Fix FaustGPU::pop_front, FaustGPU::power_iteration2.

This fix makes the following script https://gitlab.inria.fr/faustgrp/faust/snippets/652 to work on GPU (it wasn't working before).
Update to gpu_mod@8dde3d98.

gpu_mod @ 8dde3d98

-Subproject commit 6e39c0eb659e3af17b8e4b34b33b43c80c19b3d1
+Subproject commit 8dde3d98af29c4f4d6399db4c31bfb99ad2ffa81

src/faust_linear_operator/CPU/faust_TransformHelper.hpp

@@ -1256,6 +1256,7 @@ namespace Faust {
 	template<typename FPP>
 		double TransformHelper<FPP,Cpu>::spectralNorm(const int nbr_iter_max, double threshold, int &flag) const
 		{
+//			std::cout << "TransformHelper<FPP,Cpu>::spectralNorm" << std::endl;
 #ifdef USE_GPU_MOD
 			if(gpu_faust != nullptr)
 			{
@@ -1285,6 +1286,7 @@ namespace Faust {
 					AtA->enable_gpu_meth_for_mul(); //TODO: it should not be needed if A is already GPU enabled
 					FPP maxAbsValue = std::sqrt(AtA->gpu_faust->power_iteration(nbr_iter_max, threshold, flag));
 					delete AtA;
+					return absValue(maxAbsValue);
 				}
 				else
 				{

src/faust_linear_operator/CPU/faust_gpu_mod_gen.hpp.in

@@ -99,6 +99,8 @@ namespace Faust
 		{
 			if(use_ref_man)
 				ref_man.release(*(cpu_mat_ptrs.begin()));
+			auto marr_funcs = (gm_MatArrayFunc_@GM_SCALAR@*) this->marr_funcs;
+			marr_funcs->erase_at(gpu_mat_arr, 0, !use_ref_man);
 			cpu_mat_ptrs.erase(cpu_mat_ptrs.begin());
 		}
 	}
@@ -247,6 +249,7 @@ namespace Faust
 			{
 				push_back(m);
 			}
+//			std::cout << "FaustGPU() factors size, matarray size: " << factors.size() << " " << marr_funcs->size(gpu_mat_arr) << std::endl;
 		}
 
 	template <>
@@ -399,6 +402,10 @@ namespace Faust
 				if(xk2_bufsz < mat->getNbRow()) // xk2 buf receives the result of F*xk
 					xk2_bufsz = mat->getNbRow();
 			}
+			// make sure buffer will be larger enough by setting the lower size buffer to the larger size
+			//TODO: this seems to correct a bug but it should be further investigated
+			if(xk_bufsz > xk2_bufsz) xk2_bufsz = xk_bufsz;
+			else if(xk2_bufsz > xk_bufsz) xk_bufsz = xk2_bufsz;
 			auto xk = dsm_funcs->togpu_bufsz(this->ncols, 1, (@GM_SCALAR@*) reinterpret_cast<@GM_REINTERPRET_CAST_SCALAR@*>(xk_cpu.getData()), xk_bufsz, 1);
 			auto xk_norm = dsm_funcs->togpu_bufsz(this->ncols, 1, (@GM_SCALAR@*) reinterpret_cast<@GM_REINTERPRET_CAST_SCALAR@*>(xk_cpu.getData()), xk_bufsz, 1); // the CPU data is irrelevant for xk_norm but it doesn't matter
 			auto xk2 = dsm_funcs->togpu_bufsz(this->nrows, 1, (@GM_SCALAR@*) reinterpret_cast<@GM_REINTERPRET_CAST_SCALAR@*>(xk2_cpu.getData()), xk2_bufsz, 1);