diff --git a/build.sh b/build.sh
index ac398082687..200cff10295 100755
--- a/build.sh
+++ b/build.sh
@@ -568,6 +568,24 @@ build_lite()
     cd ${BASEPATH}/mindspore/lite/build
     write_commit_file
 
+    if [[ "${local_lite_platform}" == "arm32" ]]; then
+      if [[ "${TOOLCHAIN_FILE}" && "${TOOLCHAIN_NAME}" ]]; then
+        RUN_TESTCASES="off"
+        COMPILE_MINDDATA_LITE="off"
+        CMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_FILE}
+        CMAKE_TOOLCHAIN_NAME=${TOOLCHAIN_NAME}
+      else
+        CMAKE_TOOLCHAIN_FILE=${ANDROID_NDK}/build/cmake/android.toolchain.cmake
+        ANDROID_NATIVE_API_LEVEL="19"
+        CMAKE_ANDROID_NDK=${ANDROID_NDK}
+        CMAKE_ANDROID_ABI="armeabi-v7a"
+        CMAKE_ANDROID_TOOLCHAIN_NAME="clang"
+        CMAKE_ANDROID_STL=${MSLITE_ANDROID_STL}
+        CMAKE_BUILD_TYPE=${LITE_BUILD_TYPE}
+        ENABLE_FP16="on"
+      fi
+    fi
+
     if [[ "${local_lite_platform}" == "arm64" ]]; then
       if [ "$(uname)" == "Darwin" ]; then
         cmake -DCMAKE_TOOLCHAIN_FILE=${BASEPATH}/cmake/lite_ios.cmake -DARCHS="arm64" -DENABLE_BITCODE=0                   \
@@ -593,10 +611,10 @@ build_lite()
       else
         checkndk
         echo "default link libc++_static.a, export MSLITE_ANDROID_STL=c++_shared to link libc++_shared.so"
-        cmake -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" -DANDROID_NATIVE_API_LEVEL="19"          \
-              -DANDROID_NDK="${ANDROID_NDK}" -DANDROID_ABI="armeabi-v7a" -DANDROID_TOOLCHAIN_NAME="clang"                          \
-              -DANDROID_STL=${MSLITE_ANDROID_STL}  -DCMAKE_BUILD_TYPE=${LITE_BUILD_TYPE} -DBUILD_MINDDATA=${COMPILE_MINDDATA_LITE} \
-              -DPLATFORM_ARM32="on" -DENABLE_NEON="on"  -DENABLE_FP16="on" -DCMAKE_INSTALL_PREFIX=${BASEPATH}/output/tmp           \
+        cmake -DCMAKE_TOOLCHAIN_FILE=${CMAKE_TOOLCHAIN_FILE} -DTOOLCHAIN_NAME=${CMAKE_TOOLCHAIN_NAME} -DANDROID_NATIVE_API_LEVEL=${ANDROID_NATIVE_API_LEVEL}          \
+              -DANDROID_NDK=${CMAKE_ANDROID_NDK} -DANDROID_ABI=${CMAKE_ANDROID_ABI} -DANDROID_TOOLCHAIN_NAME=${CMAKE_ANDROID_TOOLCHAIN_NAME}                    \
+              -DANDROID_STL=${CMAKE_ANDROID_STL}  -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DBUILD_MINDDATA=${COMPILE_MINDDATA_LITE} \
+              -DPLATFORM_ARM32="on" -DENABLE_NEON="on"  -DENABLE_FP16=${ENABLE_FP16} -DCMAKE_INSTALL_PREFIX=${BASEPATH}/output/tmp           \
               -DMS_VERSION_MAJOR=${VERSION_MAJOR} -DMS_VERSION_MINOR=${VERSION_MINOR} -DMS_VERSION_REVISION=${VERSION_REVISION}    \
               -DENABLE_ASAN=${ENABLE_ASAN} -DENABLE_VERBOSE=${ENABLE_VERBOSE} "${BASEPATH}/mindspore/lite"
       fi
diff --git a/cmake/external_libs/flatbuffers.cmake b/cmake/external_libs/flatbuffers.cmake
index 78288263c1d..8ec9d6730f6 100644
--- a/cmake/external_libs/flatbuffers.cmake
+++ b/cmake/external_libs/flatbuffers.cmake
@@ -18,7 +18,8 @@ mindspore_add_pkg(flatbuffers
         EXE flatc
         URL ${REQ_URL}
         MD5 ${MD5}
-        CMAKE_OPTION -DFLATBUFFERS_BUILD_TESTS=OFF -DCMAKE_INSTALL_LIBDIR=lib)
+        CMAKE_OPTION -DCMAKE_C_COMPILER=${FLATC_GCC_COMPILER} -DCMAKE_CXX_COMPILER=${FLATC_GXX_COMPILER}
+        -DFLATBUFFERS_BUILD_TESTS=OFF -DCMAKE_INSTALL_LIBDIR=lib)
 
 include_directories(${flatbuffers_INC})
 add_library(mindspore::flatbuffers ALIAS flatbuffers::flatbuffers)
diff --git a/cmake/package_lite.cmake b/cmake/package_lite.cmake
index 43ae08f1b63..a6e513ef315 100644
--- a/cmake/package_lite.cmake
+++ b/cmake/package_lite.cmake
@@ -15,6 +15,7 @@ set(MIND_DATA_INC_DIR ${RUNTIME_PKG_NAME}/inference/include/dataset)
 set(TURBO_DIR ${RUNTIME_PKG_NAME}/inference/third_party/libjpeg-turbo)
 set(SECUREC_DIR ${RUNTIME_PKG_NAME}/inference/third_party/securec)
 set(MINDSPORE_LITE_LIB_NAME libmindspore-lite)
+set(MINDSPORE_CORE_LIB_NAME libmindspore_core)
 set(BENCHMARK_NAME benchmark)
 set(BENCHMARK_ROOT_DIR ${RUNTIME_PKG_NAME}/tools/benchmark)
 
