diff --git a/src/graph/backend/dnnl/common.cpp b/src/graph/backend/dnnl/common.cpp
index e678de34098..f4e33655578 100644
--- a/src/graph/backend/dnnl/common.cpp
+++ b/src/graph/backend/dnnl/common.cpp
@@ -646,7 +646,7 @@ dnnl::memory::format_tag get_format_tag(const dnnl::memory::desc &md) {
     return format_tag;
 }
 
-size_t generate_constant_cache_key(
+size_t generate_constant_md_hash(
         size_t part_id, const std::vector<dnnl::memory::desc> &const_mds) {
     size_t key = 0;
     key = hash_combine(key, part_id);
diff --git a/src/graph/backend/dnnl/common.hpp b/src/graph/backend/dnnl/common.hpp
index f53450bb373..5f3ad3634f7 100644
--- a/src/graph/backend/dnnl/common.hpp
+++ b/src/graph/backend/dnnl/common.hpp
@@ -141,7 +141,7 @@ std::string get_format_tag_str(const dnnl::memory::desc &md);
 
 dnnl::memory::format_tag get_format_tag(const dnnl::memory::desc &md);
 
-size_t generate_constant_cache_key(
+size_t generate_constant_md_hash(
         size_t part_id, const std::vector<dnnl::memory::desc> &const_mds);
 
