[Bug tree-optimization/63302] New: [4.9,5.0 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[Bug tree-optimization/63302] New: [4.9,5.0 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

            Bug ID: 63302
           Summary: [4.9,5.0 Regression] Code with 64-bit long long
                    constants is miscompiled on 32-bit host
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: danglin at gcc dot gnu.org
              Host: hppa-unknown-linux-gnu
            Target: hppa-unknown-linux-gnu
             Build: hppa-unknown-linux-gnu

The following function from pa.c is miscompiled when generating a hppa64
cross compiler:

dave@mx3210:~/hardware/ldil$ cat ldil.c
#define HOST_WIDE_INT long long
int
pa_ldil_cint_p (HOST_WIDE_INT ival)
{
 HOST_WIDE_INT x = ival & (((HOST_WIDE_INT) -1 << 31) | 0x7ff);

 return x == 0 || x == ((HOST_WIDE_INT) -1 << 31);
}

gcc-4.9 -S -O1 ldil.c:

dave@mx3210:~/hardware/ldil$ cat ldil.s
    .LEVEL 1.1
    .text
    .align 4
.globl pa_ldil_cint_p
    .type    pa_ldil_cint_p, @function
pa_ldil_cint_p:
    .PROC
    .CALLINFO FRAME=0,NO_CALLS
    .ENTRY
    extru %r26,31,11,%r28
    comiclr,<> 0,%r28,%r28
    ldi 1,%r28
    bv,n %r0(%r2)
    .EXIT
    .PROCEND
    .size    pa_ldil_cint_p, .-pa_ldil_cint_p
    .ident    "GCC: (Debian 4.9.1-14) 4.9.1"

The high order bits are not checked in the assembly code.  We have the
following in ldil.c.170r.expand:

pa_ldil_cint_p (long long int ival)
{
  int _6;
  long long int _8;
  _Bool _9;

;;   basic block 2, loop depth 0
;;    pred:       ENTRY
  _8 = ival_1(D) & 2047;
  _9 = _8 == 0;
  _6 = (int) _9;
  return _6;
;;    succ:       EXIT

}

This breaks 64-bit kernel build.  4.8 branch is ok.

[Bug tree-optimization/63302] [4.9,5.0 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #1 from John David Anglin <danglin at gcc dot gnu.org> ---
This is with:

dave@mx3210:~/debian/gcc/gcc/gcc-4.9-4.9.1$ gcc-4.9 -v
Using built-in specs.
COLLECT_GCC=/usr/bin/gcc-4.9.real
COLLECT_LTO_WRAPPER=/usr/lib/gcc/hppa-linux-gnu/4.9/lto-wrapper
Target: hppa-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian 4.9.1-14+b1'
--with-bugurl=file:///usr/share/doc/gcc-4.9/README.Bugs
--enable-languages=c,c++,java,d,fortran,objc,obj-c++ --prefix=/usr
--program-suffix=-4.9 --enable-shared --enable-linker-build-id
--libexecdir=/usr/lib --without-included-gettext --enable-threads=posix
--with-gxx-include-dir=/usr/include/c++/4.9 --libdir=/usr/lib --enable-nls
--with-sysroot=/ --enable-clocale=gnu --enable-libstdcxx-debug
--enable-libstdcxx-time=yes --enable-gnu-unique-object --disable-libssp
--disable-libitm --disable-libsanitizer --disable-libquadmath --enable-plugin
--with-system-zlib --disable-browser-plugin --enable-java-awt=gtk
--enable-gtk-cairo --with-java-home=/usr/lib/jvm/java-1.5.0-gcj-4.9-hppa/jre
--enable-java-home --with-jvm-root-dir=/usr/lib/jvm/java-1.5.0-gcj-4.9-hppa
--with-jvm-jar-dir=/usr/lib/jvm-exports/java-1.5.0-gcj-4.9-hppa
--with-arch-directory=parisc --with-ecj-jar=/usr/share/java/eclipse-ecj.jar
--enable-objc-gc --enable-multiarch --disable-libstdcxx-pch
--enable-checking=release --build=hppa-linux-gnu --host=hppa-linux-gnu
--target=hppa-linux-gnu
Thread model: posix
gcc version 4.9.1 (Debian 4.9.1-14+b1)