@@ -253,6 +254,10 @@ elseif(PLATFORM_ARM32)
                     ${RUNTIME_COMPONENT_NAME})
         endif()
     endif()
+    if(MSLITE_ENABLE_NNIE AND TARGET_HIMIX200)
+        install(FILES ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie.so
+                DESTINATION ${RUNTIME_PKG_NAME}/providers/3516D COMPONENT ${RUNTIME_COMPONENT_NAME})
+    endif()
 elseif(WIN32)
     get_filename_component(CXX_DIR ${CMAKE_CXX_COMPILER} PATH)
     file(GLOB LIB_LIST ${CXX_DIR}/libstdc++-6.dll ${CXX_DIR}/libwinpthread-1.dll
@@ -393,6 +398,39 @@ else()
         install(FILES ${TOP_DIR}/mindspore/lite/build/tools/cropper/cropper_mapping_npu.cfg
                 DESTINATION ${CROPPER_ROOT_DIR} COMPONENT ${RUNTIME_COMPONENT_NAME})
     endif()
+    if(NOT SUPPORT_TRAIN AND MSLITE_ENABLE_NNIE)
+        install(DIRECTORY ${TOP_DIR}/mindspore/lite/build/schema/ DESTINATION ${CONVERTER_ROOT_DIR}/include/schema
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(DIRECTORY ${TOP_DIR}/mindspore/core/abstract/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/abstract
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(DIRECTORY ${TOP_DIR}/mindspore/core/base/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/base
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(DIRECTORY ${TOP_DIR}/mindspore/core/ir/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/ir
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(DIRECTORY ${TOP_DIR}/mindspore/core/ops/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/ops
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(DIRECTORY ${TOP_DIR}/mindspore/core/utils/ DESTINATION ${CONVERTER_ROOT_DIR}/include/core/utils
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.h")
+        install(FILES ${TOP_DIR}/mindspore/lite/build/tools/converter/mindspore_core/${MINDSPORE_CORE_LIB_NAME}.a
+                DESTINATION ${CONVERTER_ROOT_DIR}/lib COMPONENT ${RUNTIME_COMPONENT_NAME})
+        install(FILES ${TOP_DIR}/mindspore/lite/build/securec/src/libsecurec.a DESTINATION ${CONVERTER_ROOT_DIR}/lib
+                COMPONENT ${RUNTIME_COMPONENT_NAME})
+        file(GLOB PROTOBUF_LIB_PATH ${TOP_DIR}/mindspore/lite/build/.mslib/protobuf_*/lib/libprotobuf.a)
+        install(FILES ${PROTOBUF_LIB_PATH} DESTINATION ${CONVERTER_ROOT_DIR}/lib
+                COMPONENT ${RUNTIME_COMPONENT_NAME})
+        install(FILES ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_converter.so
+                DESTINATION ${CONVERTER_ROOT_DIR}/providers/3516D/ COMPONENT ${RUNTIME_COMPONENT_NAME})
+        install(FILES ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_data_process.so
+                DESTINATION ${CONVERTER_ROOT_DIR}/providers/3516D COMPONENT ${RUNTIME_COMPONENT_NAME})
+        install(FILES ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/libnnie_mapper.so
+                DESTINATION ${CONVERTER_ROOT_DIR}/providers/3516D COMPONENT ${RUNTIME_COMPONENT_NAME})
+        install(DIRECTORY ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/
+                DESTINATION ${CONVERTER_ROOT_DIR}/providers/3516D/third_party/opencv-4.2.0
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.so*")
+        install(DIRECTORY ${TOP_DIR}/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/
+                DESTINATION ${CONVERTER_ROOT_DIR}/providers/3516D/third_party/protobuf-3.9.0
+                COMPONENT ${RUNTIME_COMPONENT_NAME} FILES_MATCHING PATTERN "*.so*")
+    endif()
 endif()
 
 if(CMAKE_SYSTEM_NAME MATCHES "Windows")
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/arg_min_max_parameter.h b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/arg_min_max_parameter.h
index bb2cb50a060..ab6f74b29b3 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/arg_min_max_parameter.h
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/arg_min_max_parameter.h
@@ -31,7 +31,9 @@ typedef struct ArgElement {
     int32_t i_data_;
     float f_data_;
 #ifdef ENABLE_ARM
+#ifndef SUPPORT_NNIE
     float16_t f16_data_;
+#endif
 #endif
   } data_;
 } ArgElement;
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/arithmetic_fp32.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/arithmetic_fp32.c
index 1f87f7e921e..8f56c33b80c 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/arithmetic_fp32.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/arithmetic_fp32.c
@@ -97,8 +97,13 @@ int ElementLogicalAnd(const float *in0, const float *in1, float *out, int size)
   uint32x4_t mask = vmovq_n_u32(((uint32_t)(1u << 31) - 1));
   uint32x4_t zeros = vdupq_n_u32(0);
   for (; index <= size - 4; index += C4NUM) {
+#ifndef SUPPORT_NNIE
     uint32x4_t vin0 = vandq_u32(vreinterpretq_s32_f32(vld1q_f32(in0 + index)), mask);
     uint32x4_t vin1 = vandq_u32(vreinterpretq_s32_f32(vld1q_f32(in1 + index)), mask);
+#else
+    uint32x4_t vin0 = vandq_u32(vreinterpretq_u32_f32(vld1q_f32(in0 + index)), mask);
+    uint32x4_t vin1 = vandq_u32(vreinterpretq_u32_f32(vld1q_f32(in1 + index)), mask);
+#endif
     float32x4_t vout = vbslq_f32(vceqq_u32(vandq_u32(vin0, vin1), zeros), vfalse, vtrue);
     vst1q_f32(out + index, vout);
   }
@@ -133,8 +138,13 @@ int ElementLogicalOr(const float *in0, const float *in1, float *out, int size) {
   uint32x4_t mask = vmovq_n_u32(((uint32_t)(1u << 31) - 1));
   uint32x4_t zeros = vdupq_n_u32(0);
   for (; index <= size - 4; index += C4NUM) {
+#ifndef SUPPORT_NNIE
     uint32x4_t vin0 = vandq_u32(vreinterpretq_s32_f32(vld1q_f32(in0 + index)), mask);
     uint32x4_t vin1 = vandq_u32(vreinterpretq_s32_f32(vld1q_f32(in1 + index)), mask);
+#else
+    uint32x4_t vin0 = vandq_u32(vreinterpretq_u32_f32(vld1q_f32(in0 + index)), mask);
+    uint32x4_t vin1 = vandq_u32(vreinterpretq_u32_f32(vld1q_f32(in1 + index)), mask);
+#endif
     float32x4_t vout = vbslq_f32(vceqq_u32(vorrq_u32(vin0, vin1), zeros), vfalse, vtrue);
     vst1q_f32(out + index, vout);
   }
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd_fp32.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd_fp32.c
index ec493c718ee..24068add231 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd_fp32.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/deconv_winograd_fp32.c
@@ -219,6 +219,7 @@ void TiledC4MatmulFp32(float *dst, const float *src, const float *weight, size_t
 #endif
 
