Add GPU elementwise multiplication for MatDense and a unit test.

Update to gpu_mod@379f19c4

gpu_mod @ 379f19c4

-Subproject commit befc13821982df29cc542f8912d281a72880b867
+Subproject commit 379f19c4a8ec5d6170da6ba23847c9ac73ea1987

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

@@ -460,6 +460,32 @@ void test_mul_vec()
 	assert(err < 1e-3);
 }
 
+void test_eltwise_mul_vec()
+{
+	cout << "test MatDense<FPP,GPU2>::eltwise_mul(const Vect<FPP,GPU2> &vec)" << endl;
+	auto size = 2048;
+	auto nrows = 1024;
+	auto ncols = size;
+	auto vec_mat = MatDense<FPP, Cpu>::randMat(size, 1);
+	Faust::Vect<FPP, Cpu> v_cpu(size, vec_mat->getData());
+	Faust::Vect<FPP, GPU2> v_gpu(size, vec_mat->getData());
+	auto cpu_mat = Faust::MatDense<FPP,Cpu>::randMat(nrows,ncols);
+	MatDense<FPP,GPU2> gpu_mat(nrows, ncols, cpu_mat->getData());
+	gpu_mat.eltwise_mul(v_gpu);
+	auto cpu_mat_out = gpu_mat.tocpu();
+	cpu_mat_out.Display();
+	// test each element
+	for(int i=0; i < cpu_mat->getNbRow();i++)
+		for(int j=0; j < cpu_mat->getNbCol(); j++)
+		{
+			auto test_elt = cpu_mat_out.getData()[j*cpu_mat_out.getNbRow()+i];
+			auto ref_elt = cpu_mat->getData()[j*cpu_mat_out.getNbRow()+i] * v_cpu[i];
+			assert(std::abs(test_elt - ref_elt) < 1e-6);
+		}
+	delete cpu_mat;
+	delete vec_mat;
+}
+
 void test_gpu_matsparse2matdense()
 {
 	cout << "test MatDense<FPP,GPU2>::MatDense<FPP,GPU2>(MatSparse<FPP,GPU2>&)" << endl;
@@ -504,6 +530,7 @@ void test_real()
 int main(int argc, char** argv)
 {
 	Faust::enable_gpu_mod();
+	test_eltwise_mul_vec();
 	test_real();
 	test_mul_gpu_dense();
 	test_resize();

src/faust_linear_operator/GPU2/faust_MatDense_gpu.cpp.in

@@ -103,6 +103,13 @@ namespace Faust
 			return out_v;
 		}
 
+	template<>
+	void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::eltwise_mul(const Vect<@FAUST_SCALAR_FOR_GM@, GPU2> &gpu_vec)
+	{
+			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
+			dsm_funcs->elt_wise_mul(this->gpu_mat, gpu_vec.gpu_mat);
+	}
+
 	template<>
 		void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::multiply(const MatSparse<@FAUST_SCALAR_FOR_GM@, Cpu> &other, MatDense<@FAUST_SCALAR_FOR_GM@,Cpu>& output, const char op_this)
 		{

src/faust_linear_operator/GPU2/faust_MatDense_gpu.h

@@ -65,6 +65,11 @@ namespace Faust
 				//TODO: void add(MatSparse<FPP,GPU2> const& A);
 				// vec = this * vec
 				Vect<FPP, Cpu> multiply(const Vect<FPP, Cpu> &vec);
+				/**
+				 * *this = *this * vec (element-wise multiplication)
+				 * possible broadcasting
+				 */
+				void eltwise_mul(const Vect<FPP, GPU2> &vec);
 				//  other = (*this) * other
 				void multiply(const MatDense<FPP, GPU2> &other, const char op_this='N');
 				//  other = (*this) * other