RFR : PPC64 : Need support for VSR spills in ppc.ad

[Doerr, Martin]

Hi Michihiro,

thanks for the new webrev. Looks good. We will test it.
Please add the bug id to the subject of RFRs. This one is 8194861.

https://bugs.openjdk.java.net/browse/JDK-8194861

The bug could be treated as P3 bug because SuperwordUseVSX is currently broken and leads to crashes.
But I'm ok with leaving it switched off for jdk10 and keeping the bug targeted to jdk11.

Best regards,
Martin


From: Michihiro Horie [mailto:HORIE at jp.ibm.com]
Sent: Donnerstag, 11. Januar 2018 05:01
To: Doerr, Martin <martin.doerr at sap.com>
Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; gromero at linux.vnet.ibm.com; hotspot-dev at openjdk.java.net; ppc-aix-port-dev at openjdk.java.net
Subject: RE: RFR : PPC64 : Need support for VSR spills in ppc.ad


Hi Martin,

>Not to be used by MachConstantBaseNode.
Thank you for the confirmation.

>I still suggest to add "iRegLdst tmp" with "effect(TEMP tmp)" to repl4F_immF_Ex.
Sure, I updated code with a tmp register instead of R19:
http://cr.openjdk.java.net/~mhorie/8194861/webrev.02/

Best regards,
--
Michihiro,
IBM Research - Tokyo

[Inactive hide details for "Doerr, Martin" ---2018/01/11 02:19:30---Hi Michihiro, > R19 is commented out in bits64_constant_tabl]"Doerr, Martin" ---2018/01/11 02:19:30---Hi Michihiro, > R19 is commented out in bits64_constant_table_base so as not to be used:

From: "Doerr, Martin" <martin.doerr at sap.com<mailto:martin.doerr at sap.com>>
To: Michihiro Horie <HORIE at jp.ibm.com<mailto:HORIE at jp.ibm.com>>
Cc: "Lindenmaier, Goetz" <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>>, "gromero at linux.vnet.ibm.com<mailto:gromero at linux.vnet.ibm.com>" <gromero at linux.vnet.ibm.com<mailto:gromero at linux.vnet.ibm.com>>, "hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>" <hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>>, "ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>" <ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>>
Date: 2018/01/11 02:19
Subject: RE: RFR : PPC64 : Need support for VSR spills in ppc.ad

________________________________



Hi Michihiro,

> R19 is commented out in bits64_constant_table_base so as not to be used:
Not to be used by MachConstantBaseNode. I don't see how this should help.

I still suggest to add "iRegLdst tmp" with "effect(TEMP tmp)" to repl4F_immF_Ex.

Best regards,
Martin


From: Michihiro Horie [mailto:HORIE at jp.ibm.com]
Sent: Mittwoch, 10. Januar 2018 16:29
To: Doerr, Martin <martin.doerr at sap.com<mailto:martin.doerr at sap.com>>
Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>>; gromero at linux.vnet.ibm.com<mailto:gromero at linux.vnet.ibm.com>; hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>; ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>
Subject: RE: RFR : PPC64 : Need support for VSR spills in ppc.ad

Hi Martin,

Thanks a lot for your review.

>I wonder how the content of R19 should get preserved.
R19 is commented out in bits64_constant_table_base so as not to be used:
reg_class bits64_constant_table_base(
:
R18_H, R18,
/*R19_H, R19*/
R20_H, R20,
:
);

Although bits64_constant_table_base is not directly referred from anywhere, it seems to be used at the following line in ppc.ad:
const RegMask& MachConstantBaseNode::_out_RegMask = BITS64_CONSTANT_TABLE_BASE_mask();
(When I remove the declaration of bits64_constant_table_base, a build error arises at this line telling the lack of the declaration.)

>Shouldn't repl4F_immF_Ex use a temp register instead of R19?
Thank you for the suggestion, which makes sense very much. I think I can declare temp registers in repl4F_immF_Ex.
I will try this approach if using R19 does not make sense.

