Fix #137382: Vulkan: Memory leak when rendering

Each time when rendering an image a new context is created. When the
context is destroyed it can still contain a render graph, which ownership
isn't transferred back to the device. Resulting in an increase of several
MB per render. The render graph is cleanly destroyed during quiting as
there is a master list of created render graphs.

Fixed by moving the ownership of the render graph back to the device
when a context is unregistered.

Pull Request: https://projects.blender.org/blender/blender/pulls/138095

source/blender/gpu/vulkan/vk_context.hh

@@ -27,6 +27,7 @@ class VKBatch;
 class VKStateManager;
 class VKShader;
 class VKThreadData;
+class VKDevice;
 
 enum RenderGraphFlushFlags {
   NONE = 0,
@@ -37,6 +38,8 @@ enum RenderGraphFlushFlags {
 ENUM_OPERATORS(RenderGraphFlushFlags, RenderGraphFlushFlags::WAIT_FOR_COMPLETION);
 
 class VKContext : public Context, NonCopyable {
+  friend class VKDevice;
+
  private:
   VkExtent2D vk_extent_ = {};
   VkSurfaceFormatKHR swap_chain_format_ = {};

source/blender/gpu/vulkan/vk_device.cc

@@ -464,6 +464,14 @@ void VKDevice::context_register(VKContext &context)
 
 void VKDevice::context_unregister(VKContext &context)
 {
+  if (context.render_graph_.has_value()) {
+    render_graph::VKRenderGraph &render_graph = context.render_graph();
+    context.render_graph_.reset();
+    BLI_assert_msg(render_graph.is_empty(),
+                   "Unregistering a context that still has an unsubmitted render graph.");
+    render_graph.reset();
+    BLI_thread_queue_push(unused_render_graphs_, &render_graph);
+  }
   orphaned_data.move_data(context.discard_pool, timeline_value_ + 1);
   contexts_.remove(contexts_.first_index_of(std::reference_wrapper(context)));
 }