diff --git a/paddle/math/tests/TensorCheck.h b/paddle/math/tests/TensorCheck.h
index 6ca303cc7c55cede5e6f51e27598a87241e4274b..f19933d1f6bc36868362c59312d5c46b41489995 100644
--- a/paddle/math/tests/TensorCheck.h
+++ b/paddle/math/tests/TensorCheck.h
@@ -21,60 +21,29 @@ using namespace std;     // NOLINT
 
 namespace autotest {
 
-class CheckEqual {
+class AssertEqual {
 public:
-  inline int operator()(real a, real b) {
-    if (a != b) {
-      return 1;
-    }
-
-    return 0;
-  }
-};
+  AssertEqual(real err = 0) : err_(err) {}
 
-class CheckWithErr {
-public:
-  CheckWithErr() {
-#ifndef PADDLE_TYPE_DOUBLE
-    err_ = 1e-5;
-#else
-    err_ = 1e-10;
-#endif
-  }
-
-  inline int operator()(real a, real b) {
-    if (std::fabs(a - b) > err_) {
-      if ((std::fabs(a - b) / std::fabs(a)) > (err_ / 10.0f)) {
-        return 1;
+  inline bool operator()(real a, real b) {
+    if (err_ == 0) {
+      if (a != b) {
+        return false;
+      }
+    } else {
+      if (std::fabs(a - b) > err_) {
+        if ((std::fabs(a - b) / std::fabs(a)) > (err_ / 10.0f)) {
+          return false;
+        }
       }
     }
-    return 0;
+    return true;
   }
 
 private:
   real err_;
 };
 
-template<typename Check>
-void TensorCheck(Check op, const CpuMatrix& matrix1, const CpuMatrix& matrix2) {
-  CHECK(matrix1.getHeight() == matrix2.getHeight());
-  CHECK(matrix1.getWidth() == matrix2.getWidth());
-
-  int height = matrix1.getHeight();
-  int width = matrix1.getWidth();
-  const real* data1 = matrix1.getData();
-  const real* data2 = matrix2.getData();
-  int count = 0;
-  for (int i = 0; i < height; i++) {
-    for (int j = 0; j < width; j++) {
-      real a = data1[i * width + j];
-      real b = data2[i * width + j];
-      count += op(a, b);
-    }
-  }
-  EXPECT_EQ(count, 0) << "There are " << count << " different element.";
-}
-
 template <typename Tensor>
 class CopyToCpu;
 
@@ -101,10 +70,36 @@ private:
   CpuMatrix arg_;
 };
 
-template<typename Tensor1, typename Tensor2>
-extern void TensorCheckErr(const Tensor1& tensor1, const Tensor2& tensor2) {
+template<typename AssertEq>
+void TensorCheck(AssertEq compare,
+                 const CpuMatrix& matrix1,
+                 const CpuMatrix& matrix2) {
+  CHECK(matrix1.getHeight() == matrix2.getHeight());
+  CHECK(matrix1.getWidth() == matrix2.getWidth());
+
+  int height = matrix1.getHeight();
+  int width = matrix1.getWidth();
+  const real* data1 = matrix1.getData();
+  const real* data2 = matrix2.getData();
+  int count = 0;
+  for (int i = 0; i < height; i++) {
+    for (int j = 0; j < width; j++) {
+      real a = data1[i * width + j];
+      real b = data2[i * width + j];
+      if (!compare(a, b)) {
+        count++;
+      }
+    }
+  }
+  EXPECT_EQ(count, 0) << "There are " << count << " different element.";
+}
+
+template<typename AssertEq, typename Tensor1, typename Tensor2>
+extern void TensorCheck(AssertEq compare,
+                        const Tensor1& tensor1,
+                        const Tensor2& tensor2) {
   TensorCheck(
-    CheckWithErr(),
+    compare,
     CopyToCpu<Tensor1>(tensor1).copiedArg(),
     CopyToCpu<Tensor2>(tensor2).copiedArg());
 }
diff --git a/paddle/math/tests/TestUtils.h b/paddle/math/tests/TestUtils.h
index 5c9049cacc2aa14a94dab93437f38ec29832b5a0..ffd113454277b02c35aba5d654d885bd9d70a0fe 100644
--- a/paddle/math/tests/TestUtils.h
+++ b/paddle/math/tests/TestUtils.h
@@ -107,12 +107,16 @@ R call(C& obj, R (FC::*f)(FArgs...), Args&&... args) {
   return (obj.*f)(args...);
 }
 
