diff --git a/paddle/fluid/distributed/index_dataset/CMakeLists.txt b/paddle/fluid/distributed/index_dataset/CMakeLists.txt
index 6edb9834d49eb752a88bacf3cf7d0cd72f081100..98bc0a0ad4a26f10319ae4650620fe51cb0fdcff 100644
--- a/paddle/fluid/distributed/index_dataset/CMakeLists.txt
+++ b/paddle/fluid/distributed/index_dataset/CMakeLists.txt
@@ -1,9 +1,9 @@
 proto_library(index_dataset_proto SRCS index_dataset.proto)
 cc_library(index_wrapper SRCS index_wrapper.cc DEPS index_dataset_proto fs)
 if(WITH_MKLDNN)
-  cc_library(index_sampler SRCS index_sampler.cc DEPS xxhash index_wrapper mkldnn)
+  cc_library(index_sampler SRCS index_sampler.cc DEPS xxhash index_wrapper eigen3 mkldnn)
 else()
-  cc_library(index_sampler SRCS index_sampler.cc DEPS xxhash index_wrapper)
+  cc_library(index_sampler SRCS index_sampler.cc DEPS xxhash index_wrapper eigen3)
 endif()
 if(WITH_PYTHON)
   py_proto_compile(index_dataset_py_proto SRCS index_dataset.proto)
diff --git a/paddle/fluid/operators/concat_op_mlu.cc b/paddle/fluid/operators/concat_op_mlu.cc
index f7a1cae72be5a42d15d2e89663010489f529962a..6325d1ccda8c83e0ef1f1b735b3fe99596e03639 100644
--- a/paddle/fluid/operators/concat_op_mlu.cc
+++ b/paddle/fluid/operators/concat_op_mlu.cc
@@ -45,7 +45,8 @@ class ConcatMLUKernel : public framework::OpKernel<T> {
         ins_dims[i] = ins[i]->dims();
       }
 
-      framework::DDim out_dims = ComputeAndCheckShape(true, ins_dims, axis);
+      framework::DDim out_dims =
+          pten::funcs::ComputeAndCheckShape(true, ins_dims, axis);
       out->Resize(out_dims);
     }
     const int axis_t = axis;
diff --git a/paddle/fluid/operators/fill_constant_op_mlu.cc b/paddle/fluid/operators/fill_constant_op_mlu.cc
index d161a857d6c1778e8136702564dae2582c8a0465..c265977715ff147bf360cafb92a86538ce828d90 100644
--- a/paddle/fluid/operators/fill_constant_op_mlu.cc
+++ b/paddle/fluid/operators/fill_constant_op_mlu.cc
@@ -63,7 +63,8 @@ class FillConstantMLUKernel : public framework::OpKernel<T> {
       framework::Tensor mlu_tensor;
       auto tmp_place = value_tensor->place();
       if (platform::is_mlu_place(tmp_place)) {
-        TensorCopySync(*value_tensor, platform::CPUPlace(), &mlu_tensor);
+        framework::TensorCopySync(*value_tensor, platform::CPUPlace(),
+                                  &mlu_tensor);
         tensor_data = mlu_tensor.data<T>();
       }
       value = tensor_data[0];
diff --git a/paddle/fluid/operators/mlu/CMakeLists.txt b/paddle/fluid/operators/mlu/CMakeLists.txt
index 59fab48b271d556c84d4f502022d41ec6d830e0a..efd6aeb8eeb1c35290e69e49f1a6b5cd75a4880a 100644
--- a/paddle/fluid/operators/mlu/CMakeLists.txt
+++ b/paddle/fluid/operators/mlu/CMakeLists.txt
@@ -1,5 +1,5 @@
 
