diff --git a/mpp/codec/mpp_dec_no_thread.cpp b/mpp/codec/mpp_dec_no_thread.cpp
index 0872e4b7..cc50c87e 100644
--- a/mpp/codec/mpp_dec_no_thread.cpp
+++ b/mpp/codec/mpp_dec_no_thread.cpp
@@ -136,7 +136,6 @@ MPP_RET mpp_dec_decode(MppDec ctx, MppPacket packet)
      * 6. copy data to hardware buffer
      */
     if (!status->dec_pkt_copy_rdy) {
-        void *dst = mpp_buffer_get_ptr(task->hal_pkt_buf_in);
         void *src = mpp_packet_get_data(task_dec->input_packet);
         size_t length = mpp_packet_get_length(task_dec->input_packet);
 
@@ -144,7 +143,8 @@ MPP_RET mpp_dec_decode(MppDec ctx, MppPacket packet)
         mpp_assert(task_dec->input_packet);
 
         dec_dbg_detail("detail: %p copy to hw length %d\n", dec, length);
-        memcpy(dst, src, length);
+        mpp_buffer_write(task->hal_pkt_buf_in, 0, src, length);
+        mpp_buffer_sync_partial_end(task->hal_pkt_buf_in, 0, length);
 
         mpp_buf_slot_set_flag(packet_slots, task_dec->input, SLOT_CODEC_READY);
         mpp_buf_slot_set_flag(packet_slots, task_dec->input, SLOT_HAL_INPUT);
diff --git a/mpp/codec/mpp_dec_normal.cpp b/mpp/codec/mpp_dec_normal.cpp
index bbc16b3f..da8b65aa 100644
--- a/mpp/codec/mpp_dec_normal.cpp
+++ b/mpp/codec/mpp_dec_normal.cpp
@@ -440,11 +440,11 @@ static MPP_RET try_proc_dec_task(Mpp *mpp, DecTask *task)
      * 6. copy prepared stream to hardware buffer
      */
     if (!task->status.dec_pkt_copy_rdy) {
-        void *dst = mpp_buffer_get_ptr(hal_buf_in);
         void *src = mpp_packet_get_data(task_dec->input_packet);
         size_t length = mpp_packet_get_length(task_dec->input_packet);
 
-        memcpy(dst, src, length);
+        mpp_buffer_write(hal_buf_in, 0, src, length);
+        mpp_buffer_sync_partial_end(hal_buf_in, 0, length);
         mpp_buf_slot_set_flag(packet_slots, task_dec->input, SLOT_CODEC_READY);
         mpp_buf_slot_set_flag(packet_slots, task_dec->input, SLOT_HAL_INPUT);
         task->status.dec_pkt_copy_rdy = 1;
@@ -571,7 +571,7 @@ static MPP_RET try_proc_dec_task(Mpp *mpp, DecTask *task)
     /* 10. whether the frame buffer group is internal or external */
     if (NULL == mpp->mFrameGroup) {
         mpp_log("mpp_dec use internal frame buffer group\n");
-        mpp_buffer_group_get_internal(&mpp->mFrameGroup, MPP_BUFFER_TYPE_ION);
+        mpp_buffer_group_get_internal(&mpp->mFrameGroup, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
     }
 
     /* 10.1 look for a unused hardware buffer for output */
@@ -1064,6 +1064,7 @@ void *mpp_dec_advanced_thread(void *data)
         mpp_port_poll(output, MPP_POLL_BLOCK);
         mpp_port_dequeue(output, &mpp_task);
         mpp_task_meta_set_frame(mpp_task, KEY_OUTPUT_FRAME, frame);
+        mpp_buffer_sync_ro_begin(mpp_frame_get_buffer(frame));
 
         // setup output task here
         mpp_port_enqueue(output, mpp_task);
diff --git a/mpp/mpp.cpp b/mpp/mpp.cpp
index 93fd4a58..4f47c253 100644
--- a/mpp/mpp.cpp
+++ b/mpp/mpp.cpp
@@ -160,7 +160,7 @@ MPP_RET Mpp::init(MppCtxType type, MppCodingType coding)
             mOutputTimeout = MPP_POLL_NON_BLOCK;
 
         if (mCoding != MPP_VIDEO_CodingMJPEG) {
-            mpp_buffer_group_get_internal(&mPacketGroup, MPP_BUFFER_TYPE_ION);
+            mpp_buffer_group_get_internal(&mPacketGroup, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
             mpp_buffer_group_limit_config(mPacketGroup, 0, 3);
 
             mpp_task_queue_setup(mInputTaskQueue, 4);
@@ -497,9 +497,15 @@ MPP_RET Mpp::get_frame(MppFrame *frame)
     }
 