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

John David Anglin <danglin at gcc dot gnu.org> changed:

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

--- Comment #2 from John David Anglin <danglin at gcc dot gnu.org> ---
Introduced in r203627:

2013-10-15  Zhenqiang Chen  <zhenqiang.chen@arm.com>

        * tree-ssa-reassoc.c: Include rtl.h and tm_p.h.
        (optimize_range_tests_1): New function,
        extracted from optimize_range_tests.
        (optimize_range_tests_xor): Similarly.
        (optimize_range_tests_diff): New function.
        (optimize_range_tests): Use optimize_range_tests_1.

It was fixed on trunk in wide-int merge but it still is present on 4.9
branch.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com

--- Comment #3 from Jeffrey A. Law <law at redhat dot com> ---
Ugh.  Clearly merging the wide-int back to 4.9 wouldn't make sense.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #4 from Jeffrey A. Law <law at redhat dot com> ---
Have you tried with the current 4.9.2 bits?  I get the following with a 32bit
host cross to hppa64-unknown-linux-gnu:



       depdi 0,52,20,%r26
        depdi,z 1,32,4,%r28
        add,l %r26,%r28,%r28
        depdi 0,32,1,%r28
        or,*>= %r0,%r28,%r28
        subi 0,%r28,%r28
        ldo -1(%r28),%r28
        bve (%r2)
        extrd,u %r28,0,1,%r28


Which is identical to what I get on the trunk.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #5 from dave.anglin at bell dot net ---
On 26-Sep-14, at 3:46 PM, law at redhat dot com wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302
>
> --- Comment #4 from Jeffrey A. Law <law at redhat dot com> ---
> Have you tried with the current 4.9.2 bits?  I get the following  
> with a 32bit
> host cross to hppa64-unknown-linux-gnu:


The wrong code is actually generated by the native 32-bit hppa-unknown- 
linux-gnu compiler.

This results in the hppa64-unknown-linux-gnu cross compiler being  
miscompiled and incorrect
ldil instructions in its output.  The hppa64-unknown-linux-gnu cross  
compiler correctly compiles
the test case.

In my regression search, I got wrong code with r215536 from the 4.9  
branch.  Then,
I switched to checking trunk.  I just checked r215649 from 4.9 branch  
and still get the following
incorrect code for the testcase:

         extru %r26,31,11,%r28
         comiclr,<> 0,%r28,%r28
         ldi 1,%r28
         bv,n %r0(%r2)

Note that this is 32-bit code.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[zhenqiang.chen at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #6 from Zhenqiang Chen <zhenqiang.chen at arm dot com> ---
I double checked the function optimize_range_tests_diff. Overall, I think it
does the right thing. X86 and ARM work correctly. The ldil.c.169t.optimized is

  <bb 2>:
  x_2 = ival_1(D) & -2147481601;
  _8 = x_2 + 2147483648;
  _9 = _8 & -2147483649;
  _10 = _9 == 0;
  _6 = (int) _10;
  return _6;

If we can not fix the wide-int issue, I will create a patch to workaround it
for 4.9 since I never expected optimize_range_tests_diff can work between a
positive value and a negative value.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #7 from dave.anglin at bell dot net ---
On 28-Sep-14, at 1:30 AM, zhenqiang.chen at arm dot com wrote:

