[Bug ipa/98265] New: gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[Bug ipa/98265] New: gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[kartikmohta at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

            Bug ID: 98265
           Summary: gcc-10 has significantly worse code generated with -O2
                    compared to -O1 (or gcc-9 -O2) when using the Eigen
                    C++ library
           Product: gcc
           Version: 10.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: kartikmohta at gmail dot com
                CC: marxin at gcc dot gnu.org
  Target Milestone: ---

I was checking the generated assembly for the following simple code using the
Eigen C++ library:

#include <Eigen/Core>

Eigen::Matrix<float, 3, 1> f(float x, float y, float z, float scale)
{
  return Eigen::Matrix<float, 3, 1>(x, y, z) * scale;
}

The Eigen::Matrix structure can be thought of as just a wrapper around an array
in this case. The preprocessed file is attached.


When compiled with -O1 optimization, this generates the expected reasonable
looking code, but when using -O2 it looks like the generated code is extremely
bad. The code generated by gcc-9.3 with -O2 also looks good. A quick comparison
of the code generated by gcc 10.2 vs gcc 9.3 can be seen at
https://godbolt.org/z/186c19.


Upon bisecting this, it seems like the offending changes are from the pair of
commits which changed some of the inliner params:

commit 1e83bd7003e03160b7d71fb959111e89b53446ab
Author: Jan Hubicka <hubicka@gcc.gnu.org>
Date:   Sat Nov 23 05:13:23 2019

    Convert inliner to new param infrastructure

commit 9340d34599e6d5e7a6f3614de44b2c578b180c1b
Author: Jan Hubicka <hubicka@ucw.cz>
Date:   Sat Nov 23 05:11:25 2019

    Convert inliner to function specific param infrastructure


By playing with the optimization parameters, I saw that using either
-fno-partial-inlining or --param early-inlining-insns=14 fixes the generated
code with -O2.

Upon further investigation, it looks like earlier there were two separate
params for early-inlining-insns for -O3 and -O2 but in the consider_split
function (ipa-split.c) only the value for -O3 was used irrespective of the
optimization level. This code was not changed when the inliner params changes
were made leading to a much smaller value of param_early_inlining_insns being
used in consider_split for -O2 now, which may be causing problems with inlining
in this case.

[Bug ipa/98265] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[kartikmohta at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #1 from Kartik Mohta <kartikmohta at gmail dot com> ---
Created attachment 49755
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49755&action=edit
preprocessed file

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|gcc-10 has significantly    |[10/11 Regression] gcc-10
                   |worse code generated with   |has significantly worse
                   |-O2 compared to -O1 (or     |code generated with -O2
                   |gcc-9 -O2) when using the   |compared to -O1 (or gcc-9
                   |Eigen C++ library           |-O2) when using the Eigen
                   |                            |C++ library
                 CC|                            |hubicka at gcc dot gnu.org
   Target Milestone|---                         |10.3
           Keywords|                            |missed-optimization

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
The TU might be too small so we run into inline limits too quickly?

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[kartikmohta at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #3 from Kartik Mohta <kartikmohta at gmail dot com> ---
This is a simple example to demonstrate the problem I've noticed in a bigger
program. Do the inlining limits depend on the size of the TU?

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
Created attachment 49947
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49947&action=edit
Reduced test-case

I reduced a test-case where GCC 10 does not inline all in fn called
'should_inline'.

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-01-12
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |hubicka at gcc dot gnu.org

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

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

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
We do not inline CwiseNullaryOp because it uses comdat local symbols.
This is because we do split the function and the .part stays local.

At least we should recompute if function calls comdat local after comdat local
function is inlined.

IPA function summary for Eigen::Matrix<float, 3, 1> should_inline(float, float,
float, float)/2 inlinable fp_expression
  global time:     36.000000
  self size:       21
  global size:     29
  min size:       24
  self stack:      16
  global stack:    20
    size:19.000000, time:18.500000
    size:4.500000, time:3.500000,  executed if:(not inlined)
  calls:
    operator*.isra/90 inlined
      freq:1.00
      Stack frame offset 16, callee self size 4
      Eigen::CwiseNullaryOp< <template-parameter-1-1>, <template-parameter-1-2>
>::CwiseNullaryOp(long int, long int, Eigen::scalar_constant_op<float>) [with
<template-parameter-1-1> = Eigen::scalar_constant_op<float>; PlainObjectType =
Eigen::Matrix<float, 4, 5, 6, 4, 5>]/20 callee refers to comdat-local symbols
        freq:1.00 loop depth: 0 size: 5 time: 14 callee size: 6 stack: 0
         op0 is compile time invariant
         op0 points to local or readonly memory
         op1 is compile time invariant
         op2 is compile time invariant
         op2 points to local or readonly memory

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #6 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
I am testing
diff --git a/gcc/cif-code.def b/gcc/cif-code.def
index 2f430cf1c39..39b89da155f 100644
--- a/gcc/cif-code.def
+++ b/gcc/cif-code.def
@@ -125,7 +125,7 @@ DEFCIFCODE(OPTIMIZATION_MISMATCH, CIF_FINAL_ERROR,
           N_("optimization level attribute mismatch"))

