https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87288
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Known to work| |7.3.1
Target Milestone|--- |8.3
Summary|Segfault after const_cast |[8/9 Regression] Segfault
|with "-O2 |after const_cast with "-O2
|-ftree-loop-vectorize" but |-ftree-loop-vectorize" but
|_without_ "-mavx" |_without_ "-mavx"
Known to fail| |8.2.1
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
-fno-tree-bit-ccp happens to fix the bug but it looks latent (same bogus
value-range). Value-ranges appear from DOM and the [1, ...] one first appears
with the
vect pass:
t.C.160t.ifcvt: # RANGE ~[2147483648, 18446744071562067967]
t.C.160t.ifcvt: # RANGE [0, 18446744073709551600]
t.C.160t.ifcvt: # RANGE [-2147483647, 2147483647]
t.C.161t.vect: # RANGE [1, 2147483647] NONZERO 2147483647
t.C.161t.vect: # RANGE ~[2147483648, 18446744071562067967]
t.C.161t.vect: # RANGE [0, 18446744073709551600]
t.C.161t.vect: # RANGE [-2147483647, 2147483647]
t.C.161t.vect: # RANGE ~[2147483648, 18446744071562067967]
t.C.161t.vect: # RANGE [0, 18446744073709551600]
t.C.161t.vect: # RANGE [-2147483647, 2147483647]
it's set here:
if (!is_gimple_val (niters_vector))
{
var = create_tmp_var (type, "bnd");
gimple_seq stmts = NULL;
niters_vector = force_gimple_operand (niters_vector, &stmts, true, var);
gsi_insert_seq_on_edge_immediate (pe, stmts);
/* Peeling algorithm guarantees that vector loop bound is at least ONE,
we set range information to make niters analyzer's life easier. */
if (stmts != NULL && log_vf)
set_range_info (niters_vector, VR_RANGE,
wi::to_wide (build_int_cst (type, 1)),
wi::to_wide (fold_build2 (RSHIFT_EXPR, type,
TYPE_MAX_VALUE (type),
log_vf)));
and the loop is
<bb 5> [local count: 105119325]:
niters.0_25 = (unsigned int) n_15;
ni_gap.1_36 = niters.0_25 + 4294967295;
# RANGE [1, 2147483647] NONZERO 2147483647
bnd.2_37 = ni_gap.1_36 >> 1;
<bb 4> [local count: 567644349]:
# ivtmp_50 = PHI <ivtmp_51(6), 0(5)>
ivtmp_51 = ivtmp_50 + 1;
if (ivtmp_51 >= bnd.2_37)
goto <bb 12>; [16.67%]
else
goto <bb 6>; [83.33%]
<bb 6> [local count: 473036958]:
goto <bb 4>; [100.00%]
which looks good according to the comment. So the number of iterations
_is_ bnd.2_37 - 1 (that number may be zero).
