Implement MatBSR<FPP, GPU2>::get_nrows/get_ncols and add their unit tests.

Update to gpu_mod@259cefc5

gpu_mod @ 259cefc5

-Subproject commit 3904f39eabcdf828cd474fad55e8b85882281d90
+Subproject commit 259cefc5c3e3c8e6579ab2ac7ebce2dda194a8b1

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

@@ -248,6 +248,57 @@ void test_set_zeros()
 	delete cpu_bsr_mat;
 }
 
+void test_get_nrows_ncols()
+{
+	cout << "test_get_nrows_ncols" << endl;
+	auto nrows = 10;
+	auto ncols = 8;
+	auto bnrows = 2;
+	auto bncols = 2;
+	auto bnnz = 2;
+	auto cpu_bsr_mat = Faust::MatBSR<FPP, Cpu>::randMat(nrows, ncols, bnrows, bncols, bnnz);
+	Faust::MatBSR<FPP, GPU2> gpu_bsr_mat(*cpu_bsr_mat);
+
+	auto gpu_rows = dynamic_cast<MatSparse<FPP, GPU2>*>(gpu_bsr_mat.get_rows(4, 3));
+	auto cpu_rows = cpu_bsr_mat->get_rows(4, 3);
+	MatSparse<FPP, Cpu> test_rows;
+	gpu_rows->tocpu(test_rows);
+	auto diff = MatDense<FPP, Cpu>(test_rows);
+	diff -= MatDense<FPP, Cpu>(*cpu_rows);
+	assert(diff.norm() < 1e-6);
+	delete gpu_rows;
+	delete cpu_rows;
+
+
+	auto gpu_cols = dynamic_cast<MatSparse<FPP, GPU2>*>(gpu_bsr_mat.get_cols(4, 3));
+	auto cpu_cols = cpu_bsr_mat->get_cols(4, 3);
+	MatSparse<FPP, Cpu> test_cols;
+	gpu_cols->tocpu(test_cols);
+	diff = MatDense<FPP, Cpu>(test_cols);
+	diff -= MatDense<FPP, Cpu>(*cpu_cols);
+	assert(diff.norm() < 1e-6);
+	delete gpu_cols;
+	delete cpu_cols;
+
+	faust_unsigned_int row_ids[3] = {1, 4, 6};
+	gpu_rows = dynamic_cast<MatSparse<FPP, GPU2>*>(gpu_bsr_mat.get_rows(row_ids, 3));
+	cpu_rows = cpu_bsr_mat->get_rows(row_ids, 3);
+	gpu_rows->tocpu(test_rows);
+	diff = MatDense<FPP, Cpu>(test_rows);
+	diff -= MatDense<FPP, Cpu>(*cpu_rows);
+	assert(diff.norm() < 1e-6);
+
+	faust_unsigned_int col_ids[3] = {1, 4, 6};
+	gpu_cols = dynamic_cast<MatSparse<FPP, GPU2>*>(gpu_bsr_mat.get_cols(col_ids, 3));
+	cpu_cols = cpu_bsr_mat->get_cols(col_ids, 3);
+	gpu_cols->tocpu(test_cols);
+	diff = MatDense<FPP, Cpu>(test_cols);
+	diff -= MatDense<FPP, Cpu>(*cpu_cols);
+	assert(diff.norm() < 1e-6);
+
+	cout << "OK" << endl;
+}
+
 
 int main()
 {
@@ -263,6 +314,7 @@ int main()
 	test_norm();
 	test_infos();
 	test_set_zeros();
+	test_get_nrows_ncols();
 }
 
 

src/faust_linear_operator/GPU2/faust_MatBSR_gpu.hpp

@@ -64,27 +64,51 @@ namespace Faust
 		}
 
 	template<typename FPP>
-		MatGeneric<FPP,GPU2>* MatBSR<FPP, GPU2>::get_rows(faust_unsigned_int row_id_start, faust_unsigned_int num_rows) const
+		Faust::MatGeneric<FPP,GPU2>* Faust::MatBSR<FPP,GPU2>::get_rows(faust_unsigned_int row_id_start, faust_unsigned_int num_rows) const
 		{
-			//TODO: implement (maybe by moving into .cpp.in
+			//TODO: pure GPU impl.
+			MatBSR<FPP,Cpu> cpu_copy;
+			tocpu(cpu_copy);
+			auto cpu_rows = cpu_copy.get_rows(row_id_start, num_rows);
+			auto gpu_rows = new MatSparse<FPP,GPU2>(*cpu_rows);
+			delete cpu_rows;
+			return gpu_rows;
 		}
 
 	template<typename FPP>
-		MatGeneric<FPP,GPU2>* MatBSR<FPP, GPU2>::get_rows(faust_unsigned_int* row_ids, faust_unsigned_int num_rows) const
+		Faust::MatGeneric<FPP,GPU2>* Faust::MatBSR<FPP,GPU2>::get_rows(faust_unsigned_int* row_ids, faust_unsigned_int num_rows) const
 		{
-			//TODO: implement (maybe by moving into .cpp.in
+			//TODO: pure GPU impl.
+			MatBSR<FPP,Cpu> cpu_copy;
+			tocpu(cpu_copy);
+			auto cpu_rows = cpu_copy.get_rows(row_ids, num_rows);
+			auto gpu_rows = new MatSparse<FPP,GPU2>(*cpu_rows);
+			delete cpu_rows;
+			return gpu_rows;
 		}
 
 	template<typename FPP>
-		Faust::MatGeneric<FPP,GPU2>* MatBSR<FPP, GPU2>::get_cols(faust_unsigned_int col_id_start, faust_unsigned_int num_cols) const
+		Faust::MatGeneric<FPP,GPU2>* Faust::MatBSR<FPP,GPU2>::get_cols(faust_unsigned_int col_id_start, faust_unsigned_int num_cols) const
 		{
-			//TODO: implement (maybe by moving into .cpp.in
+			//TODO: pure GPU impl.
+			MatBSR<FPP,Cpu> cpu_copy;
+			tocpu(cpu_copy);
+			auto cpu_cols = cpu_copy.get_cols(col_id_start, num_cols);
+			auto gpu_cols = new MatSparse<FPP,GPU2>(*cpu_cols);
+			delete cpu_cols;
+			return gpu_cols;
 		}
 
 	template<typename FPP>
-		Faust::MatGeneric<FPP,GPU2>* MatBSR<FPP, GPU2>::get_cols(faust_unsigned_int* col_ids, faust_unsigned_int num_cols) const
+		Faust::MatGeneric<FPP,GPU2>* Faust::MatBSR<FPP,GPU2>::get_cols(faust_unsigned_int* col_ids, faust_unsigned_int num_cols) const
 		{
-			//TODO: implement (maybe by moving into .cpp.in
+			//TODO: pure GPU impl.
+			MatBSR<FPP,Cpu> cpu_copy;
+			tocpu(cpu_copy);
+			auto cpu_cols = cpu_copy.get_cols(col_ids, num_cols);
+			auto gpu_cols = new MatSparse<FPP,GPU2>(*cpu_cols);
+			delete cpu_cols;
+			return gpu_cols;
 		}
 
 	template<typename FPP>