Add GPU2 MatButterfly get_gpu_mat_ptr, conjugate, adjoint, Display and unit tests.

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

@@ -182,6 +182,69 @@ void test_transpose(MatButterfly<FPP, GPU2> &gpu_bm, MatButterfly<FPP, Cpu> &cpu
 	cout << "OK" << endl;
 }
 
+void test_conjugate(MatButterfly<FPP, GPU2> &gpu_bm, MatButterfly<FPP, Cpu> &cpu_bm)
+{
+	cout << "Test MatButterfly<FPP, GPU2>::conjugate()" << endl;
+	auto size = cpu_bm.getNbRow();
+	auto rmat = MatDense<FPP, Cpu>::randMat(size, size);
+	MatDense<FPP, Cpu> ref_mat(*rmat);
+	MatDense<FPP, GPU2> test_mat(*rmat);
+	auto gpu_clone = gpu_bm.clone();
+	auto cpu_clone = cpu_bm.Clone();
+	cpu_clone->conjugate();
+	cpu_clone->multiply(ref_mat, 'N');
+	gpu_clone->conjugate();
+	gpu_clone->multiply(test_mat, 'N');
+	auto test_mat_cpu = test_mat.tocpu();
+	assert(verifyMatEq(test_mat_cpu, ref_mat));
+	// undo the conjugate
+	cpu_clone->conjugate();
+	cpu_clone->multiply(ref_mat, 'N');
+	gpu_clone->conjugate();
+	gpu_clone->multiply(test_mat, 'N');
+	test_mat_cpu = test_mat.tocpu();
+	assert(verifyMatEq(test_mat_cpu, ref_mat));
+	delete rmat;
+	delete gpu_clone;
+	delete cpu_clone;
+	cout << "OK" << endl;
+}
+
+void test_adjoint(MatButterfly<FPP, GPU2> &gpu_bm, MatButterfly<FPP, Cpu> &cpu_bm)
+{
+	cout << "Test MatButterfly<FPP, GPU2>::adjoint()" << endl;
+	auto size = cpu_bm.getNbRow();
+	auto rmat = MatDense<FPP, Cpu>::randMat(size, size);
+	MatDense<FPP, Cpu> ref_mat(*rmat);
+	MatDense<FPP, GPU2> test_mat(*rmat);
+	auto gpu_clone = gpu_bm.clone();
+	auto cpu_clone = cpu_bm.Clone();
+	cpu_clone->adjoint();
+	cpu_clone->multiply(ref_mat, 'N');
+	gpu_clone->adjoint();
+	gpu_clone->multiply(test_mat, 'N');
+	auto test_mat_cpu = test_mat.tocpu();
+	assert(verifyMatEq(test_mat_cpu, ref_mat));
+	// undo the adjoint
+	cpu_clone->adjoint();
+	cpu_clone->multiply(ref_mat, 'N');
+	gpu_clone->adjoint();
+	gpu_clone->multiply(test_mat, 'N');
+	test_mat_cpu = test_mat.tocpu();
+	assert(verifyMatEq(test_mat_cpu, ref_mat));
+	delete rmat;
+	delete gpu_clone;
+	delete cpu_clone;
+	cout << "OK" << endl;
+}
+
+void test_display(MatButterfly<FPP, GPU2> &gpu_bm)
+{
+	cout << "Test MatButterfly<FPP, GPU2>::Display()" << endl;
+	gpu_bm.Display();
+	cout << "OK" << endl;
+}
+
 int main(int argc, char** argv)
 {
 	Faust::enable_gpu_mod();
@@ -208,6 +271,9 @@ int main(int argc, char** argv)
 	test_clone(gpu_bm, cpu_bm);
 	test_getNonZeros(gpu_bm, cpu_bm);
 	test_transpose(gpu_bm, cpu_bm);
+	test_conjugate(gpu_bm, cpu_bm);
+	test_adjoint(gpu_bm, cpu_bm);
+	test_display(gpu_bm);
 
 	return 0;
 }

src/faust_linear_operator/GPU2/faust_MatButterfly_gpu.h

@@ -33,9 +33,9 @@ namespace Faust
 			MatButterfly<FPP,GPU2>* Clone(const bool isOptimize=false) const;
 			void transpose();
 			void init_transpose();
-			/*  void* get_gpu_mat_ptr() const;
-			  void conjugate();
-			  void adjoint();*/
+			void conjugate();
+			void adjoint();
+			void* get_gpu_mat_ptr() const {return d1.getData();} // not only one pointer, choose d1 arbitrarily 
 			//! \brief Returns a sub-group of rows of this matrix as the same type of matrix
 			MatSparse<FPP,GPU2>* get_rows(faust_unsigned_int row_id_start, faust_unsigned_int num_rows) const;
 			//! \brief Returns a sub-group of rows of this matrix as the same type of matrix
@@ -45,11 +45,12 @@ namespace Faust
 			//! \brief Returns a sub-group of columns of this matrix as the same type of matrix
 			MatSparse<FPP,GPU2>* get_cols(faust_unsigned_int* col_ids, faust_unsigned_int num_cols) const;
 
-			/*void Display() const;*/
+			void Display() const;
 			Real<FPP> norm() const;
 			void multiply(MatDense<FPP, GPU2> &other, const char op_this);
 			void multiply(MatSparse<FPP, GPU2> &other, const char op_this);
 			MatSparse<FPP, GPU2> toMatSparse() const;
+			~MatButterfly();
 		};
 
 }

src/faust_linear_operator/GPU2/faust_MatButterfly_gpu.hpp

@@ -26,6 +26,8 @@ namespace Faust
 	template<typename FPP>
 		void MatButterfly<FPP, GPU2>::multiply(MatDense<FPP, GPU2> &other, const char op_this)
 		{
+			if(op_this != 'N' && op_this != 'T')
+				throw std::runtime_error("MatButtermfly::multiply only handle 'N' and 'T' for op_this");
 			bool use_d2t = is_transp ^ op_this == 'T';
 			butterfly_diag_prod(other, d1, use_d2t?d2t:d2, subdiag_ids);
 		}
@@ -153,4 +155,48 @@ namespace Faust
 				d2t =  cpu_d2t;
 			}
 		}
+
+	template<typename FPP>
+		void MatButterfly<FPP, GPU2>::conjugate()
+		{
+			d1.conjugate();
+			d2.conjugate();
+			if(d2t.size() > 0)
+				d2t.conjugate();
+		}
+
+
+	template<typename FPP>
+		void MatButterfly<FPP, GPU2>::adjoint()
+		{
+			transpose();
+			conjugate();
+		}
+
+
+	template<typename FPP>
+		void MatButterfly<FPP, GPU2>::Display() const
+		{
+			//TODO: adjust consistently with MatGeneric Display (using to_string)
+			std::cout << "MatButterfly on GPU: ";
+			std::cout << "D1: ";
+			d1.tocpu().Display();
+			std::cout << "D2: ";
+			d2.tocpu().Display();
+			if(d2t.size() > 0)
+			{
+				std::cout << "D2T: ";
+				d2t.tocpu().Display();
+			}
+			std::cout << "subdiag_ids: ";
+			for(int i=0;i < d2.size();i++)
+				std::cout << subdiag_ids[i] << " ";
+			std::cout << std::endl;
+		}
+
+	template<typename FPP>
+		MatButterfly<FPP, GPU2>::~MatButterfly()
+		{
+			delete subdiag_ids;
+		}
 }