-template <std::size_t... I, typename C, typename R, typename ...Args>
-void BaseMatrixCompare(R (C::*f)(Args...)) {
+template <bool ApplyRow, bool ApplyCol,
+          std::size_t... I, typename C, typename R, typename ...Args,
+          typename AssertEq>
+void BaseMatrixCompare(R (C::*f)(Args...), AssertEq compare) {
   for (auto height : {1, 11, 73, 128, 200, 330}) {
-    for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
-      CpuMatrix obj1(height, width);
-      GpuMatrix obj2(height, width);
+    for (auto width : {1, 3, 32, 100, 512, 1000}) {
+      CpuMatrix obj1(ApplyCol ? 1 : height,
+                     ApplyRow ? 1 : width);
+      GpuMatrix obj2(ApplyCol ? 1 : height,
+                     ApplyRow ? 1 : width);
       init(obj1);
       copy(obj2, obj1);
 
@@ -132,7 +136,7 @@ void BaseMatrixCompare(R (C::*f)(Args...)) {
       call(obj1, f, std::get<I>(tuple1)...);
       call(obj2, f, std::get<I>(tuple2)...);
 
-      TensorCheckErr(obj1, obj2);
+      TensorCheck(compare, obj1, obj2);
     }
   }
 }
@@ -144,6 +148,39 @@ void BaseMatrixCompare(R (C::*f)(Args...)) {
   static_assert(sizeof...(I) == sizeof...(Args),
     "size of parameter packs are not equal");
 
-  autotest::BaseMatrixCompare<I...>(f);
+#ifndef PADDLE_TYPE_DOUBLE
+  autotest::AssertEqual compare(1e-5);
+#else
+  autotest::AssertEqual compare(1e-10);
+#endif
+
+  autotest::BaseMatrixCompare<false, false, I...>(f, compare);
+}
+
+template <std::size_t... I, typename C, typename R, typename ...Args>
+void BaseMatrixApplyRow(R (C::*f)(Args...)) {
+  static_assert(sizeof...(I) == sizeof...(Args),
+    "size of parameter packs are not equal");
+
+#ifndef PADDLE_TYPE_DOUBLE
+  autotest::AssertEqual compare(1e-3);
+#else
+  autotest::AssertEqual compare(1e-8);
+#endif
+
+  autotest::BaseMatrixCompare<true, false, I...>(f, compare);
+}
+
+template <std::size_t... I, typename C, typename R, typename ...Args>
+void BaseMatrixApplyCol(R (C::*f)(Args...)) {
+  static_assert(sizeof...(I) == sizeof...(Args),
+    "size of parameter packs are not equal");
+
+#ifndef PADDLE_TYPE_DOUBLE
+  autotest::AssertEqual compare(1e-3);
+#else
+  autotest::AssertEqual compare(1e-8);
+#endif
+  autotest::BaseMatrixCompare<false, true, I...>(f, compare);
 }
 
diff --git a/paddle/math/tests/test_BaseMatrix.cpp b/paddle/math/tests/test_BaseMatrix.cpp
index c06580ca82e1d3cb78eee692b4d10b3792a2062b..99ab640f48f3b13b758be2eef60fe27f37e66413 100644
--- a/paddle/math/tests/test_BaseMatrix.cpp
+++ b/paddle/math/tests/test_BaseMatrix.cpp
@@ -26,25 +26,224 @@ limitations under the License. */
 using namespace paddle;  // NOLINT
 using namespace std;     // NOLINT
 
