add speed test and unit test for DiffLogSoftmax

src/cudamatrix/cu-matrix-speed-test.cc

@@ -458,7 +458,6 @@ template<typename Real> void TestCuMatrixMulRowsGroupMat(int32 dim) {
             << dim << ", speed was " << gflops << " gigaflops.";
 }
 
-
 template<typename Real> void TestCuMatrixDiffSoftmax(int32 dim) {
   BaseFloat time_in_secs = 0.025;
   CuMatrix<Real> M(dim, dim), N(dim, dim), L(dim, dim);
@@ -477,6 +476,24 @@ template<typename Real> void TestCuMatrixDiffSoftmax(int32 dim) {
             << dim << ", speed was " << gflops << " gigaflops.";
 }
 
+template<typename Real> void TestCuMatrixDiffLogSoftmax(int32 dim) {
+  BaseFloat time_in_secs = 0.025;
+  CuMatrix<Real> M(dim, dim), N(dim, dim), L(dim, dim);
+  M.SetRandn();
+  N.SetRandn();
+  L.SetRandn();
+  Timer tim;
+  int32 iter = 0;
+  for (; tim.Elapsed() < time_in_secs; iter++) {
+    N.DiffLogSoftmaxPerRow(M, L);
+  }
+
+  BaseFloat fdim = dim;
+  BaseFloat gflops = (fdim * fdim * iter) / (tim.Elapsed() * 1.0e+09);
+  KALDI_LOG << "For CuMatrix::DiffLogSoftmaxPerRow" << NameOf<Real>() << ", for dim = "
+            << dim << ", speed was " << gflops << " gigaflops.";
+}
+
 template<typename Real> void TestCuMatrixSoftmax(int32 dim) {
   BaseFloat time_in_secs = 0.025;
   CuMatrix<Real> M(dim, dim), N(dim, dim);
@@ -1021,6 +1038,8 @@ template<typename Real> void CudaMatrixSpeedTest() {
     TestCuMatrixSoftmax<Real>(sizes[s]);
   for (int32 s = 0; s < ns; s++)
     TestCuMatrixDiffSoftmax<Real>(sizes[s]);
+  for (int32 s = 0; s < ns; s++)
+    TestCuMatrixDiffLogSoftmax<Real>(sizes[s]);
   for (int32 s = 0; s < ns; s++)
     TestCuMatrixLogSoftmax<Real>(sizes[s]);
   for (int32 s = 0; s < ns; s++)

src/cudamatrix/cu-matrix-test.cc

@@ -2047,6 +2047,43 @@ static void UnitTestCuDiffSoftmax() {
 }
 
 
+template<typename Real>
+static void UnitTestCuDiffLogSoftmax() {
+  int m = 100, n = 111;
+  Matrix<Real> Hi(m, n);
+  Matrix<Real> Ho(m, n);
+  Matrix<Real> Hy(m, n);
+  Hi.SetRandn();
+  RandZeroToOneMatrix(&Hy);
+
+  CuMatrix<Real> Di(m, n);
+  CuMatrix<Real> Do(m, n);
+  CuMatrix<Real> Dy(m, n);
+  Di.CopyFromMat(Hi);
+  Dy.CopyFromMat(Hy);
+
+  //gpu
+  Do.DiffLogSoftmaxPerRow(Dy, Di);
+  //cpu
+  {
+    const MatrixBase<Real> &Y(Hy), &E(Hi);
+    MatrixBase<Real> &D(Ho);
+    D.CopyFromMat(Y);
+    D.ApplyExp();                           // exp(y)
+    Vector<Real> E_sum(D.NumRows());        // Initializes to zero
+    E_sum.AddColSumMat(1.0, E);             // Sum(e)
+    D.MulRowsVec(E_sum);                    // exp(y) Sum(e)
+    D.Scale(-1.0);                          // - exp(y) Sum(e)
+    D.AddMat(1.0, E, kNoTrans);             // e - exp(y_i) Sum(e)
+  }
+
+  Matrix<Real> Ho2(m, n);
+  Do.CopyToMat(&Ho2);
+
+  AssertEqual(Ho, Ho2);
+}
+
+
 template<typename Real>
 static void UnitTestCuSoftmax() {
 
@@ -2645,6 +2682,7 @@ template<typename Real> void CudaMatrixUnitTest() {
   UnitTestCuMatrixAdd2<Real>();
   UnitTestCuDiffSigmoid<Real>();
   UnitTestCuDiffSoftmax<Real>();
+  UnitTestCuDiffLogSoftmax<Real>();
   UnitTestCuMatrixGroupPnorm<Real>();
   UnitTestCuMatrixGroupPnormDeriv<Real>();
   UnitTestCuMatrixDiffGroupPnorm<Real>();