[Bug c/61108] New: gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[Bug c/61108] New: gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[corydoras at ridiculousfish dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

            Bug ID: 61108
           Summary: gcc 4.8.2 generates incorrect integer arithmetic at O3
                    (ok at O2)
           Product: gcc
           Version: 4.8.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: corydoras at ridiculousfish dot com

(My first time filing a gcc bug, apologies if I screw it up)

I have a loop with a high multiply and some arithmetic. gcc generates code that
computes the wrong value as the result, under -O3 on x86-64. It computes the
correct value under -O2. The key difference appears to be that the loop is
vectorized under -O3.

To reproduce: compile the following as 'gcc -O3 test.c' When run, it will
abort. Now compile as 'gcc -O2 test.c' and it will not abort.

#include <stdlib.h>
#include <stdint.h>

#define ITERS 8

__attribute__((noinline))
uint32_t sum(const uint32_t *vals)
{
    uint32_t result = 0;
    int i;
    for (i=0; i < ITERS; i++)
    {
        uint32_t val = vals[i];
        uint32_t q = (val * 613566757LLU) >> 32;
        uint32_t t = ((val + q) >> 1);
        result += t;
    }
    return result;
}

int main(void)
{
    uint32_t vals[ITERS] = {};
    vals[1] = 7;
    if (sum(vals) != 4)
        abort();
    return 0;
}





> gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/4.8/lto-wrapper
Target: x86_64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Ubuntu 4.8.2-19ubuntu1'
--with-bugurl=file:///usr/share/doc/gcc-4.8/README.Bugs
--enable-languages=c,c++,java,go,d,fortran,objc,obj-c++ --prefix=/usr
--program-suffix=-4.8 --enable-shared --enable-linker-build-id
--libexecdir=/usr/lib --without-included-gettext --enable-threads=posix
--with-gxx-include-dir=/usr/include/c++/4.8 --libdir=/usr/lib --enable-nls
--with-sysroot=/ --enable-clocale=gnu --enable-libstdcxx-debug
--enable-libstdcxx-time=yes --enable-gnu-unique-object --disable-libmudflap
--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-4.8-amd64/jre --enable-java-home
--with-jvm-root-dir=/usr/lib/jvm/java-1.5.0-gcj-4.8-amd64
--with-jvm-jar-dir=/usr/lib/jvm-exports/java-1.5.0-gcj-4.8-amd64
--with-arch-directory=amd64 --with-ecj-jar=/usr/share/java/eclipse-ecj.jar
--enable-objc-gc --enable-multiarch --disable-werror --with-arch-32=i686
--with-abi=m64 --with-multilib-list=m32,m64,mx32 --with-tune=generic
--enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu
--target=x86_64-linux-gnu
Thread model: posix
gcc version 4.8.2 (Ubuntu 4.8.2-19ubuntu1)

[Bug tree-optimization/61108] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The tree level looks fine at least to me:
  vect_val_8.15_127 = MEM[(const uint32_t *)vectp_vals.14_85];
  vect_patt_47.16_128 = WIDEN_MULT_EVEN_EXPR <vect_val_8.15_127, { 613566757,
613566757, 613566757, 613566757 }>;
  vect_patt_47.16_129 = WIDEN_MULT_ODD_EXPR <vect_val_8.15_127, { 613566757,
613566757, 613566757, 613566757 }>;
  vect__11.18_130 = vect_patt_47.16_128 >> 32;
  vect__11.18_131 = vect_patt_47.16_129 >> 32;
  vect_q_12.19_132 = VEC_PACK_TRUNC_EXPR <vect__11.18_130, vect__11.18_131>;
  vect__13.20_133 = vect_q_12.19_132 + vect_val_8.15_127;
  vect_t_14.21_134 = vect__13.20_133 >> 1;

But the assembly code looks broken:
    leaq    (%rdi,%r8,4), %r8
    movdqa    .LC0(%rip), %xmm1
    cmpl    $2, %r10d
    movdqa    (%r8), %xmm2
    movdqa    %xmm2, %xmm0
    movdqa    %xmm2, %xmm3
    pmuludq    %xmm1, %xmm0
    psrlq    $32, %xmm0
    psrlq    $32, %xmm3 <---- where did this come from?
    pmuludq    %xmm1, %xmm3
    psrlq    $32, %xmm3
    shufps    $136, %xmm3, %xmm0
    paddd    %xmm2, %xmm0
    psrld    $1, %xmm0

[Bug target/61108] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-05-08
          Component|tree-optimization           |target
     Ever confirmed|0                           |1

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The problem looks related to WIDEN_MULT_EVEN_EXPR/WIDEN_MULT_ODD_EXPR .

If we add an assignment from t to an array, we get
WIDEN_MULT_LO_EXPR/WIDEN_MULT_HI_EXPR which works.

So in the end this is a target issue since this is correct at tree level and it
looks like it is just an expansion of that is causing the issue.

[Bug target/61108] [4.8 Regression] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

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

[Bug target/61108] [4.8 Regression] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
      Known to work|                            |4.7.3, 4.9.0
   Target Milestone|---                         |4.8.3
            Summary|gcc 4.8.2 generates         |[4.8 Regression] gcc 4.8.2
                   |incorrect integer           |generates incorrect integer
                   |arithmetic at O3 (ok at O2) |arithmetic at O3 (ok at O2)
      Known to fail|                            |4.8.0, 4.8.2

[Bug target/61108] [4.8 Regression] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

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

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
__attribute__ ((noinline)) unsigned int
sum (const unsigned int *x)
{
  unsigned int r = 0;
  int i;
  for (i = 0; i < 8; i++)
    {
      unsigned int v = x[i];
      unsigned int q = (v * 613566757ULL) >> 32;
      unsigned int t = ((v + q) >> 1);
      r += t;
    }
  return r;
}

int
main (void)
{
  unsigned int x[8] = { 0, 7, 0, 0, 0, 0, 0, 0 };
  if (sum (x) != 4)
    __builtin_abort ();
  return 0;
}

So, I guess we want to backport PR60656 fix to 4.8 branch (both r209138 and
r209363), and add this new testcase to 4.8/4.9/trunk.

[Bug target/61108] [4.8 Regression] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.3                       |4.8.4

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 4.8.3 is being released, adjusting target milestone.

[Bug target/61108] [4.8 Regression] gcc 4.8.2 generates incorrect integer arithmetic at O3 (ok at O2)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61108

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.4                       |4.8.5

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.8.4 has been released.