vkd3d: Add implementation of a queue timeline.

The purpose of this is to get a high-level glance of:

- CPU / GPU overlap
- Latency
- Being able to correlate performance against events which may lead to
  stutter such as resource allocations, etc.

The format is the trivial JSON format supported by Chromium
chrome://tracing.

Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>

libs/vkd3d/device.c

@@ -3278,6 +3278,7 @@ static void d3d12_device_destroy(struct d3d12_device *device)
 
     vkd3d_cleanup_format_info(device);
     vkd3d_memory_info_cleanup(&device->memory_info, device);
+    vkd3d_queue_timeline_trace_cleanup(&device->queue_timeline_trace);
     vkd3d_shader_debug_ring_cleanup(&device->debug_ring, device);
 #ifdef VKD3D_ENABLE_BREADCRUMBS
     vkd3d_breadcrumb_tracer_cleanup_barrier_hashes(&device->breadcrumb_tracer);
@@ -8296,13 +8297,16 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
     if (FAILED(hr = vkd3d_shader_debug_ring_init(&device->debug_ring, device)))
         goto out_cleanup_meta_ops;
 
+    if (FAILED(hr = vkd3d_queue_timeline_trace_init(&device->queue_timeline_trace)))
+        goto out_cleanup_debug_ring;
+
     vkd3d_scratch_pool_init(device);
 
 #ifdef VKD3D_ENABLE_BREADCRUMBS
     vkd3d_breadcrumb_tracer_init_barrier_hashes(&device->breadcrumb_tracer);
     if (vkd3d_config_flags & VKD3D_CONFIG_FLAG_BREADCRUMBS)
         if (FAILED(hr = vkd3d_breadcrumb_tracer_init(&device->breadcrumb_tracer, device)))
-            goto out_cleanup_debug_ring;
+            goto out_cleanup_queue_timeline_trace;
 #endif
 
     if (vkd3d_descriptor_debug_active_qa_checks())
@@ -8349,9 +8353,11 @@ static HRESULT d3d12_device_init(struct d3d12_device *device,
 #ifdef VKD3D_ENABLE_BREADCRUMBS
     if (vkd3d_config_flags & VKD3D_CONFIG_FLAG_BREADCRUMBS)
         vkd3d_breadcrumb_tracer_cleanup(&device->breadcrumb_tracer, device);
-out_cleanup_debug_ring:
+out_cleanup_queue_timeline_trace:
     vkd3d_breadcrumb_tracer_cleanup_barrier_hashes(&device->breadcrumb_tracer);
+    vkd3d_queue_timeline_trace_cleanup(&device->queue_timeline_trace);
 #endif
+out_cleanup_debug_ring:
     vkd3d_shader_debug_ring_cleanup(&device->debug_ring, device);
 out_cleanup_meta_ops:
     vkd3d_meta_ops_cleanup(&device->meta_ops, device);

libs/vkd3d/heap.c

@@ -317,6 +317,9 @@ static HRESULT d3d12_heap_init(struct d3d12_heap *heap, struct d3d12_device *dev
         heap->allocation.chunk == NULL /* not suballocated */ &&
         (device->memory_properties.memoryTypes[heap->allocation.device_allocation.vk_memory_type].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
 
+    vkd3d_queue_timeline_trace_register_instantaneous(&device->queue_timeline_trace,
+            VKD3D_QUEUE_TIMELINE_TRACE_STATE_TYPE_HEAP_ALLOCATION, desc->SizeInBytes);
+
     d3d12_device_add_ref(heap->device);
     return S_OK;
 }

libs/vkd3d/memory.c

@@ -448,7 +448,7 @@ static HRESULT vkd3d_memory_transfer_queue_flush_locked(struct vkd3d_memory_tran
             vkd3d_queue_add_wait(queue_family->queues[i],
                     NULL,
                     queue->vk_semaphore,
-                    queue->next_signal_value);
+                    queue->next_signal_value, 0);
         }
     }
 

libs/vkd3d/meson.build

@@ -62,7 +62,8 @@ vkd3d_src = [
   'vkd3d_main.c',
   'raytracing_pipeline.c',
   'acceleration_structure.c',
-  'swapchain.c'
+  'swapchain.c',
+  'queue_timeline.c'
 ]
 
 if enable_renderdoc