> I double checked the function optimize_range_tests_diff. Overall, I  
> think it
> does the right thing. X86 and ARM work correctly. The ldil.c. 
> 169t.optimized is
>
>  <bb 2>:
>  x_2 = ival_1(D) & -2147481601;
>  _8 = x_2 + 2147483648;
>  _9 = _8 & -2147483649;
>  _10 = _9 == 0;
>  _6 = (int) _10;
>  return _6;


This is what I see on the trunk, but 4.9 is wrong.  Possibly, there is  
a transformation
after optimize_range_tests_diff where things go wrong on 4.9.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #8 from dave.anglin at bell dot net ---
On 28-Sep-14, at 10:34 AM, dave.anglin at bell dot net wrote:

> This is what I see on the trunk, but 4.9 is wrong.  Possibly, there is
> a transformation
> after optimize_range_tests_diff where things go wrong on 4.9.

It seems the difference actually occurs in reassoc1.

4.9:

   <bb 2>:
   x_2 = ival_1(D) & -2147481601;
   _3 = x_2 == 0;
   _8 = x_2 & 2147483647;
   _9 = _8 == 0;
   _4 = x_2 == -2147483648;
   _5 = _9;
   _6 = (int) _5;
   return _6;

trunk:

   <bb 2>:
   x_2 = ival_1(D) & -2147481601;
   _3 = x_2 == 0;
   _8 = x_2 + 2147483648;
   _9 = _8 & -2147483649;
   _10 = _9 == 0;
   _4 = x_2 == -2147483648;
   _5 = _10;
   _6 = (int) _5;
   return _6;

The preceding dse1 is same for both branches:

   <bb 2>:
   x_2 = ival_1(D) & -2147481601;
   _3 = x_2 == 0;
   _4 = x_2 == -2147483648;
   _5 = _3 | _4;
   _6 = (int) _5;
   return _6;

--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #9 from dave.anglin at bell dot net ---
On 28-Sep-14, at 1:30 AM, zhenqiang.chen at arm dot com wrote:

> X86 and ARM work correctly.

I suspect this is because both have "need_64bit_hwint=yes" in  
config.gcc.

--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[zhenqiang.chen at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #10 from Zhenqiang Chen <zhenqiang.chen at arm dot com> ---
(In reply to dave.anglin from comment #8)
> On 28-Sep-14, at 10:34 AM, dave.anglin at bell dot net wrote:
> 
> > This is what I see on the trunk, but 4.9 is wrong.  Possibly, there is
> > a transformation
> > after optimize_range_tests_diff where things go wrong on 4.9.
> 
> It seems the difference actually occurs in reassoc1.
> 
> 4.9:
> 
>    <bb 2>:
>    x_2 = ival_1(D) & -2147481601;
>    _3 = x_2 == 0;
>    _8 = x_2 & 2147483647;
>    _9 = _8 == 0;
>    _4 = x_2 == -2147483648;
>    _5 = _9;
>    _6 = (int) _5;
>    return _6;

This is not optimized by my patch: optimize_range_tests_diff

> trunk:
> 
>    <bb 2>:
>    x_2 = ival_1(D) & -2147481601;
>    _3 = x_2 == 0;
>    _8 = x_2 + 2147483648;
>    _9 = _8 & -2147483649;
>    _10 = _9 == 0;
>    _4 = x_2 == -2147483648;
>    _5 = _10;
>    _6 = (int) _5;
>    return _6;

This is optimized by the patch. optimize_range_tests_diff is the only change I
added for the patch. Others were just reorganization.

Can you show more detail dumps with -fdump-tree-reassoc1-details?

Thanks!
-Zhenqiang

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #11 from dave.anglin at bell dot net ---
On 28-Sep-14, at 9:17 PM, zhenqiang.chen at arm dot com wrote:

> Can you show more detail dumps with -fdump-tree-reassoc1-details?

Attached is 4.9 dump with more details.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #12 from dave.anglin at bell dot net ---
On 28-Sep-14, at 9:17 PM, zhenqiang.chen at arm dot com wrote:

> Can you show more detail dumps with -fdump-tree-reassoc1-details?