 #ifdef ENABLE_ARM32
+#ifndef SUPPORT_NNIE
 void DeConvWgMergeArm32(const float *src_ptr, float *dst_ptr, size_t src_step, size_t dst_step) {
   asm volatile(
     "mov r11, %[src_ptr]\n"
@@ -277,6 +278,66 @@ void DeConvWgMergeArm32(const float *src_ptr, float *dst_ptr, size_t src_step, s
     : "r8", "r10", "r11", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
   return;
 }
+#else
+void DeConvWgMergeArm32(const float *src_ptr, float *dst_ptr, size_t src_step, size_t dst_step) {
+  asm volatile(
+    "mov r7, %[src_ptr]\n"
+    "mov r8, %[dst_ptr]\n"
+    "mov r10, r8\n"
+
+    "vld1.32 {q0}, [r7], %[src_step]\n"
+    "vld1.32 {q1}, [r8], %[dst_step]\n"
+    "vld1.32 {q2}, [r7], %[src_step]\n"
+    "vld1.32 {q3}, [r8], %[dst_step]\n"
+
+    "vadd.f32 q0, q0, q1\n"
+    "vld1.32 {q8}, [r7], %[src_step]\n"
+    "vadd.f32 q2, q2, q3\n"
+
+    "vst1.32 {q0}, [r10], %[dst_step]\n"
+    "vst1.32 {q2}, [r10], %[dst_step]\n"
+
+    "vld1.32 {q9}, [r8], %[dst_step]\n"
+
+    "vld1.32 {q10}, [r7], %[src_step]\n"
+
+    "vadd.f32 q8, q8, q9\n"
+    "vld1.32 {q11}, [r8], %[dst_step]\n"
+    "vadd.f32 q10, q10, q11\n"
+
+    "vld1.32 {q0}, [r7], %[src_step]\n"
+    "vst1.32 {q8}, [r10], %[dst_step]\n"
+    "vst1.32 {q10}, [r10], %[dst_step]\n"
+
+    "vld1.32 {q1}, [r8], %[dst_step]\n"
+
+    "vld1.32 {q2}, [r7], %[src_step]\n"
+    "vld1.32 {q3}, [r8], %[dst_step]\n"
+
+    "vadd.f32 q0, q0, q1\n"
+    "vadd.f32 q2, q2, q3\n"
+
+    "vst1.32 {q0}, [r10], %[dst_step]\n"
+    "vst1.32 {q2}, [r10], %[dst_step]\n"
+
+    "vld1.32 {q8}, [r7], %[src_step]\n"
+    "vld1.32 {q9}, [r8], %[dst_step]\n"
+
+    "vld1.32 {q10}, [r7], %[src_step]\n"
+    "vld1.32 {q11}, [r8], %[dst_step]\n"
+
+    "vadd.f32 q8, q8, q9\n"
+    "vadd.f32 q10, q10, q11\n"
+
+    "vst1.32 {q8}, [r10], %[dst_step]\n"
+    "vst1.32 {q10}, [r10], %[dst_step]\n"
+
+    :
+    : [ src_ptr ] "r"(src_ptr), [ dst_ptr ] "r"(dst_ptr), [ src_step ] "r"(src_step), [ dst_step ] "r"(dst_step)
+    : "r8", "r10", "r7", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
+  return;
+}
+#endif
 #endif
 
