Is

Y(1:N) = a*X(1:N) + Y(1:N) 

any better/worse than

use blas
call daxpy(N, a, X, 1, Y, 1)

in a modern Fortran (e.g. GCC) in terms of performance? The former syntax is definitely much more readable. Or do/can modern compilers implicitly call system-optimized BLAS in such cases?

Is the answer universal for other BLAS subroutines that can be expressed in Fortran vector expressions, and for different compilers?

There is a lot of info about BLAS routines vs basic loops. I could not find any simple answer for the question above.

Thank you!