8038953,Add sanity tests for BMI1 and LZCNT instructions
Anton Ivanov
anton.ivanov at oracle.com
Tue Apr 1 17:51:55 UTC 2014
Hi Vladimir,
They test different things,
Fillip's tests - verify results to be the same for compiler and
interpreter (but imagine compiler do not use intrinsic at all, test will
pass too, because they will do exactly the same operations)
My tests verify only that certain asm instructions are presented in
native code emitted by compiler (but do not verify the correctness of
execution)
Combination of these 2 sets of tests makes sure that new instructions
are really used, and used correctly.
It also was not easy to modify Fillip's tests because they were
developed in parallel
combination of these 2 sets give us
On 01.04.2014 21:03, Vladimir Kozlov wrote:
> I mean why do you need to create separate tests and not use (add new
> code) existing one?
> For example, you added bmi/sanity/AddnTestI.java when there is already
> bmi/TestAndnI.java.
>
> Thanks,
> Vladimir
>
> On 4/1/14 9:57 AM, Vladimir Kozlov wrote:
>> Anton,
>>
>> How your changes relate to Filipp's?:
>>
>> http://hg.openjdk.java.net/jdk9/hs-comp/hotspot/rev/15d507abfc7a
>>
>> Thanks,
>> Vladimir
>>
>> On 4/1/14 9:40 AM, Anton Ivanov wrote:
>>> Hi all,
>>>
>>> Please review the patch
>>>
>>> New tests for using of blsi,blsr,andn,blsmsk,lzcnt,tzcnt
>>> instructions were developed.
>>> They verify that instructions above are really presented in emitted
>>> native code on supported hardware
>>>
>>> webrev:
>>> http://cr.openjdk.java.net/~iignatyev/aaivanov/8038953/webrev.00
>>> jbs: https://bugs.openjdk.java.net/browse/JDK-8038953
>>> testing: test/compiler/intrinsics/bmi/sanity
>>>
--
Best regards,
Anton Ivanov
More information about the hotspot-compiler-dev
mailing list