 /* We can't inline because the callee refers to comdat-local symbols.  */
-DEFCIFCODE(USES_COMDAT_LOCAL, CIF_FINAL_ERROR,
+DEFCIFCODE(USES_COMDAT_LOCAL, CIF_FINAL_NORMAL,
           N_("callee refers to comdat-local symbols"))

 /* We can't inline because of mismatched caller/callee

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jan Hubicka <hubicka@gcc.gnu.org>:

https://gcc.gnu.org/g:e7fd3b783238d034018443e43a58ff87908b4db6

commit r11-7940-ge7fd3b783238d034018443e43a58ff87908b4db6
Author: Jan Hubicka <jh@suse.cz>
Date:   Wed Mar 31 22:44:20 2021 +0200

    Make USES_COMDAT_LOCAL CIF_FINAL_NORMAL

    USES_COMDAT_LOCAL is incorrectly defined as CIF_FINAL_ERROR which makes
inliner
    to mis some inlines of functions in comdat section that was previously
split.

    2021-03-31  Jan Hubicka  <hubicka@ucw.cz>

            PR ipa/98265
            * cif-code.def (USES_COMDAT_LOCAL): Make CIF_FINAL_NORMAL.

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Jan Hubicka
<hubicka@gcc.gnu.org>:

https://gcc.gnu.org/g:42c22a4d724b4a4b0183f4412c3d42c9cca29d30

commit r10-9646-g42c22a4d724b4a4b0183f4412c3d42c9cca29d30
Author: Jan Hubicka <jh@suse.cz>
Date:   Wed Mar 31 22:44:20 2021 +0200

    Make USES_COMDAT_LOCAL CIF_FINAL_NORMAL

    USES_COMDAT_LOCAL is incorrectly defined as CIF_FINAL_ERROR which makes
inliner
    to mis some inlines of functions in comdat section that was previously
split.

    2021-03-31  Jan Hubicka  <hubicka@ucw.cz>

            PR ipa/98265
            * cif-code.def (USES_COMDAT_LOCAL): Make CIF_FINAL_NORMAL.

    (cherry picked from commit e7fd3b783238d034018443e43a58ff87908b4db6)

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jan Hubicka <hubicka@gcc.gnu.org>:

https://gcc.gnu.org/g:3064fc21aa29d8e04b23c0b52dc4f67de1da6b2f

commit r11-7948-g3064fc21aa29d8e04b23c0b52dc4f67de1da6b2f
Author: Jan Hubicka <jh@suse.cz>
Date:   Thu Apr 1 12:11:39 2021 +0200

    Add testcase for PR98265

    gcc/testsuite/ChangeLog:

    2021-04-01  Jan Hubicka  <hubicka@ucw.cz>

            PR ipa/98265
            * gcc.dg/tree-ssa/pr98265.C: New test.

[Bug ipa/98265] [10/11 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #10 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Trunk now generates on the unreduced testcase:
        .file   "test.cpp"
        .text
        .p2align 4
        .globl  _Z1fffff
        .type   _Z1fffff, @function
_Z1fffff:
.LFB6287:
        .cfi_startproc
        mulss   %xmm3, %xmm0
        movq    %rdi, %rax
        mulss   %xmm3, %xmm1
        mulss   %xmm3, %xmm2
        movss   %xmm0, (%rdi)
        movss   %xmm1, 4(%rdi)
        movss   %xmm2, 8(%rdi)
        ret
        .cfi_endproc
.LFE6287:
        .size   _Z1fffff, .-_Z1fffff
        .ident  "GCC: (GNU) 11.0.1 20210331 (experimental)"
        .section        .note.GNU-stack,"",@progbits

[Bug ipa/98265] [10 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[10/11 Regression] gcc-10   |[10 Regression] gcc-10 has
                   |has significantly worse     |significantly worse code
                   |code generated with -O2     |generated with -O2 compared
                   |compared to -O1 (or gcc-9   |to -O1 (or gcc-9 -O2) when
                   |-O2) when using the Eigen   |using the Eigen C++ library
                   |C++ library                 |
             Status|RESOLVED                    |ASSIGNED
      Known to work|                            |11.0
         Resolution|FIXED                       |---

[Bug ipa/98265] [10 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
Indeed fixed everywhere.

[Bug ipa/98265] [10 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265
Bug 98265 depends on bug 99989, which changed state.

Bug 99989 Summary: [11 regression] -Wmaybe-uninitialized warning breaks bootstrap on riscv64
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99989

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED

[Bug ipa/98265] [10 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |cuzdav at gmail dot com

--- Comment #12 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
*** Bug 80726 has been marked as a duplicate of this bug. ***

[Bug ipa/98265] [10 Regression] gcc-10 has significantly worse code generated with -O2 compared to -O1 (or gcc-9 -O2) when using the Eigen C++ library

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98265

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tobi at gcc dot gnu.org

--- Comment #13 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 95859 has been marked as a duplicate of this bug. ***