-TEST(BaseMatrix, apply) {
-  // member function with no argument
-  BaseMatrixCompare(&BaseMatrix::neg);
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)().
+ */
+TEST(BaseMatrix, void) {
+  typedef void (BaseMatrix::*FunctionProto)();
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXCOMPARE(neg);
+  BASEMATRIXCOMPARE(exp);
+  BASEMATRIXCOMPARE(log);
+  BASEMATRIXCOMPARE(sqrt);
+  BASEMATRIXCOMPARE(square);
+  BASEMATRIXCOMPARE(reciprocal);
+  BASEMATRIXCOMPARE(abs);
+  BASEMATRIXCOMPARE(sign);
+  BASEMATRIXCOMPARE(zero);
+  BASEMATRIXCOMPARE(one);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)(real).
+ */
+TEST(BaseMatrix, real) {
+  typedef void (BaseMatrix::*FunctionProto)(real);
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare<0>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXCOMPARE(pow);
+  BASEMATRIXCOMPARE(subScalar);
+  BASEMATRIXCOMPARE(mulScalar);
+  BASEMATRIXCOMPARE(divScalar);
+  BASEMATRIXCOMPARE(assign);
+  BASEMATRIXCOMPARE(add);
+  BASEMATRIXCOMPARE(biggerThanScalar);
+  BASEMATRIXCOMPARE(downClip);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)(real, real).
+ */
+TEST(BaseMatrix, real_real) {
+  typedef void (BaseMatrix::*FunctionProto)(real, real);
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXCOMPARE(add);
+  BASEMATRIXCOMPARE(clip);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&).
+ */
+TEST(BaseMatrix, BaseMatrix) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&);
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare<0>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXCOMPARE(assign);
+  BASEMATRIXCOMPARE(add);
+  BASEMATRIXCOMPARE(relu);
+  BASEMATRIXCOMPARE(reluDerivative);
+  BASEMATRIXCOMPARE(softrelu);
+  BASEMATRIXCOMPARE(softreluDerivative);
+  BASEMATRIXCOMPARE(brelu);
+  BASEMATRIXCOMPARE(breluDerivative);
+  BASEMATRIXCOMPARE(square);
+  BASEMATRIXCOMPARE(squareDerivative);
+  BASEMATRIXCOMPARE(tanh);
+  BASEMATRIXCOMPARE(tanhDerivative);
+
+  BASEMATRIXCOMPARE(reciprocal);
+  BASEMATRIXCOMPARE(reciprocalDerivative);
+  BASEMATRIXCOMPARE(abs);
+  BASEMATRIXCOMPARE(absDerivative);
+  BASEMATRIXCOMPARE(sigmoid);
+  BASEMATRIXCOMPARE(sigmoidDerivative);
+  BASEMATRIXCOMPARE(expDerivative);
+  BASEMATRIXCOMPARE(sign);
+  BASEMATRIXCOMPARE(exp);
+  BASEMATRIXCOMPARE(log);
+  BASEMATRIXCOMPARE(sqrt);
+  BASEMATRIXCOMPARE(dotMul);
+  BASEMATRIXCOMPARE(dotMulSquare);
+  BASEMATRIXCOMPARE(dotSquareMul);
+
+  BASEMATRIXCOMPARE(addColVector);
+  BASEMATRIXCOMPARE(addRowVector);
+  BASEMATRIXCOMPARE(mulRowVector);
+  BASEMATRIXCOMPARE(divRowVector);
+  BASEMATRIXCOMPARE(addP2P);
+  BASEMATRIXCOMPARE(invSqrt);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&, real).
+ */
+TEST(BaseMatrix, BaseMatrix_real) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, real);
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
 
