diff --git a/src/Function1D.cpp b/src/Function1D.cpp
index b9b1a55..4d4d781 100644
--- a/src/Function1D.cpp
+++ b/src/Function1D.cpp
@@ -753,26 +753,26 @@ void Aurora::padding(Matrix &aMatrix, int aIndex, double aValue)
     aMatrix = Matrix::New(newData,size,1,1,aMatrix.getValueType());
 }
 
-Matrix Aurora::convertfp16tofloat(const Matrix& aMatrix)
+Matrix Aurora::convertfp16tofloat(short* aData, int aRows, int aColumns)
 {
 
-    auto input = aMatrix.getData();
+    auto input = aData;
+    size_t size = aRows*aColumns;
+    size_t quaterSize = size/4;
     //uint16变换为float(32位)输出大小翻倍
-    auto output = malloc(aMatrix.getDataSize()*4);
-    size_t rows = aMatrix.getDataSize()*sizeof(double)/sizeof(short);
-    size_t total_count = aMatrix.getDataSize();
+    auto output = malloc(size);
     
-    // #pragma omp parallel for
-    for (size_t i = 0; i < total_count; i+=8) {
+    #pragma omp parallel for
+    for (size_t i = 0; i < quaterSize; i+=8) {
         //循环展开以避免过度的线程调用
-        if (i  < total_count)::convertValue2((short*)(input+i), output + (i) * 4);
-        if (i+1  < total_count)::convertValue2((short*)(input+i+1), output + (i+1) * 4);
-        if (i+2  < total_count)::convertValue2((short*)(input+i+2), output + (i+2) * 4);
-        if (i+3  < total_count)::convertValue2((short*)(input+i+3), output + (i+3) * 4);
-        if (i+4  < total_count)::convertValue2((short*)(input+i+4), output + (i+4) * 4);
-        if (i+5  < total_count)::convertValue2((short*)(input+i+5), output + (i+5) * 4);
-        if (i+6  < total_count)::convertValue2((short*)(input+i+6), output + (i+6) * 4);
-        if (i+7  < total_count)::convertValue2((short*)(input+i+7), output + (i+7) * 4);
+        if (i  < quaterSize)::convertValue2((short*)(input+i*4), output + (i) * 4);
+        if (i+1  < quaterSize)::convertValue2((short*)(input+(i+1)*4), output + (i+1) * 4);
+        if (i+2  < quaterSize)::convertValue2((short*)(input+(i+2)*4), output + (i+2) * 4);
+        if (i+3  < quaterSize)::convertValue2((short*)(input+(i+3)*4), output + (i+3) * 4);
+        if (i+4  < quaterSize)::convertValue2((short*)(input+(i+4)*4), output + (i+4) * 4);
+        if (i+5  < quaterSize)::convertValue2((short*)(input+(i+5)*4), output + (i+5) * 4);
+        if (i+6  < quaterSize)::convertValue2((short*)(input+(i+6)*4), output + (i+6) * 4);
+        if (i+7  < quaterSize)::convertValue2((short*)(input+(i+7)*4), output + (i+7) * 4);
     }
-    return Matrix::New(output,aMatrix.getDimSize(0),aMatrix.getDimSize(1),aMatrix.getDimSize(2));
+    return Matrix::New(output,aRows,aColumns,1);
 }
diff --git a/src/Function1D.h b/src/Function1D.h
index c26dafe..cddd088 100644
--- a/src/Function1D.h
+++ b/src/Function1D.h
@@ -113,7 +113,7 @@ namespace Aurora {
      */
     void padding(Matrix& aMatrix, int aIndex, double aValue);
 
-    Matrix convertfp16tofloat(const Matrix& aMatrix);
+    Matrix convertfp16tofloat(short* aData, int aRows, int aColumns);
 };
 
 
diff --git a/test/Function1D_Test.cpp b/test/Function1D_Test.cpp
index 7949da1..a8b51f2 100644
--- a/test/Function1D_Test.cpp
+++ b/test/Function1D_Test.cpp
@@ -513,12 +513,14 @@ TEST_F(Function1D_Test, convertfp16tofloat) {
 
     size_t count = 0;
     auto input = m.readint16("input",count);
-    auto ma = Aurora::Matrix::copyFromRawData((double*)input.get(),count/4);
-    auto resultM = Aurora::convertfp16tofloat(ma);
+    for(int i = 0; i<1000; i++){
+        auto resultM = Aurora::convertfp16tofloat(input.get(),count,1);
+    }
+    auto resultM = Aurora::convertfp16tofloat(input.get(),count,1);
     auto result = resultM.getData();
     auto output = m.read("output");
     for (size_t i = 0; i<count; i++) {
-        EXPECT_EQ(result[i], output.getData()[i])<<"index:"<<i<<",input:"<< ((short*)ma.getData())[i]<<",input2:"<<input.get()[i];
+        EXPECT_EQ(result[i], output.getData()[i])<<"index:"<<i<<",input:"<<input.get()[i];
     }
 
 }
diff --git a/test/Matrix_Test.cpp b/test/Matrix_Test.cpp
index 55deaf0..6c16cd4 100644
--- a/test/Matrix_Test.cpp
+++ b/test/Matrix_Test.cpp
@@ -115,7 +115,7 @@ TEST_F(Matrix_Test, MatrixCreate) {
     {
         double *tempData = new double[9];
         Aurora::Matrix C = Aurora::Matrix::fromRawData(tempData, 3, 1, 3);
-        EXPECT_EQ(dataD, C.getData());
+        EXPECT_EQ(tempData, C.getData());
         EXPECT_EQ(9, C.getDataSize());
         EXPECT_EQ(3, C.getDims());
         EXPECT_EQ(3, C.getDimSize(0));