[Haskell-iPhone] FFI calls into Haskell taking a long time

[James Parker]

Hi,

I’ve been looking into this some more. It looks like generatePrime <http://hackage.haskell.org/package/crypto-numbers-0.2.7/docs/src/Crypto-Number-Prime.html#generatePrime> is causing the slowdown. I manually added some benchmarking and logging to this function and its dependencies (I removed the gmp code). Here is the output from the iOS version: 

http://lpaste.net/6160729019554725888 <http://lpaste.net/6160729019554725888>

Here is the output run natively on my machine: 

http://lpaste.net/2971101072195584000 <http://lpaste.net/2971101072195584000>

As you can see, the iOS version is much slower (about 75x). What differences are there when cross-compiling for iOS that might explain this difference? It looks like the iOS version might be using integer-simple instead of gmp? Are there any other differences?

Thanks!

James

> On Apr 8, 2016, at 2:47 PM, James Parker <jp at jamesparker.me> wrote:
> 
> Hi,
> 
> Sorry, it took a while to get running on the device. Yes, I still get the slowdown on the device. XCode is reporting 99-100% CPU usage while memory stays pretty consistent (around 18MB). This is similar to what I was seeing on the simulator. 
> 
> Are there any tools to benchmark where time is being spent on the device? Is there some way to enable -enable-library-profiling? 
> 
> Thanks,
> 
> James
> 
>> On Apr 6, 2016, at 7:45 PM, Manuel M T Chakravarty <chak at justtesting.org> wrote:
>> 
>> Did you try on the device? The simulator is a strange beast and its performance may not be representative of what you get on a real device.
>> 
>> Manuel
>> 
>>> James Parker <jp at jamesparker.me>:
>>> 
>>> Hi,
>>> 
>>> I’m exporting a Haskell function that generates RSA keys for an iOS application. I’m using the crypto-pubkey and crypto-random packages. Unfortunately, the key generation is taking an extremely long time (around 5 minutes) when run in the simulator (XCode 7.2.1). When running Haskell code through Criterion, key generation only takes 110ms. When benchmarking a c program that links the Haskell FFI library (on my x86 laptop), key generation takes about 130ms. 
>>> 
>>> Does anyone have any ideas why this is so much slower on the simulator? I thought maybe there was an issue with running out of randomness, so I tried creating a fixed seed by using `createTestEntropyPool`, but this did not make any difference. I’ve included the relevant key generate function below. 
>>> 
>>> gen :: MonadRandom m => m (RSA.PublicKey, RSA.PrivateKey)
>>> gen = withCPRG $ \prg -> return $ RSA.generate prg 256 65537
>>> 
>>> instance MonadRandom IO where
>>>  type MonadCPRG IO = SystemRNG
>>> 
>>>  withCPRG f = do
>>>      entropy <- createEntropyPool
>>>      (res, _) <- f $ cprgCreate entropy
>>>      return res
>>> 
>>> I’d appreciate any help.
>>> 
>>> Thanks!
>>> 
>>> James
>>> _______________________________________________
>>> iPhone mailing list
>>> iPhone at haskell.org
>>> http://mail.haskell.org/cgi-bin/mailman/listinfo/iphone
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/iphone/attachments/20160415/d70f453c/attachment.html>