 void DeConvWgMerge(const float *src, float *dst, size_t src_stride, size_t dst_stride, size_t count) {
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_fp32.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_fp32.c
index f4cfaccdc34..33328b41f1c 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_fp32.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/matmul_fp32.c
@@ -420,6 +420,7 @@ void RowMajor2Col8Major_arm64(const float *src_c, float *dst_c, size_t col) {
 }
 #endif
 #ifdef ENABLE_ARM32
+#ifndef SUPPORT_NNIE
 void RowMajor2Col8Major_arm32(const float *src_c, float *dst_c, size_t col) {
   size_t stride = col * sizeof(float);
   asm volatile(
@@ -462,6 +463,50 @@ void RowMajor2Col8Major_arm32(const float *src_c, float *dst_c, size_t col) {
     : "r10", "r11", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
   return;
 }
+#else
+void RowMajor2Col8Major_arm32(const float *src_c, float *dst_c, size_t col) {
+  size_t stride = col * sizeof(float);
+  asm volatile(
+    "mov r10, %[src_c]\n"
+    "mov r7, %[dst_c]\n"
+
+    "vld1.32 {q0}, [r10], %[stride]\n"
+    "vld1.32 {q2}, [r10], %[stride]\n"
+    "vld1.32 {q4}, [r10], %[stride]\n"
+    "vld1.32 {q6}, [r10], %[stride]\n"
+
+    "vtrn.32 d0, d4\n"
+    "vtrn.32 d1, d5\n"
+    "vtrn.32 d8, d12\n"
+    "vtrn.32 d9, d13\n"
+
+    "vld1.32 {q1}, [r10], %[stride]\n"
+    "vld1.32 {q3}, [r10], %[stride]\n"
+    "vld1.32 {q5}, [r10], %[stride]\n"
+    "vld1.32 {q7}, [r10], %[stride]\n"
+
+    "vswp d1, d8\n"
+    "vswp d5, d12\n"
+
+    "vtrn.32 d2, d6\n"
+    "vtrn.32 d3, d7\n"
+    "vtrn.32 d10, d14\n"
+    "vtrn.32 d11, d15\n"
+
+    "vswp d3, d10\n"
+    "vswp d7, d14\n"
+
+    "vst1.32 {q0, q1}, [r7]!\n"
+    "vst1.32 {q2, q3}, [r7]!\n"
+    "vst1.32 {q4, q5}, [r7]!\n"
+    "vst1.32 {q6, q7}, [r7]!\n"
+
+    :
+    : [ dst_c ] "r"(dst_c), [ src_c ] "r"(src_c), [ stride ] "r"(stride)
+    : "r10", "r7", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
+  return;
+}
+#endif
 #endif
 void RowMajor2Col8Major(const float *src_ptr, float *dst_ptr, size_t row, size_t col) {
   size_t row8 = row / C8NUM * C8NUM;
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/gemm.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/gemm.c
index 8815498f930..4de2a8ad69b 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/gemm.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/gemm.c
@@ -353,6 +353,7 @@ static void RowMajor2Col8MajorStride(const float *src_ptr, float *dst_ptr, size_
           "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29",
           "v30", "v31");
 #elif ENABLE_ARM32
+#ifndef SUPPORT_NNIE
       /* 8x4 row-major to col-major */
       size_t stride = col * sizeof(float);
       asm volatile(
@@ -393,6 +394,48 @@ static void RowMajor2Col8MajorStride(const float *src_ptr, float *dst_ptr, size_
         :
         : [ dst_c ] "r"(dst_c), [ src_c ] "r"(src_c), [ stride ] "r"(stride)
         : "r10", "r11", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
+#else
+      /* 8x4 row-major to col-major */
+      size_t stride = col * sizeof(float);
+      asm volatile(
+        "mov r10, %[src_c]\n"
+        "mov r7, %[dst_c]\n"
+
+        "vld1.32 {q0}, [r10], %[stride]\n"
+        "vld1.32 {q2}, [r10], %[stride]\n"
+        "vld1.32 {q4}, [r10], %[stride]\n"
+        "vld1.32 {q6}, [r10], %[stride]\n"
+
+        "vtrn.32 d0, d4\n"
+        "vtrn.32 d1, d5\n"
+        "vtrn.32 d8, d12\n"
+        "vtrn.32 d9, d13\n"
+
+        "vld1.32 {q1}, [r10], %[stride]\n"
+        "vld1.32 {q3}, [r10], %[stride]\n"
+        "vld1.32 {q5}, [r10], %[stride]\n"
+        "vld1.32 {q7}, [r10], %[stride]\n"
+
+        "vswp d1, d8\n"
+        "vswp d5, d12\n"
+
+        "vtrn.32 d2, d6\n"
+        "vtrn.32 d3, d7\n"
+        "vtrn.32 d10, d14\n"
+        "vtrn.32 d11, d15\n"
+
+        "vswp d3, d10\n"
+        "vswp d7, d14\n"
+
+        "vst1.32 {q0, q1}, [r7]!\n"
+        "vst1.32 {q2, q3}, [r7]!\n"
+        "vst1.32 {q4, q5}, [r7]!\n"
+        "vst1.32 {q6, q7}, [r7]!\n"
+
+        :
+        : [ dst_c ] "r"(dst_c), [ src_c ] "r"(src_c), [ stride ] "r"(stride)
+        : "r10", "r7", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
+#endif
 #else
       for (int tr = 0; tr < 8; tr++) {
         for (int tc = 0; tc < 4; tc++) {
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/pooling_grad.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/pooling_grad.c
index 6084d4ab637..f2ac7ccba8f 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/pooling_grad.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32_grad/pooling_grad.c
@@ -146,7 +146,8 @@ void MaxPoolingGrad(const float *input_ptr, const float *dy_ptr, float *output_p
 #ifdef ENABLE_ARM
               uint32x4_t index = {val_idx, val_idx + 1, val_idx + 2, val_idx + 3};
               float32x4_t in = vld1q_f32(inPtr + val_idx);
-              max_idx = MaxIndex(in, &max_val, index, max_idx);
+              max_idx = vreinterpretq_u32_s32(
+                MaxIndex(in, &max_val, vreinterpretq_s32_u32(index), vreinterpretq_s32_u32(max_idx)));
 #else
               float val[4] = {inPtr[val_idx], inPtr[val_idx + 1], inPtr[val_idx + 2], inPtr[val_idx + 3]};
               for (int i = 0; i < 4; i++) {
diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/matmul_int8.c b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/matmul_int8.c
index ac80749c0cb..ae0c0ef96ff 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/matmul_int8.c
+++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/matmul_int8.c
@@ -137,10 +137,10 @@ void RowMajor2Row16x4MajorInt8(int8_t *src_ptr, int8_t *dst_ptr, int row, int co
         "vld1.8 {q2}, [r0], r2 \n"
         "vld1.8 {q3}, [r0], r2 \n"
 
-        "vst1.32 q0, [r1], r3 \n"
-        "vst1.32 q1, [r1], r3 \n"
-        "vst1.32 q2, [r1], r3 \n"
-        "vst1.32 q3, [r1], r3 \n"
+        "vst1.32 {d0, d1}, [r1], r3 \n"
+        "vst1.32 {d2, d3}, [r1], r3 \n"
+        "vst1.32 {d4, d5}, [r1], r3 \n"
+        "vst1.32 {d6, d7}, [r1], r3 \n"
 
         :
         : [ dst_c ] "r"(dst_c), [ src_c ] "r"(src_c), [ col_offset ] "r"(col_offset)
diff --git a/mindspore/core/base/base.cc b/mindspore/core/base/base.cc
index 07ed252e962..94c73127368 100644
--- a/mindspore/core/base/base.cc
+++ b/mindspore/core/base/base.cc
@@ -17,6 +17,7 @@
 #include "base/base.h"
 #include <atomic>
 #include <mutex>
+#include <string>
 #include <unordered_map>
 
