[Bug tree-optimization/86632] New: Incorrect value copied into output array with -O3 ftree-loop-vectorize

[ketan.surender at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86632

            Bug ID: 86632
           Summary: Incorrect value copied into output array with -O3
                    ftree-loop-vectorize
           Product: gcc
           Version: 6.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ketan.surender at gmail dot com
  Target Milestone: ---

I am observing incorrect results for the following function with -O3

static void mDiscBlocks2_repro_output(void)
{
  /* local scratch DWork variables */
  int32_T ForEach_itr_p;
  real_T rtb_ImpAsg_InsertedFor_Out1_a_d[6];
  real_T rtb_ImpAsg_InsertedFor_Out2_a_o[3];
  int32_T i;

  for (ForEach_itr_p = 0; ForEach_itr_p < 3; ForEach_itr_p++) {
    rtb_ImpAsg_InsertedFor_Out2_a_o[ForEach_itr_p] =
      mDiscBlocks2_repro_DW.CoreSubsys[ForEach_itr_p].Memory1_PreviousInput;
    rtb_ImpAsg_InsertedFor_Out1_a_d[ForEach_itr_p << 1] =
      mDiscBlocks2_repro_DW.CoreSubsys[ForEach_itr_p].Memory_PreviousInput[0];
    rtb_ImpAsg_InsertedFor_Out1_a_d[1 + (ForEach_itr_p << 1)] =
      mDiscBlocks2_repro_DW.CoreSubsys[ForEach_itr_p].Memory_PreviousInput[1];
  }

  /* KS REQUIRED */
  for (i = 0; i < 6; i++) {
    mDiscBlocks2_repro_Y.Out14[i] = rtb_ImpAsg_InsertedFor_Out1_a_d[i];
  }

  /* KS REQUIRED */
  mDiscBlocks2_repro_Y.Out15[0] = rtb_ImpAsg_InsertedFor_Out2_a_o[0];
  mDiscBlocks2_repro_Y.Out15[1] = rtb_ImpAsg_InsertedFor_Out2_a_o[1];
  mDiscBlocks2_repro_Y.Out15[2] = rtb_ImpAsg_InsertedFor_Out2_a_o[2];
}

This code copies some global data to a local array, then copies the local array
to a global. 

mDiscBlocks2_repro_DW.CoreSubsys[0-2].Memory1_PreviousInput -->
rtb_ImpAsg_InsertedFor_Out2_a_o[0-2] -->  mDiscBlocks2_repro_Y.Out15[0-2]
mDiscBlocks2_repro_DW.CoreSubsys[0-2].Memory_PreviousInput[0] -->
rtb_ImpAsg_InsertedFor_Out1_a_d[0,2,4] --> mDiscBlocks2_repro_Y.Out14[0,2,4]
mDiscBlocks2_repro_DW.CoreSubsys[0-2].Memory_PreviousInput[1] -->
rtb_ImpAsg_InsertedFor_Out1_a_d[1,3,5] --> mDiscBlocks2_repro_Y.Out14[1,3,5]

For the global 'mDiscBlocks2_repro_Y.Out14' I am observing the incorrect value
at index 2. The issue goes away if I add the switch -fno-tree-loop-vectorize. 

I looked at the generated asm a little and can see the incorrect assignment.
For some reason it writes to element 2 before writing the remaining elements.

Here is my gcc info

Using built-in specs.
COLLECT_GCC=[SNIP]/glnxa64/gcc-6.3.0/bin/gcc
COLLECT_LTO_WRAPPER=[SNIP]/glnxa64/gcc-6.3.0/bin/../libexec/gcc/x86_64-pc-linux-gnu/6.3.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: [SNIP]/sources/gcc-6.3/configure
--with-gmp=[SNIP]/glnxa64/gcc-6.3/gmp-4.3 --with-mpfr=[SNIP]/gcc-6.3/mpfr
--with-mpc=[SNIP]/gcc-6.3/mpc --enable-languages=c,c++,fortran
--with-bugurl=[SNIP],_Debugging --enable-shared --enable-linker-build-id
--enable-plugin --enable-checking=release --enable-multiarch --enable-gold
--enable-ld=default --enable-libstdcxx-time=no
--prefix=[SNIP]/glnxa64/gcc-6.3.0 --with-pkgversion='MW GCC 6.3.0-GLIBC2.12'
--with-tune=generic --with-system-zlib --enable-multilib
--with-multilib-list=m32,m64 --with-arch-directory=amd64 --with-arch-32=i586
--with-abi=m64
Thread model: posix
gcc version 6.3.0 (MW GCC 6.3.0-GLIBC2.12)