RFR (S): 7174532: jdk/test/java/lang/Math/WorstCaseTests.java failing on x86
Vladimir Kozlov
vladimir.kozlov at oracle.com
Mon Jun 11 07:10:33 PDT 2012
Sorry, Roland
Nevermind my comment. Your code is good, I did not realize first that you save/restore current precision. Please, only
verify HyperbolicTest.java test and you can push this fix.
Thanks,
Vladimir
On 6/11/12 6:59 AM, Vladimir Kozlov wrote:
> Roland,
>
> There is orl(Address, int) version of instruction so you don't need to load into rax.
>
> I forgot yo ask, did you verify the test HyperbolicTest.java? If yes, please, close new bug 7175838
> as duplicate.
>
> Thanks,
> Vladimir
>
> On 6/11/12 4:25 AM, Roland Westrelin wrote:
>>> Yes, windows may set different fpu_ctw. May be instead of verification we should save/restore current control word. I
>>> surprise that we don't have problems with Log and Trigonometric instructions.
>>
>> Here is a new version:
>> http://cr.openjdk.java.net/~roland/7174532/webrev.01/
>>
>>> An other thing to test is to run 32bit VM with UseSSE=0.
>>
>> Testing with UseSSE=0 went fine.
>>
>> Roland.
More information about the hotspot-compiler-dev
mailing list