It is many years since I last tried to use FORTRAN. I am now trying to use
gfortran to work through some examples from a book. Unfortunately I am being
hampered by a bug in gfortran. It usually causes SIGSEGV.
I couldn't spot this bug in a search, but this is probably due to my
inexperience.
$ gfortran --version
GNU Fortran (GCC) 4.2.1 (Ubuntu 4.2.1-5ubuntu4)
Copyright (C) 2007 Free Software Foundation, Inc.
If I compile the following (using only -Wall), all appears to be o.k.:
Subroutine fn(f, b, y, m)
dimension f(m,m)
y = b / f(1,1)
return
end
However, change it to this:
Subroutine fn(f, b, y, m)
dimension f(m,m), b(m), y(m)
y(l) = b(1) / f(1,1)
return
end
and the generated code is wrong.
The key point is the treatment of the index of y(1). It appears that it is
intended to be copied to the local stack frame and used later. In the (second)
version above, the copying to the local stack frame is skipped in the generated
code.
The following assembler code is generated from the second version by
$ gfortran -S subs2.f
The instruction which references an uninitialised part of the stack is
indicated by "*******". It appears that the code would be correct if $1 (i.e.
the index of y(1)) had been previously stored in -4(%ebp).
.file "subs2.f"
.text
.globl fn_
.type fn_, @function
fn_:
pushl %ebp
movl %esp, %ebp
subl $24, %esp
movl 20(%ebp), %eax
movl (%eax), %eax
testl %eax, %eax
js .L2
jmp .L4
.L2:
.L4:
movl 20(%ebp), %eax
movl (%eax), %eax
testl %eax, %eax
js .L5
jmp .L7
.L5:
.L7:
movl 20(%ebp), %eax
movl (%eax), %eax
movl %eax, -20(%ebp)
cmpl $0, -20(%ebp)
js .L8
movl -20(%ebp), %eax
movl %eax, -24(%ebp)
jmp .L10
.L8:
movl $0, -24(%ebp)
.L10:
movl -24(%ebp), %eax
movl %eax, -20(%ebp)
movl 20(%ebp), %eax
movl (%eax), %eax
imull -20(%ebp), %eax
testl %eax, %eax
js .L11
jmp .L13
.L11:
.L13:
movl -20(%ebp), %edx
notl %edx
movl -4(%ebp), %ecx *******
subl $1, %ecx
movl 12(%ebp), %eax
flds (%eax)
movl -20(%ebp), %eax
addl $1, %eax
leal (%eax,%edx), %edx
movl 8(%ebp), %eax
flds (%eax,%edx,4)
fdivrp %st, %st(1)
movl 16(%ebp), %eax
fstps (%eax,%ecx,4)
leave
ret
.size fn_, .-fn_
.ident "GCC: (GNU) 4.2.1 (Ubuntu 4.2.1-5ubuntu4)"
.section .note.GNU-stack,"",@progbits
--
Summary: Incorrect code generated by gfortran (missing code)
Product: gcc
Version: 4.2.1
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: fortran
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: billm at melbpc dot org dot au
GCC build triplet: 4.2.1-5ubuntu4
GCC host triplet: Linux Ubuntu
GCC target triplet: ix86
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34234
