[9] RFR (L) 8143012: CRC32 Intrinsics support on SPARC

james cheng james.cheng at oracle.com
Fri Nov 20 17:32:50 UTC 2015

StubRoutines::Sparc::_crc_by128_masks[] seems not being used directly,
but majority of the comment for it would apply to the CRC32_CONST_* in


On 11/20/2015 7:55 AM, Ahmed Khawaja wrote:
> By having separate tables it gives us the freedom to use different types of loads, which is something we are interested in doing in the future (possibly when newer
> instructions are available) and merging them now would only make that more difficult in the future. Also I believe the way the code is written now, only one table is
> instantiated during build/runtime, so there is no memory overhead. I believe keeping it as is is advantageous from a code maintenance perspective and incurs no overhead.
> Ahmed
> On 11/20/2015 9:51 AM, Vladimir Kozlov wrote:
>> Are they really the same (little vs big endians)?
>> We have crc32 implementation on other platforms too
>> http://hg.openjdk.java.net/jdk9/jdk9/hotspot/rev/3b81bc9fe683
>> Thanks,
>> Vladimir
>> On 11/20/15 7:41 AM, Roland Westrelin wrote:
>>>> http://cr.openjdk.java.net/~kvn/8143012/webrev/
>>> That looks good to me but aren’t:
>>> Aren’t StubRoutines::Sparc::_crc_by128_masks[] and StubRoutines::Sparc::_crc_table[] the same as StubRoutines::x86::_crc_by128_masks[] and
>>> StubRoutines::x86::_crc_table[] and can’t we do better?
>>> Roland.

More information about the hotspot-compiler-dev mailing list