Fix GPU2 missing multiply signature (issue #189).

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.h

@@ -45,7 +45,8 @@ namespace Faust
 				TransformHelper<FPP,GPU2>* multiply(const TransformHelper<FPP,GPU2>*);
 				Vect<FPP,GPU2> multiply(const Faust::Vect<FPP,GPU2>& a, const bool transpose=false, const bool conjugate=false);
 				Vect<FPP,Cpu> multiply(const Vect<FPP,Cpu> &x, const bool transpose=false, const bool conjugate=false);
-				void multiply(const FPP* x, FPP* y, const bool transpose=false, const bool conjugate=false);
+				void multiply(const FPP* cpu_x, FPP* cpu_y, const bool transpose=false, const bool conjugate=false);
+				void multiply(const FPP* cpu_x, int x_ncols, FPP* cpu_y, const bool transpose=false, const bool conjugate=false);
 				Real<FPP> normFro() const;
 				Real<FPP> normL1() const;
 				Real<FPP> normInf() const;

src/faust_linear_operator/GPU2/faust_TransformHelper_gpu.hpp

@@ -339,6 +339,24 @@ namespace Faust
 			v.tocpu(cpu_out_buf);
 		}
 
+	template<typename FPP>
+		void TransformHelper<FPP,GPU2>::multiply(const FPP* cpu_x_buf, int x_ncols, FPP* cpu_out_buf, const bool transpose /* deft to false */, const bool conjugate/*=false*/)
+		{
+			this->is_transposed ^= transpose;
+			this->is_conjugate ^= conjugate;
+			int32_t x_nrows;
+			if(this->is_transposed)
+				x_nrows = this->getNbRow();
+			else
+				x_nrows = this->getNbCol();
+			MatDense<FPP,GPU2> gpu_x(x_nrows, x_ncols, cpu_x_buf);
+			MatDense<FPP,GPU2> gpu_M = this->multiply(gpu_x, transpose, conjugate); //TODO: handle transpose and conjugate
+			// TODO: fix this function, it works until here then it segfaults or gives a cuda error with tocpu (even if I use a cpu matdense set locally)
+			this->is_transposed ^= transpose;
+			this->is_conjugate ^= conjugate;
+			gpu_M.tocpu(cpu_out_buf);
+		}
+
 	template<typename FPP>
 		TransformHelper<FPP,GPU2>* TransformHelper<FPP,GPU2>::multiply(const FPP& a)
 		{

wrapper/python/src/FaustCoreCpp.hpp

@@ -177,6 +177,7 @@ void FaustCoreCpp<FPP,DEV>::multiply(FPP* value_y,int nbrow_y,int nbcol_y, const
         {
             // assuming that x and y are pointers to memory allocated to the proper
             // sizes
+//            Y = this->transform->multiply(value_x);
             this->transform->multiply(value_x, value_y);
         }
         else
@@ -185,13 +186,12 @@ void FaustCoreCpp<FPP,DEV>::multiply(FPP* value_y,int nbrow_y,int nbcol_y, const
             Faust::Vect<FPP,Cpu> Y;
 
             Y = this->transform->multiply(X);
-            Y = this->transform->multiply(value_x);
             memcpy(value_y,Y.getData(),sizeof(FPP)*nbrow_y);
         }
     }
     else
     {
-        if(this->transform->get_mul_order_opt_mode() == Faust::DEFAULT)
+        if(this->transform->get_mul_order_opt_mode() == Faust::DEFAULT && DEV == Cpu /*tmp fix to GPU2 that fails in multiply below */)
         {
             //assuming that value_x and value_y are allocated properly (to the good
             //sizes) in numpy world
@@ -208,6 +208,7 @@ void FaustCoreCpp<FPP,DEV>::multiply(FPP* value_y,int nbrow_y,int nbcol_y, const
         }
     }
 }
+
 template<typename FPP, FDevice DEV>
  FaustCoreCpp<FPP,DEV>* FaustCoreCpp<FPP,DEV>::polyNext() const
 {