Getting rid of OpenBLAS compilation warnings.

CMakeLists.txt

@@ -44,10 +44,12 @@ MESSAGE("  Boost_LIBRARY_DIRS: ${Boost_LIBRARY_DIRS}")
 find_package( TBB CONFIG )
 
 # If not found, use FindTBB.cmake
-#if ("${TBB_LIBRARIES}" STREQUAL "" OR NOT ${TBB_LIBRARIES})
 if ("${TBB_LIBRARIES}" STREQUAL "")
 	MESSAGE("TBB not found in CONFIG, searching with FindTBB.cmake.") 
     find_package( TBB REQUIRED )
+	if ("${TBB_LIBRARIES}" STREQUAL "")
+		MESSAGE(FATAL_ERROR "TBB not found")
+	endif()
 else()
 	MESSAGE("TBB found in CONFIG: ${TBB_LIBRARIES}") 
 endif()

lib/local/LandmarkDetector/src/CCNF_patch_expert.cpp

@@ -450,7 +450,8 @@ void CCNF_patch_expert::ResponseOpenBlas(const cv::Mat_<float> &area_of_interest
 	// Perform matrix multiplication in OpenBLAS (fortran call)
 	float alpha1 = 1.0;
 	float beta1 = 0.0;
-	sgemm_("N", "N", &normalized_input.cols, &weight_matrix.rows, &weight_matrix.cols, &alpha1, (float*)normalized_input.data, &normalized_input.cols, (float*)weight_matrix.data, &weight_matrix.cols, &beta1, (float*)neuron_resp_full.data, &normalized_input.cols);
+	char* N = "N";
+	sgemm_(N, N, &normalized_input.cols, &weight_matrix.rows, &weight_matrix.cols, &alpha1, (float*)normalized_input.data, &normalized_input.cols, (float*)weight_matrix.data, &weight_matrix.cols, &beta1, (float*)neuron_resp_full.data, &normalized_input.cols);
 
 	// Above is a faster version of this
 	//cv::Mat_<float> neuron_resp_full = this->weight_matrix * normalized_input;
@@ -494,7 +495,7 @@ void CCNF_patch_expert::ResponseOpenBlas(const cv::Mat_<float> &area_of_interest
 	// Perform matrix multiplication in OpenBLAS (fortran call)
 	alpha1 = 1.0;
 	beta1 = 0.0;
-	sgemm_("N", "N", &resp_vec_f.cols, &Sigmas[s_to_use].rows, &Sigmas[s_to_use].cols, &alpha1, (float*)resp_vec_f.data, &resp_vec_f.cols, (float*)Sigmas[s_to_use].data, &Sigmas[s_to_use].cols, &beta1, (float*)out.data, &resp_vec_f.cols);
+	sgemm_(N, N, &resp_vec_f.cols, &Sigmas[s_to_use].rows, &Sigmas[s_to_use].cols, &alpha1, (float*)resp_vec_f.data, &resp_vec_f.cols, (float*)Sigmas[s_to_use].data, &Sigmas[s_to_use].cols, &beta1, (float*)out.data, &resp_vec_f.cols);
 
 	// Above is a faster version of this
 	//cv::Mat out = Sigmas[s_to_use] * resp_vec_f;

lib/local/LandmarkDetector/src/CEN_patch_expert.cpp

@@ -217,7 +217,8 @@ void CEN_patch_expert::Response(const cv::Mat_<float> &area_of_interest, cv::Mat
 		// Perform matrix multiplication in OpenBLAS (fortran call)
 		float alpha1 = 1.0;
 		float beta1 = 0.0;
-		sgemm_("N", "N", &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
+		char* N = "N";
+		sgemm_(N, N, &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
 
 		// The above is a faster version of this, by calling the fortran version directly
 		//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);
@@ -489,7 +490,8 @@ void CEN_patch_expert::ResponseInternal(cv::Mat_<float>& response)
 		// Perform matrix multiplication in OpenBLAS (fortran call)
 		float alpha1 = 1.0;
 		float beta1 = 0.0;
-		sgemm_("N", "N", &resp.cols, &weights[layer].rows, &weights[layer].cols, &alpha1, m1, &resp.cols, m2, &weights[layer].cols, &beta1, m3, &resp.cols);
+		char* N = "N";
+		sgemm_(N, N, &resp.cols, &weights[layer].rows, &weights[layer].cols, &alpha1, m1, &resp.cols, m2, &weights[layer].cols, &beta1, m3, &resp.cols);
 
