diff --git a/src/configure b/src/configure
index 7e5ccbf3d..4206839ec 100755
--- a/src/configure
+++ b/src/configure
@@ -451,7 +451,7 @@ function linux_check_dynamic {
     echo "Atlas found in $dir";
     return 0;
   else
-      echo "No libatlas.so in $dir";
+      echo "... no libatlas.so in $dir";
       return 1;
   fi
 }
diff --git a/src/cudamatrix/cu-kernels.cu b/src/cudamatrix/cu-kernels.cu
index d1c2f28dc..4f5248a38 100644
--- a/src/cudamatrix/cu-kernels.cu
+++ b/src/cudamatrix/cu-kernels.cu
@@ -276,8 +276,8 @@ static void _copy_col_from_vec(Real* mat, const Real* v, int col, MatrixDim d) {
 template<typename Real>
 __global__
 static void _apply_exp(Real* mat, MatrixDim d) {
-  int32_cuda i = blockIdx.y * blockDim.y + threadIdx.y;
-  int32_cuda j = blockIdx.x * blockDim.x + threadIdx.x;
+  int32_cuda i = blockIdx.x * blockDim.x + threadIdx.x;
+  int32_cuda j = blockIdx.y * blockDim.y + threadIdx.y;
   int32_cuda index = i + j * d.stride;
   if ( i < d.cols && j < d.rows ) {
     mat[index] = exp(mat[index]);
diff --git a/src/cudamatrix/cu-matrix-test.cc b/src/cudamatrix/cu-matrix-test.cc
index c760837f2..11c3744a1 100644
--- a/src/cudamatrix/cu-matrix-test.cc
+++ b/src/cudamatrix/cu-matrix-test.cc
@@ -143,6 +143,28 @@ static void UnitTestCuMatrixApplyLog() {
 }
 
 
+/*
+ * CuMatrix
+ */
+template<typename Real> 
+static void UnitTestCuMatrixApplyExp() {
+  int32 M = 100 + rand() % 200, N = 100 + rand() % 200;
+  Matrix<Real> H(M, N);
+  H.SetRandn();
+  H.MulElements(H); // make numbers positive
+
+  CuMatrix<Real> D(H);
+
+  D.ApplyExp();
+  H.ApplyExp();
+
+  Matrix<Real> H2(D);
+
+  AssertEqual(H,H2);
+}
+
+
+
 template<typename Real> 
 static void UnitTestCuMatrixSigmoid() {
   for (int32 i = 0; i < 3; i++) {
@@ -1852,6 +1874,7 @@ template<typename Real> void CudaMatrixUnitTest() {
   UnitTestCuMatrixCopyCross<Real>();
   UnitTestCuMatrixCopyCross2<Real>();
   UnitTestCuMatrixApplyLog<Real>();
+  UnitTestCuMatrixApplyExp<Real>();
   UnitTestCuMatrixSetRandn<Real>();
   UnitTestCuMatrixSetRandUniform<Real>();
   UnitTestCuMatrixScale<Real>();
diff --git a/tools/Makefile b/tools/Makefile
index fd0c256e9..4794162bc 100644
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -1,8 +1,6 @@
 # SHELL += -x
 
-CXX ?= g++
-CXXFLAGS ?=
-LDFLAGS ?=
+CXX = g++
 
 # On Mac OS 10.9, g++ is actually clang in disguise which by default uses the
 # new c++ standard library libc++. Since openfst uses stuff from the tr1
@@ -12,7 +10,7 @@ ifeq ($(findstring clang,$(COMPILER)),clang)
 	CXXFLAGS += -stdlib=libstdc++
 	LDFLAGS += -stdlib=libstdc++
 endif
-	
+
 all: check_required_programs sph2pipe atlas irstlm_tgt sclite_tgt openfst_tgt
 
 
@@ -109,7 +107,7 @@ irstlm_tgt: irstlm_compiled
 .PHONY: irstlm_compiled
 irstlm_compiled: irstlm/Makefile
 		cd irstlm/; \
-		$(MAKE); $(MAKE) install	
+		make; $(MAKE) install	
 
 irstlm/Makefile: irstlm/.patched
 		cd irstlm; \
@@ -123,7 +121,7 @@ irstlm/.patched: | irstlm
 		-cd irstlm;\
 		patch --verbose -N -p0 < ../interpolatedwrite-5.60.02.patch; \
 		patch --verbose -N -p0 < ../irstlm.patch; \
-		touch $@ 		
+		touch .patched		
 
 irstlm:
 		svn -r 398 co --non-interactive --trust-server-cert https://svn.code.sf.net/p/irstlm/code/trunk irstlm