[aarch64] assert(is_power_of_2(x)) failed: x must be a power of 2: 0xffffffff80000000

[Schmidt, Lutz]

Hi,

I would locate the real bug in is_power_of_2(jint). Why?

The function is passed in a jint (= signed int) of value 0x8000_0000. This clearly is a negative number. There does not exist a negative number which is a power of 2.

Separate from that, exact_log2(jint) sign-extends the passed-in value to long, creating a bit pattern which obviously is not a power of two.

A possible fix would include:
 - make is_power_of_2() and exact_log_2() sign-aware.
 - call is_power_of_2() and exact_log_2() with arguments of proper type (juint in this case).

Regards,
Lutz


On 24.02.20, 15:18, "hotspot-compiler-dev on behalf of Andrew Haley" <hotspot-compiler-dev-bounces at openjdk.java.net on behalf of aph at redhat.com> wrote:

    On 2/24/20 1:47 PM, Doerr, Martin wrote:
    > I've created an issue:
    > https://bugs.openjdk.java.net/browse/JDK-8239861
    >
    > Failues were observed in e.g. jck tests api/java_awt/Font/LineBreakMeasurer.
    
    This is caused by "8183574: Unify the is_power_of_2 functions".
    
    I think there's a real bug in the new exact_log2().
    is_power_of_2((jint)0x8000_0000) returns true but
    exact_log2((jint)0x8000_0000) throws an exception. This sounds wrong
    to me. We should work around it in the AArch64 back end by converting
    the operand to an unsigned type, but I wonder what else broke.
    
    Here:
    
    instruct far_cmpI_branch_bit(cmpOpEqNe cmp, iRegIorL2I op1, immI op2, immI0 op3, label labl) %{
      match(If cmp (CmpI (AndI op1 op2) op3));
      predicate(is_power_of_2(n->in(2)->in(1)->in(2)->get_int()));
      effect(USE labl);
    
      ins_cost(BRANCH_COST);
      format %{ "tb$cmp   $op1, $op2, $labl" %}
      ins_encode %{
        Label* L = $labl$$label;
        Assembler::Condition cond = (Assembler::Condition)$cmp$$cmpcode;
        int bit = exact_log2($op2$$constant);
    
    Note that we check that $op2 is a power of two before we call
    exact_log2().
    
    -- 
    Andrew Haley  (he/him)
    Java Platform Lead Engineer
    Red Hat UK Ltd. <https://www.redhat.com>
    https://keybase.io/andrewhaley
    EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671