 #ifndef NDEBUG
diff --git a/src/graph/backend/dnnl/kernels/batch_norm.cpp b/src/graph/backend/dnnl/kernels/batch_norm.cpp
index eb9aadc54ef..48afa45df65 100644
--- a/src/graph/backend/dnnl/kernels/batch_norm.cpp
+++ b/src/graph/backend/dnnl/kernels/batch_norm.cpp
@@ -109,7 +109,7 @@ status_t batch_norm_fwd_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -135,7 +135,8 @@ status_t batch_norm_fwd_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -205,7 +206,8 @@ status_t batch_norm_fwd_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -281,7 +283,8 @@ status_t batch_norm_fwd_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/batch_norm.hpp b/src/graph/backend/dnnl/kernels/batch_norm.hpp
index 258487b796b..3abfd6f438b 100644
--- a/src/graph/backend/dnnl/kernels/batch_norm.hpp
+++ b/src/graph/backend/dnnl/kernels/batch_norm.hpp
@@ -46,7 +46,7 @@ struct batch_norm_fwd_t : public kernel_base_t {
     memory_planner_t memory_planner_;
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     batch_norm_fwd_t() {
diff --git a/src/graph/backend/dnnl/kernels/conv.cpp b/src/graph/backend/dnnl/kernels/conv.cpp
index 60dd968f42b..685e0fa83fa 100644
--- a/src/graph/backend/dnnl/kernels/conv.cpp
+++ b/src/graph/backend/dnnl/kernels/conv.cpp
@@ -134,7 +134,7 @@ status_t conv_fwd_t<quantized>::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -202,7 +202,7 @@ status_t conv_bwd_data_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
diff --git a/src/graph/backend/dnnl/kernels/conv_base.cpp b/src/graph/backend/dnnl/kernels/conv_base.cpp
index 256e1426471..280f5974949 100644
--- a/src/graph/backend/dnnl/kernels/conv_base.cpp
+++ b/src/graph/backend/dnnl/kernels/conv_base.cpp
@@ -63,7 +63,8 @@ status_t conv_base_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -133,7 +134,8 @@ status_t conv_base_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -209,7 +211,8 @@ status_t conv_base_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/conv_base.hpp b/src/graph/backend/dnnl/kernels/conv_base.hpp
index 877ac5604b2..08728d63e44 100644
--- a/src/graph/backend/dnnl/kernels/conv_base.hpp
+++ b/src/graph/backend/dnnl/kernels/conv_base.hpp
@@ -47,7 +47,7 @@ struct conv_base_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     conv_base_t() {
diff --git a/src/graph/backend/dnnl/kernels/conv_transpose.cpp b/src/graph/backend/dnnl/kernels/conv_transpose.cpp
index 330e22b8050..425990556e0 100644
--- a/src/graph/backend/dnnl/kernels/conv_transpose.cpp
+++ b/src/graph/backend/dnnl/kernels/conv_transpose.cpp
@@ -114,7 +114,7 @@ status_t conv_transpose_fwd_t<quantized>::compile_impl(
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -183,7 +183,7 @@ status_t conv_transpose_bwd_data_t::compile_impl(
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
diff --git a/src/graph/backend/dnnl/kernels/eltwise.cpp b/src/graph/backend/dnnl/kernels/eltwise.cpp
index e8abe4a368c..72ede8ca406 100644
--- a/src/graph/backend/dnnl/kernels/eltwise.cpp
+++ b/src/graph/backend/dnnl/kernels/eltwise.cpp
@@ -91,7 +91,7 @@ status_t eltwise_fwd_t<quantized>::compile_impl(
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -138,7 +138,8 @@ status_t eltwise_fwd_t<quantized>::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -209,7 +210,8 @@ status_t eltwise_fwd_t<quantized>::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -286,7 +288,8 @@ status_t eltwise_fwd_t<quantized>::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/eltwise.hpp b/src/graph/backend/dnnl/kernels/eltwise.hpp
index 537ac67805e..cc1b318ab38 100644
--- a/src/graph/backend/dnnl/kernels/eltwise.hpp
+++ b/src/graph/backend/dnnl/kernels/eltwise.hpp
@@ -49,7 +49,7 @@ struct eltwise_fwd_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     eltwise_fwd_t() {
diff --git a/src/graph/backend/dnnl/kernels/group_norm.cpp b/src/graph/backend/dnnl/kernels/group_norm.cpp
index 2b11861c1e4..4b39f7500ee 100644
--- a/src/graph/backend/dnnl/kernels/group_norm.cpp
+++ b/src/graph/backend/dnnl/kernels/group_norm.cpp
@@ -97,7 +97,7 @@ status_t group_norm_fwd_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -143,7 +143,8 @@ status_t group_norm_fwd_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -213,7 +214,8 @@ status_t group_norm_fwd_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -289,7 +291,8 @@ status_t group_norm_fwd_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/group_norm.hpp b/src/graph/backend/dnnl/kernels/group_norm.hpp
index e4c2192b9f0..ab571415201 100644
--- a/src/graph/backend/dnnl/kernels/group_norm.hpp
+++ b/src/graph/backend/dnnl/kernels/group_norm.hpp
@@ -48,7 +48,7 @@ struct group_norm_fwd_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     group_norm_fwd_t() {
diff --git a/src/graph/backend/dnnl/kernels/kernel_base.cpp b/src/graph/backend/dnnl/kernels/kernel_base.cpp
index 798ddaebaa3..cfb5b2f50b1 100644
--- a/src/graph/backend/dnnl/kernels/kernel_base.cpp
+++ b/src/graph/backend/dnnl/kernels/kernel_base.cpp
@@ -48,14 +48,9 @@ size_t kernel_base_t::encode_constant_cache_key(
     // Encode the constant memory address into cache key for differentiation
     size_t encoded_cache_key = cache_key;
     for (const auto &in : inputs) {
-        if (in.get_logical_tensor().property
-                == dnnl_graph_tensor_property_t::
-                        dnnl_graph_tensor_property_constant) {
-            auto data_handle = in.get_data_handle();
-            if (data_handle != nullptr) {
-                encoded_cache_key = hash_combine(encoded_cache_key,
-                        reinterpret_cast<uintptr_t>(data_handle));
-            }
+        if (logical_tensor_wrapper_t(in.get_logical_tensor()).is_constant()) {
+            encoded_cache_key = hash_combine(encoded_cache_key,
+                    reinterpret_cast<uintptr_t>(in.get_data_handle()));
         }
     }
     return encoded_cache_key;
diff --git a/src/graph/backend/dnnl/kernels/large_partition.cpp b/src/graph/backend/dnnl/kernels/large_partition.cpp
index a8d8393ac42..e60234882c3 100644
--- a/src/graph/backend/dnnl/kernels/large_partition.cpp
+++ b/src/graph/backend/dnnl/kernels/large_partition.cpp
@@ -222,7 +222,7 @@ status_t larger_partition_kernel_t::compile_impl(
