RFR(XS): 8239856: [ntintel] asserts about copying unaligned array element

[Doerr, Martin]

Hi Chris,

according to arraycopy.hpp,
“arraycopy operations are implicitly atomic on each array element.”

This requires 8 Byte alignment for jlong and jdouble.

I don’t want to give up this property just because Windows 32 bit doesn’t align them this way by default.
All other supported platforms do it right by default.

Best regards,
Martin



From: Chris Plummer <chris.plummer at oracle.com>
Sent: Montag, 24. Februar 2020 21:52
To: Doerr, Martin <martin.doerr at sap.com>; OpenJDK Serviceability <serviceability-dev at openjdk.java.net>
Subject: Re: RFR(XS): 8239856: [ntintel] asserts about copying unaligned array element

Hi Martin,

I'm not so sure I agree with the approach to this fix, nor for the one already done for JDK-8220348. Shouldn't a user be expected to be able to pass a jlong variable to SetLongArrayRegion() without the need for any special platform dependent modifiers added to the declaration of the variable?

cheers,

Chris

On 2/24/20 5:51 AM, Doerr, Martin wrote:
Hi,

reposting on serviceability-dev (was core-libs-dev before).

Bug:
https://bugs.openjdk.java.net/browse/JDK-8239856

Webrev:
http://cr.openjdk.java.net/~mdoerr/8239856_win32_long_double_align/webrev.00/

Thanks for the review, Thomas!

Best regards,
Martin


From: Thomas Stüfe <thomas.stuefe at gmail.com><mailto:thomas.stuefe at gmail.com>
Sent: Montag, 24. Februar 2020 14:41
To: Doerr, Martin <martin.doerr at sap.com><mailto:martin.doerr at sap.com>
Cc: core-libs-dev at openjdk.java.net<mailto:core-libs-dev at openjdk.java.net>; Lindenmaier, Goetz <goetz.lindenmaier at sap.com><mailto:goetz.lindenmaier at sap.com>; Langer, Christoph <christoph.langer at sap.com><mailto:christoph.langer at sap.com>
Subject: Re: RFR(XS): 8239856: [ntintel] asserts about copying unaligned array element

Oh okay. Then it looks okay to me.

Cheers, Thomas

On Mon, Feb 24, 2020 at 12:56 PM Doerr, Martin <martin.doerr at sap.com<mailto:martin.doerr at sap.com>> wrote:
Hi Thomas,

thanks for the quick review.

ATTRIBUTE_ALIGNED is defined in hotspot. I can’t use it for src/jdk.jdwp.agent/share/native/libjdwp/ArrayReferenceImpl.c.

Christoph had already suggested to make it available for core libs, too, but I haven’t found a good place for it.

Best regards,
Martin


From: Thomas Stüfe <thomas.stuefe at gmail.com<mailto:thomas.stuefe at gmail.com>>
Sent: Montag, 24. Februar 2020 12:52
To: Doerr, Martin <martin.doerr at sap.com<mailto:martin.doerr at sap.com>>
Cc: core-libs-dev at openjdk.java.net<mailto:core-libs-dev at openjdk.java.net>; Lindenmaier, Goetz <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>>; Langer, Christoph <christoph.langer at sap.com<mailto:christoph.langer at sap.com>>
Subject: Re: RFR(XS): 8239856: [ntintel] asserts about copying unaligned array element

Hi Martin,

maybe use ATTRIBUTE_ALIGNED instead?

Cheers, Thomas

On Mon, Feb 24, 2020 at 12:44 PM Doerr, Martin <martin.doerr at sap.com<mailto:martin.doerr at sap.com>> wrote:
Hi,

we had fixed stack array alignment for Windows 32 bit with JDK-8220348.
However, there are also stack allocated jlong and jdouble used as source for SetLongArrayRegion and SetDoubleArrayRegion with insufficient alignment for this platform.

Here’s my proposed fix:
http://cr.openjdk.java.net/~mdoerr/8239856_win32_long_double_align/webrev.00/

Please review.

Best regards,
Martin


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/serviceability-dev/attachments/20200225/420236dc/attachment-0001.htm>