[Bug c++/107163] New: Compile time regression when using templated base classes, virtual method, and Wall

[Bug c++/107163] New: Compile time regression when using templated base classes, virtual method, and Wall

[cfsteefel at arista dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

            Bug ID: 107163
           Summary: Compile time regression when using templated base
                    classes, virtual method, and Wall
           Product: gcc
           Version: 11.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: cfsteefel at arista dot com
  Target Milestone: ---

When using templates to create a recursive base class structure, if the base
class uses a virtual method and -Wall is enabled, g++-11 shows much worse
compile time compared to g++-8.

Consider the following code:
# 0 "test.cpp"
# 0 "<built-in>"
# 0 "<command-line>"
# 1 "/usr/include/stdc-predef.h" 1 3 4
# 0 "<command-line>" 2
# 1 "test.cpp"


class BaseType  {
   virtual int doSomething() {
      return 1;
   }
};

template< int Seq >
class DerivedType : public DerivedType< Seq - 1 > {
};

template<>
class DerivedType< -1 > : public BaseType {
};

int main() {
   auto * d = new DerivedType< COUNT >();
   delete d;
   return 0;
}

Compiled as:
g++-11 test.cpp -Wall -DCOUNT=200
g++-8 test.cpp -Wall -DCOUNT=200

In g++-11, with COUNT=100, the time to compile is around 2 seconds, while with
COUNT=200, the time is around 1 minute, and COUNT=400 took around 28 minutes.

For comparison, compiling the same code with g++-8 took around .3 seconds, 1.6
seconds, and 22 seconds, respectively.

Removing the virtual will lead to the compile time being near instant again, as
will removing the Wall flag (with g++-11).

Other details:
Compiler versions:
g++ (GCC) 11.3.1 20220421 (Red Hat 11.3.1-2)
g++ (GCC) 8.4.0 20200304 (Red Hat 8.4.0-4)
System:
Centos7 based linux
Compiler output:
test.cpp: In function 'int main()':
test.cpp:20:4: warning: deleting object of polymorphic class type
'DerivedType<100>' which has non-virtual destructor might cause undefined
behavior [-Wdelete-non-virtual-dtor]
   20 |    delete d;
      |    ^~~~~~~~
(fixing the warning changes nothing, and seemed to double the compile time).

extra g++-11/8 configuration options:
%{O*:%{!fomit-frame-pointer:-fno-omit-frame-pointer}} \
%{O*:%{!fno-wrapv:-fwrapv}} \
%{O*:%{!freorder-blocks-and-partition:-fno-reorder-blocks-and-partition}} \
%{mfpmath=sse:-msse2} \
%{!fsemantic-interposition:-fno-semantic-interposition} \
%{g:%{!g0:-gdwarf-4}} \


Using godbolt, I can see similar timing for g++-10 and g++-8, so this seems to
be a g++-11 specific regression.

[Bug c++/107163] Compile time regression when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2022-10-05
            Summary|Compile time regression     |Compile time regression
                   |when using templated base   |when using templated base
                   |classes, virtual method,    |classes, virtual method,
                   |and Wall                    |and Wall since
                   |                            |r10-2823-g6a07489267e55084
     Ever confirmed|0                           |1
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |marxin at gcc dot gnu.org

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
With -std=c++17 likely started with r10-2823-g6a07489267e55084.

[Bug c++/107163] [10/11/12/13 Regression] huge Compile time when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |10.5
            Summary|Compile time regression     |[10/11/12/13 Regression]
                   |when using templated base   |huge Compile time when
                   |classes, virtual method,    |using templated base
                   |and Wall since              |classes, virtual method,
                   |r10-2823-g6a07489267e55084  |and Wall since
                   |                            |r10-2823-g6a07489267e55084

[Bug c++/107163] [10/11/12/13 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Yes, using -std=gnu++14 makes GCC 11 as fast as GCC 8.

[Bug c++/107163] [10/11/12/13 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug c++/107163] [10/11/12/13 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu.org
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jason at gcc dot gnu.org

--- Comment #3 from Jason Merrill <jason at gcc dot gnu.org> ---
This isn't specific to virtual functions; the same slowdown can be seen with

struct BaseType  {
  int i;
};

template< int Seq >
class DerivedType : public DerivedType< Seq - 1 > { };

template<>
class DerivedType< -1 > : public BaseType { };

int main() {
  DerivedType< COUNT > d;
  d.i = 42;
}

The reference to the base member is represented with a long sequence of
COMPONENT_REFs, like
d.derived<5>.derived<4>.derived<3>.derived<2>.derived<1>.derived<0>.derived<-1>.base.i
= 42

and with r10-2823 we check sequence points across each of those .s, with high
algorithmic complexity.  Checking across COMPONENT_REF seems unnecessary, since
it just selects a subobject of the LHS, there's no additional evaluation.

