RFR(M): 8152172: PPC64: Support AES intrinsics

Wed Mar 30 10:29:31 UTC 2016

Hi everybody,

here finally comes the updated version of Hiroshi's change (sorry for
the delay):

http://cr.openjdk.java.net/~simonis/webrevs/2016/8152172.hs/
http://cr.openjdk.java.net/~simonis/webrevs/2016/8152172.jdk/

With regard to the previous version it contains three small updates:
- changed the type of AESCrypt::sessionK from Object[] to int[][]
- also updated the type of the AESCrypt::sessionK field to "[[I" in
LibraryCallKit::get_key_start_from_aescrypt_object() where it gets
loaded.
- protected the generation of the AES-stubs by "if (UseAES)" tp
prevent them from beeing generated on ppc64 big endian.

We need a sponsor to push these two changes in sync to hs-comp.

Thank you and best regards,
Volker

On Tue, Mar 29, 2016 at 10:58 PM, Anthony Scarpino
<anthony.scarpino at oracle.com> wrote:
> Volker,
>
> Changing this with your other changes in hs-comp repo is best.. Just let me
> know when you are ready for code review on this piece.
>
> thanks
>
> Tony
>
>
>
>
> On 03/29/2016 06:05 AM, Volker Simonis wrote:
>>
>> Hi Anthony, Vladimir,
>>
>> thanks for your evaluation. Should I open a new bug for the change in
>> AESCrypt or can we do it under the same bug id like the ppc-specific
>> AES intrinisc (i.e. "8152172: PPC64: Support AES intrinsics") ?
>>
>> I also think that the AESCrypt::sessionK type change change, although
>> it is a class library change, should be done right in hs-comp together
>> with the new ppc64 AES intrinsics. Do you agree?
>>
>> Regards,
>> Volker
>>
>>
>> On Mon, Mar 28, 2016 at 7:41 PM, Anthony Scarpino
>> <anthony.scarpino at oracle.com> wrote:
>>>
>>> [Switching to security-dev, core-lib-dev was bcc'ed]
>>>
>>> I don't see any problem changing it to int[][].  Being an Object[] is odd
>>> in
>>> my opinion.
>>>
>>> Tony
>>>
>>>
>>> On 03/25/2016 04:00 PM, Vladimir Kozlov wrote:
>>>>
>>>>
>>>> This question is for core libs.
>>>>
>>>> For JIT to generate optimal code we want to change type of
>>>> AESCrypt::sessionK from Object[] to int[][].
>>>> Otherwise JIT have to generate checkcast code to make sure that elements
>>>> of sessionK array are int[].
>>>> I see that sessionK field is private so we are not changing public API.
>>>>
>>>> Thanks,
>>>> Vladimir
>>>>
>>>> On 3/24/16 6:17 PM, Hiroshi H Horii wrote:
>>>>>
>>>>>
>>>>> Hi Vladimir,
>>>>>
>>>>> Thank you for your reviewing.
>>>>>
>>>>> Is it possible to modify the type of sessionK from Object[] to int[][]?
>>>>>
>>>>> We can guess, the type was designed for flexibility. However, only
>>>>> int[] is used
>>>>> to store elements of sessionK, so far. In addition, this field is
>>>>> private.
>>>>> Though adding checkcast is another solution, it produces overheads.
>>>>>
>>>>> I attached an additional patch (generated with hg diff -g) to jdk
>>>>> directory.
>>>>> I would like to ask thoughts about this change of java code.
>>>>>
>>>>>
>>>>>
>>>>> Regards,
>>>>> Hiroshi
>>>>> -----------------------
>>>>> Hiroshi Horii,
>>>>> IBM Research - Tokyo
>>>>>
>>>>>
>>>>> Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote on 03/24/2016
>>>>> 09:09:11:
>>>>>
>>>>>   > From: Vladimir Kozlov <vladimir.kozlov at oracle.com>
>>>>>   > To: Hiroshi H Horii/Japan/IBM at IBMJP, "hotspot-compiler-
>>>>>   > dev at openjdk.java.net" <hotspot-compiler-dev at openjdk.java.net>
>>>>>   > Cc: "Doerr, Martin" <martin.doerr at sap.com>, Tim Ellison
>>>>>   > <Tim_Ellison at uk.ibm.com>, "Simonis, Volker"
>>>>> <volker.simonis at sap.com>
>>>>>   > Date: 03/24/2016 09:19
>>>>>   > Subject: Re: RFR(M): 8152172: PPC64: Support AES intrinsics
>>>>>   >
>>>>>   > I think we need to add gen_checkcast for objAESCryptKey node since
>>>>>   > java code has no guarantee that sessionK array elements are
>>>>>   > initialized with int[]. Or we should modify java code to declare
>>>>> sessionK