 IF(WITH_MLU)
-    cc_library(mlu_baseop SRCS mlu_baseop.cc DEPS neuware_lib framework_proto xxhash)
+    cc_library(mlu_baseop SRCS mlu_baseop.cc DEPS neuware_lib device_context)
     cc_test(activation_op_mlu_test SRCS activation_op_mlu_test.cc DEPS op_registry activation_op scope device_context executor)
 ENDIF()
diff --git a/paddle/fluid/operators/split_op_mlu.cc b/paddle/fluid/operators/split_op_mlu.cc
index c569c9bf091335a01dfb2d70808cb6ce0bb66812..d6d3f4c99aa052aa87fb6365c77d3cba8a2f5cab 100644
--- a/paddle/fluid/operators/split_op_mlu.cc
+++ b/paddle/fluid/operators/split_op_mlu.cc
@@ -58,7 +58,6 @@ class SplitMLUKernel : public framework::OpKernel<T> {
     std::vector<void*> vct_tensor;
     std::vector<MLUCnnlTensorDesc> output_descs;
     std::vector<cnnlTensorDescriptor_t> desc_vector;
-    auto place = ctx.GetPlace();
     for (size_t i = 0; i < outs.size(); i++) {
       outs[i]->mutable_data<T>(ctx.GetPlace());
       output_descs.emplace_back(MLUCnnlTensorDesc(
diff --git a/paddle/fluid/operators/sum_op_mlu.cc b/paddle/fluid/operators/sum_op_mlu.cc
index e2cd649722b2444bb7a032eac18760e582db71d8..59f3a424479bb5c971d9e86e4be4283f45a2b79c 100644
--- a/paddle/fluid/operators/sum_op_mlu.cc
+++ b/paddle/fluid/operators/sum_op_mlu.cc
@@ -33,7 +33,7 @@ class SumMLUKernel : public framework::OpKernel<T> {
       auto place = ctx.GetPlace();
       int ins_size = static_cast<int>(ins.size());
       if (ins_size == 1) {
-        TensorCopy(*ins[0], place, out);
+        framework::TensorCopy(*ins[0], place, out);
         return;
       }
 
diff --git a/paddle/fluid/platform/CMakeLists.txt b/paddle/fluid/platform/CMakeLists.txt
index a151c824a220fba6e0b72c1f7869f22b48a30dab..7dc07942d44b14014513f441238418239fd29abb 100644
--- a/paddle/fluid/platform/CMakeLists.txt
+++ b/paddle/fluid/platform/CMakeLists.txt
@@ -55,6 +55,12 @@ ENDIF()
 cc_library(place SRCS place.cc DEPS enforce boost pten_place)
 cc_test(place_test SRCS place_test.cc DEPS place glog gflags)
 
+IF(WITH_MKLDNN)
+    set(MKLDNN_CTX_DEPS mkldnn)
+ELSE()
+    set(MKLDNN_CTX_DEPS)
+ENDIF()
+
 add_subdirectory(device)
 add_subdirectory(dynload)
 add_subdirectory(stream)
@@ -85,12 +91,6 @@ IF(WITH_MLU)
     set(MLU_CTX_DEPS mlu_device_context)
 ENDIF()
 
-IF(WITH_MKLDNN)
-    set(MKLDNN_CTX_DEPS mkldnn)
-ELSE()
-    set(MKLDNN_CTX_DEPS)
-ENDIF()
-
 IF(WITH_ASCEND_CL OR WITH_MLU)
 cc_library(stream_callback_manager SRCS stream_callback_manager.cc DEPS simple_threadpool enforce)
 ENDIF()
diff --git a/paddle/fluid/platform/device/CMakeLists.txt b/paddle/fluid/platform/device/CMakeLists.txt
index c5fe211470949e4023bb28c6f0e9d1888d66a1fc..2cd068badf51e8a3176de4ec80700ce7057862d1 100644
--- a/paddle/fluid/platform/device/CMakeLists.txt
+++ b/paddle/fluid/platform/device/CMakeLists.txt
@@ -1,10 +1,12 @@
+# GPU
+IF(WITH_GPU OR WITH_ROCM)
+  add_subdirectory(gpu)
+ENDIF()
+
 # XPU
 IF(WITH_XPU)
   add_subdirectory(xpu)
 ENDIF()
-IF(WITH_GPU OR WITH_ROCM)
-  add_subdirectory(gpu)
-ENDIF()
 
 # NPU
 IF(WITH_ASCEND OR WITH_ASCEND_CL)
diff --git a/paddle/fluid/platform/device/mlu/CMakeLists.txt b/paddle/fluid/platform/device/mlu/CMakeLists.txt
index a4584f54637a615d79995e1e27303128b4202b5e..e8b794a03e38db5c2f5c5f471b4ed419a4884b43 100644
--- a/paddle/fluid/platform/device/mlu/CMakeLists.txt
+++ b/paddle/fluid/platform/device/mlu/CMakeLists.txt
@@ -1,10 +1,10 @@
-IF(WITH_MLU)
-    cc_test(mlu_enforce_test SRCS enforce_test.cc DEPS stringpiece)
 
