This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug target/85756] [9 regression] wrong code at -Os on x86-64-linux-gnu in 32-bit mode


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85756

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
            Version|unknown                     |9.0
   Last reconfirmed|                            |2018-05-12
          Component|tree-optimization           |target
                 CC|                            |ebotcazou at gcc dot gnu.org,
                   |                            |jakub at gcc dot gnu.org
             Blocks|                            |85683
     Ever confirmed|0                           |1
            Summary|wrong code at -Os on        |[9 regression] wrong code
                   |x86-64-linux-gnu in 32-bit  |at -Os on x86-64-linux-gnu
                   |mode                        |in 32-bit mode
   Target Milestone|---                         |9.0
           Severity|normal                      |major

--- Comment #1 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
This was introduced by:

        PR target/85683
        * config/i386/i386.md: Add peepholes for mem {+,-,&,|,^}= x; mem != 0
        after cmpelim optimization.

--- pr85756_0.s 2018-05-12 12:23:29.214331432 +0200
+++ pr85756.s   2018-05-12 12:23:38.950611436 +0200
@@ -37,10 +37,9 @@
        testw   %ax, %ax
        movl    %edx, f
        setne   %al
-       subl    -44(%ebp), %esi
-       setne   %cl
-       movl    %esi, -44(%ebp)
+       subl    %esi, -44(%ebp)
        movl    %edx, %esi
+       setne   %cl
        orl     %ecx, %eax
        movzbl  %al, %eax
        xorl    %eax, %esi

The transformation swaps the operands of the subtraction.  It's also
responsible for the miscompilation of the Ada runtime (g-catiio.o) in 32-bit
mode.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85683
[Bug 85683] [8 Regression] GCC 8 stopped using RMW (Read Modify Write)
instructions on x86[_64]

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]