RFR: 8250745: Fix a potential bug on AVX512 machines with assert(eval_map.contains(n)) failed: absent

[Vladimir Ivanov]

FYI test results are clean (hs-precheckin-comp,hs-tier1,hs-tier2).

Best regards,
Vladimir Ivanov

On 29.07.2020 12:20, Vladimir Ivanov wrote:
>> Webrev: http://cr.openjdk.java.net/~jiefu/8250745/webrev.00/
> 
> Looks good.
> 
> FTR the bug was introduced by JDK-8241040, but I don't see a way it can 
> be hit by auto-vectorizer: before it kicks in, scalar code is strongly 
> normalized and constants are pushed to the right. It leads to the shape 
> where (Replicate -1) is always the second input of bitwise NOT shape 
> (XorV v (Replicate -1)). Since there are no GVN transformations 
> happening for vector nodes, both left-hand and right-hand variants 
> become possible with Vector API.
> 
> Best regards,
> Vladimir Ivanov