 namespace mindspore {
@@ -32,7 +33,7 @@ uint32_t Base::GetTypeId(const char *const type_name) {
   if (it != t->map.end()) {
     return it->second;
   }
-  uint32_t tid = ++(t->type_counter);
+  uint32_t tid = std::hash<std::string>()(type_name);
   t->map[type_name] = tid;
   return tid;
 }
diff --git a/mindspore/core/base/core_ops.h b/mindspore/core/base/core_ops.h
index e1f3a92bb8f..9461cb80675 100644
--- a/mindspore/core/base/core_ops.h
+++ b/mindspore/core/base/core_ops.h
@@ -589,6 +589,8 @@ inline const PrimitivePtr kPrimIf = std::make_shared<Primitive>("If");
 inline const PrimitivePtr kPrimAvgPoolFusion = std::make_shared<Primitive>("AvgPoolFusion");
 inline const PrimitivePtr kPrimMaxPoolFusion = std::make_shared<Primitive>("MaxPoolFusion");
 inline const PrimitivePtr kPrimActivation = std::make_shared<Primitive>("Activation");
+inline const PrimitivePtr kPrimPReLUFusion = std::make_shared<Primitive>("PReLUFusion");
+inline const PrimitivePtr kPrimCustom = std::make_shared<Primitive>("Custom");
 inline const PrimitivePtr kPrimTopKFusion = std::make_shared<Primitive>("TopKFusion");
 inline const PrimitivePtr kPrimTileFusion = std::make_shared<Primitive>("TileFusion");
 inline const PrimitivePtr kPrimReduceFusion = std::make_shared<Primitive>("ReduceFusion");
diff --git a/mindspore/core/mindrt/src/thread/inter_threadpool.cc b/mindspore/core/mindrt/src/thread/inter_threadpool.cc
index b1928e6eeb6..143ef6d1ae8 100644
--- a/mindspore/core/mindrt/src/thread/inter_threadpool.cc
+++ b/mindspore/core/mindrt/src/thread/inter_threadpool.cc
@@ -42,6 +42,7 @@ void InterThreadPool::ThreadAsyncRun(Worker *worker) {
 }
 
 void InterThreadPool::ActorThreadRun() {
+#ifndef SUPPORT_NNIE
   ActorReference actor;
   {
     std::unique_lock<std::mutex> _l(actor_mutex_);
@@ -54,6 +55,7 @@ void InterThreadPool::ActorThreadRun() {
   }
   actor->Run();
   finish_cond_var_.notify_one();
+#endif
 }
 
 void InterThreadPool::EnqueReadyActor(const ActorReference &actor) {
diff --git a/mindspore/core/mindrt/src/thread/threadpool.cc b/mindspore/core/mindrt/src/thread/threadpool.cc
index b8556e9c7a3..a2d5d6555d8 100644
--- a/mindspore/core/mindrt/src/thread/threadpool.cc
+++ b/mindspore/core/mindrt/src/thread/threadpool.cc
@@ -100,7 +100,7 @@ void ThreadPool::KernelThreadRun(Worker *worker) {
 int ThreadPool::ParallelLaunch(const Func &func, Content content, int task_num) {
   // distribute task to the KernelThread and the free ActorThread,
   // if the task num is greater than the KernelThread num
-  Task task = Task(func, content);
+  Task task = {func, content};
   Worker *curr = CurrentWorker();
   if (inter_thread_num_ == thread_num_ || curr == nullptr) {
     SyncRunTask(&task, task_num);
diff --git a/mindspore/lite/CMakeLists.txt b/mindspore/lite/CMakeLists.txt
index db8f7bebbaf..a648de09456 100644
--- a/mindspore/lite/CMakeLists.txt
+++ b/mindspore/lite/CMakeLists.txt
@@ -3,7 +3,12 @@ project(Lite)
 
 set(BUILD_LITE "on")
 
-if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.3.0)
+if(TOOLCHAIN_NAME STREQUAL "himix200")
+  set(TARGET_HIMIX200 on)
+  add_compile_definitions(SUPPORT_NNIE)
+endif()
+
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 7.3.0 AND NOT TARGET_HIMIX200)
     message(FATAL_ERROR "GCC version ${CMAKE_CXX_COMPILER_VERSION} must not be less than 7.3.0")
 endif()
 
@@ -22,6 +27,7 @@ option(MSLITE_ENABLE_AVX "enable AVX instruction set, only x86_64 support" off)
 option(MSLITE_ENABLE_CONVERTER "enable converter, only x86_64 support" on)
 option(MSLITE_ENABLE_TOOLS "enable tools" on)
 option(MSLITE_ENABLE_TESTCASES "enable testcase" off)
+option(MSLITE_ENABLE_NNIE "enable NNIE" off)
 
 # Option that can be configured through manually
 option(ENABLE_ARM82_A32 "if build fp16 on platform_arm32" off)
@@ -55,6 +61,9 @@ endif()
 if(DEFINED ENV{MSLITE_ENABLE_TESTCASES})
     set(MSLITE_ENABLE_TESTCASES $ENV{MSLITE_ENABLE_TESTCASES})
 endif()
+if(DEFINED ENV{MSLITE_ENABLE_NNIE})
+    set(MSLITE_ENABLE_NNIE $ENV{MSLITE_ENABLE_NNIE})
+endif()
 
 if(PLATFORM_ARM64 OR PLATFORM_ARM32)
     set(PLATFORM_ARM "on")
@@ -197,6 +206,9 @@ else()
     -Wno-deprecated-declarations -Wno-missing-braces ${CMAKE_C_FLAGS}")
     set(CMAKE_CXX_FLAGS "-fPIC -fPIE -D_FORTIFY_SOURCE=2 -O2 -Wall -Werror -fstack-protector-strong -Wno-attributes \
     -Wno-deprecated-declarations -Wno-missing-braces -Wno-overloaded-virtual ${CMAKE_CXX_FLAGS}")
+    if(TARGET_HIMIX200)
+        set(CMAKE_CXX_FLAGS "-Wno-error=maybe-uninitialized ${CMAKE_CXX_FLAGS}")
+    endif()
     if(NOT WIN32)
         set(CMAKE_SHARED_LINKER_FLAGS "-Wl,-z,relro,-z,now -Wl,-z,noexecstack ${CMAKE_SHARED_LINKER_FLAGS}")
         set(CMAKE_EXE_LINKER_FLAGS "-Wl,-z,relro,-z,now -Wl,-z,noexecstack ${CMAKE_EXE_LINKER_FLAGS}")
diff --git a/mindspore/lite/src/CMakeLists.txt b/mindspore/lite/src/CMakeLists.txt
index 6db52ab4452..b580fe8bcd9 100644
--- a/mindspore/lite/src/CMakeLists.txt
+++ b/mindspore/lite/src/CMakeLists.txt
@@ -14,6 +14,9 @@ if(PLATFORM_ARM32 OR PLATFORM_ARM64)
         -fdata-sections -ffast-math -fno-rtti -fno-exceptions")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}  -fomit-frame-pointer -fstrict-aliasing -ffunction-sections \
         -fdata-sections -ffast-math -fno-rtti -fno-exceptions")
+        if(TARGET_HIMIX200)
+            string(REPLACE "-fno-rtti " "" CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
+        endif()
     endif()
     if("${CMAKE_BUILD_TYPE}" STREQUAL "Release" AND APPLE)
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fstrict-aliasing -ffunction-sections \
@@ -25,7 +28,10 @@ if(PLATFORM_ARM32 OR PLATFORM_ARM64)
     endif()
 endif()
 
