public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Patrick Palka <ppalka@redhat.com>
To: Jason Merrill <jason@redhat.com>
Cc: Patrick Palka <ppalka@redhat.com>, gcc-patches@gcc.gnu.org
Subject: Re: [PATCH] c++: unnecessary instantiation of constexpr var [PR99130]
Date: Wed, 7 Sep 2022 16:40:29 -0400 (EDT)	[thread overview]
Message-ID: <c7dc289e-6667-631f-ea1d-52a8671a81a6@idea> (raw)
In-Reply-To: <38d432fb-f9db-6f0f-1587-1b8c0f5c75e7@redhat.com>

On Wed, 7 Sep 2022, Jason Merrill wrote:

> On 9/7/22 15:41, Patrick Palka wrote:
> > Here the use of the constexpr member/variable specialization 'value'
> > from within an unevaluated context causes us to overeagerly instantiate
> > it, via maybe_instantiate_decl called from mark_used, despite only its
> > declaration not its definition being needed.
> 
> If the issue is with unevaluated context, maybe maybe_instantiate_decl should
> guard the call to decl_maybe_constant_var_p with !cp_unevaluated_operand?

Hmm, that seems to work too.  But IIUC this would mean in an evaluated
(but non-constexpr) context we'd continue to instantiate constexpr
variables _immediately_ rather than ideally allowing mark_used to
postpone their instantiation until the end of TU processing (which is
what happens with the below approach).

Another benefit of the below approach is that from within a template
definition we we now avoid instantiation altogether e.g. for

  template<class T> constexpr int value = /* blah */;

  template<class T>
  int f() { return value<int>; }

we no longer instantiate value<int> which IIUC is consistent with how we
handle other kinds of specializations used within a template definition.
So making mark_used no longer instantiate constexpr variables immediately
(in both evaluated and unevaluated contexts) seems to yield the most
benefits.

