[Bug middle-end/67452] New: LTO ICE with -fopenmp-simd

["jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67452

            Bug ID: 67452
           Summary: LTO ICE with -fopenmp-simd
           Product: gcc
           Version: 5.2.1
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: jakub at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org
                CC: hubicka at gcc dot gnu.org, rguenth at gcc dot gnu.org
  Target Milestone: ---

On the gomp-4_1-branch the libgomp.c/for-5.c and libgomp.c++/for-13.C testcases
ICE with intelmic emul offloading, I've reduced it down to -O2 -fopenmp-simd
-flto on the following testcase:
float b[3][3];

__attribute__((used, noinline)) void
foo ()
{
  int v1, v2;
#pragma omp simd collapse(2)
  for (v1 = 0; v1 < 3; v1++)
    for (v2 = 0; v2 < 3; v2++)
      b[v1][v2] = 2.5;
}

int
main ()
{
  asm volatile ("" : : "r" (b) : "memory");
  foo ();
  asm volatile ("" : : "r" (b) : "memory");
  return 0;
}

The ICE is:
In function âfooâ:
lto1: internal compiler error: in operator[], at vec.h:714
0x73a7a4 vec<tree_node*, va_heap, vl_embed>::operator[](unsigned int)
../../gcc/vec.h:714
0x73949c vec<tree_node*, va_heap, vl_ptr>::operator[](unsigned int)
../../gcc/vec.h:1180
0xedf193 streamer_tree_cache_get_tree
../../gcc/tree-streamer.h:111
0xee3487 streamer_get_pickled_tree(lto_input_block*, data_in*)
../../gcc/tree-streamer-in.c:1105
0xae9d42 lto_input_tree_1(lto_input_block*, data_in*, LTO_tags, unsigned int)
../../gcc/lto-streamer-in.c:1428
0xae9f7b lto_input_tree(lto_input_block*, data_in*)
../../gcc/lto-streamer-in.c:1474
0xee1da5 lto_input_ts_decl_minimal_tree_pointers
../../gcc/tree-streamer-in.c:689
0xee31c4 streamer_read_tree_body(lto_input_block*, data_in*, tree_node*)
../../gcc/tree-streamer-in.c:1046
0xae99b2 lto_read_tree_1
../../gcc/lto-streamer-in.c:1304
0xae9abd lto_read_tree
../../gcc/lto-streamer-in.c:1338
0xae9f12 lto_input_tree_1(lto_input_block*, data_in*, LTO_tags, unsigned int)
../../gcc/lto-streamer-in.c:1457
0xae9b37 lto_input_scc(lto_input_block*, data_in*, unsigned int*, unsigned
int*)
../../gcc/lto-streamer-in.c:1362
0xae9f61 lto_input_tree(lto_input_block*, data_in*)
../../gcc/lto-streamer-in.c:1472
0xae828a input_cfg
../../gcc/lto-streamer-in.c:863
0xae8e71 input_function
../../gcc/lto-streamer-in.c:1079
0xae9683 lto_read_body_or_constructor
../../gcc/lto-streamer-in.c:1232
0xae993c lto_input_function_body(lto_file_decl_data*, cgraph_node*, char
const*)
../../gcc/lto-streamer-in.c:1278
0x7f4d60 cgraph_node::get_untransformed_body()
../../gcc/cgraph.c:3286
0x800c84 cgraph_node::expand()
../../gcc/cgraphunit.c:1892
0x801395 expand_all_functions
../../gcc/cgraphunit.c:2056
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.
lto-wrapper: fatal error: ./xgcc returned 1 exit status
compilation terminated.
/usr/bin/ld: lto-wrapper failed
collect2: error: ld returned 1 exit status

The problem is that early optimizations eliminate simduid.3 uses and during
release_ssa pass it is removed from cfun->local_decls, but remains in
loop->simduid, and the LTO streaming doesn't expect to have trees streamed too
late (during output_cfg) apparently.