Remove some locks from progress update

Removes mutex locks when updating pixel counts etc. Also, when
in background mode text status updates are removed.

@@ -887,7 +887,6 @@ typedef enum  {
 
 typedef struct _nvrtcProgram* nvrtcProgram;
 
-

@@ -681,7 +682,6 @@ static int cuewNvrtcInit(void) {
   return result;
 }
 
-

@@ -68,6 +68,7 @@ typedef void* DynamicLibrary;
 
 static DynamicLibrary cuda_lib;
 static DynamicLibrary nvrtc_lib;
+static DynamicLibrary nvtx_lib;

@@ -125,6 +125,7 @@ void BlenderSession::create_session()
   session = new Session(session_params, scene_params);
   session->progress.set_update_callback(function_bind(&BlenderSession::tag_redraw, this));
   session->progress.set_cancel_callback(function_bind(&BlenderSession::test_cancel, this));
+  session->progress.set_updates(!background);

@@ -14,1 +13,4 @@
+                   const Device *device_)
+    : BVH(params_, geometry_, objects_), device(device_)
 {
+  // Resize the sub-bvh container to match the number of devices

@@ -149,6 +149,15 @@ void CPUDevice::mem_copy_to(device_memory &mem)
   }
 }
 
+void CPUDevice::mem_copy_to(device_memory &mem, size_t, size_t offset)

@@ -31,0 +30,4 @@
+
+  int flags = CUEW_INIT_CUDA;
+
+  int cuew_result = cuewInit(flags);

@@ -548,3 +548,3 @@
   const CUDAContextScope scope(this);
 
-  cuda_assert(cuMemcpyHtoD((CUdeviceptr)device_pointer, host_pointer, size));
+  cuda_assert(cuMemcpyHtoD((CUdeviceptr)device_pointer + offset,

@@ -551,0 +554,4 @@
+
+void CUDADevice::host_mem_alloc(size_t size, int aligment, void **p_mem) {
+  CUDAContextScope scope(this);
+  CUresult result = cuMemAllocHost(p_mem, size);

@@ -551,0 +555,4 @@
+void CUDADevice::host_mem_alloc(size_t size, int aligment, void **p_mem) {
+  CUDAContextScope scope(this);
+  CUresult result = cuMemAllocHost(p_mem, size);
+  if(result != CUDA_SUCCESS) {

@@ -566,8 +584,13 @@ void CUDADevice::mem_alloc(device_memory &mem)
 void CUDADevice::mem_copy_to(device_memory &mem)

@@ -568,3 +586,2 @@
   if (mem.type == MEM_GLOBAL) {
-    global_free(mem);
-    global_alloc(mem);
+    if ((mem.device_size < mem.memory_size()) || (!mem.device_pointer)) {

@@ -584,0 +607,4 @@
+void CUDADevice::mem_copy_to(device_memory &mem, size_t size, size_t offset)
+{
+  if (mem.type == MEM_GLOBAL) {
+    if ((mem.device_size < mem.memory_size()) || (!mem.device_pointer)) {

@@ -766,0 +774,4 @@
+  }
+}
+
+void GPUDevice::generic_copy_to(device_memory &mem, size_t size, size_t offset)

@@ -8,3 +8,3 @@
 
 #include "bvh/params.h"
-
+#include "scene/scene.h"

@@ -217,2 +222,4 @@
     return 0;
   }
+  
+  virtual device_ptr find_matching_mem(device_ptr key, Device * /*sub*/)

@@ -290,8 +300,11 @@ class Device {
   friend class DeviceServer;
   friend class device_memory;
 
+  virtual void host_mem_alloc(size_t size, int alignment, void **p_mem);

@@ -369,17 +379,78 @@ template<typename T> class device_vector : public device_memory {
     assert(data_elements > 0);
   }
 
+  device_vector(Device *device,

@@ -41,3 +45,3 @@
       /* Always add CPU devices at the back since GPU devices can change
        * host memory pointers, which CPU uses as device pointer. */
-      SubDevice *sub;
+      SubDevice *sub = new SubDevice;

@@ -49,2 +52,2 @@
-        devices.emplace_front();
-        sub = &devices.front();
+        devices.emplace_back(sub);
+        // find first CPU device and swop it with the new device

