[ET-VK] Move save_cache from Runtime dtor to model destroy

backends/vulkan/runtime/VulkanBackend.cpp

@@ -621,6 +621,10 @@ class VulkanBackend final : public ::executorch::runtime::BackendInterface {
   void destroy(DelegateHandle* handle) const override {
     if (handle != nullptr) {
       ComputeGraph* compute_graph = static_cast<ComputeGraph*>(handle);
+      compute_graph->context()
+          ->adapter_ptr()
+          ->compute_pipeline_cache()
+          .save_cache();
       // ComputeGraph is not trivially destructible. Since
       // this was constructed manually in init(), we must destroy it manually
       // here.

backends/vulkan/runtime/vk_api/Pipeline.cpp

@@ -406,8 +406,6 @@ ComputePipelineCache::~ComputePipelineCache() {
     return;
   }
 
-  save_cache();
-
   vkDestroyPipelineCache(device_, pipeline_cache_, nullptr);
   pipeline_cache_ = VK_NULL_HANDLE;
 }
@@ -433,12 +431,12 @@ void ComputePipelineCache::purge() {
 }
 
 std::vector<char> ComputePipelineCache::load_cache() {
-  // Return if path is not specified; this means the optimization is disabled
+  // No optimization if path is unspecified
   if (cache_data_path_.empty()) {
     return {};
   }
 
-  // Return if file doesn't exist; this is expected on the first model-load
+  // Return if file doesn't exist; this is expected on first model-load
   std::ifstream file(cache_data_path_, std::ios::binary | std::ios::ate);
   if (file.fail()) {
     return {};
@@ -454,6 +452,17 @@ std::vector<char> ComputePipelineCache::load_cache() {
 }
 
 void ComputePipelineCache::save_cache() {
+  // No optimization if path is unspecified
+  if (cache_data_path_.empty()) {
+    return;
+  }
+
+  // Return if file exists; the cache is already saved
+  std::ifstream ifile(cache_data_path_);
+  if (ifile.good()) {
+    return;
+  }
+
   size_t size{};
   vkGetPipelineCacheData(device_, pipeline_cache_, &size, nullptr);
 

backends/vulkan/runtime/vk_api/Pipeline.h

@@ -269,9 +269,10 @@ class ComputePipelineCache final {
     }
   };
 
+  void save_cache();
+
  private:
   std::vector<char> load_cache();
-  void save_cache();
 
   // Multiple threads could potentially be adding entries into the cache, so use
   // a mutex to manage access

backends/vulkan/runtime/vk_api/Runtime.cpp

@@ -265,7 +265,7 @@ std::unique_ptr<Runtime> init_global_vulkan_runtime() {
   };
 
   try {
-    return std::make_unique<Runtime>(Runtime(default_config));
+    return std::make_unique<Runtime>(default_config);
   } catch (...) {
   }
 
@@ -323,16 +323,6 @@ Runtime::~Runtime() {
   instance_ = VK_NULL_HANDLE;
 }
 
-Runtime::Runtime(Runtime&& other) noexcept
-    : config_(other.config_),
-      instance_(other.instance_),
-      adapters_(std::move(other.adapters_)),
-      default_adapter_i_(other.default_adapter_i_),
-      debug_report_callback_(other.debug_report_callback_) {
-  other.instance_ = VK_NULL_HANDLE;
-  other.debug_report_callback_ = {};
-}
-
 uint32_t Runtime::create_adapter(const Selector& selector) {
   VK_CHECK_COND(
       !device_mappings_.empty(),

backends/vulkan/runtime/vk_api/Runtime.h

@@ -51,7 +51,7 @@ class Runtime final {
   Runtime(const Runtime&) = delete;
   Runtime& operator=(const Runtime&) = delete;
 
-  Runtime(Runtime&&) noexcept;
+  Runtime(Runtime&&) = delete;
   Runtime& operator=(Runtime&&) = delete;
 
   ~Runtime();