Add MatDense<FPP,GPU2>::add(MatSparse<FPP,Cpu>) and a test (update to gpu_mod@862ed4d).

Add also missing MatDense<FPP,Cpu>::add(MatSparse<FPP,Cpu>) function.

gpu_mod @ 862ed4dc

-Subproject commit 666a08aec40867e9ab7b3ab68c735f16abfc5e7e
+Subproject commit 862ed4dc10b7d7e66329396c7023d9f213d8fd59

misc/test/src/C++/test_matdense_gpu_mod.cpp.in

@@ -7,6 +7,14 @@
 using namespace std;
 using namespace Faust;
 
+double calc_err(MatDense<double, GPU2> &gpu_mat, MatDense<double, Cpu> &cpu_mat)
+{
+	auto gpu_mat_to_cpu = gpu_mat.tocpu();
+	auto err_diff = cpu_mat;
+	err_diff -= gpu_mat_to_cpu;
+	auto err = err_diff.norm()/cpu_mat.norm();
+	return err;
+}
 
 void test_mul_gpu_dense()
 {
@@ -277,6 +285,8 @@ void test_abs()
 	cout << "OK" << endl;
 }
 
+
+
 void test_add()
 {
 	cout << "test add" << endl;
@@ -292,6 +302,24 @@ void test_add()
 	err_diff -= gpu_mat1_to_cpu;
 	auto err = err_diff.norm()/cpu_mat1->norm();
 	assert(err < 1e-6);
+	Faust::MatSparse<double, Cpu> *cpu_mat3_sp = Faust::MatSparse<double, Cpu>::randMat(nrows, ncols, .2);
+	cpu_mat3_sp->Display();
+	cout << cpu_mat3_sp->to_string(false, true) << endl;
+	assert(calc_err(gpu_mat1, *cpu_mat1) < 1e-6);
+	cout << "--- gpu_mat1 after" << endl;
+	cout << gpu_mat1.tocpu().to_string(false, true) << endl;
+	cout << "--- cpu_mat1 before" << endl;
+	cout << cpu_mat1->to_string(false, true) << endl;
+	cout << "---" << endl;
+	cpu_mat1->add(*cpu_mat3_sp);
+	gpu_mat1.add(*cpu_mat3_sp);
+	cout << "--- cpu_mat1 after" << endl;
+	cout << cpu_mat1->to_string(false, true) << endl;
+	cout << "--- gpu_mat1 after" << endl;
+	cout << gpu_mat1.tocpu().to_string(false, true) << endl;
+	auto err_add_sp = calc_err(gpu_mat1, *cpu_mat1);
+	cout << err_add_sp << endl;
+	assert(err_add_sp < 1e-6);
 	cout << "OK" << endl;
 }
 

src/faust_linear_operator/CPU/faust_MatDense.h

@@ -383,6 +383,9 @@ void spgemm(const Faust::MatSparse<FPP,Cpu> & A,const Faust::MatDense<FPP,Cpu> &
 
         //! \brief (*this) = (*this) + A
         void add(MatDense<FPP,Cpu> const& A);
+		//
+        //! \brief (*this) = (*this) + A
+        void add(MatSparse<FPP,Cpu> const& A);
 
         //!  \brief (*this) = (*this) - A
         void sub(MatDense<FPP,Cpu> const& A);

src/faust_linear_operator/CPU/faust_MatDense.hpp

@@ -672,6 +672,24 @@ void Faust::MatDense<FPP,Cpu>::add(Faust::MatDense<FPP,Cpu> const& A)
 #ifdef __COMPILE_TIMERS__
 	t_add.stop();
 #endif
+}
+
+	template<typename FPP>
+void Faust::MatDense<FPP,Cpu>::add(Faust::MatSparse<FPP,Cpu> const& A)
+{
+#ifdef __COMPILE_TIMERS__
+	t_add.start();
+#endif
+	if ((this->getNbCol() != A.getNbCol()) || (this->getNbRow() != A.getNbRow()))
+	{
+		handleError(m_className, "add : matrix dimension not equal");
+	}
+	mat += A.mat;
+	isZeros = false;
+	this->is_identity = false;
+#ifdef __COMPILE_TIMERS__
+	t_add.stop();
+#endif
 }
 
 	template<typename FPP>
@@ -735,7 +753,7 @@ std::string Faust::MatDense<FPP,Cpu>::to_string(const bool transpose /* set to f
 		str <<"zeros matrix flag" <<std::endl;
 	else
 	{
-		if (displaying_small_mat_elts && this->dim1*this->dim2 < 100)
+		if (displaying_small_mat_elts && this->dim1*this->dim2 < 1000)
 		{
 			for (int i=0 ; i<this->dim1 ; i++)
 			{

src/faust_linear_operator/GPU2/faust_MatDense_gpu.h

@@ -16,12 +16,15 @@ namespace Faust
 			public:
 				MatDense(const faust_unsigned_int nbRow, const faust_unsigned_int nbCol, const FPP* data = nullptr, const bool no_alloc=false);
 
-				void add(MatDense<FPP,Cpu> const& A);
-				void add(MatDense<FPP,GPU2> const& A);
+				// *this = *this + A
+				void add(const MatDense<FPP,Cpu> & A);
+				void add(const MatDense<FPP,GPU2> & A);
+				void add(const MatSparse<FPP,Cpu> & A);
+				//TODO: void add(MatSparse<FPP,GPU2> const& A);
 				// vec = this * vec
 				Vect<FPP, Cpu> multiply(const Vect<FPP, Cpu> &vec);
 				void multiply(MatDense<FPP, GPU2> &other, const char op_this='N');
-				void multiply(MatDense<FPP,Cpu> &other, const char op_this='N');
+				void multiply(MatDense<FPP, Cpu> &other, const char op_this='N');
 //				void multiply(MatSparse<FPP, Cpu> &other, MatDense<FPP, GPU2>& output, const char op_this='N');
 				void multiply(const MatSparse<FPP, Cpu> &other, MatDense<FPP, Cpu>& output, const char op_this='N');
 				void multiply(const MatSparse<FPP, Cpu> &other, MatDense<FPP, GPU2>& output, const char op_this='N');

src/faust_linear_operator/GPU2/faust_MatDense_gpu.hpp.in

@@ -218,3 +218,10 @@ void Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::add(MatDense<@FAUST_SCALAR_FO
 	auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
 	dsm_funcs->add_cpu_dsm(gpu_mat, A.getData(), A.getNbRow(), A.getNbCol());
 }
+
+template<>
+void Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::add(MatSparse<@FAUST_SCALAR_FOR_GM@,Cpu> const& A)
+{
+	auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
+	dsm_funcs->add_cpu_spm(this->gpu_mat, A.getNbRow(), A.getNbCol(), A.getNonZeros(), A.getRowPtr(), A.getColInd(), A.getValuePtr());
+}