https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110628
--- Comment #6 from Jan Hubicka <hubicka at ucw dot cz> ---
The mismatch happens on:
void foo (unsigned int x)
{
if (x != 0x800 && x != 0x810)
abort ();
}
It is bug in reassoc turning:
void foo (unsigned int x)
{
;; basic block 2, loop depth 0, count 1073741824 (estimated locally, freq
1.0000), maybe hot
;; prev block 0, next block 3, flags: (NEW, REACHABLE, VISITED)
;; pred: ENTRY [always] count:1073741824 (estimated locally, freq
1.0000) (FALLTHRU,EXECUTABLE)
if (x_1(D) != 2048)
goto <bb 3>; [66.00%]
else
goto <bb 5>; [34.00%]
;; succ: 3 [66.0% (guessed)] count:708669600 (estimated locally, freq
0.6600) (TRUE_VALUE,EXECUTABLE)
;; 5 [34.0% (guessed)] count:365072224 (estimated locally, freq
0.3400) (FALSE_VALUE,EXECUTABLE)
;; basic block 3, loop depth 0, count 708669600 (estimated locally, freq
0.6600), maybe hot
;; prev block 2, next block 4, flags: (NEW, REACHABLE, VISITED)
;; pred: 2 [66.0% (guessed)] count:708669600 (estimated locally, freq
0.6600) (TRUE_VALUE,EXECUTABLE)
if (x_1(D) != 2064)
goto <bb 4>; [0.00%]
else
goto <bb 5>; [100.00%]
;; succ: 4 [never] count:0 (precise, freq 0.0000)
(TRUE_VALUE,EXECUTABLE)
;; 5 [always] count:708669600 (estimated locally, freq 0.6600)
(FALSE_VALUE,EXECUTABLE)
;; basic block 4, loop depth 0, count 0 (precise, freq 0.0000), probably
never executed
;; prev block 3, next block 5, flags: (NEW, REACHABLE, VISITED)
;; pred: 3 [never] count:0 (precise, freq 0.0000)
(TRUE_VALUE,EXECUTABLE)
abort ();
;; succ:
to:
void foo (unsigned int x)
{
unsigned int _4;
_Bool _5;
;; basic block 2, loop depth 0, count 1073741824 (estimated locally, freq
1.0000), maybe hot
;; prev block 0, next block 3, flags: (NEW, REACHABLE, VISITED)
;; pred: ENTRY [always] count:1073741824 (estimated locally, freq
1.0000) (FALLTHRU,EXECUTABLE)
_4 = x_1(D) & 4294967279;
_5 = _4 != 2048;
if (_5 != 0)
goto <bb 3>; [66.00%]
else
goto <bb 4>; [34.00%]
;; succ: 3 [66.0% (guessed)] count:708669600 (estimated locally, freq
0.6600) (TRUE_VALUE,EXECUTABLE)
;; 4 [34.0% (guessed)] count:365072224 (estimated locally, freq
0.3400) (FALSE_VALUE,EXECUTABLE)
;; basic block 3, loop depth 0, count 0 (precise, freq 0.0000), probably
never executed
;; Invalid sum of incoming counts 708669600 (estimated locally, freq 0.6600),
should be 0 (precise, freq 0.0000)
;; prev block 2, next block 4, flags: (NEW, REACHABLE, VISITED)
;; pred: 2 [66.0% (guessed)] count:708669600 (estimated locally, freq
0.6600) (TRUE_VALUE,EXECUTABLE)
abort ();
;; succ:
;; basic block 4, loop depth 0, count 1073741824 (estimated locally, freq
1.0000), maybe hot
;; Invalid sum of incoming counts 365072224 (estimated locally, freq 0.3400),
should be 1073741824 (estimated locally, freq 1.0000)
;; prev block 3, next block 1, flags: (NEW, REACHABLE, VISITED)
;; pred: 2 [34.0% (guessed)] count:365072224 (estimated locally, freq
0.3400) (FALSE_VALUE,EXECUTABLE)
return;
So it combines two conditionals together but does not update the
outgoing probabilitis of the conditional.
On x86-64 we have already in cfg dump:
_1 = x != 2048;
_2 = x != 2064;
_3 = _1 & _2;
if (_3 != 0)
goto <bb 3>; [INV]
else
I wonder why optimization diverges here?
