I’m running an older Fortran code on MacOS Big Sur with apple silicon. There is a compiler bug that I can not identify specifically. I’d like to find it, so I can make sure the GNU developers know about it. Here is the code: GitHub - Nicholaswogan/ImpactAtmosphere at dev compile with cmake in the …

Here are three errors that I found. There are three places in EvolveAtmFort.f90 where workspace arrays WA are allocated. Those should have been declared as type double precision, whereas you have declared them as type integer. As a consequence, the workspace passed in the second and third calls is s…

Are you sure that lmdif2 is bug free? If it has an out of bounds access, it might write to the address of N. Did you run with debug options and/or valgrind?

The procedure argument passed to LMDIF itself ends up calling LMDIF. You have tried to circumvent recursion by using two copies of the Netlib lmdif1.f, each of those calls subroutine LMDIF, and the compiler does not know that subroutine LMDIF itself is recursive. You have to unravel this recursion …

@kargl . There are experimental releases for MacOS Apple M1. The goal here is to help identify those bugs. @MarDie valgrind says everything is OK. @mecej4 I think you are right. There is some recursion going on, which might be messing things up. I’m going to look more carefully at this.

@mecej4 thanks so much! This is really helpful. After fixing the wa variable, it works on my MacOS M1 machine. I should have written module interfaces for the old minpack routine. This bug would have been caught by the compiler if I had. I was suspicious of compiler bug because I found one recently…

Another observation about the algorithmic choices in the program may be helpful; it has nothing to do with the bug caused by the under-allocation of the work array wa. You are calling lmdif2() from rhs and rhs_verbose. The purpose of each of these calls is to solve a single (scalar) nonlinear equat…

You don’t need to use the old minpack routines anymore. See: jacobwilliams/minpack: Modernized Minpack (github.com) Also, if you want fzero (or any other root solver), you can use: jacobwilliams/roots-fortran: A modern Fortran library for finding the roots of continuous scalar functions of a si…

You can also use our modern Fortran wrappers for the minpack routines here: GitHub - certik/minpack: Library for solving nonlinear equations and nonlinear least squares problems

One of the non-linear solves has 1 unknown and the other has 3 unknowns. So I could apply fzero to the one unknown solve.

Help me find this gfortran bug for ARM machines

Help

Topic		Replies	Views
Let's get LFortran to compile Minpack	8	2263	January 30, 2023
Does sometimes -O2, -O3, -Ofast flags cause segment 11 error on Mac M1 chip?	8	945	March 14, 2022
Compiling gfortran in Macbook (M4 Pro) Help	12	505	September 2, 2025
Why am I getting a segmentation fault here (M1 Mac)? Help	10	7059	December 22, 2021
Can not operate Mordern minpack from github fortran-lang/minpack Help	3	496	March 26, 2023

Help me find this gfortran bug for ARM machines

Related topics