https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115825
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
We also apply a 2/3 factor for an estimate on code reduction due to followup
optimization (we've worked on this not long ago - it seems to be critical for
some testcases derived from SPEC for example). We're likely hitting
static bool
try_unroll_loop_completely (class loop *loop,
edge exit, tree niter, bool may_be_zero,
enum unroll_level ul,
HOST_WIDE_INT maxiter,
dump_user_location_t locus, bool allow_peel,
bool cunrolli)
{
...
unsigned HOST_WIDE_INT ninsns = size.overall;
unsigned HOST_WIDE_INT unr_insns
= estimated_unrolled_size (&size, n_unroll);
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, " Loop size: %d\n", (int) ninsns);
fprintf (dump_file, " Estimated size after unrolling: %d\n",
(int) unr_insns);
}
/* If the code is going to shrink, we don't need to be extra
cautious on guessing if the unrolling is going to be
profitable.
Move from estimated_unrolled_size to unroll small loops. */
if (unr_insns * 2 / 3
/* If there is IV variable that will become constant, we
save one instruction in the loop prologue we do not
account otherwise. */
<= ninsns + (size.constant_iv != false))
;
but note the estimate is already suspiciously low.
Btw, on trunk I see
Loop 1 iterates 8 times.
Loop 1 iterates at most 8 times.
Loop 1 likely iterates at most 8 times.
Estimating sizes for loop 1
BB: 4, after_exit: 0
size: 2 if (i_3 != 0)
Exit condition will be eliminated in peeled copies.
Exit condition will be eliminated in last copy.
Constant conditional.
BB: 3, after_exit: 1
size: 1 v ={v} i_3;
size: 0 i.0_1 = (unsigned char) i_3;
Induction variable computation will be folded away.
size: 1 _2 = i.0_1 + 254;
Induction variable computation will be folded away.
size: 0 i_7 = (char) _2;
Induction variable computation will be folded away.
size: 4-3, last_iteration: 2-2
Loop size: 4
Estimated size after unrolling: 8
Not unrolling loop 1: size would grow.
I don't have a avr cross on the 14 branch to cross-check - can you paste
more context from the cunroll[i] details dump?
