fix memory profile build (#16177)

### Description
<!-- Describe your changes. -->
This PR is to fix the build break when onnxruntime_ENABLE_MEMORY_PROFILE
is on


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
This PR is to fix the build break when onnxruntime_ENABLE_MEMORY_PROFILE
is on.
It fixes this issue
#16124

Co-authored-by: Lei Cao <[email protected]>

onnxruntime/core/framework/execution_frame.cc

@@ -421,14 +421,14 @@ ExecutionFrame::ExecutionFrame(gsl::span<const int> feed_mlvalue_idxs, gsl::span
               // Planning of one memory type should only happen once.
 #if !defined(ORT_MINIMAL_BUILD) && defined(ORT_MEMORY_PROFILE)
               ORT_ENFORCE(
-                  static_activation_memory_sizes_in_byte_.find(location.name) ==
+                  static_activation_memory_sizes_in_byte_.find(location.ToString()) ==
                       static_activation_memory_sizes_in_byte_.end(),
                   "Memory type ",
-                  location.name,
+                  location.ToString(),
                   " should only appear once.");
               // static_activation_memory_in_bytes_ is max virtual memory size the planner computes.
               // Memory dynamically allocated when executing kernels is not recorded using this field.
-              static_activation_memory_sizes_in_byte_[location.name] = peak_size;
+              static_activation_memory_sizes_in_byte_[location.ToString()] = peak_size;
 #endif
               // the memory pattern buffer will leave in the whole execution.
 #ifdef ORT_ENABLE_STREAM
@@ -609,7 +609,7 @@ Status ExecutionFrame::AllocateMLValueTensorSelfOwnBufferHelper(OrtValue& ort_va
     // Dynamic activation size would be accessed by multiple threads
     // if parallel executor is used.
     std::unique_lock<std::mutex> lock(mtx_);
-    dynamic_activation_memory_sizes_in_byte_[location.name] += size;
+    dynamic_activation_memory_sizes_in_byte_[location.ToString()] += size;
     session_state_.GetMemoryProfiler()->GetMemoryInfo().SetDynamicAllocation(ort_value_index);
 #endif
   }

onnxruntime/core/framework/memory_info.cc

@@ -130,7 +130,7 @@ void PrintInforPerTensor(const MemoryInfo::AllocInfoPerTensor& alloc_info, const
   std::cout << "Index: " << alloc_info.mlvalue_index << ", ";
   std::cout << "Reuse inplace: " << alloc_info.inplace_reuse << ", ";
   std::cout << "Alloc type: " << alloc_info.alloc_kind << ", ";
-  std::cout << "Location: " << alloc_info.location.name << ", ";
+  std::cout << "Location: " << alloc_info.location.ToString() << ", ";
   std::cout << "lifetime: (" << alloc_info.lifetime_interval.first << ", " << alloc_info.lifetime_interval.second << "), ";
   std::cout << "planned block: (" << mem_info.planned_block.offset_ << ", "
             << (mem_info.planned_block.offset_ + mem_info.planned_block.size_) << "), ";
@@ -142,24 +142,24 @@ void PrintInforPerTensor(const MemoryInfo::AllocInfoPerTensor& alloc_info, const
 
 void MemoryInfo::PrintMemoryInfoForLocation(const OrtDevice::DeviceType location) {
   for (const auto& map : tensors_memory_info_map_) {
-    if (map.first.device.Type() != location) continue;
-    std::cout << "Initializer in " << map.first.name << "\n";
+    if (map.first.Type() != location) continue;
+    std::cout << "Initializer in " << map.first.ToString() << "\n";
     const auto& initailizer_map = map.second.at(MapType::Initializer);
     for (const auto& item : initailizer_map) {
-      if (AllocPlan(item.first)->location.device.Type() != location) continue;
+      if (AllocPlan(item.first)->location.Type() != location) continue;
       PrintInforPerTensor(*AllocPlan(item.first), item.second, initailizer_map.GetAllocAddress(item.first));
     }
-    std::cout << "\nStatic Activation in " << map.first.name << "\n";
+    std::cout << "\nStatic Activation in " << map.first.ToString() << "\n";
     const auto& static_activation_map = map.second.at(MapType::StaticActivation);
     for (const auto& item : static_activation_map) {
-      if (AllocPlan(item.first)->location.device.Type() != location) continue;
+      if (AllocPlan(item.first)->location.Type() != location) continue;
       PrintInforPerTensor(*AllocPlan(item.first), item.second, static_activation_map.GetAllocAddress(item.first));
     }
 
-    std::cout << "\nDynamic Activation in " << map.first.name << "\n";
+    std::cout << "\nDynamic Activation in " << map.first.ToString() << "\n";
     const auto& dynamic_activation_map = map.second.at(MapType::DynamicActivation);
     for (const auto& item : dynamic_activation_map) {
-      if (AllocPlan(item.first)->location.device.Type() != location) continue;
+      if (AllocPlan(item.first)->location.Type() != location) continue;
       PrintInforPerTensor(*AllocPlan(item.first), item.second, dynamic_activation_map.GetAllocAddress(item.first));
     }
   }
@@ -273,10 +273,10 @@ void MemoryProfiler::CreateEvents(const std::string& p_name,
   // Create Event for each tensor
   auto& time_step_trace = GetMemoryInfo().time_step_trace_;
   for (const auto& location_map : GetMemoryInfo().tensors_memory_info_map_) {
-    const OrtMemoryInfo& memory_info = location_map.first;
+    const OrtDevice& memory_info = location_map.first;
     const auto& maptype_to_map_mapping = location_map.second;
 
-    if (memory_info.device.Type() != device_type) continue;
+    if (memory_info.Type() != device_type) continue;
 
     // If there is no tensor of a map_type, we skip creating event for that map_type
     if (maptype_to_map_mapping.find(map_type) == maptype_to_map_mapping.end()) continue;

onnxruntime/core/framework/memory_info.h

@@ -122,7 +122,7 @@ class MemoryInfo {
     bool inplace_reuse{false};
     OrtValueIndex reused_buffer{0};  // The index of the reused tensor, if no reuse, it is its own tensor.
     AllocKind alloc_kind{AllocKind::kAllocate};
-    OrtMemoryInfo location;
+    OrtDevice location;
   };
 
   struct AllocationSummary {
@@ -185,7 +185,7 @@ class MemoryInfo {
   }
 
   // Key: The map type. E.g., initializer, activation. Value: A map from the tensor index to its memory information
-  std::map<OrtMemoryInfo, std::map<MapType, MemoryInfoMap> > tensors_memory_info_map_;
+  std::map<OrtDevice, std::map<MapType, MemoryInfoMap> > tensors_memory_info_map_;
 
   // Key: The tensor index. Value: The Allocation information per tensor
   std::map<OrtValueIndex, AllocInfoPerTensor> tensor_alloc_info_map_;