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -248,7 +248,8 @@ status_t larger_partition_kernel_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -317,7 +318,8 @@ status_t larger_partition_kernel_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -391,7 +393,8 @@ status_t larger_partition_kernel_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/large_partition.hpp b/src/graph/backend/dnnl/kernels/large_partition.hpp
index b6fc59d63a7..dc3a903945e 100644
--- a/src/graph/backend/dnnl/kernels/large_partition.hpp
+++ b/src/graph/backend/dnnl/kernels/large_partition.hpp
@@ -48,7 +48,7 @@ class larger_partition_kernel_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
     std::once_flag once_flag_;
     subgraph_visualizer_t vis_;
diff --git a/src/graph/backend/dnnl/kernels/layer_norm.cpp b/src/graph/backend/dnnl/kernels/layer_norm.cpp
index f7a4d084987..9b7c234ad20 100644
--- a/src/graph/backend/dnnl/kernels/layer_norm.cpp
+++ b/src/graph/backend/dnnl/kernels/layer_norm.cpp
@@ -94,7 +94,7 @@ status_t layer_norm_fwd_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -140,7 +140,8 @@ status_t layer_norm_fwd_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -210,7 +211,8 @@ status_t layer_norm_fwd_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -286,7 +288,8 @@ status_t layer_norm_fwd_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/layer_norm.hpp b/src/graph/backend/dnnl/kernels/layer_norm.hpp
index 7d104117b7f..169c1e54022 100644
--- a/src/graph/backend/dnnl/kernels/layer_norm.hpp
+++ b/src/graph/backend/dnnl/kernels/layer_norm.hpp
@@ -48,7 +48,7 @@ struct layer_norm_fwd_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     layer_norm_fwd_t() {
diff --git a/src/graph/backend/dnnl/kernels/matmul.cpp b/src/graph/backend/dnnl/kernels/matmul.cpp
index 80146cdbc96..c0583359cde 100644
--- a/src/graph/backend/dnnl/kernels/matmul.cpp
+++ b/src/graph/backend/dnnl/kernels/matmul.cpp
@@ -145,7 +145,7 @@ status_t matmul_t<quantized>::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -193,7 +193,8 @@ status_t matmul_t<quantized>::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -264,7 +265,8 @@ status_t matmul_t<quantized>::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -341,7 +343,8 @@ status_t matmul_t<quantized>::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/matmul.hpp b/src/graph/backend/dnnl/kernels/matmul.hpp
index a88f71bcf60..9a3d6be11b2 100644
--- a/src/graph/backend/dnnl/kernels/matmul.hpp
+++ b/src/graph/backend/dnnl/kernels/matmul.hpp
@@ -49,7 +49,7 @@ struct matmul_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     matmul_t() {
diff --git a/src/graph/backend/dnnl/kernels/pool.cpp b/src/graph/backend/dnnl/kernels/pool.cpp
index 9703ffe0495..3db019e97f5 100644
--- a/src/graph/backend/dnnl/kernels/pool.cpp
+++ b/src/graph/backend/dnnl/kernels/pool.cpp
@@ -122,7 +122,7 @@ status_t pooling_fwd_t<quantized>::compile_impl(
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -170,7 +170,8 @@ status_t pooling_fwd_t<quantized>::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -241,7 +242,8 @@ status_t pooling_fwd_t<quantized>::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -318,7 +320,8 @@ status_t pooling_fwd_t<quantized>::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/pool.hpp b/src/graph/backend/dnnl/kernels/pool.hpp
index 7a58f51d75d..57884e7394e 100644
--- a/src/graph/backend/dnnl/kernels/pool.hpp
+++ b/src/graph/backend/dnnl/kernels/pool.hpp
@@ -49,7 +49,7 @@ struct pooling_fwd_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     pooling_fwd_t() {
diff --git a/src/graph/backend/dnnl/kernels/quantize.cpp b/src/graph/backend/dnnl/kernels/quantize.cpp
index 31388ef5d10..fbc90bb73fa 100644
--- a/src/graph/backend/dnnl/kernels/quantize.cpp
+++ b/src/graph/backend/dnnl/kernels/quantize.cpp
@@ -88,7 +88,7 @@ status_t quantize_dequantize_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -134,7 +134,8 @@ status_t quantize_dequantize_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -209,7 +210,8 @@ status_t quantize_dequantize_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -285,7 +287,8 @@ status_t quantize_dequantize_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/quantize.hpp b/src/graph/backend/dnnl/kernels/quantize.hpp
index 5c434a111af..369b197af88 100644
--- a/src/graph/backend/dnnl/kernels/quantize.hpp
+++ b/src/graph/backend/dnnl/kernels/quantize.hpp
@@ -46,7 +46,7 @@ struct quantize_dequantize_t : public kernel_base_t {
     memory_planner_t memory_planner_;
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     quantize_dequantize_t() {
diff --git a/src/graph/backend/dnnl/kernels/reorder.cpp b/src/graph/backend/dnnl/kernels/reorder.cpp
index b4f38f8376c..e513ab7be9e 100644
--- a/src/graph/backend/dnnl/kernels/reorder.cpp
+++ b/src/graph/backend/dnnl/kernels/reorder.cpp
@@ -105,7 +105,7 @@ status_t reorder_t<quantized>::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -153,7 +153,8 @@ status_t reorder_t<quantized>::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -224,7 +225,8 @@ status_t reorder_t<quantized>::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -301,7 +303,8 @@ status_t reorder_t<quantized>::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/reorder.hpp b/src/graph/backend/dnnl/kernels/reorder.hpp
index a4e3e16408b..9993c5458fe 100644
--- a/src/graph/backend/dnnl/kernels/reorder.hpp
+++ b/src/graph/backend/dnnl/kernels/reorder.hpp
@@ -49,7 +49,7 @@ struct reorder_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     reorder_t() {
diff --git a/src/graph/backend/dnnl/kernels/select.cpp b/src/graph/backend/dnnl/kernels/select.cpp
index 3cbb75f2aff..9990937d9ba 100644
--- a/src/graph/backend/dnnl/kernels/select.cpp
+++ b/src/graph/backend/dnnl/kernels/select.cpp
@@ -88,7 +88,7 @@ status_t select_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -134,7 +134,8 @@ status_t select_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -204,7 +205,8 @@ status_t select_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -280,7 +282,8 @@ status_t select_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/select.hpp b/src/graph/backend/dnnl/kernels/select.hpp
index 58c9ef6abe6..fc4259bd28f 100644
--- a/src/graph/backend/dnnl/kernels/select.hpp
+++ b/src/graph/backend/dnnl/kernels/select.hpp
@@ -48,7 +48,7 @@ struct select_t : public kernel_base_t {
 
