RFR(L): 8069539: RSA acceleration

[Viswanathan, Sandhya]

Hi Sergey,

Thanks for these performance numbers. It is very encouraging!

Best Regards,
Sandhya

-----Original Message-----
From: Sergey Kuksenko [mailto:sergey.kuksenko at oracle.com] 
Sent: Friday, March 13, 2015 6:31 AM
To: Viswanathan, Sandhya; hotspot-compiler-dev at openjdk.java.net
Subject: Re: RFR(L): 8069539: RSA acceleration

Fine Sandhya,

Besides SPECjvm2008 I did quick test on microbenchmarks.
Draft results (on Haswell):
BigInteger.modPow speedup    -  40-50%
BigInteger.nextProbablyPrime -  35%
BigInteger.pow               -  20-25%

BigInteger.TEN.pow()         - up to 35% speedup
(I checked this because of BigInteger.TEN pows frequently used in 
BigDecimal)

And even "new BigDecimal(double)" - got 8%-10% speedup.

Later I will provide more precise (and verified) results.

On 03/12/2015 12:36 AM, Viswanathan, Sandhya wrote:
> Hi All,
>
> We would like to contribute the Crypto RSA acceleration optimization
> from Intel.
>
> The contribution is towards the JEP 8046943 and Bug ID:  8069539.
>
> Please review this patch:
>
> JEP: https://bugs.openjdk.java.net/browse/JDK-8046943
>
> Bug-id: https://bugs.openjdk.java.net/browse/JDK-8069539
>
> webrev: https://bugs.openjdk.java.net/secure/attachment/26024/webrev.zip
>
> The optimization is achieved through intrinsics for BigInteger
> squareToLen and mulAdd methods.
>
> We see ~50% improvement for SPECjvm2008 crypto.rsa with this patch.
>
> Nils Eliasson has kindly offered to sponsor this patch.
>
> Best Regards,
>
> Sandhya
>

-- 
Best regards,
Sergey Kuksenko