Since in the code I use the matmul
intrinsic function to multiply two matrices, I was looking into ways of optimizing this (without invoking the subroutine dgemm
, since I would like to keep the code at a basic level). So I came across this excellent post by @zaikunzhang. One suggestion given in the comments to that post by @ivanpribec was to compile with
ifort -O3 -xHost -qopenmp -qmkl=parallel -heap-arrays 40 -qopt-matmul test_matmul.f90
In my example I am compiling with these flags:
SWITCH = /O3 /fast /Qipo /Qmkl=parallel /Qopenmp /Qparallel -qopt-matmul
The problem is that ifort does not recognize -qopt-matmul
. In particular, it gives this warning:
ifort: command line warning #10006: ignoring unknown option ‘/qopt-matmul’
I am a bit puzzled by this…