diff --git a/amexec/build.sh b/amexec/build.sh
index d6052fd..3399fa8 100755
--- a/amexec/build.sh
+++ b/amexec/build.sh
@@ -1,7 +1,7 @@
 rm -rf build
 mkdir -p build/intermediates/host
 cd build/intermediates/host
-cmake ../../../src -DCMAKE_CXX_COMPILER=g++ -DCMAKE_SKIP_RPATH=TRUE
+cmake ../../../src -DCMAKE_CXX_COMPILER=aarch64-linux-gnu-g++ -DCMAKE_SKIP_RPATH=TRUE
 make
 cd ../../../out
 mv main amexec
diff --git a/amexec/inc/utils.h b/amexec/inc/utils.h
index f5d8d2c..6fc942c 100755
--- a/amexec/inc/utils.h
+++ b/amexec/inc/utils.h
@@ -70,6 +70,8 @@ public:
     static double printDiffTime(time_t begin, time_t end);
 	
     static double InferenceTimeAverage(double *x, int len);
+	
+	static double InferenceTimeAverageWithoutFirst(double *x, int len);
 };
  
 #pragma once
diff --git a/amexec/out/amexec b/amexec/out/amexec
index 2f4ccb0..6673ffb 100755
Binary files a/amexec/out/amexec and b/amexec/out/amexec differ
diff --git a/amexec/src/model_process.cpp b/amexec/src/model_process.cpp
index 4f01763..8711298 100755
--- a/amexec/src/model_process.cpp
+++ b/amexec/src/model_process.cpp
@@ -260,11 +260,21 @@ void ModelProcess::OutputModelResult(std::string& s,std::string& modelName,size_
         aclDataType datatype = aclmdlGetOutputDataType(modelDesc_, i);
         void *dims = nullptr;
         aclmdlIODims *dim = nullptr;
-        aclError ret = aclrtMallocHost(&dims, sizeof(aclmdlIODims));
-        if (ret != ACL_ERROR_NONE) {
-            ERROR_LOG("aclrtMallocHost failed, ret[%d]", ret);
-            return;
-        }
+        aclError ret = ACL_ERROR_NONE;
+        if (!g_isDevice) { 
+            ret = aclrtMallocHost(&dims, sizeof(aclmdlIODims));
+            if (ret != ACL_ERROR_NONE) {
+                ERROR_LOG("aclrtMallocHost failed, ret[%d]", ret);
+                return;
+            }
+		}
+		else {
+			ret = aclrtMalloc(&dims, sizeof(aclmdlIODims), ACL_MEM_MALLOC_NORMAL_ONLY);
+			if (ret != ACL_ERROR_NONE) {
+                ERROR_LOG("malloc device buffer failed, ret[%d]", ret);
+                return;
+            }
+		}
         dim = reinterpret_cast<aclmdlIODims*>(dims);
         ret = aclmdlGetOutputDims(modelDesc_, i, dim);
         if (ret != ACL_ERROR_NONE) {
diff --git a/amexec/src/sample_process.cpp b/amexec/src/sample_process.cpp
index cbdc214..4186c29 100755
--- a/amexec/src/sample_process.cpp
+++ b/amexec/src/sample_process.cpp
@@ -64,7 +64,7 @@ Result SampleProcess::InitResource()
  
     // get run mode
     aclrtRunMode runMode;
-   ret = aclrtGetRunMode(&runMode);
+    ret = aclrtGetRunMode(&runMode);
     if (ret != ACL_ERROR_NONE) {
         ERROR_LOG("acl get run mode failed");
         return FAILED;
@@ -151,7 +151,10 @@ Result SampleProcess::Process(vector<string>& params, vector<string>& input_file
     }
     double infer_time_ave = Utils::InferenceTimeAverage(inference_time, loop);
     printf("Inference average time: %f ms\n", infer_time_ave);
-
+	if (loop > 1){
+		double infer_time_ave_without_first = Utils::InferenceTimeAverageWithoutFirst(inference_time, loop);
+		printf("Inference average time without first time: %f ms\n", infer_time_ave_without_first);
+	} 
     processModel.DestroyInput();
 
     // release model input buffer
diff --git a/amexec/src/utils.cpp b/amexec/src/utils.cpp
index 2941a9c..528c64d 100755
--- a/amexec/src/utils.cpp
+++ b/amexec/src/utils.cpp
@@ -51,6 +51,7 @@ void* Utils::ReadBinFile(std::string fileName, uint32_t &fileSize)
             return nullptr;
         }
     }
+		
     binFile.read(static_cast<char *>(binFileBufferData), binFileBufferLen);
     binFile.close();
     fileSize = binFileBufferLen;
@@ -180,7 +181,7 @@ double Utils::printDiffTime(time_t begin, time_t end)
 {
     double diffT = difftime(begin, end);
     printf("The inference time is: %f millisecond\n", 1000*diffT);
-    return diffT*1000;
+    return diffT * 1000;
 }
 
 double Utils::InferenceTimeAverage(double *x, int len)
@@ -188,5 +189,16 @@ double Utils::InferenceTimeAverage(double *x, int len)
     double sum = 0;
     for (int i = 0; i < len; i++)
         sum += x[i];
-    return sum/len;
+    return sum / len;
+}
+
+double Utils::InferenceTimeAverageWithoutFirst(double *x, int len)
+{
+    double sum = 0;
+    for (int i = 0; i < len; i++)
+		if (i !=0){
+		    sum += x[i];
+		}
+        
+    return sum / (len - 1);
 }