-set(API_SRC
+if(TARGET_HIMIX200)
+    set(API_SRC)
+else()
+    set(API_SRC
         ${CORE_DIR}/utils/status.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/cxx_api/cell.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/cxx_api/serialization.cc
@@ -35,7 +41,8 @@ set(API_SRC
         ${CMAKE_CURRENT_SOURCE_DIR}/cxx_api/model/model_impl.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/cxx_api/graph/graph.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/cxx_api/tensor/tensor_impl.cc
-)
+        )
+endif()
 
 if(SUPPORT_NPU)
     include_directories(${DDK_PATH})
@@ -138,6 +145,16 @@ if(ENABLE_MINDRT)
         ${CMAKE_CURRENT_SOURCE_DIR}/lite_mindrt.cc
         ${CMAKE_CURRENT_SOURCE_DIR}/mindrt_executor.cc
         )
+elseif(TARGET_HIMIX200)
+    include_directories(${CORE_DIR}/mindrt)
+    include_directories(${CORE_DIR}/mindrt/include)
+    include_directories(${CORE_DIR}/mindrt/src)
+    set(LITE_SRC
+        ${LITE_SRC}
+        ${CORE_DIR}/mindrt/src/thread/core_affinity.cc
+        ${CORE_DIR}/mindrt/src/thread/inter_threadpool.cc
+        ${CORE_DIR}/mindrt/src/thread/threadpool.cc
+        )
 endif()
 
 add_subdirectory(ops)
@@ -199,7 +216,7 @@ if(SUPPORT_NPU)
     target_link_libraries(mindspore-lite npu_kernel_mid)
     target_link_libraries(mindspore-lite_static npu_kernel_mid)
 endif()
-if(PLATFORM_ARM32 OR PLATFORM_ARM64)
+if(PLATFORM_ARM32 OR PLATFORM_ARM64 AND NOT TARGET_HIMIX200)
     target_link_libraries(mindspore-lite log)
     target_link_libraries(mindspore-lite_static log)
 endif()
@@ -222,12 +239,15 @@ if(SUPPORT_TRAIN)
   target_link_libraries(mindspore-lite-train_static minddata-lite mindspore-lite)
 endif()
 
-if("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
-    if(NOT APPLE AND PLATFORM_ARM)
-        set(NDK_STRIP
-            "${ANDROID_NDK}/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/aarch64-linux-android/bin/strip")
+if(NOT APPLE AND PLATFORM_ARM AND NOT TARGET_HIMIX200)
+    set(NDK_STRIP
+        "${ANDROID_NDK}/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/aarch64-linux-android/bin/strip")
+endif()
+
+if(NOT APPLE AND "${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+    if(PLATFORM_ARM)
         add_custom_command(TARGET mindspore-lite POST_BUILD COMMAND ${NDK_STRIP}
-                            ${CMAKE_BINARY_DIR}/src/libmindspore-lite*.so)
+                ${CMAKE_BINARY_DIR}/src/libmindspore-lite*.so)
     elseif(NOT WIN32)
         add_custom_command(TARGET mindspore-lite POST_BUILD COMMAND strip ${CMAKE_BINARY_DIR}/src/libmindspore-lite*.so)
     endif()
diff --git a/mindspore/lite/src/common/log_adapter.cc b/mindspore/lite/src/common/log_adapter.cc
index 434ddd3522f..44231a23871 100644
--- a/mindspore/lite/src/common/log_adapter.cc
+++ b/mindspore/lite/src/common/log_adapter.cc
@@ -93,11 +93,9 @@ const char *EnumStrForMsLogLevel(MsLogLevel level) {
 
