I use Intel OneAPI 2022.0.3, linked with Intel MKL cluster library as below
I have a simple code as below,
!include "_rms.fi"
program avx
implicit none
!include "mkl_vml.f90"
integer, parameter :: dp = kind(0.d0)
real(dp) :: t1, t2, r
call cpu_time(t1)
r = f(100000000)
call cpu_time(t2)
print *, "Time", t2-t1
print *, r
contains
real(dp) function f(N) result(r)
integer, intent(in) :: N
integer :: i
real(dp) :: j(N)
call vdsin(N,(dble([(i,i=1,N)])),j)
r = sum(j)
!r = sum(sin(dble([(i,i=1,N)])))
return
end function
end program
I found that using MKL's vdsin is 2 times slower than simply using sin.
I mean, if I do
r = sum(sin(dble([(i,i=1,N)])))
it is two times faster than using MKL’s vdsin as below
call vdsin(N,(dble([(i,i=1,N)])),j)
r = sum(j)
Does anyone know why MKL's vdsin is slower than the intrinsic sin? Or how to correctly use MKL's vdsin or MKL in general?
Thanks much in advance!
PS.
A similar post is here below too,
https://fortran-lang.discourse.group/t/why-mkls-vdsin-is-slower-than-the-intrinsic-sin/3108/3
Link Copied
