From: Steve Kargl <sgk@troutmask.apl.washington.edu>
To: Mikael Morin <morin-mikael@orange.fr>
Cc: fortran@gcc.gnu.org
Subject: Re: Need a hint or more likely help
Date: Sun, 11 Feb 2024 09:40:04 -0800 [thread overview]
Message-ID: <ZckGdHM7rs7hG90m@troutmask.apl.washington.edu> (raw)
In-Reply-To: <c96a1fa3-dda8-4299-9145-82ea580ca996@orange.fr>
On Sun, Feb 11, 2024 at 11:56:52AM +0100, Mikael Morin wrote:
> Hello,
>
> Le 11/02/2024 à 03:00, Steve Kargl a écrit :
> > All, consider this simple code:
> >
> > module foo
> > contains
> > subroutine bar
> > character(len=:), allocatable :: s(:)
> > call bah(s)
> > end subroutine bar
> > end module foo
> >
> > If one compiles with -fdump-tree-original, one see (with some pruning)
> >
> > void bar ()
> > {
> > integer(kind=8) .s;
> > struct array01_character(kind=1) s;
> >
> > The above two lines seem to be ok.
> >
> > bitsizetype D.4319;
> > sizetype D.4320;
> >
> > try
> > {
> > D.4319 = (bitsizetype) (sizetype) NON_LVALUE_EXPR <.s> * 8;
> > D.4320 = (sizetype) NON_LVALUE_EXPR <.s>;
> > s.data = 0B;
> > s.dtype = {.elem_len=(unsigned long) .s, .version=0, .rank=1, .type=6};
> > bah ((character(kind=1)[0:][1:.s] * restrict) s.data, .s);
> > }
> >
> > This is bad. .s is undefined. I've trace this to trans-array.cc:11531
> >
> > if (sym->ts.type == BT_CHARACTER
> > && !INTEGER_CST_P (sym->ts.u.cl->backend_decl))
> > {
> > gfc_conv_string_length (sym->ts.u.cl, NULL, &init);
> > gfc_trans_vla_type_sizes (sym, &init);
> >
> > The problem here is that sym->ts.u.cl->length == NULL. If I change
> > the conditional to
> >
> > if (sym->ts.type == BT_CHARACTER
> > && sym->ts.u.cl->length
> > && !INTEGER_CST_P (sym->ts.u.cl->backend_decl))
> >
> No, I think sym->ts.u.cl->length == NULL is correct for deferred length.
> It is set only when there is an expression defining the length:
> character(len=size(somevar)) :: s(:)
Yes, you are correct. I spent some time reading gfc_conv_string_length()
and does appear to address the NULL case.
> > then the option -fdump-tree-original produces
> >
> > void bar ()
> > {
> > integer(kind=8) .s;
> > struct array01_character(kind=1) s;
> >
> > try
> > {
> > s.data = 0B;
> > s.dtype = {.elem_len=(unsigned long) .s, .version=0, .rank=1, .type=6};
> > bah ((character(kind=1)[0:][1:.s] * restrict) s.data, .s);
> > }
> >
> > which looks good except I don't know what the reference to .s here
> > means. Is this correct or should we set .s to zero by artificially
> > setting sym->ts.u.cl->length to, say, zero length?
> >
> Setting the variable .s to zero should be correct in any case, but it's not
> the same as setting the expression.
> I think you can use
> if (sym->ts.deferred)
> gfc_add_modify (
> &init,
> sym->ts.u.cl->backend_decl,
> build_zero_cst (TREE_TYPE (sym->ts.u.cl->backen_decl))
> );
> to set .s to zero.
>
> Actually, setting to any value should be correct (including -1), as the
> value should not be used as long as the data pointer is NULL.
>
> What is not clear is whether the values of D.4319 and D.4320 need update
> when memory is allocated for s. I think those variables are referenced in
> the type of s (but it's not visible in the dump), and may need update when
> the value of .s changes.
>
> I hope it helps.
Indeed, the above appears to be the solution. I looked at
the first 9 dumps from -fdump-tree-all, and by time we get
to a.f90.022t.ssa the temp variables D.4310 and D.4320 seem
to be optimized out/reaped. Thanks for the help. I tend
to forget how the trans-* files work, and sometimes need
a reminder.
--
Steve
next prev parent reply other threads:[~2024-02-11 17:40 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-02-11 2:00 Steve Kargl
2024-02-11 7:02 ` Steve Kargl
2024-02-11 10:56 ` Mikael Morin
2024-02-11 17:40 ` Steve Kargl [this message]
2024-02-12 3:14 ` Steve Kargl
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZckGdHM7rs7hG90m@troutmask.apl.washington.edu \
--to=sgk@troutmask.apl.washington.edu \
--cc=fortran@gcc.gnu.org \
--cc=morin-mikael@orange.fr \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).