some more stuff using linear allocator

external/imgui/imgui_user.inl

@@ -905,7 +905,7 @@ namespace ImGuiEx {
 	}
 
 	// copy-pasted from imgui with alwaysautoresize flag removed
-	bool ImGuiEx::BeginResizableMenu(const char* label, const char* icon, bool enabled)
+	bool BeginResizableMenu(const char* label, const char* icon, bool enabled)
 	{
 		ImGuiWindow* window = GetCurrentWindow();
 		if (window->SkipItems)

src/engine/allocators.cpp

@@ -316,14 +316,23 @@ static u32 roundUp(u32 val, u32 align) {
 
 void* LinearAllocator::allocate_aligned(size_t size, size_t align) {
 	ASSERT(size < 0xffFFffFF);
-	const u32 start = roundUp(m_end, (u32)align);
-	if (start + size > m_commited) {
-		const u32 commited = roundUp(start + (u32)size, 4096);
-		ASSERT(commited < m_reserved);
-		os::memCommit(m_mem + m_commited, commited - m_commited);
-		m_commited = commited;
+	u32 start;
+	for (;;) {
+		const u32 end = m_end;
+		start = roundUp(end, (u32)align);
+		if (compareAndExchange(&m_end, u32(start + size), end)) break;
 	}
-	m_end = start + (u32)size;
+
+	if (start + size <= m_commited) return m_mem + start;
+
+	MutexGuard guard(m_mutex);
+	if (start + size <= m_commited) return m_mem + start;
+
+	const u32 commited = roundUp(start + (u32)size, 4096);
+	ASSERT(commited < m_reserved);
+	os::memCommit(m_mem + m_commited, commited - m_commited);
+	m_commited = commited;
+
 	return m_mem + start;
 }
 
@@ -337,14 +346,18 @@ void* LinearAllocator::reallocate_aligned(void* ptr, size_t size, size_t align)
 
 void* LinearAllocator::allocate(size_t size) {
 	ASSERT(size < 0xffFFffFF);
-	const u32 start = m_end;
-	if (start + size > m_commited) {
-		const u32 commited = roundUp(start + (u32)size, 4096);
-		ASSERT(commited < m_reserved);
-		os::memCommit(m_mem + m_commited, commited - m_commited);
-		m_commited = commited;
-	}
-	m_end = start + (u32)size;
+	const u32 start = atomicAdd(&m_end, (u32)size);
+
+	if (start + size <= m_commited) return m_mem + start;
+
+	MutexGuard guard(m_mutex);
+	if (start + size <= m_commited) return m_mem + start;
+
+	const u32 commited = roundUp(start + (u32)size, 4096);
+	ASSERT(commited < m_reserved);
+	os::memCommit(m_mem + m_commited, commited - m_commited);
+	m_commited = commited;
+
 	return m_mem + start;
 }
 

src/engine/allocators.h

@@ -41,7 +41,7 @@ private:
 	volatile i32 m_allocation_count;
 };
 
-struct LinearAllocator : IAllocator {
+struct LUMIX_ENGINE_API LinearAllocator : IAllocator {
 	LinearAllocator(u32 reserved);
 	~LinearAllocator();
 
@@ -56,8 +56,9 @@ struct LinearAllocator : IAllocator {
 private:
 	u32 m_commited;
 	u32 m_reserved;
-	u32 m_end;
+	volatile i32 m_end;
 	u8* m_mem;
+	Mutex m_mutex;
 };
 
 // only single allocation, can be used by Array<T> to allocate on stack

src/renderer/pipeline.cpp

@@ -682,7 +682,7 @@ struct PipelineImpl final : Pipeline
 		
 		static_assert(sizeof(Page) == PageAllocator::PAGE_SIZE);
 
-		AutoInstancer(IAllocator& allocator, PageAllocator& page_allocator)
+		AutoInstancer(LinearAllocator& allocator, PageAllocator& page_allocator)
 			: instances(allocator)
 			, page_allocator(page_allocator)
 		{
@@ -806,7 +806,7 @@ struct PipelineImpl final : Pipeline
 	};
 
 	struct View {
-		View(IAllocator& allocator, PageAllocator& page_allocator) 
+		View(LinearAllocator& allocator, PageAllocator& page_allocator) 
 			: sorter(allocator, page_allocator)
 			, instancers(allocator)
 			, buckets(allocator)
@@ -1459,8 +1459,8 @@ struct PipelineImpl final : Pipeline
 				pipeline->m_renderer.endProfileBlock();
 				while (instanced_meshes) {
 					InstancedMeshes* i = instanced_meshes;
+					i->~InstancedMeshes();
 					instanced_meshes = instanced_meshes->next;
-					LUMIX_DELETE(*allocator, i);
 				}
 			}
 