 		// The above is a faster version of this, by calling the fortran version directly
 		//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);

lib/local/LandmarkDetector/src/CNN_utils.cpp

@@ -522,7 +522,8 @@ namespace LandmarkDetector
 		float alpha = 1.0f;
 		float beta = 0.0f;
 		// Call fortran directly (faster)
-		sgemm_("N", "N", &m2_cols, &num_rows, &pre_alloc_im2col.cols, &alpha, m2, &m2_cols, m1, &pre_alloc_im2col.cols, &beta, m3, &m2_cols);
+		char* N = "N";
+		sgemm_(N, N, &m2_cols, &num_rows, &pre_alloc_im2col.cols, &alpha, m2, &m2_cols, m1, &pre_alloc_im2col.cols, &beta, m3, &m2_cols);
 
 		// Above is equivalent to out = pre_alloc_im2col * weight_matrix;
 

lib/local/LandmarkDetector/src/LandmarkDetectorModel.cpp

@@ -1120,7 +1120,8 @@ float CLNF::NU_RLMS(cv::Vec6f& final_global, cv::Mat_<float>& final_local, const
 		// Perform matrix multiplication in OpenBLAS (fortran call)
 		float alpha1 = 1.0;
 		float beta1 = 1.0;
-		sgemm_("N", "N", &J.cols, &J_w_t.rows, &J_w_t.cols, &alpha1, (float*)J.data, &J.cols, (float*)J_w_t.data, &J_w_t.cols, &beta1, (float*)Hessian.data, &J.cols);
+		char* N = "N";
+		sgemm_(N, N, &J.cols, &J_w_t.rows, &J_w_t.cols, &alpha1, (float*)J.data, &J.cols, (float*)J_w_t.data, &J_w_t.cols, &beta1, (float*)Hessian.data, &J.cols);
 
 		// Above is a fast (but ugly) version of 
 		// cv::Mat_<float> Hessian = J_w_t * J + regTerm;

lib/local/LandmarkDetector/src/PDM.cpp

@@ -146,7 +146,8 @@ void PDM::CalcShape3D(cv::Mat_<float>& out_shape, const cv::Mat_<float>& p_local
 	int p_local_cols = p_local.cols;
 	int princ_comp_rows = princ_comp.rows;
 	int princ_comp_cols = princ_comp.cols;
-	sgemm_("N", "N", &p_local_cols, &princ_comp_rows, &princ_comp_cols, &alpha1, (float*)p_local.data, &p_local_cols, (float*)princ_comp.data, &princ_comp_cols, &beta1, (float*)out_shape.data, &p_local_cols);
+	char* N = "N";
+	sgemm_(N, N, &p_local_cols, &princ_comp_rows, &princ_comp_cols, &alpha1, (float*)p_local.data, &p_local_cols, (float*)princ_comp.data, &princ_comp_cols, &beta1, (float*)out_shape.data, &p_local_cols);
 
 	// Above is a fast (but ugly) version of 
 	// out_shape = mean_shape + princ_comp * p_local;	 
@@ -645,7 +646,8 @@ void PDM::CalcParams(cv::Vec6f& out_params_global, cv::Mat_<float>& out_params_l
 		// Perform matrix multiplication in OpenBLAS (fortran call)
 		float alpha1 = 1.0;
 		float beta1 = 1.0;
-		sgemm_("N", "N", &J.cols, &J_w_t.rows, &J_w_t.cols, &alpha1, (float*)J.data, &J.cols, (float*)J_w_t.data, &J_w_t.cols, &beta1, (float*)Hessian.data, &J.cols);
+		char* N = "N";
+		sgemm_(N, N, &J.cols, &J_w_t.rows, &J_w_t.cols, &alpha1, (float*)J.data, &J.cols, (float*)J_w_t.data, &J_w_t.cols, &beta1, (float*)Hessian.data, &J.cols);
 
 		// Above is a fast (but ugly) version of 
 		// cv::Mat_<float> Hessian2 = J_w_t * J + regularisations;