[11] RFR(S): 8203196: C1 emits incorrect code due to integer overflow in _tableswitch keys

[Tobias Hartmann]

Thanks, Martin!

Best regards,
Tobias

On 18.05.2018 15:13, Doerr, Martin wrote:
> Hi Tobias,
> 
> thanks for fixing. Undefined behavior of signed integer overflow is really annoying in C/C++.
> Fix looks good to me.
> 
> Best regards,
> Martin
> 
> 
> -----Original Message-----
> From: hotspot-compiler-dev [mailto:hotspot-compiler-dev-bounces at openjdk.java.net] On Behalf Of Tobias Hartmann
> Sent: Freitag, 18. Mai 2018 12:03
> To: hotspot compiler <hotspot-compiler-dev at openjdk.java.net>
> Subject: [11] RFR(S): 8203196: C1 emits incorrect code due to integer overflow in _tableswitch keys
> 
> Hi,
> 
> please review the following patch:
> https://bugs.openjdk.java.net/browse/JDK-8203196
> http://cr.openjdk.java.net/~thartmann/8203196/webrev.00/
> 
> C1 incorrectly compiles a _tableswitch instruction due to an integer overflow of x->hi_key() in
> Canonicalizer::do_TableSwitch():
>   0 0 i5 2147483647
> . 2 0 6 tableswitch i5
>                    case 2147483647: B1
>                    default : B2
> canonicalized to:
> . 2 0 7 goto B2
> 
> We compute hi_key = _lo_key + length() - 1 = INT_MAX + (2 - 1) - 1 = INT_MAX + 1 - 1 which may
> overflow depending on how the C++ compiler translates it (fails with fastdebug, works with
> slowdebug). I've added parentheses and an assert for sanity checking.
> 
> This problem was introduced by the fix for JDK-8200303. Before, we would use an if instead of a
> switch [1].
> 
> Thanks,
> Tobias
> 
> [1] http://hg.openjdk.java.net/jdk/jdk/rev/d84f06a0cae1#l3.8
>