 void LogWriter::OutputLog(const std::ostringstream &msg) const {
   if (IsPrint(log_level_)) {
-#ifdef ENABLE_ARM
-#if defined(__ANDROID__) || defined(ANDROID)
+#if defined(ENABLE_ARM) && (defined(__ANDROID__) || defined(ANDROID))
     __android_log_print(GetAndroidLogLevel(log_level_), ANDROID_LOG_TAG, "[%s:%d] %s] %s", location_.file_,
                         location_.line_, location_.func_, msg.str().c_str());
-#endif
 #else
     printf("%s [%s:%d] %s] %s\n", EnumStrForMsLogLevel(log_level_), location_.file_, location_.line_, location_.func_,
            msg.str().c_str());
diff --git a/mindspore/lite/src/lite_session.h b/mindspore/lite/src/lite_session.h
index 7ec7ebe48be..e1b71f5be1b 100644
--- a/mindspore/lite/src/lite_session.h
+++ b/mindspore/lite/src/lite_session.h
@@ -139,7 +139,7 @@ class LiteSession : public session::LiteSession {
   std::unordered_map<Tensor *, Tensor *> graph_output_map_; /* <calculate-tensor,  graph-output-tensor> */
   Executor *executor_ = nullptr;
   Model *model_ = nullptr;
-  std::atomic<bool> is_running_ = false;
+  std::atomic<bool> is_running_ = {false};
   bool is_train_session_ = false;
   friend class TransferSession;
 #if SUPPORT_NPU
diff --git a/mindspore/lite/src/runtime/inner_allocator.h b/mindspore/lite/src/runtime/inner_allocator.h
index 7319df38745..0c40677f878 100644
--- a/mindspore/lite/src/runtime/inner_allocator.h
+++ b/mindspore/lite/src/runtime/inner_allocator.h
@@ -52,7 +52,7 @@ class DefaultAllocator : public Allocator {
   void UnLock();
   bool ReuseMemory(size_t free_size, size_t size);
   struct MemBuf {
-    std::atomic_int ref_count_ = 0;
+    std::atomic_int ref_count_ = {0};
     size_t size;
     void *buf;
   };
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp32/nonzero_fp32.cc b/mindspore/lite/src/runtime/kernel/arm/fp32/nonzero_fp32.cc
index 9a69d982109..60f267d1efc 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp32/nonzero_fp32.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp32/nonzero_fp32.cc
@@ -47,7 +47,7 @@ int NonZeroCPUKernel::Run() {
   }
   auto non_zero_nums = out_tensor->shape()[1];
   int non_zero_count = 0;
-  std::vector coordiate_values(in_tensor->shape().size(), 0);
+  std::vector<int> coordiate_values(in_tensor->shape().size(), 0);
   for (int i = 0; i < in_tensor->ElementsNum(); i += 1) {
     if (input_data[i]) {
       for (size_t j = 0; j < input_dim_size; j++) {
diff --git a/mindspore/lite/src/sub_graph_kernel.h b/mindspore/lite/src/sub_graph_kernel.h
index a117f26a527..8530911e5b1 100644
--- a/mindspore/lite/src/sub_graph_kernel.h
+++ b/mindspore/lite/src/sub_graph_kernel.h
@@ -145,7 +145,7 @@ class CpuFp32SubGraph : public CpuSubGraph {
                   std::vector<LiteKernel *> nodes, Kernel *kernel)
       : CpuSubGraph(std::move(in_kernels), std::move(out_kernels), std::move(nodes), kernel) {
     subgraph_type_ = kCpuFP32SubGraph;
-    static std::atomic_int index = 0;
+    static std::atomic_int index = {0};
     this->set_name("CpuFP32SubGraph" + std::to_string(index++));
     desc_.data_type = kNumberTypeFloat32;
   }
diff --git a/mindspore/lite/src/tensor.h b/mindspore/lite/src/tensor.h
index f26fe4f875c..334e670f19f 100644
--- a/mindspore/lite/src/tensor.h
+++ b/mindspore/lite/src/tensor.h
@@ -217,8 +217,8 @@ class Tensor : public mindspore::tensor::MSTensor {
   std::vector<int> shape_;
   schema::Format format_;
   Category category_;
-  std::atomic_int ref_count_ = 0;
-  int init_ref_count_ = 0;
+  std::atomic_int ref_count_ = {0};
+  size_t init_ref_count_ = 0;
   std::vector<QuantArg> quant_params_;
   std::vector<float> quant_clusters_;
   AllocatorPtr allocator_ = nullptr;
diff --git a/mindspore/lite/tools/benchmark/CMakeLists.txt b/mindspore/lite/tools/benchmark/CMakeLists.txt
index 2be7b049b6d..20a74c3217b 100644
--- a/mindspore/lite/tools/benchmark/CMakeLists.txt
+++ b/mindspore/lite/tools/benchmark/CMakeLists.txt
@@ -12,7 +12,7 @@ add_executable(benchmark
 
 add_dependencies(benchmark fbs_src)
 
