JDK10/RFR(L): 8144448: Avoid placing CTI immediately following or preceding RDPC instruction.
patric.hedlin at oracle.com
Fri Jun 16 13:58:58 UTC 2017
On 2017-06-15 19:51, Vladimir Kozlov wrote:
> assembler_sparc.cpp - may be use BytesPerInstWord instead of 4.
Indeed... sloppy me.
> In sparc.ad in MachConstantBaseNode::emit() can you use nop() in case
> disp == 0? To avoid changing O7 reg.
Sure, updated accordingly.
Thanks for your time,
> On 6/15/17 9:06 AM, Patric Hedlin wrote:
>> Dear all,
>> I would like to ask for help to review the following change/update:
>> Issue: https://bugs.openjdk.java.net/browse/JDK-8144448
>> Webrev: http://cr.openjdk.java.net/~neliasso/phedlin/tr8144448/
>> Prerequisite: https://bugs.openjdk.java.net/browse/JDK-8181853
>> *** As a comment to the discussion on how to simplify processing of
>> "mundane" changes,
>> this change/update comes with an additional prerequisite (patch)
>> whitespace and two lingering uses of 'NOT_LP64' and 'LP64_ONLY'.
>> 8144448: Avoid placing CTI immediately following or preceding RDPC
>> Approach taken here is to handle 'rdpc' in the same manner as
>> 'cbcond', using
>> a simple scheme to prohibit the assembler from emitting any
>> 'rdpc' instruction
>> back-to-back with other CTI ('rdpc' itself included), inserting
>> 'nop' as needed.
>> This change is applied to all generations of SPARC cores event
>> though it is the
>> SPARC Core S5 that is the actual target. Benchmarking on T4 and
>> M7 suggests that
>> there is no penalty. This choice (which is subject to change)
>> has been made in
>> order to give the update some mileage while waiting for Core S5
>> hardware to be
>> available in regular testing.
>> Mostly tested on JDK9 (jtreg/RBT/hotspot/tier0-comp).
>> Testing on JDK10 (jtreg/RBT/hotspot/precheckin-comp).
>> Best regards,
More information about the hotspot-dev