@@ -2696,12 +2696,12 @@ struct PipelineImpl final : Pipeline
 
 	// TODO optimize
 	struct FillClustersJob : Renderer::RenderJob {
-		FillClustersJob(IAllocator& allocator)
-			: m_clusters(allocator)
-			, m_map(allocator)
+		FillClustersJob(LinearAllocator& current_frame_allocator, IAllocator& allocator)
+			: m_clusters(current_frame_allocator)
+			, m_map(current_frame_allocator)
 			, m_point_lights(allocator)
-			, m_env_probes(allocator)
-			, m_refl_probes(allocator)
+			, m_env_probes(current_frame_allocator)
+			, m_refl_probes(current_frame_allocator)
 		{}
 
 		void setup() override {
@@ -2772,6 +2772,7 @@ struct PipelineImpl final : Pipeline
 
 			const Span<const ReflectionProbe> scene_refl_probes = scene->getReflectionProbes();
 			const Span<EntityRef> refl_probe_entities = scene->getReflectionProbesEntities();
+			refl_probes.reserve(scene_refl_probes.length());
 			for (u32 i = 0, c = scene_refl_probes.length(); i < c; ++i) {
 				const ReflectionProbe& refl_probe = scene_refl_probes[i];
 				if (!refl_probe.flags.isSet(ReflectionProbe::ENABLED)) continue;
@@ -2795,6 +2796,7 @@ struct PipelineImpl final : Pipeline
 
 			const Span<const EnvironmentProbe> scene_env_probes = scene->getEnvironmentProbes();
 			const Span<EntityRef> env_probe_entities = scene->getEnvironmentProbesEntities();
+			env_probes.reserve(scene_env_probes.length());
 			for (u32 i = 0, c = scene_env_probes.length(); i < c; ++i) {
 					const EnvironmentProbe& env_probe = scene_env_probes[i];
 					if (!env_probe.flags.isSet(EnvironmentProbe::ENABLED)) continue;
@@ -3347,7 +3349,10 @@ struct PipelineImpl final : Pipeline
 		PROFILE_FUNCTION();
 		Array<UniquePtr<View>>& views = pipeline->m_views;
 		UniquePtr<View>& view = views.emplace();
-		view = UniquePtr<View>::create(pipeline->m_allocator, pipeline->m_allocator, pipeline->m_renderer.getEngine().getPageAllocator());
+		LinearAllocator& allocator = pipeline->m_renderer.getCurrentFrameAllocator();
+		view = UniquePtr<View>::create(allocator
+			, allocator
+			, pipeline->m_renderer.getEngine().getPageAllocator());
 		view->cp = cp;
 		if (views.size() > 1) {
 			views[views.size() - 2]->instanced_meshes->next = view->instanced_meshes;
@@ -3383,13 +3388,7 @@ struct PipelineImpl final : Pipeline
 			view->layer_to_bucket[bucket.layer] = i;
 		}
 
-		struct JobData {
-			PipelineImpl* pipeline;
-			u32 view_id;
-			IAllocator* allocator;
-		};
-
-		PrepareViewJob& job = pipeline->m_renderer.createJob<PrepareViewJob>(pipeline->m_allocator);
+		PrepareViewJob& job = pipeline->m_renderer.createJob<PrepareViewJob>(allocator);
 		job.m_pipeline = pipeline;
 		job.m_camera_params = cp;
 		job.m_view = view.get();
@@ -4110,7 +4109,8 @@ struct PipelineImpl final : Pipeline
 
 	void fillClusters(LuaWrapper::Optional<CameraParams> cp) {
 		PROFILE_FUNCTION();
-		FillClustersJob& job = m_renderer.createJob<FillClustersJob>(m_allocator);
+		LinearAllocator& current_frame_allocator = m_renderer.getCurrentFrameAllocator();
+		FillClustersJob& job = m_renderer.createJob<FillClustersJob>(current_frame_allocator, m_allocator);
 		
 		job.m_pipeline = this;
 		if (cp.valid) {
@@ -4641,8 +4641,9 @@ struct PipelineImpl final : Pipeline
 		PagedListIterator<const CullResult> iterator(view.renderables);
 
 		view.instancers.reserve(jobs::getWorkersCount());
+		LinearAllocator& allocator = m_renderer.getCurrentFrameAllocator();
 		for (u8 i = 0; i < jobs::getWorkersCount(); ++i) {
-			view.instancers.emplace(m_allocator, m_renderer.getEngine().getPageAllocator());
+			view.instancers.emplace(allocator, m_renderer.getEngine().getPageAllocator());
 		}
 
 		const float time_delta = m_renderer.getEngine().getLastTimeDelta();

src/renderer/renderer.cpp

@@ -1287,6 +1287,8 @@ struct RendererImpl final : Renderer
 			jobs::setGreen(&m_gpu_frame->can_setup);
 		}
 	}
+	
+	LinearAllocator& getCurrentFrameAllocator() { return m_cpu_frame->job_allocator; }
 
 	void waitForCommandSetup() override
 	{

src/renderer/renderer.h

@@ -109,6 +109,8 @@ struct LUMIX_RENDERER_API Renderer : IPlugin {
 		deallocJob(&job);
 	}
 
+	virtual struct LinearAllocator& getCurrentFrameAllocator() = 0;
+
 protected:
 	virtual void* allocJob(u32 size, u32 align) = 0;
 	virtual void deallocJob(void* ptr) = 0;