@@ -51,0 +54,4 @@
+        // to keep the CPU devices at the end of the vector.
+        int last = devices.size() - 1;
+        int o = last;
+        while ((o > 0) && (devices[o - 1]->device->info.type == DEVICE_CPU)) {

@@ -1041,2 +637,2 @@
-    }
-  }
+  /* copy svm attributes to device */
+  dscene->attributes_map.copy_to_device_if_modified();

@@ -2088,3 +1487,1 @@
-    if (geom->is_mesh()) {
-      Mesh *mesh = static_cast<Mesh *>(geom);
-      mesh->subd_attributes.clear_modified();
+    size_t n_scenes = scene->dscenes.size();

@@ -2091,0 +1487,4 @@
+    size_t n_scenes = scene->dscenes.size();
+    // Parallel upload the geometry data to the devices and
+    // calculate or refit the BVHs
+    tbb::parallel_for(size_t(0),

@@ -2091,0 +1489,4 @@
+    // calculate or refit the BVHs
+    tbb::parallel_for(size_t(0),
+                      n_scenes,
+                      [this,

@@ -162,0 +185,4 @@
+  size_t *motion_vert_offsets;
+};
+
+struct AttributeSizes {

@@ -0,0 +1,1045 @@
+/* SPDX-License-Identifier: Apache-2.0

@@ -0,0 +34,4 @@
+/*
+ * Clears all tags used to indicate the the shader needs to be updated.
+ */
+void GeometryManager::clearShaderUpdateTags(Scene *scene)

@@ -0,0 +121,4 @@
+
+  progress.set_status("Updating Mesh", "Computing attributes");
+
+  // SHOULD NOT ALLOC ONLY ALLOC IF MORE SPACE IS NEEDED

@@ -660,11 +660,8 @@ void ObjectManager::device_update_transforms(DeviceScene *dscene, Scene *scene,
                  }
                });
 
-  if (progress.get_cancel()) {

@@ -36,0 +37,4 @@
+ * checks the progress for if a cancel has been requested and also
+ * the device to see if an error has occurred.
+ */
+bool Scene::progressErrorCheck(Progress &progress, Device *device) {

@@ -36,0 +39,4 @@
+ */
+bool Scene::progressErrorCheck(Progress &progress, Device *device) {
+  bool status = false;
+  if (!background && progress.get_updates()) {

@@ -104,1 +116,4 @@
 {
+  /* Create a DeviceScene for each device */
+  device->foreach_device([this](Device *sub_device) {
+    DeviceScene *sub_dscene = new DeviceScene(sub_device);

@@ -107,0 +127,4 @@
+  mesh_times = new double[device_count];
+  attrib_times = new double[device_count];
+  object_bvh_times = new double[device_count];
+  scene_bvh_times = new double[device_count];

@@ -17,3 +28,4 @@
 
 CCL_NAMESPACE_BEGIN
 
+class SpinLock {

@@ -41,2 +93,4 @@
     error_message = "";
     cancel_cb = function_null;
+    updates = true;
+    //    update_call = true;

@@ -14,1 +13,4 @@
+                   const Device *device_)
+    : BVH(params_, geometry_, objects_), device(device_)
 {
+  // Resize the sub-bvh container to match the number of devices

@@ -548,3 +548,3 @@
   const CUDAContextScope scope(this);
 
-  cuda_assert(cuMemcpyHtoD((CUdeviceptr)device_pointer, host_pointer, size));
+  cuda_assert(cuMemcpyHtoD((CUdeviceptr)device_pointer + offset,

@@ -551,0 +554,4 @@
+
+void *CUDADevice::host_mem_alloc(size_t size, int aligment) {
+  void *p_mem = NULL;
+  CUDAContextScope scope(this);

@@ -581,6 +607,29 @@ void CUDADevice::mem_copy_to(device_memory &mem)
   }
 }
 
+void CUDADevice::mem_copy_to(device_memory &mem, size_t size, size_t offset)

@@ -8,3 +8,3 @@
 
 #include "bvh/params.h"
-
+#include "scene/scene.h"

@@ -377,4 +449,2 @@
   {
     size_t new_size = size(width, height, depth);
-
-    if (new_size != data_size) {