     if (mFrmOut->list_size()) {
+        MppBuffer buffer;
+
         mFrmOut->del_at_head(&frm, sizeof(frame));
         mFrameGetCount++;
         notify(MPP_OUTPUT_DEQUEUE);
+
+        buffer = mpp_frame_get_buffer(frm);
+        if (buffer)
+            mpp_buffer_sync_ro_begin(buffer);
     } else {
         // NOTE: Add signal here is not efficient
         // This is for fix bug of stucking on decoder parser thread
@@ -530,6 +536,7 @@ MPP_RET Mpp::get_frame_noblock(MppFrame *frame)
     mFrmOut->lock();
     if (mFrmOut->list_size()) {
         mFrmOut->del_at_head(&first, sizeof(frame));
+        mpp_buffer_sync_ro_begin(mpp_frame_get_buffer(first));
         mFrameGetCount++;
     }
     mFrmOut->unlock();
@@ -559,6 +566,7 @@ MPP_RET Mpp::decode(MppPacket packet, MppFrame *frame)
 
         if (mFrmOut->list_size()) {
             mFrmOut->del_at_head(frame, sizeof(*frame));
+            mpp_buffer_sync_ro_begin(mpp_frame_get_buffer(*frame));
             mFrameGetCount++;
             return MPP_OK;
         }
@@ -578,6 +586,7 @@ MPP_RET Mpp::decode(MppPacket packet, MppFrame *frame)
 
             if (mFrmOut->list_size()) {
                 mFrmOut->del_at_head(frame, sizeof(*frame));
+                mpp_buffer_sync_ro_begin(mpp_frame_get_buffer(*frame));
                 mFrameGetCount++;
                 frm_rdy = 1;
             }
diff --git a/test/mpi_dec_mt_test.c b/test/mpi_dec_mt_test.c
index eace7048..97f02624 100644
--- a/test/mpi_dec_mt_test.c
+++ b/test/mpi_dec_mt_test.c
@@ -154,7 +154,7 @@ void *thread_output(void *arg)
 
             if (NULL == data->frm_grp) {
                 /* If buffer group is not set create one and limit it */
-                ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION);
+                ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
                 if (ret) {
                     mpp_err("get mpp buffer group failed ret %d\n", ret);
                     break;
diff --git a/test/mpi_dec_multi_test.c b/test/mpi_dec_multi_test.c
index cf3fce3a..84e83652 100644
--- a/test/mpi_dec_multi_test.c
+++ b/test/mpi_dec_multi_test.c
@@ -158,7 +158,7 @@ static int multi_dec_simple(MpiDecMultiCtx *data)
 
                     if (NULL == data->frm_grp) {
                         /* If buffer group is not set create one and limit it */
-                        ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION);
+                        ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
                         if (ret) {
                             mpp_err("get mpp buffer group failed ret %d\n", ret);
                             break;
@@ -438,7 +438,7 @@ void* multi_dec_decode(void *cmd_ctx)
         RK_U32 hor_stride = MPP_ALIGN(width, 16);
         RK_U32 ver_stride = MPP_ALIGN(height, 16);
 
-        ret = mpp_buffer_group_get_internal(&dec_ctx->frm_grp, MPP_BUFFER_TYPE_ION);
+        ret = mpp_buffer_group_get_internal(&dec_ctx->frm_grp, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
         if (ret) {
             mpp_err("failed to get buffer group for input frame ret %d\n", ret);
             goto MPP_TEST_OUT;
diff --git a/test/mpi_dec_test.c b/test/mpi_dec_test.c
index bace4ceb..2d035248 100644
--- a/test/mpi_dec_test.c
+++ b/test/mpi_dec_test.c
@@ -206,7 +206,7 @@ static int dec_simple(MpiDecLoopData *data)
 
                     if (NULL == data->frm_grp) {
                         /* If buffer group is not set create one and limit it */
-                        ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION);
+                        ret = mpp_buffer_group_get_internal(&data->frm_grp, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
                         if (ret) {
                             mpp_err("%p get mpp buffer group failed ret %d\n", ctx, ret);
                             break;
@@ -580,7 +580,7 @@ int dec_decode(MpiDecTestCmd *cmd)
         RK_U32 hor_stride = MPP_ALIGN(width, 16);
         RK_U32 ver_stride = MPP_ALIGN(height, 16);
 
-        ret = mpp_buffer_group_get_internal(&data.frm_grp, MPP_BUFFER_TYPE_ION);
+        ret = mpp_buffer_group_get_internal(&data.frm_grp, MPP_BUFFER_TYPE_ION | MPP_BUFFER_FLAGS_CACHABLE);
         if (ret) {
             mpp_err("failed to get buffer group for input frame ret %d\n", ret);
             goto MPP_TEST_OUT;