4216	&& a.dstColor == b.dstColor
4217	&& a.srcAlpha == b.srcAlpha
4218	&& a.dstAlpha == b.dstAlpha
4219	&& a.opColor == b.opColor
4220	&& a.opAlpha == b.opAlpha
4221	&& a.colorWrite == b.colorWrite
4222	&& a.cullMode == b.cullMode
4223	&& a.usesScissor == b.usesScissor
4224	&& a.stencilTest == b.stencilTest
4225	&& a.sampleCount == b.sampleCount
4226	&& a.drawMode == b.drawMode
4227	&& a.lineWidth == b.lineWidth
4228	&& a.polygonMode == b.polygonMode
4229	&& a.multiViewCount == b.multiViewCount;
4230	}
4231
4232	bool operator!=(const GraphicsState &a, const GraphicsState &b) noexcept
4233	{
4234	return !(a == b);
4235	}
4236
4237	size_t qHash(const GraphicsState &s, size_t seed) noexcept
4238	{
4239	// do not bother with all fields
4240	return seed
4241	+ s.depthTest * 1000
4242	+ s.depthWrite * 100
4243	+ s.depthFunc
4244	+ s.blending * 10
4245	+ s.srcColor
4246	+ s.cullMode
4247	+ s.usesScissor
4248	+ s.stencilTest
4249	+ s.sampleCount
4250	+ s.multiViewCount;
4251	}
4252
4253	bool operator==(const GraphicsPipelineStateKey &a, const GraphicsPipelineStateKey &b) noexcept
4254	{
4255	return a.state == b.state
4256	&& a.sms->materialShader == b.sms->materialShader
4257	&& a.renderTargetDescription == b.renderTargetDescription
4258	&& a.srbLayoutDescription == b.srbLayoutDescription;
4259	}
4260
4261	bool operator!=(const GraphicsPipelineStateKey &a, const GraphicsPipelineStateKey &b) noexcept
4262	{
4263	return !(a == b);
4264	}
4265
4266	size_t qHash(const GraphicsPipelineStateKey &k, size_t seed) noexcept
4267	{
4268	return qHash(s: k.state, seed)
4269	^ qHash(key: k.sms->materialShader)
4270	^ k.extra.renderTargetDescriptionHash
4271	^ k.extra.srbLayoutDescriptionHash;
4272	}
4273
4274	bool operator==(const ShaderKey &a, const ShaderKey &b) noexcept
4275	{
4276	return a.type == b.type
4277	&& a.renderMode == b.renderMode
4278	&& a.multiViewCount == b.multiViewCount;
4279	}
4280
4281	bool operator!=(const ShaderKey &a, const ShaderKey &b) noexcept
4282	{
4283	return !(a == b);
4284	}
4285
4286	size_t qHash(const ShaderKey &k, size_t seed) noexcept
4287	{
4288	return qHash(t: k.type, seed) ^ int(k.renderMode) ^ k.multiViewCount;
4289	}
4290
4291	Visualizer::Visualizer(Renderer *renderer)
4292	: m_renderer(renderer),
4293	m_visualizeMode(VisualizeNothing)
4294	{
4295	}
4296
4297	Visualizer::~Visualizer()
4298	{
4299	}
4300
4301	#define QSGNODE_DIRTY_PARENT (QSGNode::DirtyNodeAdded \
4302	| QSGNode::DirtyOpacity \
4303	| QSGNode::DirtyMatrix \
4304	| QSGNode::DirtyNodeRemoved)
4305
4306	void Visualizer::visualizeChangesPrepare(Node *n, uint parentChanges)
4307	{
4308	uint childDirty = (parentChanges | n->dirtyState) & QSGNODE_DIRTY_PARENT;
4309	uint selfDirty = n->dirtyState | parentChanges;
4310	if (n->type() == QSGNode::GeometryNodeType && selfDirty != 0)
4311	m_visualizeChangeSet.insert(key: n, value: selfDirty);
4312	SHADOWNODE_TRAVERSE(n) {
4313	visualizeChangesPrepare(n: child, parentChanges: childDirty);
4314	}
4315	}
4316
4317	} // namespace QSGBatchRenderer
4318
4319	QT_END_NAMESPACE
4320
4321	#include "moc_qsgbatchrenderer_p.cpp"
4322