debug about getting ptr device

Author: zhangxiaoli73

src/ATen/native/xpu/Copy.cpp

@@ -71,8 +71,10 @@ void memcpyAsync(
   Device dst_device = iter.device(0);
   Device src_device = iter.device(1);
   if (dst_device == src_device) {
+    std::cout << "zl_debug: go to same device and specialized kernel" << std::endl;
     copy_kernel(iter);
   } else {
+    std::cout << "zl_debug: go to sycl copy kernel" << std::endl;
     TORCH_INTERNAL_ASSERT(p2p_enabled == true);
     auto dst = (char*)iter.data_ptr(0);
     auto src = (char*)iter.data_ptr(1);

src/xccl/XPUSymmetricMemory.cpp

@@ -132,7 +132,9 @@ at::Tensor XPUSymmetricMemory::get_buffer(
       " bytes)");
   auto data_ptr = reinterpret_cast<uint8_t*>(buffers_[rank]) +
       storage_offset * element_size;
-  auto device = c10::Device(c10::DeviceType::XPU, local_device_idx_);
+  // check the device of this device buffer
+  auto ptr_to_device_id = c10::xpu::get_device_idx_from_pointer(data_ptr);
+  auto device = c10::Device(c10::DeviceType::XPU, ptr_to_device_id);
   auto options = at::TensorOptions().dtype(dtype).device(device);
 
   return at::for_blob(data_ptr, sizes)