-  // If the member function are overloaded, use static_cast to specify which
-  // member function need be test.
-  BaseMatrixCompare(
-    static_cast<void (BaseMatrix::*)()>(&BaseMatrix::exp));
-  BaseMatrixCompare(
-    static_cast<void (BaseMatrix::*)()>(&BaseMatrix::sqrt));
+  BASEMATRIXCOMPARE(addBias);
+  BASEMATRIXCOMPARE(add);
+  BASEMATRIXCOMPARE(sub);
+  BASEMATRIXCOMPARE(pow);
+  BASEMATRIXCOMPARE(addScalar);
+  BASEMATRIXCOMPARE(subScalar);
+  BASEMATRIXCOMPARE(mulScalar);
+  BASEMATRIXCOMPARE(divScalar);
+  BASEMATRIXCOMPARE(scalarDiv);
+  BASEMATRIXCOMPARE(addSquare);
+
+  BASEMATRIXCOMPARE(isEqualTo);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&, BaseMatrix&).
+ */
+TEST(BaseMatrix, BaseMatrix_BaseMatrix) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, BaseMatrix&);
+  #define BASEMATRIXCOMPARE(function) \
+    BaseMatrixCompare<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXCOMPARE(softCrossEntropy);
+  BASEMATRIXCOMPARE(softCrossEntropyBp);
+  BASEMATRIXCOMPARE(binaryLabelCrossEntropy);
+  BASEMATRIXCOMPARE(binaryLabelCrossEntropyBp);
+  BASEMATRIXCOMPARE(sub);
+  BASEMATRIXCOMPARE(add2);
+  BASEMATRIXCOMPARE(dotMul);
+  BASEMATRIXCOMPARE(dotDiv);
+  BASEMATRIXCOMPARE(logisticRegressionLoss);
+  BASEMATRIXCOMPARE(logisticRegressionLossBp);
+  BASEMATRIXCOMPARE(biggerThan);
+  BASEMATRIXCOMPARE(max);
+  BASEMATRIXCOMPARE(dotMulSquare);
+  BASEMATRIXCOMPARE(dotSquareSquare);
+
+  #undef BASEMATRIXCOMPARE
+}
+
+/**
+ * Test aggregate member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&).
+ */
+TEST(Aggregate, BaseMatrix) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&);
+  #define BASEMATRIXAPPLYROW(function) \
+    BaseMatrixApplyRow<0>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+ #define BASEMATRIXAPPLYCOL(function) \
+    BaseMatrixApplyCol<0>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXAPPLYROW(maxRows);
+  BASEMATRIXAPPLYROW(minRows);
+
+  BASEMATRIXAPPLYCOL(sumCols);
+  BASEMATRIXAPPLYCOL(maxCols);
+  BASEMATRIXAPPLYCOL(minCols);
+
+  #undef BASEMATRIXAPPLYROW
+  #undef BASEMATRIXAPPLYCOL
+}
+
+/**
+ * Test aggregate member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&, BaseMatrix&).
+ */
+TEST(Aggregate, BaseMatrix_BaseMatrix) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, BaseMatrix&);
+  #define BASEMATRIXAPPLYROW(function) \
+    BaseMatrixApplyRow<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  #define BASEMATRIXAPPLYCOL(function) \
+    BaseMatrixApplyCol<0, 1>(static_cast<FunctionProto>(&BaseMatrix::function));
+
+  BASEMATRIXAPPLYCOL(addDotMulVMM);
+
+  #undef BASEMATRIXAPPLYROW
+  #undef BASEMATRIXAPPLYCOL
+}
+
+/**
+ * Test aggregate member functions which prototype is
+ * void (BaseMatrix::*)(BaseMatrix&, real, real).
+ */
+TEST(Aggregate, BaseMatrix_real_real) {
+  typedef void (BaseMatrix::*FunctionProto)(BaseMatrix&, real, real);
+  #define BASEMATRIXAPPLYROW(function) \
+    BaseMatrixApplyRow<0, 1, 2>(\
+      static_cast<FunctionProto>(&BaseMatrix::function));
 
