Compute the elementwise multplication manually (without eigen) using OpenMP to...

Compute the elementwise multplication manually (without eigen) using OpenMP to parallelize and avoiding to reorder the multiplied vector (rather indexing it).

cf. issue #275.

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

@@ -66,8 +66,13 @@ class ButterflyMat
 
 	Vec<FPP> multiply(Vec<FPP>& x) const
 	{
-		Vec<FPP> z;
-		z = d1.array() * x.array() + d2.array() * x(subdiag_ids, Eigen::placeholders::all).array();
+		Vec<FPP> z(x.rows());
+		//auto y = x(subdiag_ids, Eigen::placeholders::all).array();
+		#pragma omp parallel for
+		for(int i=0;i < x.rows(); i++)
+			z[i] = d1[i] * x[i] + d2[i] * x[subdiag_ids[i]];
+//			z[i] = d1[i] * x[i] + d2[i] * y[i];
+//		z = d1.array() * x.array() + d2.array() * x(subdiag_ids, Eigen::placeholders::all).array();
 		return z;
 	}
 
@@ -104,8 +109,12 @@ class ButterflyPermFaust
 
 	Vec<FPP> multiply(Vec<FPP>& x) const
 	{
-		Vec<FPP> y = x(bitrev_perm, Eigen::placeholders::all).array();
-		y = perm_d.array() * y.array();
+//		Vec<FPP> y = x(bitrev_perm, Eigen::placeholders::all).array();
+//		y = perm_d.array() * y.array();
+		Vec<FPP> y(x.rows());
+		#pragma omp parallel for
+		for(int i=0;i < x.rows(); i++)
+			y[i] = perm_d[i] * x[bitrev_perm[i]];
 		int i = csr_faust.size()-2;
 		for(auto fac: opt_factors)
 		{