-if(PLATFORM_ARM32 OR PLATFORM_ARM64)
+if(PLATFORM_ARM32 OR PLATFORM_ARM64 AND NOT TARGET_HIMIX200)
     if(SUPPORT_NPU AND ANDROID_STL STREQUAL "c++_static")
         target_link_libraries(benchmark mindspore-lite mindspore::json c++_shared)
     else()
diff --git a/mindspore/lite/tools/converter/CMakeLists.txt b/mindspore/lite/tools/converter/CMakeLists.txt
index 867e6d4cc06..53d030baa48 100644
--- a/mindspore/lite/tools/converter/CMakeLists.txt
+++ b/mindspore/lite/tools/converter/CMakeLists.txt
@@ -125,6 +125,8 @@ set(LITE_SRC
         ${SRC_DIR}/tensor.cc
         ${SRC_DIR}/ms_tensor.cc
         ${SRC_DIR}/tensorlist.cc
+        ${SRC_DIR}/registry/kernel_interface_registry.cc
+        ${SRC_DIR}/registry/kernel_interface.cc
         ${SRC_DIR}/kernel_registry.cc
         ${SRC_DIR}/inner_kernel.cc
         ${SRC_DIR}/lite_kernel.cc
diff --git a/mindspore/lite/tools/converter/converter_flags.cc b/mindspore/lite/tools/converter/converter_flags.cc
index 0b0863ff948..e3ebb98d1b6 100644
--- a/mindspore/lite/tools/converter/converter_flags.cc
+++ b/mindspore/lite/tools/converter/converter_flags.cc
@@ -45,7 +45,10 @@ Flags::Flags() {
   AddFlag(&Flags::bitNumIn, "bitNum", "Weight quantization bitNum", "8");
   AddFlag(&Flags::quantWeightSizeStr, "quantWeightSize", "Weight quantization size threshold", "0");
   AddFlag(&Flags::quantWeightChannelStr, "quantWeightChannel", "Channel threshold for weight quantization", "16");
-  AddFlag(&Flags::configFile, "configFile", "Configuration for post-training, offline split op to parallel", "");
+  AddFlag(&Flags::configFile, "configFile",
+          "Configuration for post-training, offline split op to parallel,"
+          "and converter for nnie",
+          "");
   AddFlag(&Flags::trainModelIn, "trainModel",
           "whether the model is going to be trained on device. "
           "true | false",
diff --git a/mindspore/lite/tools/optimizer/common/format_utils.cc b/mindspore/lite/tools/optimizer/common/format_utils.cc
index 443875247cd..a597df43aca 100644
--- a/mindspore/lite/tools/optimizer/common/format_utils.cc
+++ b/mindspore/lite/tools/optimizer/common/format_utils.cc
@@ -56,6 +56,7 @@
 #include "ops/op_utils.h"
 #include "ops/quant_dtype_cast.h"
 #include "ops/resize.h"
+#include "ops/roi_pooling.h"
 #include "ops/sgd.h"
 #include "ops/space_to_batch.h"
 #include "ops/space_to_batch_nd.h"
@@ -88,6 +89,7 @@ static const std::unordered_map<std::string, std::vector<size_t>> NHWCOpMap = {
   {ops::kNamePReLUFusion, {1}},
   {ops::kNameResize, {1}},
   {ops::kNameResizeGrad, {}},
+  {ops::kNameROIPooling, {1}},
   {ops::kNameSGD, {2}},
   {ops::kNameSpaceToBatch, {1}},
   {ops::kNameSpaceToBatchND, {1}},
diff --git a/mindspore/lite/tools/optimizer/graph/node_infershape.cc b/mindspore/lite/tools/optimizer/graph/node_infershape.cc
index 382ab530788..1e2812a96ee 100644
--- a/mindspore/lite/tools/optimizer/graph/node_infershape.cc
+++ b/mindspore/lite/tools/optimizer/graph/node_infershape.cc
@@ -132,45 +132,49 @@ STATUS NodeInferShape::InferShape(const CNodePtr &cnode) {
     fbb.Clear();
     return lite::RET_ERROR;
   }
-  auto parameter_gen = lite::PopulateRegistry::GetInstance()->GetParameterCreator(prim->value_type(), lite::SCHEMA_CUR);
-  if (parameter_gen == nullptr) {
-    MS_LOG(ERROR) << "PopulateParameter return nullptr, type: " << schema::EnumNamePrimitiveType(prim->value_type());
-    FreeTensors(&inputs);
-    FreeTensors(&outputs);
-    fbb.Clear();
-    return lite::RET_ERROR;
-  }
-  auto parameter = parameter_gen(prim);
-  if (parameter == nullptr) {
-    MS_LOG(ERROR) << "parameter is nullptr.";
-    FreeTensors(&inputs);
-    FreeTensors(&outputs);
-    fbb.Clear();
-    return lite::RET_ERROR;
-  }
-  RectifyFormat(cnode, inputs, fmk_type_);
-  auto status = KernelInferShape(inputs, outputs, parameter);
-  if (status == lite::RET_OK) {
-    anf_prim->AddAttr(kInferDone, MakeValue<bool>(true));
-  }
-  if (status == lite::RET_OK || status == lite::RET_INFER_INVALID) {
-    auto set_status = SetCNodeAbstract(cnode, outputs, status);
-    if (set_status != lite::RET_OK) {
-      MS_LOG(ERROR) << "set CNode abstract failed: " << cnode->fullname_with_scope();
+  auto ret = KernelInferShape(inputs, outputs, prim, {});
+  if (ret == lite::RET_NOT_SUPPORT) {
+    auto parameter_gen =
+      lite::PopulateRegistry::GetInstance()->GetParameterCreator(prim->value_type(), lite::SCHEMA_CUR);
+    if (parameter_gen == nullptr) {
+      MS_LOG(ERROR) << "PopulateParameter return nullptr, type: " << schema::EnumNamePrimitiveType(prim->value_type());
       FreeTensors(&inputs);
       FreeTensors(&outputs);
-      free(parameter);
       fbb.Clear();
-      return set_status;
+      return lite::RET_ERROR;
     }
-  } else {
-    MS_LOG(ERROR) << "infer shape failed.";
+    auto parameter = parameter_gen(prim);
+    if (parameter == nullptr) {
+      MS_LOG(ERROR) << "parameter is nullptr.";
+      FreeTensors(&inputs);
+      FreeTensors(&outputs);
+      fbb.Clear();
+      return lite::RET_ERROR;
+    }
+    RectifyFormat(cnode, inputs, fmk_type_);
+    ret = KernelInferShape(inputs, outputs, parameter);
+    if (ret == lite::RET_OK) {
+      anf_prim->AddAttr(kInferDone, MakeValue<bool>(true));
+    }
+    if (ret == lite::RET_OK || ret == lite::RET_INFER_INVALID) {
+      auto set_status = SetCNodeAbstract(cnode, outputs, ret);
+      if (set_status != lite::RET_OK) {
+        MS_LOG(ERROR) << "set CNode abstract failed: " << cnode->fullname_with_scope();
+        FreeTensors(&inputs);
+        FreeTensors(&outputs);
+        free(parameter);
+        fbb.Clear();
+        return set_status;
+      }
+    } else {
+      MS_LOG(ERROR) << "infer shape failed.";
+    }
+    FreeTensors(&inputs);
+    FreeTensors(&outputs);
+    free(parameter);
   }
-  FreeTensors(&inputs);
-  FreeTensors(&outputs);
-  free(parameter);
   fbb.Clear();
-  return status;
+  return ret;
 }
 
 std::vector<int> NodeInferShape::GetInputShape(const CNodePtr &cnode, size_t index) {
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_converter.so b/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_converter.so
new file mode 100755
index 00000000000..054eab0e436
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_converter.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_data_process.so b/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_data_process.so
new file mode 100755
index 00000000000..f08583752af
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/libmslite_nnie_data_process.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/libnnie_mapper.so b/mindspore/lite/tools/providers/NNIE/3516D/libnnie_mapper.so
new file mode 100755
index 00000000000..84c48f905be
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/libnnie_mapper.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so
new file mode 100644
index 00000000000..9a116c05b8b
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2
new file mode 100644
index 00000000000..9a116c05b8b
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2.0 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2.0
new file mode 100644
index 00000000000..9a116c05b8b
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_core.so.4.2.0 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so
new file mode 100644
index 00000000000..b07fe08ce29
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2
new file mode 100644
index 00000000000..b07fe08ce29
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2.0 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2.0
new file mode 100644
index 00000000000..b07fe08ce29
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgcodecs.so.4.2.0 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so
new file mode 100644
index 00000000000..52b46fda4a9
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2
new file mode 100644
index 00000000000..52b46fda4a9
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2.0 b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2.0
new file mode 100644
index 00000000000..52b46fda4a9
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/opencv-4.2.0/lib/libopencv_imgproc.so.4.2.0 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so
new file mode 100644
index 00000000000..507bf79a191
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so.3.9.0.0 b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so.3.9.0.0
new file mode 100644
index 00000000000..507bf79a191
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotobuf.so.3.9.0.0 differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so
new file mode 100644
index 00000000000..cf792792cdc
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so differ
diff --git a/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so.3.9.0.0 b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so.3.9.0.0
new file mode 100644
index 00000000000..cf792792cdc
Binary files /dev/null and b/mindspore/lite/tools/providers/NNIE/3516D/protobuf-3.9.0/lib/libprotoc.so.3.9.0.0 differ