@@ -369,1 +379,4 @@
 
+  device_vector(Device *device,
+                const char *name,
+                void *p_mem,

@@ -833,0 +856,4 @@
+    }
+    else {
+      generic_copy_to(mem, size, offset);
+    }

@@ -41,3 +45,3 @@
       /* Always add CPU devices at the back since GPU devices can change
        * host memory pointers, which CPU uses as device pointer. */
-      SubDevice *sub;
+      SubDevice *sub = new SubDevice;

@@ -51,0 +54,4 @@
+        // to keep the CPU devices at the end of the vector.
+        int last = devices.size() - 1;
+        int o = last;
+        while ((o > 0) && (devices[o - 1]->device->info.type == DEVICE_CPU)) {

@@ -162,0 +185,4 @@
+  size_t *motion_vert_offsets;
+};
+
+struct AttributeSizes {

@@ -36,0 +37,4 @@
+ * checks the progress for if a cancel has been requested and also
+ * the device to see if an error has occurred.
+ */
+bool Scene::progressErrorCheck(Progress &progress, Device *device) {

@@ -36,0 +39,4 @@
+ */
+bool Scene::progressErrorCheck(Progress &progress, Device *device) {
+  bool status = false;
+  if (!background && progress.get_updates()) {

@@ -107,0 +127,4 @@
+  mesh_times = new double[device_count];
+  attrib_times = new double[device_count];
+  object_bvh_times = new double[device_count];
+  scene_bvh_times = new double[device_count];

@@ -17,3 +28,4 @@
 
 CCL_NAMESPACE_BEGIN
 
+class SpinLock {

@@ -41,2 +93,4 @@
     error_message = "";
     cancel_cb = function_null;
+    updates = true;
+    //    update_call = true;

@@ -0,0 +1,407 @@
+/* SPDX-License-Identifier: Apache-2.0

@@ -0,0 +34,4 @@
+/*
+ * Clears all tags used to indicate the the shader needs to be updated.
+ */
+void GeometryManager::clearShaderUpdateTags(Scene *scene)

@@ -0,0 +121,4 @@
+      sub_dscene->tri_verts.assign_mem(dscene->tri_verts);
+      sub_dscene->tri_verts.tag_modified();
+    }
+    {

@@ -125,6 +125,7 @@ void BlenderSession::create_session()
   session = new Session(session_params, scene_params);
   session->progress.set_update_callback(function_bind(&BlenderSession::tag_redraw, this));
   session->progress.set_cancel_callback(function_bind(&BlenderSession::test_cancel, this));
+  session->progress.set_updates(!headless);

@@ -66,3 +66,3 @@
   vector<Geometry *> geometry;
   vector<Object *> objects;
-
+  bool built;

@@ -24,0 +36,4 @@
+{
+  int id = device->device_number(subdevice);
+  resize_sub_bvhs_if_needed(id);
+  sub_bvhs[id] = bvh;

@@ -54,3 +53,2 @@
   BVH2 *const bvh2 = static_cast<BVH2 *>(bvh);
-  if (refit) {
-    bvh2->refit(progress);
+  static std::atomic<int> building{0};

@@ -55,2 +54,2 @@
-  if (refit) {
-    bvh2->refit(progress);
+  static std::atomic<int> building{0};
+  /* Top level BVH2 build must wait on all other BVH2 builds to finish

@@ -531,3 +536,2 @@
   if (mem.type == MEM_GLOBAL) {
-    global_free(mem);
-    global_alloc(mem);
+    if ((mem.device_size < mem.memory_size()) || (!mem.device_pointer)) {

@@ -0,0 +90,4 @@
+}
+
+/*
+ * Clears the modified tags for all elements of the device scene

@@ -0,0 +128,4 @@
+                                              DeviceScene *dscene,
+                                              const GeometrySizes *p_sizes)
+{
+  if (p_sizes->tri_size != 0) {

@@ -630,3 +630,3 @@
 }
 
-double Session::get_estimated_remaining_time() const
+double Session::get_estimated_remaining_time() //const

@@ -7,0 +11,4 @@
+#else
+#define MUTEX              thread_mutex
+#define SCOPED_LOCK(mutex) thread_scoped_lock scopedlock(mutex)
+#endif