Update the push_back functions in pyfaust wrapper and C++ core (CPU and GPU)...

Update the push_back functions in pyfaust wrapper and C++ core (CPU and GPU) in order to remove one avoidable copy for array/MatDense factor addition in a Faust.

src/faust_linear_operator/CPU/faust_TransformHelper.h

@@ -111,6 +111,7 @@ namespace Faust
 				void push_back_(Head& h, Tail&... t);
 			//
 			void push_back_();
+			void push_back(const FPP* data, int nrows, int ncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 			void push_back(const FPP* data, const int* row_ptr, const int* id_col, const int nnz, const int nrows, const int ncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 			void push_back(const FPP* bdata, const int* brow_ptr, const int* bcol_inds, const int nrows, const int ncols, const int bnnz, const int bnrows, const int bncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 

src/faust_linear_operator/CPU/faust_TransformHelper.hpp

@@ -633,6 +633,15 @@ namespace Faust {
 				if(copying) delete bsr_mat;
 			}
 
+	template<typename FPP>
+		void TransformHelper<FPP,Cpu>::push_back(const FPP* data, int nrows, int ncols, const bool optimizedCopy/*=famse*/, const bool transpose/*=famse*/, const bool conjugate/*=famse*/)
+		{
+			auto copying = optimizedCopy||transpose||conjugate;
+			auto dense_mat = new MatDense<FPP, Cpu>(data, nrows, ncols);
+			this->push_back(dense_mat, optimizedCopy, copying, transpose, conjugate);
+			if(copying) delete dense_mat;
+		}
+
 	template<typename FPP>
 		void TransformHelper<FPP,Cpu>::push_back(const FPP* data, const int* row_ptr, const int* id_col, const int nnz, const int nrows, const int ncols, const bool optimizedCopy /* false by deft */, const bool transpose/*=false*/, const bool conjugate/*=false*/)
 		{
@@ -1174,13 +1183,9 @@ template<typename FPP>
 		int num_factors = num_fac_distr(generator);
 		// create factors
 		std::vector<MatGeneric<FPP,Cpu>*> factors(num_factors);
-//		std::cout  << "num_factors:" << num_factors << std::endl;
-//		std::cout  << "bnnz :" << bnnz << std::endl;
 		for(int i=0;i<num_factors;i++)
-		{
 			factors[i] = MatBSR<FPP, Cpu>::randMat(faust_nrows, faust_ncols, bnrows, bncols, bnnz);
-//			std::cout  << "i=" << i << " num_factors: " << num_factors; factors[i]->Display();
-		}
+		// create the Faust
 		TransformHelper<FPP,Cpu>* randFaust = new TransformHelper<FPP, Cpu>(factors,1.0,false,false);
 		return randFaust;
 	}

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.h

@@ -24,6 +24,7 @@ namespace Faust
 				template<typename ...GList> TransformHelper(GList& ... t);
 #endif
 
+				void push_back(const FPP* data, int nrows, int ncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 				void push_back(const FPP* data, const int* row_ptr, const int* id_col, const int nnz, const int nrows, const int ncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 				void push_back(const FPP* bdata, const int* brow_ptr, const int* bcol_inds, const int nrows, const int ncols, const int bnnz, const int bnrows, const int bncols, const bool optimizedCopy=false, const bool transpose=false, const bool conjugate=false);
 				void push_back(const MatGeneric<FPP,GPU2>* M, const bool optimizedCopy=false, const bool copying=true, const bool transpose=false, const bool conjugate=false);

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.hpp

@@ -97,13 +97,22 @@ namespace Faust
 			this->transform->push_back(gpu_M, false);
 		}
 
+	template<typename FPP>
+		void TransformHelper<FPP,GPU2>::push_back(const FPP* data, const int nrows, const int ncols, const bool optimizedCopy/*=false*/, const bool transpose/*=false*/, const bool conjugate/*=false*/)
+		{
+			auto dense_mat = new MatDense<FPP,GPU2>(nrows, ncols, data, /* no_alloc */ false);
+			auto copying = transpose || conjugate || optimizedCopy;
+			this->push_back(dense_mat, optimizedCopy, copying, transpose, conjugate); // optimizedCopy not supported on GPU2
+			if(copying) delete dense_mat;
+		}
 
 	template<typename FPP>
 		void TransformHelper<FPP,GPU2>::push_back(const FPP* data, const int* row_ptr, const int* id_col, const int nnz, const int nrows, const int ncols, const bool optimizedCopy/*=false*/, const bool transpose/*=false*/, const bool conjugate/*=false*/)
 		{
 			auto sparse_mat = new MatSparse<FPP,GPU2>(nrows, ncols, nnz, data, row_ptr, id_col);
-			this->push_back(sparse_mat, optimizedCopy, false, transpose, conjugate);
-//			if(optimizedCopy) delete sparse_mat; // optimizedCopy not supported on GPU2
+			auto copying = transpose || conjugate || optimizedCopy;
+			this->push_back(sparse_mat, optimizedCopy, copying, transpose, conjugate); // optimizedCopy not supported on GPU2
+			if(copying) delete sparse_mat;
 		}
 
 	template<typename FPP>

wrapper/python/src/FaustCoreCpp.hpp

@@ -58,11 +58,8 @@ FaustCoreCpp<FPP, DEV>::FaustCoreCpp(Faust::TransformHelper<FPP,DEV> *th)
     template<typename FPP, FDevice DEV>
 void FaustCoreCpp<FPP,DEV>::push_back(FPP* valueMat, unsigned int nbrow, unsigned int nbcol, bool optimizedCopy /* false by deft */)
 {
-    Faust::MatDense<FPP,DEV> dense_mat(nbrow, nbcol, valueMat);
-    // Faust::MatSparse<FPP,DEV> sparse_mat(dense_mat);
-    // sparse_mat.Display();
     if(transform == nullptr) transform = new Faust::TransformHelper<FPP,DEV>();
-    this->transform->push_back(&dense_mat, optimizedCopy);
+    this->transform->push_back(valueMat, nbrow, nbcol, optimizedCopy);
 }
 
 template<typename FPP, FDevice DEV>