-    cc_library(mlu_info SRCS mlu_info.cc DEPS enforce glog monitor neuware_lib)
+if(NOT WITH_MLU)
+    return()
+endif()
 
-    cc_library(mlu_stream SRCS mlu_stream.cc DEPS boost mlu_info stream_callback_manager)
-    
-    cc_library(mlu_device_context SRCS device_context.cc DEPS mlu_stream eigen3)
-    cc_test(mlu_device_context_test SRCS device_context_test.cc DEPS mlu_device_context)
-ENDIF()
+cc_test(mlu_enforce_test SRCS enforce_test.cc DEPS stringpiece)
+cc_library(mlu_info SRCS mlu_info.cc DEPS enforce glog monitor neuware_lib)
+cc_library(mlu_stream SRCS mlu_stream.cc DEPS boost mlu_info stream_callback_manager eigen3 ${MKLDNN_CTX_DEPS})
+cc_library(mlu_device_context SRCS device_context.cc DEPS mlu_stream)
+cc_test(mlu_device_context_test SRCS device_context_test.cc DEPS mlu_device_context)
diff --git a/paddle/fluid/platform/stream/CMakeLists.txt b/paddle/fluid/platform/stream/CMakeLists.txt
index 623b069df9b571dd7a169cfe641066b2b1cae802..6a825e9077c0a47b1daa98d7da7fce6cac281cde 100644
--- a/paddle/fluid/platform/stream/CMakeLists.txt
+++ b/paddle/fluid/platform/stream/CMakeLists.txt
@@ -1,9 +1,3 @@
-IF(WITH_MKLDNN)
-    set(MKLDNN_CTX_DEPS mkldnn)
-ELSE()
-    set(MKLDNN_CTX_DEPS)
-ENDIF()
-
 IF(WITH_GPU OR WITH_ROCM)
-cc_library(cuda_stream SRCS cuda_stream.cc DEPS enforce boost ${MKLDNN_CTX_DEPS})
+    cc_library(cuda_stream SRCS cuda_stream.cc DEPS enforce boost eigen3 ${MKLDNN_CTX_DEPS})
 ENDIF()
diff --git a/paddle/scripts/paddle_build.sh b/paddle/scripts/paddle_build.sh
index 384dfbf558f42926e40f1fbe4c8611547037d180..d1fb2ba13bb65c968340e3ebc55e5a218ca962ad 100755
--- a/paddle/scripts/paddle_build.sh
+++ b/paddle/scripts/paddle_build.sh
@@ -227,6 +227,7 @@ function cmake_base() {
         -DWITH_GLOO=${gloo_flag}
         -DWITH_LITE=${WITH_LITE:-OFF}
         -DWITH_XPU=${WITH_XPU:-OFF}
+        -DWITH_MLU=${WITH_MLU:-OFF}
         -DLITE_GIT_TAG=release/v2.10
         -DWITH_UNITY_BUILD=${WITH_UNITY_BUILD:-OFF}
         -DWITH_XPU_BKCL=${WITH_XPU_BKCL:-OFF}
@@ -275,6 +276,7 @@ EOF
         -DWITH_GLOO=${gloo_flag} \
         -DLITE_GIT_TAG=release/v2.10 \
         -DWITH_XPU=${WITH_XPU:-OFF} \
+        -DWITH_MLU=${WITH_MLU:-OFF} \
         -DXPU_SDK_ROOT=${XPU_SDK_ROOT:-""} \
         -DWITH_LITE=${WITH_LITE:-OFF} \
         -DWITH_XPU_BKCL=${WITH_XPU_BKCL:-OFF} \
diff --git a/tools/dockerfile/Dockerfile.paddle-mlu-build b/tools/dockerfile/Dockerfile.paddle-mlu-build
new file mode 100644
index 0000000000000000000000000000000000000000..6fcf6b6278d75f1aeb4006c03de57b56eef23e59
--- /dev/null
+++ b/tools/dockerfile/Dockerfile.paddle-mlu-build
@@ -0,0 +1,5 @@
+FROM registry.baidubce.com/qili93/paddle-mlu:neuware-latest
+RUN apt-get install pigz -y
+RUN apt-get remove -y openjdk*
+CMD ["/bin/bash"]
+EXPOSE 22