RFR: 8352422: [ubsan] Out-of-range reported in ciMethod.cpp:917:20: runtime error: 2.68435e+09 is outside the range of representable values of type 'int' [v2]

[Emanuel Peter]

On Tue, 29 Apr 2025 16:37:02 GMT, Marc Chevalier <mchevalier at openjdk.org> wrote:

>> The double `(double)count * prof_factor * method_life / counter_life + 0.5`
>> can overflow a 32-bit int, causing UB on casting, but in practice computing
>> a wrong scale, probably.
>> 
>> We just need to compare that the cast is not going to overflow. This is possible
>> because `INT_MAX` is exactly representable in a `double`. It is also good to
>> notice that the expression `(double)count * prof_factor * method_life / counter_life + 0.5`
>> cannot overflow a `double`:
>> - `count` is a int, max value = 2^31-1 < 2.2e9
>> - `method_lie` is a int, max value < 2.2e9
>> - `prof_factor` is a float, max value < 3.5e38
>> - `counter_life` is a int, positive at this point, so min value = 1
>> So, the whole expression is bounded by 16.94e56 + 0.5, which is much smaller than the
>> max value of a double (about 1.8e308). We probably would have precision issues, but
>> it probably doesn't matter a lot.
>> 
>> The semantic I picked here is basically `min(INT_MAX, count_d)`, so it'd always fit.
>> 
>> Thanks,
>> Marc
>
> Marc Chevalier has updated the pull request incrementally with one additional commit since the last revision:
> 
>   +comment

Marked as reviewed by epeter (Reviewer).

-------------

PR Review: https://git.openjdk.org/jdk/pull/24824#pullrequestreview-2806199875