[Bug other/53742] New: bad assembler output when compiling with LTO and PGO

[Bug other/53742] New: bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

             Bug #: 53742
           Summary: bad assembler output when compiling with LTO and PGO
    Classification: Unclassified
           Product: gcc
           Version: 4.7.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ioctl@yandex.ru


LANG=C c++ -c -v -fPIC  -fno-rtti -march=prescott -mtune=prescott -mfpmath=sse
-msse2 -msse3 -fstack-protector --param=ssp-buffer-size=4 -flto
-fuse-linker-plugin -O3 -freorder-blocks-and-partition -pthread -g
-fprofile-use -fprofile-correction -fomit-frame-pointer String.cpp

...


String.s: Assembler messages:
String.s:5667: Error: invalid operands (.text.hot and .text sections) for `-'
String.s:5669: Error: invalid operands (.text.hot and .text sections) for `-'


String.cpp is from the firefox 12.

Problem appears only when using profile information.

Full output with -v and -save-temps is too long, so, I will attach it.

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #1 from ioctl at yandex dot ru 2012-06-21 09:03:45 UTC ---
Created attachment 27676
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27676
Preprocessed source

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #2 from ioctl at yandex dot ru 2012-06-21 09:04:20 UTC ---
Created attachment 27677
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27677
Profile data

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #3 from ioctl at yandex dot ru 2012-06-21 09:05:36 UTC ---
Created attachment 27678
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27678
Full c++ console output

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #4 from ioctl at yandex dot ru 2012-06-21 09:07:18 UTC ---
Created attachment 27679
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27679
Assembler output that rejected by as

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-06-21 09:12:50 UTC ---
What version of binutils are you using?  Also are you using gold or GNU ld?

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #6 from ioctl at yandex dot ru 2012-06-21 09:22:36 UTC ---
GNU assembler version 2.22.0 (i686-pc-linux-gnu) using BFD version (GNU
Binutils) 2.22.0.20120323

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #7 from ioctl at yandex dot ru 2012-06-21 09:27:46 UTC ---
I have two ld: GNU and GOLD.
Problem appears before ld call, but with such gcc keys ld GOLD must be called.

It version is GNU gold (GNU Binutils 2.22.0.20120323) 1.11.

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-06-21 09:35:16 UTC ---
Surely not using -freorder-blocks-and-partition will fix this?  The actual
bug should not be LTO or PGO related.

[Bug other/53742] bad assembler output when compiling with LTO and PGO

[ioctl at yandex dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53742

--- Comment #9 from ioctl at yandex dot ru 2012-06-21 10:07:44 UTC ---
(In reply to comment #8)
> Surely not using -freorder-blocks-and-partition will fix this?

Yes, it fixes.

>The actual
> bug should not be LTO or PGO related.

Without profile data and with -freorder-blocks-and-partition assembler eats
file silently.

If to build ff without -flto on first cycle with -fprofile-generate, and then
build without -flto on the second pass, this problem also disappears.


I have builded ff before from the same sources with the same configuration, but
without LTO. And there was no such problem.


So, it seems problem appears in case when use profile data with key
-freorder-blocks-and-partition. At that, profile must be generated by programm
builded with LTO.