Same for trunk.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[zhenqiang.chen at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #13 from Zhenqiang Chen <zhenqiang.chen at arm dot com> ---
For 4.9, some function optimizes the code as:

Optimizing range tests x_2 -[-2147483648, -2147483648] and -[0, 0]
 into (x_2 & 2147483647) != 0

For trunk, optimize_range_tests_diff optimizes the code as:

Optimizing range tests x_2 -[-2147483648, -2147483648] and -[0, 0]
 into (x_2 + 2147483648 & -2147483649) != 0

I tried to remove need_64bit_hwint=yes" from config.gcc to build i686. But it
does not build. So still need your help to idnetify the root cause.

Can you try to remove the line in tree-ssa-reassoc.c?

  any_changes |= optimize_range_tests_1 (opcode, first, length, true,
                                         ops, ranges);
And add a printf at the place:

  printf ("changes: %d\n", any_changes);

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[zhenqiang.chen at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #14 from Zhenqiang Chen <zhenqiang.chen at arm dot com> ---
Created attachment 33608
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33608&action=edit
patch

After investigation, I found I mis-use tree_log2.

Please try the attached patch. If it works, I will run all tests and send it
for community review.

Thanks!

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #15 from dave.anglin at bell dot net ---
On 29-Sep-14, at 2:43 AM, zhenqiang.chen at arm dot com wrote:

> Please try the attached patch. If it works, I will run all tests and  
> send it
> for community review.

The patch appears to fix the testcase.  I've started a full build and  
check
with 4.9 branch.  I'll also test it on hpux starting this evening.

Thanks very much for investigating.

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #16 from dave.anglin at bell dot net ---
On 9/29/2014 9:02 AM, dave.anglin at bell dot net wrote:
> I've started a full build and
> check
> with 4.9 branch.  I'll also test it on hpux starting this evening.
I see no regressions with the change on 4.9 and trunk.  Tested on 
hppa2.0w-hp-hpux11.11,
hppa64-hp-hpux11.11 and hppa-unknown-linux-gnu on 4.9.  Tested on 
hppa2.0w-hp-hpux11.11
and hppa64-hp-hpux1.11 on trunk.

Dave

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

Mikael Pettersson <mikpelinux at gmail dot com> changed:

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

--- Comment #17 from Mikael Pettersson <mikpelinux at gmail dot com> ---
This bug also affects m68k-linux, since it doesn't set need_64bit_hwint=yes.

On m68k-linux, gcc-4.8 -O1 -fomit-frame-pointer generates:

pa_ldil_cint_p:
        move.l %d2,-(%sp)
        move.l 8(%sp),%d0
        move.l 12(%sp),%d1
        and.l #-2147481601,%d1
        move.l %d0,%d2
        or.l %d1,%d2
        jeq .L3
        move.w #-1,%a0
        move.l #-2147483648,%a1
        cmp.l %a0,%d0
        jne .L4
        cmp.l %a1,%d1
.L4:
        seq %d0
        extb.l %d0
        neg.l %d0
        jra .L2
.L3:
        moveq #1,%d0
.L2:
        move.l (%sp)+,%d2
        rts

while gcc-4.9 generates:

pa_ldil_cint_p:
        move.l 8(%sp),%d0
        and.l #2047,%d0
        seq %d0
        neg.b %d0
        and.l #255,%d0
        rts

The gcc-4.9 code completely ignores the high 32-bits.

I'll try the patch in a couple of days when my current 4.9 bootstrap+regtest is
done.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-10-04
     Ever confirmed|0                           |1

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #18 from John David Anglin <danglin at gcc dot gnu.org> ---
Hi Zhenqiang,

Do you plan to submit patch to gcc-patches soon?

