tu: Add MESA_TRACE_FUNC to submit and BO alloc paths

Makes much easier to see how slow are submits and BO allocs,
which sometimes happen together.

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35637>

src/freedreno/vulkan/tu_device.cc

@@ -2234,6 +2234,7 @@ tu_u_trace_submission_data_create(
    uint32_t cmd_buffer_count,
    struct tu_u_trace_submission_data **submission_data)
 {
+   MESA_TRACE_FUNC();
    *submission_data = (struct tu_u_trace_submission_data *)
       vk_zalloc(&device->vk.alloc,
                 sizeof(struct tu_u_trace_submission_data), 8,

src/freedreno/vulkan/tu_knl.cc

@@ -37,6 +37,7 @@ tu_bo_init_new_explicit_iova(struct tu_device *dev,
                              VkMemoryPropertyFlags mem_property,
                              enum tu_bo_alloc_flags flags, const char *name)
 {
+   MESA_TRACE_FUNC();
    struct tu_instance *instance = dev->physical_device->instance;
 
    VkResult result =
@@ -88,6 +89,7 @@ tu_bo_export_dmabuf(struct tu_device *dev, struct tu_bo *bo)
 void
 tu_bo_finish(struct tu_device *dev, struct tu_bo *bo)
 {
+   MESA_TRACE_FUNC();
    struct tu_instance *instance = dev->physical_device->instance;
 
    vk_address_binding_report(&instance->vk, bo->base ? bo->base : &dev->vk.base,

src/freedreno/vulkan/tu_knl_drm_msm.cc

@@ -351,6 +351,7 @@ tu_wait_fence(struct tu_device *dev,
               int fence,
               uint64_t timeout_ns)
 {
+   MESA_TRACE_FUNC();
    /* fence was created when no work was yet submitted */
    if (fence < 0)
       return VK_SUCCESS;
@@ -390,6 +391,8 @@ tu_free_zombie_vma_locked(struct tu_device *dev, bool wait)
    if (!u_vector_length(&dev->zombie_vmas))
       return VK_SUCCESS;
 
+   MESA_TRACE_FUNC();
+
    if (wait) {
       struct tu_zombie_vma *vma = (struct tu_zombie_vma *)
             u_vector_head(&dev->zombie_vmas);
@@ -654,6 +657,7 @@ msm_bo_init(struct tu_device *dev,
             enum tu_bo_alloc_flags flags,
             const char *name)
 {
+   MESA_TRACE_FUNC();
    struct drm_msm_gem_new req = {
       .size = size,
       .flags = 0
@@ -946,9 +950,12 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
       .syncobj_stride = sizeof(struct drm_msm_gem_submit_syncobj),
    };
 
-   ret = drmCommandWriteRead(queue->device->fd,
-                             DRM_MSM_GEM_SUBMIT,
-                             &req, sizeof(req));
+   {
+      MESA_TRACE_SCOPE("DRM_MSM_GEM_SUBMIT");
+      ret = drmCommandWriteRead(queue->device->fd,
+                              DRM_MSM_GEM_SUBMIT,
+                              &req, sizeof(req));
+   }
 
    mtx_unlock(&queue->device->bo_mutex);
 

src/freedreno/vulkan/tu_queue.cc

@@ -66,6 +66,7 @@ submit_add_entries(struct tu_device *dev, void *submit,
 static VkResult
 queue_submit(struct vk_queue *_queue, struct vk_queue_submit *vk_submit)
 {
+   MESA_TRACE_FUNC();
    struct tu_queue *queue = list_entry(_queue, struct tu_queue, vk);
    struct tu_device *device = queue->device;
    bool u_trace_enabled = u_trace_should_process(&queue->device->trace_context);