Hello,

I'm running a benchmark with Eigen and MKL, and in order to get the best results I'm using information gathered from the Linpack becnmark test (hpl), to find the largest possible matrix I can use.

Since most my operations don't use a single matrix, I have to find out the total matrices that will be used in the scope, and divide my max amount of space by that.

Some of my functions require the <Matrix_Base>.array() function. Will the array create an alias? I'm assuming it will, but couldn't find any documentation on it. Just want to make sure before I run a large scale test, and crash halfway through.

Also, will a matrix expression create a tmp alias per operation?

So (if letters are matrices): E = (A + B * C - D) will have,
tmp1 = B * C, tmp2 = A + tmp1, tmp3 = tmp2 - D. E = tmp3?

Thank you!

.array() is no-op, it only provides a different API.

Regarding:
E = (A + B * C - D)

the matrix product has to be evaluated into a temporary to be efficient:

tmp = B * C;

then the other operations are evaluated through a single loop without any temp:

for(....)
E[i] = A[i] + tmp[i] - D[i];


You can avoid this tmp by rewriting it as:

E = A - D;
E.noalias() += B * C;

In this case B*C is directly accumulated into the content of E.

Most of the operations in Eigen only returns an expression with lazy evaluation, they include .transpose(), .adjoint(), etc. Only products and inverse products have to be evaluated.

Thank you!