stack overflow problem

mybiandou — Thu, 28 Sep 2023 00:41:19 GMT

Hello, I found a strange problem when I use Intel Fortran

Intel(R) Fortran Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.10.0 Build 20230609_000000

The minimal work source code is as the below,

subroutine foo(adim1,bdim1,adim2,bdim2,a,b,abdist) implicit none integer,parameter::nmax=1000 integer ,intent(in):: adim1,bdim1,adim2,bdim2 double precision ,intent(in)::a(adim1,adim2),b(bdim1,bdim2) double precision ,intent(out)::abdist(adim2,bdim2) double precision :: c(nmax,nmax) !OK !double precision :: c(adim2,bdim2) !NG print*,"adim1,bdim1,adim2,bdim2:",adim1,bdim1,adim2,bdim2 c=0.0 abdist=0.0 return end subroutine program main implicit none integer,parameter::nmax=1000,ndim=2 integer na,nb double precision a(ndim,nmax),b(ndim,nmax) integer i double precision abdist_(nmax,nmax) double precision ,allocatable :: abdist(:,:) print*,'test:' !OK case na = 500 ; nb = 250 !(1) ! NG case na = 500 ; nb = 300 !(2) do i=1,na a(:,i) = i*1.0 end do do i=1,nb b(:,i) = i*1.0 end do !reshape abdist according to the current size abdist = RESHAPE(abdist_,(/na,nb/)) call foo(ndim,ndim,na,nb,a ,b,abdist) print*,"final abdist:",abdist(na,nb) stop end

In the subroutine foo, if I declare array c

double precision :: c(adim2,bdim2)

Then, in my test main program,

when I let na=500;nb=250, it is OK.

But when I let na=500;nb=300, it gave me error.

forrtl: severe (170): Program Exception - stack overflow

But if I declare c in subroutine foo as

double precision :: c(nmax,nmax) !nmax=1000

it will also be OK.

I compile the above code using command

ifort test.f90 -check -traceback -o test.exe

gfortran has no such problems.

Please help me to solve this problem. Thank you in advance.

Re: stack overflow problem

andrew_4619 — Thu, 28 Sep 2023 10:47:06 GMT

You are exceeding the stack size. There are compiler options to change the stack size and also the heaparrays option to make certain things be created on heap rather than stack. it gets a bit complicated and is effort! As a principal it is usually better to make larger arrays allocatable and allocate them and they are then always on the heap. See if that fixes the problem.

Re: stack overflow problem

Steve_Lionel — Thu, 28 Sep 2023 15:12:25 GMT

gfortran uses the equivalent of Intel's -heap-arrays option by default.

Re: stack overflow problem

mybiandou — Thu, 28 Sep 2023 23:36:29 GMT

Thanks for your answers, andrew_461 and Steve_Lionel.
Following your help, I use -heap-arrays option and get it run ok.

ifort test.f90 -heap-arrays -check -traceback -o test.exe

topic Re: stack overflow problem in Intel® Fortran Compiler

stack overflow problem

Re: stack overflow problem

Re: stack overflow problem

Re: stack overflow problem