Implement GPU MatDense::real() and add a unit test for it.

Update to gpu_mod@97e4ce3b.

gpu_mod @ 97e4ce3b

-Subproject commit 746033d0f6406e8361b5d785873b2f664b742981
+Subproject commit 97e4ce3b76bcb4057b9e4d8dc16a8f7ee1f3cd7b

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

@@ -14,7 +14,8 @@ typedef @TEST_FPP@ FPP;
 using namespace std;
 using namespace Faust;
 
-double calc_err(MatDense<FPP, GPU2> &gpu_mat, MatDense<FPP, Cpu> &cpu_mat)
+template<typename U>
+double calc_err(MatDense<U, GPU2> &gpu_mat, MatDense<U, Cpu> &cpu_mat)
 {
 	auto gpu_mat_to_cpu = gpu_mat.tocpu();
 	auto err_diff = cpu_mat;
@@ -479,9 +480,31 @@ void test_trace()
 {
 }
 
+void test_real()
+{
+	cout << "test real()" << endl;
+	faust_unsigned_int nrows = 10, ncols = 10;
+	auto cpu_mat1 = Faust::MatDense<FPP,Cpu>::randMat(nrows,ncols);
+	cout << "cpu_mat1" << endl;
+	cpu_mat1->Display();
+	MatDense<Real<FPP>, Cpu> cpu_mat_ref;
+	cpu_mat1->real(cpu_mat_ref);
+	cout << "cpu_mat_ref" << endl;
+	cpu_mat_ref.Display();
+	cout << cpu_mat_ref.norm() << endl;
+	MatDense<FPP,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
+	MatDense<Real<FPP>, GPU2> gpu_mat2;
+	gpu_mat1.real(gpu_mat2);
+	cout << "gpu_mat2" << endl;
+	gpu_mat2.Display();
+	cout << gpu_mat2.norm() << endl;
+	assert(calc_err<Real<FPP>>(gpu_mat2, cpu_mat_ref) < 1e-6);
+}
+
 int main(int argc, char** argv)
 {
 	Faust::enable_gpu_mod();
+	test_real();
 	test_mul_gpu_dense();
 	test_resize();
 	test_setones();

src/faust_linear_operator/GPU2/faust_MatDense_gpu.cpp.in

@@ -737,4 +737,15 @@ namespace Faust
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->prox_spcol(gpu_mat, k, normalized, pos);
 		}
+
+	template<>
+		void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::real(Faust::MatDense<Real<@FAUST_SCALAR_FOR_GM@>, GPU2>& real_mat) const
+		{
+			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
+			auto real_gpu_mat = dsm_funcs->real(gpu_mat);
+			if(real_mat.gpu_mat != nullptr)
+				dsm_funcs->free(real_mat.gpu_mat);
+			real_mat.gpu_mat = real_gpu_mat;
+		}
+
 };

src/faust_linear_operator/GPU2/faust_MatDense_gpu.h

@@ -19,6 +19,7 @@ namespace Faust
 		{
 			friend Transform<FPP,GPU2>; // need to access to get_gpu_mat_ptr
 			friend MatSparse<FPP,GPU2>;
+			friend MatDense<std::complex<double>,GPU2>; // TODO limit to real function
 //			friend void gemm<>(const MatDense<FPP, GPU2> &A, const MatDense<FPP, GPU2> &B, MatDense<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB);
 //
 //			friend void gemv<>(const MatDense<FPP, GPU2> &A, const Vect<FPP, GPU2> &B, Vect<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB);
@@ -128,6 +129,7 @@ namespace Faust
 				void prox_sp(int32_t k, bool normalized=false, bool pos=false) const;
 				void prox_spcol(int32_t k, bool normalized=false, bool pos=false) const;
 				void prox_splin(int32_t k, bool normalized=false, bool pos=false) const;
+				void real(MatDense<Real<FPP>, GPU2>& real_mat) const;
 				static void gemm(const MatDense<FPP, GPU2> &A, const MatDense<FPP, GPU2> &B, MatDense<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB);
 				static void gemv(const MatDense<FPP, GPU2> &A, const Vect<FPP, GPU2> &B, Vect<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB='N');