[Bug c++/107163] [10/11/12/13 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The trunk branch has been updated by Jason Merrill <jason@gcc.gnu.org>:

https://gcc.gnu.org/g:0e8fc610fb7112deb8c33c673a52983368dde9b7

commit r13-6910-g0e8fc610fb7112deb8c33c673a52983368dde9b7
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Mar 23 15:57:39 2023 -0400

    c-family: -Wsequence-point and COMPONENT_REF [PR107163]

    The patch for PR91415 fixed -Wsequence-point to treat shifts and ARRAY_REF
    as sequenced in C++17, and COMPONENT_REF as well.  But this is unnecessary
    for COMPONENT_REF, since the RHS is just a FIELD_DECL with no actual
    evaluation, and in this testcase handling COMPONENT_REF as sequenced blows
    up fast in a deep inheritance tree.  Instead, look through it.

            PR c++/107163

    gcc/c-family/ChangeLog:

            * c-common.cc (verify_tree): Don't use sequenced handling
            for COMPONENT_REF.

    gcc/testsuite/ChangeLog:

            * g++.dg/warn/Wsequence-point-5.C: New test.

[Bug c++/107163] [10/11/12 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Jason Merrill
<jason@gcc.gnu.org>:

https://gcc.gnu.org/g:dcfa9615062c46f3c45c70654ea8fa6719a5f40c

commit r12-9444-gdcfa9615062c46f3c45c70654ea8fa6719a5f40c
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Mar 23 15:57:39 2023 -0400

    c-family: -Wsequence-point and COMPONENT_REF [PR107163]

    The patch for PR91415 fixed -Wsequence-point to treat shifts and ARRAY_REF
    as sequenced in C++17, and COMPONENT_REF as well.  But this is unnecessary
    for COMPONENT_REF, since the RHS is just a FIELD_DECL with no actual
    evaluation, and in this testcase handling COMPONENT_REF as sequenced blows
    up fast in a deep inheritance tree.  Instead, look through it.

            PR c++/107163

    gcc/c-family/ChangeLog:

            * c-common.cc (verify_tree): Don't use sequenced handling
            for COMPONENT_REF.

    gcc/testsuite/ChangeLog:

            * g++.dg/warn/Wsequence-point-5.C: New test.

[Bug c++/107163] [10/11 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

--- Comment #6 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Jason Merrill
<jason@gcc.gnu.org>:

https://gcc.gnu.org/g:e81e393cd864abcd2de02602bd51e435dc28f418

commit r10-11307-ge81e393cd864abcd2de02602bd51e435dc28f418
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Mar 23 15:57:39 2023 -0400

    c-family: -Wsequence-point and COMPONENT_REF [PR107163]

    The patch for PR91415 fixed -Wsequence-point to treat shifts and ARRAY_REF
    as sequenced in C++17, and COMPONENT_REF as well.  But this is unnecessary
    for COMPONENT_REF, since the RHS is just a FIELD_DECL with no actual
    evaluation, and in this testcase handling COMPONENT_REF as sequenced blows
    up fast in a deep inheritance tree.  Instead, look through it.

            PR c++/107163

    gcc/c-family/ChangeLog:

            * c-common.c (verify_tree): Don't use sequenced handling
            for COMPONENT_REF.

    gcc/testsuite/ChangeLog:

            * g++.dg/warn/Wsequence-point-5.C: New test.

[Bug c++/107163] [10/11 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jason Merrill
<jason@gcc.gnu.org>:

https://gcc.gnu.org/g:9ab083eb8a96b7f8baf6fe632d03aa496017e456

commit r11-10647-g9ab083eb8a96b7f8baf6fe632d03aa496017e456
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Mar 23 15:57:39 2023 -0400

    c-family: -Wsequence-point and COMPONENT_REF [PR107163]

    The patch for PR91415 fixed -Wsequence-point to treat shifts and ARRAY_REF
    as sequenced in C++17, and COMPONENT_REF as well.  But this is unnecessary
    for COMPONENT_REF, since the RHS is just a FIELD_DECL with no actual
    evaluation, and in this testcase handling COMPONENT_REF as sequenced blows
    up fast in a deep inheritance tree.  Instead, look through it.

            PR c++/107163

    gcc/c-family/ChangeLog:

            * c-common.c (verify_tree): Don't use sequenced handling
            for COMPONENT_REF.

    gcc/testsuite/ChangeLog:

            * g++.dg/warn/Wsequence-point-5.C: New test.

[Bug c++/107163] [11 Regression] huge Compile time increase when using templated base classes, virtual method, and Wall since r10-2823-g6a07489267e55084

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107163

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.