clean up code

Author: Kotomi-Du

onnxruntime/core/providers/openvino/ov_interface.cc

@@ -423,17 +423,17 @@ void StatefulOVInferRequest::PreProcessInferRequest() {
   // TODO(ankit): Address this issue and implement the fix at the appropriate layer.
   FillTensor("beam_idx", ov::element::i32, {1}, 0);
 
-  if (src_idx_val.size() > 0) {
-    ov::Tensor src_idx_tensor = ov::Tensor(ov::element::i32, {src_idx_val.size()});
-    for (int i = 0; i < src_idx_val.size(); ++i) {
-      src_idx_tensor.data<int32_t>()[i] = int32_t(src_idx_val[i]);
+  if (kv_src_indices.size() > 0) {
+    ov::Tensor src_idx_tensor = ov::Tensor(ov::element::i32, {kv_src_indices.size()});
+    for (int i = 0; i < kv_src_indices.size(); ++i) {
+      src_idx_tensor.data<int32_t>()[i] = int32_t(kv_src_indices[i]);
     }
     ovInfReq.set_tensor("src_idx", src_idx_tensor);
-    ov::Tensor dst_idx_tensor = ov::Tensor(ov::element::i32, {1, 32, dst_idx_val.size(), 96});
-    for (int i = 0; i < dst_idx_val.size(); ++i) {
+    ov::Tensor dst_idx_tensor = ov::Tensor(ov::element::i32, {1, 32, kv_dst_indices.size(), 96});
+    for (int i = 0; i < kv_dst_indices.size(); ++i) {
       for (int j = 0; j < 32; ++j) {
         for (int k = 0; k < 96; ++k) {
-          dst_idx_tensor.data<int32_t>()[(j * dst_idx_val.size() + i) * 96 + k] = int32_t(dst_idx_val[i]);
+          dst_idx_tensor.data<int32_t>()[(j * kv_dst_indices.size() + i) * 96 + k] = int32_t(kv_dst_indices[i]);
         }
       }
     }
@@ -492,25 +492,12 @@ void StatefulOVInferRequest::ReorderKVCache(const std::vector<size_t>& src_indic
   LOGS_DEFAULT(INFO) << log_tag << "ReorderKVCache: Reordering OpenVINO-internal KVCache state with "
                      << src_indices.size() << " index pairs";
 
-  // set beam_idx and dst_idx based on provided values
-  src_idx_val.clear();
-  dst_idx_val.clear();
+  kv_src_indices.clear();
+  kv_dst_indices.clear();
   for (int i = 0; i < src_indices.size(); ++i) {
-    src_idx_val.emplace_back(src_indices[i]);
-    dst_idx_val.emplace_back(dst_indices[i]);
+    kv_src_indices.emplace_back(src_indices[i]);
+    kv_dst_indices.emplace_back(dst_indices[i]);
   }
-  /*
-  // Retrieve KVCache states and reorder them based on the provided indices
-  auto states = ovInfReq.query_state();
-
-  for (auto& state : states) {
-    auto start_time = std::chrono::high_resolution_clock::now();
-    state.gather_by_axis(src_indices, dst_indices);
-    auto end_time = std::chrono::high_resolution_clock::now();
-    auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end_time - start_time).count();
-    LOGS_DEFAULT(INFO) << log_tag << "gather_by_axis: " << duration << " microseconds";
-  }
-  */
 }
 
 void StatefulOVInferRequest::RewindKVCache(size_t index) {

onnxruntime/core/providers/openvino/ov_interface.h

@@ -157,8 +157,8 @@ class StatefulOVInferRequest : public OVInferRequest {
   bool prefill_use_full_chat_history = false;
   std::vector<int64_t> cached_input_ids;
   std::vector<int64_t> cached_position_ids;
-  std::vector<int64_t> src_idx_val;
-  std::vector<int64_t> dst_idx_val;
+  std::vector<int64_t> kv_src_indices;
+  std::vector<int64_t> kv_dst_indices;
 };
 
 }  // namespace openvino_ep

onnxruntime/core/providers/openvino/ov_stateful_patch_utils.cc

@@ -121,17 +121,17 @@ void FuseCacheReorder(std::shared_ptr<ov::Model> ov_model,
                                               beam_idx,
                                               ov::opset13::Constant::create(ov::element::i64, {}, {gather_dim}));
 
-    auto update_gather_op =
+    auto updatekv_gather_op =
         std::make_shared<ov::opset13::Gather>(gather_op,
                                               src_idx,
                                               ov::opset13::Constant::create(ov::element::i64, {}, {2}));
 
-    auto update_op = std::make_shared<ov::opset12::ScatterElementsUpdate>(gather_op,
-        dst_idx, update_gather_op, ov::opset13::Constant::create(ov::element::i64, {}, {2}));
+    auto updatekv_op = std::make_shared<ov::opset12::ScatterElementsUpdate>(gather_op,
+        dst_idx, updatekv_gather_op, ov::opset13::Constant::create(ov::element::i64, {}, {2}));
 
     // Replace the source output for all consumers of the input tensor
     for (auto& consumer : consumers) {
-      consumer.replace_source_output(update_op->output(0));
+      consumer.replace_source_output(updatekv_op->output(0));
     }
   }
 
@@ -279,10 +279,10 @@ void PatchStatefulDecoder(std::shared_ptr<ov::Model> model) {
   }
 
   if (key_value_input_names.size() != key_value_output_names.size()) {
-    ORT_THROW("Found different sizes between key_value_input_names (", 
-    key_value_input_names.size(), 
-    ") and key_value_output_names (", 
-    key_value_output_names.size(), 
+    ORT_THROW("Found different sizes between key_value_input_names (",
+    key_value_input_names.size(),
+    ") and key_value_output_names (",
+    key_value_output_names.size(),
     "). They couldn't be paired.");
   }
 

onnxruntime/core/providers/openvino/ov_stateful_patch_utils.h

@@ -13,7 +13,6 @@
 
 #include "openvino/pass/manager.hpp"
 #include "openvino/pass/make_stateful.hpp"
-#include "openvino/opsets/opset3.hpp"
 #include "openvino/opsets/opset12.hpp"
 #include "openvino/opsets/opset13.hpp"