From 56eeee93aafa2d23b2b3895a03c45d3b796faa3b Mon Sep 17 00:00:00 2001
From: MirceaDan99 <mircea-aurelian.dan@intel.com>
Date: Thu, 12 Dec 2024 16:37:04 +0200
Subject: [PATCH] Prepare `BlobContainerAlignedBuffer` for `OV versioning
 metadata`

---
 .../include/intel_npu/common/blob_container.hpp   | 15 ++++++++++-----
 .../src/compiler_adapter/src/plugin_graph.cpp     |  2 +-
 src/plugins/intel_npu/src/plugin/src/plugin.cpp   | 10 +++++++---
 3 files changed, 18 insertions(+), 9 deletions(-)

diff --git a/src/plugins/intel_npu/src/common/include/intel_npu/common/blob_container.hpp b/src/plugins/intel_npu/src/common/include/intel_npu/common/blob_container.hpp
index fbceccb26824e0..13f833ddfccc88 100644
--- a/src/plugins/intel_npu/src/common/include/intel_npu/common/blob_container.hpp
+++ b/src/plugins/intel_npu/src/common/include/intel_npu/common/blob_container.hpp
@@ -46,16 +46,20 @@ class BlobContainerVector : public BlobContainer {
 
 class BlobContainerAlignedBuffer : public BlobContainer {
 public:
-    BlobContainerAlignedBuffer(const std::shared_ptr<ov::AlignedBuffer>& blobSO, size_t offset)
+    BlobContainerAlignedBuffer(const std::shared_ptr<ov::AlignedBuffer>& blobSO,
+                               size_t ovHeaderOffset,
+                               size_t metadataSize)
         : _ownershipBlob(blobSO),
-          _offset(offset) {}
+          _ovHeaderOffset(ovHeaderOffset),
+          _metadataSize(metadataSize) {}
 
     void* get_ptr() override {
-        return _ownershipBlob->get_ptr(_offset);
+        return _ownershipBlob->get_ptr(_ovHeaderOffset);
     }
 
     size_t size() const override {
-        return _ownershipBlob->size();
+        // remove OV header offset and metadata from blob size
+        return _ownershipBlob->size() - _ovHeaderOffset - _metadataSize;
     }
 
     bool release_from_memory() override {
@@ -64,7 +68,8 @@ class BlobContainerAlignedBuffer : public BlobContainer {
 
 private:
     std::shared_ptr<ov::AlignedBuffer> _ownershipBlob;
-    size_t _offset;
+    size_t _ovHeaderOffset;
+    size_t _metadataSize;
 };
 
 }  // namespace intel_npu
diff --git a/src/plugins/intel_npu/src/compiler_adapter/src/plugin_graph.cpp b/src/plugins/intel_npu/src/compiler_adapter/src/plugin_graph.cpp
index a54d14c0462587..90e531d5b940c7 100644
--- a/src/plugins/intel_npu/src/compiler_adapter/src/plugin_graph.cpp
+++ b/src/plugins/intel_npu/src/compiler_adapter/src/plugin_graph.cpp
@@ -30,7 +30,7 @@ PluginGraph::PluginGraph(const std::shared_ptr<ZeGraphExtWrappers>& zeGraphExt,
     initialize(config);
 }
 
-void PluginGraph::export_blob(std::ostream& stream) const {
+size_t PluginGraph::export_blob(std::ostream& stream) const {
     stream.write(reinterpret_cast<const char*>(_blobPtr->get_ptr()), _blobPtr->size());
 
     if (!stream) {
diff --git a/src/plugins/intel_npu/src/plugin/src/plugin.cpp b/src/plugins/intel_npu/src/plugin/src/plugin.cpp
index 29b675b9baa659..26ed6e7347f40f 100644
--- a/src/plugins/intel_npu/src/plugin/src/plugin.cpp
+++ b/src/plugins/intel_npu/src/plugin/src/plugin.cpp
@@ -790,11 +790,15 @@ std::shared_ptr<ov::ICompiledModel> Plugin::import_model(std::istream& stream, c
         CompilerAdapterFactory compilerAdapterFactory;
         auto compiler = compilerAdapterFactory.getCompiler(_backends->getIEngineBackend(), localConfig);
 
+        auto storedMeta = read_metadata_from(stream);
+        if (!storedMeta->is_compatible()) {
+            OPENVINO_THROW("Incompatible blob version!");
+        }
+
         std::unique_ptr<BlobContainer> blobPtr;
+        auto graphSize = storedMeta->get_blob_size();
 
         if (modelBuffer == nullptr) {
-            auto graphSize = getFileSize(stream);
-
             std::vector<uint8_t> blob(graphSize);
             stream.read(reinterpret_cast<char*>(blob.data()), graphSize);
             if (!stream) {
@@ -804,7 +808,7 @@ std::shared_ptr<ov::ICompiledModel> Plugin::import_model(std::istream& stream, c
 
             blobPtr = std::make_unique<BlobContainerVector>(std::move(blob));
         } else {
-            blobPtr = std::make_unique<BlobContainerAlignedBuffer>(modelBuffer, stream.tellg());
+            blobPtr = std::make_unique<BlobContainerAlignedBuffer>(modelBuffer, stream.tellg(), graphSize);
         }
 
         auto graph = compiler->parse(std::move(blobPtr), localConfig);