-  // member function with one argument
+  #define BASEMATRIXAPPLYCOL(function) \
+    BaseMatrixApplyCol<0, 1, 2>(\
+      static_cast<FunctionProto>(&BaseMatrix::function));
 
-  BaseMatrixCompare<0>(&BaseMatrix::tanh);
+  BASEMATRIXAPPLYROW(sumRows);
+  BASEMATRIXAPPLYCOL(sumCols);
 
-  BaseMatrixCompare<0>(
-    static_cast<void (BaseMatrix::*)(real)>(&BaseMatrix::assign));
-  BaseMatrixCompare<0>(
-    static_cast<void (BaseMatrix::*)(real)>(&BaseMatrix::pow));
+  #undef BASEMATRIXAPPLYROW
+  #undef BASEMATRIXAPPLYCOL
 }
 
 int main(int argc, char** argv) {
diff --git a/paddle/math/tests/test_matrixCompare.cpp b/paddle/math/tests/test_matrixCompare.cpp
index ae5bc5a86a1790ce30a8d7f83c9564f52d7cf7ea..8f96672b88e7c40d615bc316b61604538714ddc9 100644
--- a/paddle/math/tests/test_matrixCompare.cpp
+++ b/paddle/math/tests/test_matrixCompare.cpp
@@ -448,125 +448,6 @@ void testMatrixZeroAtOffset(int height, int width) {
   MatrixCheckEqual(*cpuA, *cpuTest);
 }
 
-void testMatrixBinaryAdd(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  cpuA->add(*cpuB);
-  gpuA->add(*gpuB);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-}
-
-void testMatrixAssign(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  cpuA->assign(2.5);
-  gpuA->assign(2.5);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-}
-
-void testMatrixAdd(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  cpuA->add(2.5);
-  gpuA->add(2.5);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-}
-
-void testMatrixSqrt(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  cpuA->sqrt();
-  gpuA->sqrt();
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixTanhDerivative(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  cpuA->tanhDerivative(*cpuB);
-  gpuA->tanhDerivative(*gpuB);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixTanh(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  cpuA->tanh(*cpuB);
-  gpuA->tanh(*gpuB);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixTernarySub(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-
-  cpuA->sub(*cpuB, *cpuC);
-  gpuA->sub(*gpuB, *gpuC);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckEqual(*cpuA, *outputCheck);
-}
-
 void testMatrixSumOfSquaresBp(int height, int width) {
   MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
   MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
@@ -789,18 +670,7 @@ TEST(Matrix, unary) {
     for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
       VLOG(3) << " height=" << height << " width=" << width;
 
-      // applyUnary
-      testMatrixAssign(height, width);
-      testMatrixAdd(height, width);
-      testMatrixSqrt(height, width);
-
-      // applyBinary
-      testMatrixBinaryAdd(height, width);
-      testMatrixTanh(height, width);
-      testMatrixTanhDerivative(height, width);
-
       // applyTernary
-      testMatrixTernarySub(height, width);
       testMatrixSumOfSquaresBp(height, width);
 