I changed vm_version_ppc.cpp:
webrev: http://cr.openjdk.java.net/~mhorie/8194861/webrev.01/<https://urldefense.proofpoint.com/v2/url?u=http-3A__cr.openjdk.java.net_-7Emhorie_8194861_webrev.01_&d=DwMGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=oecsIpYF-cifqq2i1JEH0Q&m=SOGOvG8yFchCvPeC1Pr5_BJNIaBdg1G1IXaEeMgwpWI&s=kFfDKuiMdVnWKBy0FOjXCTzN430PRKsHQgSCRWkdR8w&e=>

Best regards,
--
Michihiro,
IBM Research - Tokyo


----- Original message -----
From: "Doerr, Martin" <martin.doerr at sap.com<mailto:martin.doerr at sap.com>>
To: Michihiro Horie <HORIE at jp.ibm.com<mailto:HORIE at jp.ibm.com>>
Cc: "ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>" <ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>>, "hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>" <hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>>, Gustavo Romero <gromero at linux.vnet.ibm.com<mailto:gromero at linux.vnet.ibm.com>>, "Lindenmaier, Goetz" <goetz.lindenmaier at sap.com<mailto:goetz.lindenmaier at sap.com>>
Subject: RE: RFR : PPC64 : Need support for VSR spills in ppc.ad
Date: Wed, Jan 10, 2018 8:17 PM

Hi Michihiro,

thanks for implementing it.

I wonder how the content of R19 should get preserved.

Shouldn't repl4F_immF_Ex use a temp register instead of R19?

SuperwordUseVSX is still not activated in vm_version_ppc.cpp. I think we should turn it on with this change (see the TODO).

We will run tests when the R19 question is clarified.

Best regards,

Martin

From: Michihiro Horie [mailto:HORIE at jp.ibm.com]
Sent: Mittwoch, 10. Januar 2018 07:10
To: Doerr, Martin <martin.doerr at sap.com<mailto:martin.doerr at sap.com>>
Cc: ppc-aix-port-dev at openjdk.java.net<mailto:ppc-aix-port-dev at openjdk.java.net>; hotspot-dev at openjdk.java.net<mailto:hotspot-dev at openjdk.java.net>; Gustavo Romero <gromero at linux.vnet.ibm.com<mailto:gromero at linux.vnet.ibm.com>>
Subject: RFR : PPC64 : Need support for VSR spills in ppc.ad

Hi Martin,

Would you review the following change that fixes the SLP for PPC?

In this change, I added support for VSR spills. Also, I fixed how to specify registers in postalloc_expand for the float constant replication. I confirmed this change works with JTREG.

Bug: https://bugs.openjdk.java.net/browse/JDK-8194861<https://urldefense.proofpoint.com/v2/url?u=https-3A__bugs.openjdk.java.net_browse_JDK-2D8194861&d=DwMFAg&c=jf_iaSHvJObTbx-siA1ZOg&r=oecsIpYF-cifqq2i1JEH0Q&m=6pCaUQ4ll6S-hRCrDXsSKjl9NrczviVl3vOv0-KnizA&s=HtvmfHT8WviFbLWwFAnJ1KOrzvPiwqleI_inLbbJqTE&e=>
webrev: http://cr.openjdk.java.net/~mhorie/8194861/webrev.00<https://urldefense.proofpoint.com/v2/url?u=http-3A__cr.openjdk.java.net_-7Emhorie_8194861_webrev.00&d=DwMGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=oecsIpYF-cifqq2i1JEH0Q&m=SOGOvG8yFchCvPeC1Pr5_BJNIaBdg1G1IXaEeMgwpWI&s=RceHIxj8obNFdZExh6_lDeUnogfNv-qGZWuh_RRuHRQ&e=>/
(I created a webrev under jdk/hs.)

Best regards,
--
Michihiro,
IBM Research - Tokyo


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/ppc-aix-port-dev/attachments/20180112/657f9b9e/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.gif
Type: image/gif
Size: 105 bytes
Desc: image001.gif
URL: <http://mail.openjdk.java.net/pipermail/ppc-aix-port-dev/attachments/20180112/657f9b9e/image001.gif>