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

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

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

--- Comment #20 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 33722
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33722&action=edit
gcc49-pr63302.patch

IMHO better fix, we already have the right predicate, it doesn't make sense to
add another one.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #21 from Jeffrey A. Law <law at redhat dot com> ---
John, I know those PA boxes are slow, but if you could try Jakub's patch, it'd
be appreciated.

BTW, are you using strictly PA 2.0 (PA8000+) hardware?  I was pondering doing a
reimplementation of pa_combine_instructions using Bin's infrastructure for insn
fusion as part of the review process for that code and having a box where I
could do performance testing would be helpful.

I can certainly test on PA8000+ hardware (assuming mine doesn't overheat), but
fmpyadd/fmpysub tend to degrade performance on the more modern hardware.  I
guess I could look for a performance degradation and call that a victory since
that would indicate we're doing a better job at fusing the independent FP mult
and add/sub instructions.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #22 from dave.anglin at bell dot net ---
On 10/15/2014 12:19 PM, law at redhat dot com wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302
>
> --- Comment #21 from Jeffrey A. Law <law at redhat dot com> ---
> John, I know those PA boxes are slow, but if you could try Jakub's patch, it'd
> be appreciated.
I started a build and check on hppa2.0w-hp-hpux11.11.  Will take a day 
or so for full
build and check.  Jakub's patch appears to fix ldil.c testcase.
>
> BTW, are you using strictly PA 2.0 (PA8000+) hardware?  I was pondering doing a
> reimplementation of pa_combine_instructions using Bin's infrastructure for insn
> fusion as part of the review process for that code and having a box where I
> could do performance testing would be helpful.
Almost all my testing is on PA 2.0 hardware.  I have a c8000, rp3440 and 
c3750 at home.
Mostly use the rp3440.

Probably, the best machine to use for development is sibaris (4-way 1 
GHz c8000 linux).  It's
our primary buildd machine but others have used it for development 
testing.  I'll check with Helge Deller
who manages the machine and get back.
>
> I can certainly test on PA8000+ hardware (assuming mine doesn't overheat), but
> fmpyadd/fmpysub tend to degrade performance on the more modern hardware.  I
> guess I could look for a performance degradation and call that a victory since
> that would indicate we're doing a better job at fusing the independent FP mult
> and add/sub instructions.
>

Dave

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #23 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Fri Oct 17 10:50:16 2014
New Revision: 216391

URL: https://gcc.gnu.org/viewcvs?rev=216391&root=gcc&view=rev
Log:
    PR tree-optimization/63302
    * tree-ssa-reassoc.c (optimize_range_tests_xor,
    optimize_range_tests_diff): Use !integer_pow2p () instead of
    tree_log2 () < 0.

    * gcc.c-torture/execute/pr63302.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr63302.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-reassoc.c

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #24 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Fri Oct 17 10:53:36 2014
New Revision: 216392

URL: https://gcc.gnu.org/viewcvs?rev=216392&root=gcc&view=rev
Log:
    PR tree-optimization/63302
    * tree-ssa-reassoc.c (optimize_range_tests_xor,
    optimize_range_tests_diff): Use !integer_pow2p () instead of
    tree_log2 () < 0.

    * gcc.c-torture/execute/pr63302.c: New test.

Added:
    branches/gcc-4_9-branch/gcc/testsuite/gcc.c-torture/execute/pr63302.c
Modified:
    branches/gcc-4_9-branch/gcc/ChangeLog
    branches/gcc-4_9-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_9-branch/gcc/tree-ssa-reassoc.c

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

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

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

--- Comment #25 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Should be fixed now.

[Bug tree-optimization/63302] [4.9 Regression] Code with 64-bit long long constants is miscompiled on 32-bit host

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63302

--- Comment #26 from dave.anglin at bell dot net ---
On 17-Oct-14, at 7:10 AM, jakub at gcc dot gnu.org wrote:

> Should be fixed now.

Thanks, Jakub.

Dave
--
John David Anglin    dave.anglin@bell.net