RFR (S) 8176580: [ppc, s390] CRC32C: wrong checksum result in some cases

[Schmidt, Lutz]

Dear all,

may I kindly request reviews for my small bugfix? And somebody volunteering as a sponsor would be welcome as well.

Bug: https://bugs.openjdk.java.net/browse/JDK-8176580
Webrev: http://cr.openjdk.java.net/~lucy/webrevs/8176580/

Almost two weeks ago, I had contributed CRC32C intrinsic implementations for ppc and s390. The arguments passed to the intrinsics differ slightly from those passed to the CRC32 intrinsics. Instead of passing the byte array length directly, it has to be calculated from the offset and the end index values.

This difference was not accounted for in templateInterpreterGenerator_{ppc|s390}.cpp as well as in c1_LIRGenerator_{ppc|s390}.cpp. The fix just adds the missing subtraction, and adjusts some comments. For the bug to become visible, it is necessary to calculate a CRC32C checksum from a byte array being accessed with a non-zero offset. The call to the intrinsic has to originate from the interpreter or from a c1-compiled method.

This change is “platform only”. No changes to shared code.

Thank you!
Lutz

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20170314/83b2fcbe/attachment.html>