look at that subtle difference in Product.h...

the cacheOptimal is only good for large enough matrices.
When taking a block in a fixed-size (hence small) matrix,
the SizeAtCompileTime is Dynamic hence that's not a good
indicator. This example shows that the good indicator is
MaxSizeAtCompileTime.
Result: +10% speed in echelon.cpp

2 files changed, 4 insertions, 3 deletions

diff --git a/Eigen/src/Core/Product.h b/Eigen/src/Core/Product.h
index 548656209..a1ec6b8d2 100644
--- a/Eigen/src/Core/Product.h
+++ b/Eigen/src/Core/Product.h
@@ -91,8 +91,9 @@ struct ei_product_eval_mode
 {
   enum {
     SizeAtCompileTime = MatrixBase<Product<Lhs,Rhs,UnrolledDotProduct> >::SizeAtCompileTime,
+    MaxSizeAtCompileTime = MatrixBase<Product<Lhs,Rhs,UnrolledDotProduct> >::MaxSizeAtCompileTime,
     EvalMode = ( EIGEN_UNROLLED_LOOPS
-              && SizeAtCompileTime != Dynamic
+              && MaxSizeAtCompileTime != Dynamic
               && SizeAtCompileTime <= EIGEN_UNROLLING_LIMIT) ? UnrolledDotProduct : CacheOptimal,
   };
 };
diff --git a/doc/echelon.cpp b/doc/echelon.cpp
index 8e3b03cd9..d28726b0f 100644
--- a/doc/echelon.cpp
+++ b/doc/echelon.cpp
@@ -67,8 +67,8 @@ int main(int, char **)
   // now m is still a matrix of rank N-1
   cout << "Here's the matrix m:" << endl << m << endl;
 
-  cout << "Now let's echelon m:" << endl;
-  echelon(m);
+  cout << "Now let's echelon m (repeating many times for benchmarking purposes):" << endl;
+  for(int i = 0; i < 1000000; i++) echelon(m);
 
   cout << "Now m is:" << endl << m << endl;
 }