Fix #133953: Cycles oneAPI texture randomly renders black

* Do oneAPI copy optimization as part of host memory alloc and free, so
  it is properly released before host memory is freed.
* Synchronize after loading texture info, like CUDA and HIP.

https://projects.blender.org/blender/blender/pulls/134412

Author: brechtvl

intern/cycles/device/oneapi/device_impl.cpp

@@ -379,6 +379,35 @@ void *OneapiDevice::kernel_globals_device_pointer()
   return kg_memory_device_;
 }
 
+void *OneapiDevice::host_alloc(const MemoryType type, const size_t size)
+{
+  void *host_pointer = GPUDevice::host_alloc(type, size);
+
+#  ifdef SYCL_EXT_ONEAPI_COPY_OPTIMIZE
+  if (host_pointer) {
+    /* Import host_pointer into USM memory for faster host<->device data transfers. */
+    if (type == MEM_READ_WRITE || type == MEM_READ_ONLY) {
+      sycl::queue *queue = reinterpret_cast<sycl::queue *>(device_queue_);
+      sycl::ext::oneapi::experimental::prepare_for_device_copy(host_pointer, size, *queue);
+    }
+  }
+#  endif
+
+  return host_pointer;
+}
+
+void OneapiDevice::host_free(const MemoryType type, void *host_pointer, const size_t size)
+{
+#  ifdef SYCL_EXT_ONEAPI_COPY_OPTIMIZE
+  if (type == MEM_READ_WRITE || type == MEM_READ_ONLY) {
+    sycl::queue *queue = reinterpret_cast<sycl::queue *>(device_queue_);
+    sycl::ext::oneapi::experimental::release_from_device_copy(host_pointer, *queue);
+  }
+#  endif
+
+  GPUDevice::host_free(type, host_pointer, size);
+}
+
 void OneapiDevice::mem_alloc(device_memory &mem)
 {
   if (mem.type == MEM_TEXTURE) {
@@ -394,14 +423,6 @@ void OneapiDevice::mem_alloc(device_memory &mem)
                  << string_human_readable_size(mem.memory_size()) << ")";
     }
     generic_alloc(mem);
-#  ifdef SYCL_EXT_ONEAPI_COPY_OPTIMIZE
-    /* Import host_pointer into USM memory for faster host<->device data transfers. */
-    if (mem.type == MEM_READ_WRITE || mem.type == MEM_READ_ONLY) {
-      sycl::queue *queue = reinterpret_cast<sycl::queue *>(device_queue_);
-      sycl::ext::oneapi::experimental::prepare_for_device_copy(
-          mem.host_pointer, mem.memory_size(), *queue);
-    }
-#  endif
   }
 }
 
@@ -543,12 +564,6 @@ void OneapiDevice::mem_free(device_memory &mem)
     tex_free((device_texture &)mem);
   }
   else {
-#  ifdef SYCL_EXT_ONEAPI_COPY_OPTIMIZE
-    if (mem.type == MEM_READ_WRITE || mem.type == MEM_READ_ONLY) {
-      sycl::queue *queue = reinterpret_cast<sycl::queue *>(device_queue_);
-      sycl::ext::oneapi::experimental::release_from_device_copy(mem.host_pointer, *queue);
-    }
-#  endif
     generic_free(mem);
   }
 }

intern/cycles/device/oneapi/device_impl.h

@@ -88,6 +88,10 @@ class OneapiDevice : public GPUDevice {
   void tex_copy_to(device_texture &mem);
   void tex_free(device_texture &mem);
 
+  /* Host side memory, override for more efficient copies. */
+  void *host_alloc(const MemoryType type, const size_t size) override;
+  void host_free(const MemoryType type, void *host_pointer, const size_t size) override;
+
   /* Device side memory. */
   void get_device_memory_info(size_t &total, size_t &free) override;
   bool alloc_device(void *&device_pointer, const size_t size) override;

intern/cycles/device/oneapi/queue.cpp

@@ -72,7 +72,7 @@ bool OneapiDeviceQueue::enqueue(DeviceKernel kernel,
 
   /* Update texture info in case memory moved to host. */
   if (oneapi_device_->load_texture_info()) {
-    if (oneapi_device_->have_error()) {
+    if (!synchronize()) {
       return false;
     }
   }