This is when searching for a PHI for vop .MEM_NNN, but apparently there is no such PHI.  I'm afraid I'm not familiar with this part of code to know why it is trying to do that.

CCing Tom as the author of r227437.

Minimal example:
...
program main
  implicit none
  real, dimension(:,:),allocatable :: a, b, c
  real :: sm

  allocate (a(2,2), b(2,2), c(2,2))

  call random_number(a)
  call random_number(b)

  c = matmul(a,b)
  sm = sum(c)

  deallocate(a,b,c)

end program main
...

The assert happens in expand_omp_for_static_chunk when trying to expand a
region from bb71 to bb72:
...
(gdb) p region.entry.index
$1 = 71
(gdb) p region.exit.index
$2 = 72
...

In other words:
...
;;   basic block 71, loop depth 0, freq 48, maybe hot
;;    prev block 69, next block 67, flags: (NEW)
;;    pred:       69 [always]  (FALLTHRU)
  #pragma omp for schedule(static,2)
  for (ivtmp_87 = 0; ivtmp_87 < _144; ivtmp_87 =  + 1)
;;    succ:       67 [always]  (FALLTHRU)
;;                72 [never (guessed)]

   ...

;;   basic block 72, loop depth 0, freq 47, maybe hot
;;   Invalid sum of incoming frequencies 269, should be 47
;;    prev block 63, next block 70, flags: (NEW)
;;    pred:       46 [always]  (FALLTHRU)
;;                71 [never (guessed)] 
  # .MEM_88 = PHI <.MEM_86(46), .MEM_86(71)>
  #pragma omp return(nowait)
;;    succ:       70 [always]  (FALLTHRU)
...

When we're ICE-ing, we're executing a bit:
...
      /* When we redirect the edge from trip_update_bb to iter_part_bb, we                            
         remove arguments of the phi nodes in fin_bb.  We need to create                              
         appropriate phi nodes in iter_part_bb instead.  */
...

When we look at fin_bb, we see indeed that one of the arguments of the phi has been dropped.:
...
(gdb) call debug_bb (fin_bb)
;; basic block 72, loop depth 0, freq 47, maybe hot
;;  prev block 100, next block 70, flags: (NEW)
;;  pred:       98 [never (guessed)]  (FALSE_VALUE)
# .MEM_88 = PHI <.MEM_86(98)>
;;  succ:       70 [always]  (FALLTHRU)
...

But since the arguments were equal anyway, this is fine, and there's no need to do anything.

Tentative patch:
...
diff --git a/gcc/omp-expand.c b/gcc/omp-expand.c
index 929c530..089bffc 100644
--- a/gcc/omp-expand.c
+++ b/gcc/omp-expand.c
@@ -4206,6 +4206,10 @@ expand_omp_for_static_chunk (struct omp_region *region,
          source_location locus;
 
          phi = psi.phi ();
+         if (operand_equal_p (gimple_phi_arg_def (phi, 0),
+                              gimple_phi_arg_def (phi, 1), 0))
+             continue;
+
          t = gimple_phi_result (phi);
          gcc_assert (t == redirect_edge_var_map_result (vm));
 
...

(In reply to Tom de Vries from comment #3)
> Tentative patch:
> ...
> diff --git a/gcc/omp-expand.c b/gcc/omp-expand.c
> index 929c530..089bffc 100644
> --- a/gcc/omp-expand.c
> +++ b/gcc/omp-expand.c
> @@ -4206,6 +4206,10 @@ expand_omp_for_static_chunk (struct omp_region
> *region,
>           source_location locus;
>  
>           phi = psi.phi ();
> +         if (operand_equal_p (gimple_phi_arg_def (phi, 0),
> +                              gimple_phi_arg_def (phi, 1), 0))
> +             continue;
> +
>           t = gimple_phi_result (phi);
>           gcc_assert (t == redirect_edge_var_map_result (vm));
>  
> ...

As it turns out, at the point of the operand_equal_p test, we already lost the second argument of the phi.

I thought that this would trigger an assert, but apparently we don't have this:
...
@@ -4379,6 +4379,7 @@ phi_nodes_ptr (basic_block bb)
 static inline tree
 gimple_phi_arg_def (gimple *gs, size_t index)
 {
+  gcc_gimple_checking_assert (index < gimple_phi_num_args(gs));
   return gimple_phi_arg (gs, index)->def;
 }
 
...

Using redirect_edge_var_map_def instead fixes this.

Created attachment 41779 [details]
Tentative patch

https://gcc.gnu.org/ml/gcc-patches/2017-07/msg01087.html

Patch with test-case committed, marking resolved fixed.

Author: vries
Date: Wed Jul 19 06:25:40 2017
New Revision: 250335

URL: https://gcc.gnu.org/viewcvs?rev=250335&root=gcc&view=rev
Log:
Handle equal-argument loop exit phi in expand_omp_for_static_chunk

2017-07-18  Tom de Vries  <tom@codesourcery.com>

	PR middle-end/81464
	* omp-expand.c (expand_omp_for_static_chunk): Handle equal-argument loop
	exit phi.

	* gfortran.dg/pr81464.f90: New test.

Added:
    trunk/gcc/testsuite/gfortran.dg/pr81464.f90
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/omp-expand.c
    trunk/gcc/testsuite/ChangeLog