[Bug rtl-optimization/57468] New: [4.9 Regression] 26% performance drop on important benchmark after r199298.

[Bug rtl-optimization/57468] New: [4.9 Regression] 26% performance drop on important benchmark after r199298.

[ysrumyan at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57468

            Bug ID: 57468
           Summary: [4.9 Regression] 26% performance drop on important
                    benchmark after r199298.
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ysrumyan at gmail dot com

We found significant performance drop after changes in lra phase, which can be
demonstrated on example from 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55342

I will attach test-case.
And we can see that redundant spill/reload was generated for inner loop that
leads to performance degradation:

.L4:
    movl    %esi, %edx
    addl    $4, %eax
    subl    %edx, %ebx
    subl    %edx, %ecx
    movzbl    (%esp), %edx
    movb    %cl, -3(%eax)
    movl    %esi, %ecx
    movl    %esi, (%esp)  <<-- redundant spill
    movb    %bl, -4(%eax)
    movzbl    (%esp), %ebx  <<-- redundant fill
    subl    %ecx, %edx
    movb    %dl, -2(%eax)
    cmpl    %ebp, 4(%esp)
    movb    %bl, -1(%eax)
    je    .L1

[Bug rtl-optimization/57468] [4.9 Regression] 26% performance drop on important benchmark after r199298.

[ysrumyan at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57468

--- Comment #1 from Yuri Rumyantsev <ysrumyan at gmail dot com> ---
Created attachment 30224
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30224&action=edit
test-case to reproduce

It should be compiled on x86 with "-O2 -m32" options.

[Bug rtl-optimization/57468] [4.9 Regression] 26% performance drop on important benchmark after r199298.

[vmakarov at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57468

Vladimir Makarov <vmakarov at gcc dot gnu.org> changed:

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

--- Comment #2 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Thanks for reporting this.  I guess the secondary reload is generated when one
pseudo is spilled.  I'll be working on this.

[Bug rtl-optimization/57468] [4.9 Regression] 26% performance drop on important benchmark after r199298.

[izamyatin at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57468

--- Comment #3 from Igor Zamyatin <izamyatin at gmail dot com> ---
Patch is here http://gcc.gnu.org/ml/gcc-patches/2013-06/msg00357.html

[Bug rtl-optimization/57468] [4.9 Regression] 26% performance drop on important benchmark after r199298.

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57468

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Thus fixed.