       // asRowVector
@@ -931,165 +801,6 @@ TEST(Matrix, softmax) {
   }
 }
 
-void testMatrixAddDotMulVMM(int height, int width, int endCol = 0) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-  MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
-
-  MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(1, width);
-  MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  cpuC->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  gpuC->copyFrom(*cpuC);
-  cpuA1->copyFrom(*cpuA);
-  cpuB1->copyFrom(*cpuB);
-  cpuC1->copyFrom(*cpuC);
-
-  if (!endCol) {
-    cpuA->addDotMulVMM(*cpuB, *cpuC);
-    gpuA->addDotMulVMM(*gpuB, *gpuC);
-    cpuA1->addDotMulVMM2(*cpuB1, *cpuC1);
-
-    MatrixCheckErr(*cpuA, *cpuA1);
-  } else {
-    MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
-    MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
-    MatrixPtr subCpuC = cpuC->subColMatrix(0, endCol);
-    MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
-    MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
-    MatrixPtr subGpuC = gpuC->subColMatrix(0, endCol);
-    subCpuA->addDotMulVMM(*subCpuB, *subCpuC);
-    subGpuA->addDotMulVMM(*subGpuB, *subGpuC);
-  }
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixRowSum(int height, int width) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA1 = std::make_shared<GpuMatrix>(height, 1);
-  MatrixPtr gpuB1 = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-  cpuA1->copyFrom(*cpuA);
-  cpuB1->copyFrom(*cpuB);
-  gpuA1->copyFrom(*cpuA);
-  gpuB1->copyFrom(*cpuB);
-
-  cpuA->colMerge(*cpuB);
-  gpuA->colMerge(*gpuB);
-
-  cpuB1->rowSum(*cpuA1);
-  gpuB1->rowSum(*gpuA1);
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-  outputCheck->copyFrom(*gpuA1);
-  MatrixCheckErr(*cpuA1, *outputCheck);
-}
-
-void testMatrixRowMax(int height, int width, int endCol = 0) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-
-  if (!endCol) {
-    cpuB->rowMax(*cpuA);
-    gpuB->rowMax(*gpuA);
-  } else {
-    MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
-    MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
-    subCpuB->rowMax(*cpuA);
-    subGpuB->rowMax(*gpuA);
-  }
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixColSum(int height, int width, int endCol = 0) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-
-  if (!endCol) {
-    cpuA->accumulateColSum(*cpuB);
-    gpuA->accumulateColSum(*gpuB);
-  } else {
-    MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
-    MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
-    MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
-    MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
-    subCpuA->accumulateColSum(*subCpuB);
-    subGpuA->accumulateColSum(*subGpuB);
-  }
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
-void testMatrixColMax(int height, int width, int endCol = 0) {
-  MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
-  MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
-  MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
-  MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
-
-  cpuA->randomizeUniform();
-  cpuB->randomizeUniform();
-  gpuA->copyFrom(*cpuA);
-  gpuB->copyFrom(*cpuB);
-
-  if (!endCol) {
-    cpuB->colMax(*cpuA);
-    gpuB->colMax(*gpuA);
-  } else {
-    MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
-    MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
-    MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
-    MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
-    subCpuB->colMax(*subCpuA);
-    subGpuB->colMax(*subGpuA);
-  }
-
-  MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
-  outputCheck->copyFrom(*gpuA);
-  MatrixCheckErr(*cpuA, *outputCheck);
-}
-
 void testMatrixCollectBias(int height, int width) {
   MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
   MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
@@ -1182,13 +893,8 @@ TEST(Matrix, aggregate) {
   for (auto height : {1, 11, 16, 32, 64, 73, 128, 200, 1024, 2345}) {
     for (auto width : {1, 9, 16, 32, 64, 100, 512, 1000, 1024, 2453}) {
       VLOG(3) << " height=" << height << " width=" << width;
-      testMatrixRowSum(height, width);
-      testMatrixRowMax(height, width);
-      testMatrixColSum(height, width);
-      testMatrixColMax(height, width);
       testMatrixCollectBias(height, width);
       testMatrixTernaryRowDotMul(height, width);
-      testMatrixAddDotMulVMM(height, width);
 
       testMatrixSumOfSquares(height, width);
       testMatrixBinaryClassificationError(height, width);
@@ -1203,11 +909,7 @@ TEST(Matrix, aggregate2) {
       VLOG(3) << " height=" << height << " width=" << width;
 
       int endCol = rand() % width;  // NOLINT
-      testMatrixRowMax(height, width, endCol);
       testMatrixSumOfSquares(height, width, endCol);
-      testMatrixColSum(height, width, endCol);
-      testMatrixColMax(height, width, endCol);
-      testMatrixAddDotMulVMM(height, width, endCol);
     }
   }
 }