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     select_t() {
diff --git a/src/graph/backend/dnnl/kernels/softmax.cpp b/src/graph/backend/dnnl/kernels/softmax.cpp
index 1474f6abac9..d2600dd14bd 100644
--- a/src/graph/backend/dnnl/kernels/softmax.cpp
+++ b/src/graph/backend/dnnl/kernels/softmax.cpp
@@ -91,7 +91,7 @@ status_t softmax_fwd_t::compile_impl(const dnnl_partition_impl_t *part,
         return this->memory_planner_.get_exec_args_set().clone();
     };
 
-    constant_key_ = generate_constant_cache_key(part->id(),
+    const_md_hash_ = generate_constant_md_hash(part->id(),
             memory_planner_.get_exec_args_set().get_persistent_mem_desc_list());
 
     return status::success;
@@ -137,7 +137,8 @@ status_t softmax_fwd_t::execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -207,7 +208,8 @@ status_t softmax_fwd_t::sycl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
@@ -283,7 +285,8 @@ status_t softmax_fwd_t::ocl_execute_impl(const stream_t *g_stream,
 
     constant_cache_t::cached_t c_buffer;
     if (enabled_constant_cache()) {
-        size_t encoded_key = encode_constant_cache_key(inputs, constant_key_);
+        const size_t encoded_key
+                = encode_constant_cache_key(inputs, const_md_hash_);
         std::promise<constant_cache_t::cached_t> c_promise;
         constant_cache_t::value_t cached_value
                 = dnnl_constant_cache_get_or_add(p_engine_, encoded_key,
diff --git a/src/graph/backend/dnnl/kernels/softmax.hpp b/src/graph/backend/dnnl/kernels/softmax.hpp
index 72de9943455..d80f88fba9d 100644
--- a/src/graph/backend/dnnl/kernels/softmax.hpp
+++ b/src/graph/backend/dnnl/kernels/softmax.hpp
@@ -46,7 +46,7 @@ struct softmax_fwd_t : public kernel_base_t {
     memory_planner_t memory_planner_;
     std::function<std::shared_ptr<execution_args_set_t>()> resource_ctor_;
 
-    constant_cache_t::key_t constant_key_ = 0;
+    size_t const_md_hash_ = 0;
 
 public:
     softmax_fwd_t() {