>>>>>   > as int[][].
>>>>>   >
>>>>>   > Thanks,
>>>>>   > Vladimir
>>>>>   >
>>>>>   > Note, intrinsics are correctly handle case when
>>>>>   > On 3/22/16 8:47 AM, Hiroshi H Horii wrote:
>>>>>   > > Dear all:
>>>>>   > >
>>>>>   > > Can I please request reviews for the following change?
>>>>>   > > This change was created for JDK 9 to enable POWER8's AES
>>>>>   > > instructions for AES calculation.
>>>>>   > >
>>>>>   > > This request follows this discussion.
>>>>>   > > http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2016-
>>>>>   > March/021926.html
>>>>>   > >
>>>>>   > > Description:
>>>>>   > > This change adds stub routines support for single-block AES
>>>>>   > > encryption and decryption operations on the POWER8 platform.
>>>>>   > > They are available only when the application is configured to
>>>>>   > > use SunJCE crypto provider on little endian.
>>>>>   > >
>>>>>   > > These stubs make use of efficient hardware AES instructions
>>>>>   > > and thus offer significant performance improvements over
>>>>>   > > JITed code on POWER8 as on x86 and SPARC. AES stub routines
>>>>>   > > are enabled by default on POWER8 platforms that support AES
>>>>>   > > instructions (vcipher). They can be explicitly enabled or
>>>>>   > > disabled on the command-line using UseAES and
>>>>>   > > UseAESIntrinsics JVM flags. Unlike x86 and SPARC, vcipher
>>>>>   > > and vnchiper of POWER8 need the same round keys of AES.
>>>>>   > > Therefore, inline_aescrypt_Block in library_call.cpp calls the
>>>>>   > > stub with AESCrypt.sessionK[0] as round keys.
>>>>>   > >
>>>>>   > > Summary of source code changes:
>>>>>   > >
>>>>>   > >   *src/cpu/ppc/vm/assembler_ppc.hpp
>>>>>   > >   *src/cpu/ppc/vm/assembler_ppc.inline.hpp
>>>>>   > >     - Adds support for vrld instruction to rotate vector register
>>>>> values
>>>>>   > >        with left doubleword.
>>>>>   > >
>>>>>   > >   *src/cpu/ppc/vm/stubGenerator_ppc.cpp
>>>>>   > >     - Defines stubs for single-block AES encryption and
>>>>> decryption
>>>>>   > >        routines supporting all key sizes (128, 192 and 256-bit).
>>>>>   > >     - Current POWER AES decryption instructions are not
>>>>> compatible
>>>>>   > >        with SunJCE expanded decryption key format. Thus
>>>>> decryption
>>>>>   > >        stubs read the expanded encryption keys (sessionK[0]) with
>>>>>   > >        descendant order.
>>>>>   > >     - Encryption stubs use SunJCE expanded encryption key as
>>>>> their
>>>>>   > >        is no incompatibility issue between POWER8 AES encryption
>>>>>   > >        instructions and SunJCE expanded encryption keys.
>>>>>   > >
>>>>>   > >   *src/cpu/ppc/vm/vm_version_ppc.cpp
>>>>>   > >     - Detects AES capabilities of the underlying CPU by using
>>>>>   > >        has_vcipher().
>>>>>   > >     - Enables UseAES and UseAESIntrinsics flags if the underlying
>>>>>   > >        CPU supports AES instructions and neither of them is
>>>>> explicitly
>>>>>   > >        disabled on the command-line. Generate warning message if
>>>>>   > >        either of these flags are enabled on the command-line
>>>>>   > >        whereas the underlying CPU does not support AES
>>>>> instructions.
>>>>>   > >
>>>>>   > >   *src/share/vm/opto/library_call.cpp
>>>>>   > >      - Passes the first input parameter, reference to sessionK[0]
>>>>> to the
>>>>>   > >         AES stubs only on the POWER platform.
>>>>>   > >
>>>>>   > >    *src/share/vm/opto/graphKit.cpp
>>>>>   > >      - Supports T_NARROWOOP type for
>>>>> GraphKit::load_array_element.
>>>>>   > >
>>>>>   > > Bug: https://bugs.openjdk.java.net/browse/JDK-8152172
>>>>>   > > Webrev:
>>>>> http://cr.openjdk.java.net/~mdoerr/8152172_ppc64le_aes/webrev.00/
>>>>>   > >
>>>>>   > >
>>>>>   > > Regards,
>>>>>   > > Hiroshi
>>>>>   > > -----------------------
>>>>>   > > Hiroshi Horii,
>>>>>   > > IBM Research - Tokyo
>>>>>   > >
>>>>>   >
>>>>>
>>>
>