> 
> > We used to have the same issue for constexpr function specializations
> > until r6-1309-g81371eff9bc7ef made us delay their instantiation until
> > necessary during constexpr evaluation.
> > 
> > So this patch makes us avoid unnecessarily instantiating constexpr
> > variable template specializations from mark_used as well.  To that end
> > this patch pulls out the test in maybe_instantiate_decl
> > 
> >    (decl_maybe_constant_var_p (decl)
> >     || (TREE_CODE (decl) == FUNCTION_DECL
> >         && DECL_OMP_DECLARE_REDUCTION_P (decl))
> >     || undeduced_auto_decl (decl))
> > 
> > into each of its three callers (including mark_used) and refines the
> > test appropriately.  The net result is that only mark_used is changed,
> > because the other two callers, resolve_address_of_overloaded_function
> > and decl_constant_var_p, already guard the call appropriately.  And
> > presumably decl_constant_var_p will take care of instantiation when
> > needed for e.g. constexpr evaluation.
> > 
> > Bootstrapped and regteste on x86_64-pc-linux-gnu, does this look OK for
> > trunk?
> > 
> > 	PR c++/99130
> > 
> > gcc/cp/ChangeLog:
> > 
> > 	* decl2.cc (maybe_instantiate_decl): Adjust function comment.
> > 	Check VAR_OR_FUNCTION_DECL_P. Pull out the disjunction into ...
> > 	(mark_used): ... here, removing the decl_maybe_constant_var_p
> > 	part of it.
> > 
> > gcc/testsuite/ChangeLog:
> > 
> > 	* g++.dg/cpp1y/var-templ70.C: New test.
> > ---
> >   gcc/cp/decl2.cc                          | 33 ++++++++----------------
> >   gcc/testsuite/g++.dg/cpp1y/var-templ70.C | 19 ++++++++++++++
> >   2 files changed, 30 insertions(+), 22 deletions(-)
> >   create mode 100644 gcc/testsuite/g++.dg/cpp1y/var-templ70.C
> > 
> > diff --git a/gcc/cp/decl2.cc b/gcc/cp/decl2.cc
> > index 89ab2545d64..cd188813bee 100644
> > --- a/gcc/cp/decl2.cc
> > +++ b/gcc/cp/decl2.cc
> > @@ -5381,24 +5381,15 @@ possibly_inlined_p (tree decl)
> >     return true;
> >   }
> >   -/* Normally, we can wait until instantiation-time to synthesize DECL.
> > -   However, if DECL is a static data member initialized with a constant
> > -   or a constexpr function, we need it right now because a reference to
> > -   such a data member or a call to such function is not value-dependent.
> > -   For a function that uses auto in the return type, we need to instantiate
> > -   it to find out its type.  For OpenMP user defined reductions, we need
> > -   them instantiated for reduction clauses which inline them by hand
> > -   directly.  */
> > +/* If DECL is a function or variable template specialization, instantiate
> > +   its definition now.  */
> >     void
> >   maybe_instantiate_decl (tree decl)
> >   {
> > -  if (DECL_LANG_SPECIFIC (decl)
> > +  if (VAR_OR_FUNCTION_DECL_P (decl)
> > +      && DECL_LANG_SPECIFIC (decl)
> >         && DECL_TEMPLATE_INFO (decl)
> > -      && (decl_maybe_constant_var_p (decl)
> > -	  || (TREE_CODE (decl) == FUNCTION_DECL
> > -	      && DECL_OMP_DECLARE_REDUCTION_P (decl))
> > -	  || undeduced_auto_decl (decl))
> >         && !DECL_DECLARED_CONCEPT_P (decl)
> >         && !uses_template_parms (DECL_TI_ARGS (decl)))
> >       {
> > @@ -5700,15 +5691,13 @@ mark_used (tree decl, tsubst_flags_t complain)
> >         return false;
> >       }
> >   -  /* Normally, we can wait until instantiation-time to synthesize DECL.
> > -     However, if DECL is a static data member initialized with a constant
> > -     or a constexpr function, we need it right now because a reference to
> > -     such a data member or a call to such function is not value-dependent.
> > -     For a function that uses auto in the return type, we need to
> > instantiate
> > -     it to find out its type.  For OpenMP user defined reductions, we need
> > -     them instantiated for reduction clauses which inline them by hand
> > -     directly.  */
> > -  maybe_instantiate_decl (decl);
> > +  /* If DECL has a deduced return type, we need to instantiate it now to
> > +     find out its type.  For OpenMP user defined reductions, we need them
> > +     instantiated for reduction clauses which inline them by hand directly.
> > */
> > +  if (undeduced_auto_decl (decl)
> > +      || (TREE_CODE (decl) == FUNCTION_DECL
> > +	  && DECL_OMP_DECLARE_REDUCTION_P (decl)))
> > +    maybe_instantiate_decl (decl);
> >       if (processing_template_decl || in_template_function ())
> >       return true;
> > diff --git a/gcc/testsuite/g++.dg/cpp1y/var-templ70.C
> > b/gcc/testsuite/g++.dg/cpp1y/var-templ70.C
> > new file mode 100644
> > index 00000000000..80965657c32
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/cpp1y/var-templ70.C
> > @@ -0,0 +1,19 @@
> > +// PR c++/99130
> > +// { dg-do compile { target c++14 } }
> > +
> > +template<class T>
> > +struct A {
> > +  static constexpr int value = T::value;
> > +};
> > +
> > +struct B {
> > +  template<class T>
> > +  static constexpr int value = T::value;
> > +};
> > +
> > +template<class T>
> > +constexpr int value = T::value;
> > +
> > +using ty1 = decltype(A<int>::value);
> > +using ty2 = decltype(B::value<int>);
> > +using ty3 = decltype(value<int>);
> 
> 


  reply	other threads:[~2022-09-07 20:40 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-09-07 19:41 Patrick Palka
2022-09-07 19:55 ` Jason Merrill
2022-09-07 20:40   ` Patrick Palka [this message]
2022-09-08 12:46     ` Jason Merrill

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=c7dc289e-6667-631f-ea1d-52a8671a81a6@idea \
    --to=ppalka@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=jason@redhat.com \
    /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).