GPGPU and x64 Multicore Programming with Accelerator from F#

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Accelerator v2 vs TPL — Sat, 29 Dec 2012 17:39:06 GMT

It seems that Accelerator v2 is superior to Task Programming Library (msdn.microsoft.com/.../dd460717.aspx). Do you agree? Does TPL have any advantages over Accelerator v2?

It seems that Accelerator v2 should eventually team up or somehow make its way into standard .NET somehow so that vectorization and making use of SIMD/SSE and multiple GPUs..

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Joy — Sat, 26 Feb 2011 05:43:08 GMT

You need the managed assembly - Accelerator v2/bin/Managed

VisualStudio\accelerator-intro\CalculatingPI\Microsoft.Accelerator.dll

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Joy Chakraborty — Wed, 29 Sep 2010 05:26:28 GMT

I can't add a reference to Accelerator.dll in VS 2010 F# project. Any idea why?

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Satnam_Singh — Tue, 22 Dec 2009 13:44:44 GMT

Alexnader,

I think you may have missed out the final 0? The main function needs to return an int value. That would expalin the error message.

Cheers,

Satnam

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Satnam_Singh — Tue, 22 Dec 2009 10:39:29 GMT

Alexander,

I am very sorry for the problems that you are encoutering and I've had a similar report from one other person. I checked again and the code at the top of this post works for me using Visual Studio 2010 Beta 2 and Visual Studio 2008 with F# 1.9.7.8. Please can you email me at satnams@microsoft.com to let me know exactly what version of Visual Studio and F# you are using and I shall do my best to try and get to the bottom of this issue. Once again, profuse apologies for the problems you are encountering.

Cheers,

Satnam

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Alexander Solonsky — Tue, 22 Dec 2009 04:20:59 GMT

the code just doesn't work out of the box.

on x86 environment, winXP

in line

printfn "DX9 --> \r\n%A" resDX

there is an error

Error 1 Type mismatch. Expecting a 'a -> int but given a 'a -> unit The type 'int' does not match the type 'unit' C:\Documents and Settings\Administrator\My Documents\Visual Studio 2010\Projects\f1\f1\Program.fs 28

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Satnam_Singh — Thu, 17 Dec 2009 16:47:12 GMT

Hello Jurgen.

You should be able to omit the FLoatParallelArray type declaration and write:

let rec expr (i : int) =

let e = ParallelArrays.Shift(inputArray, [| -i |]) * kernel.[i]

if i = 0 then e else e + expr (i-1)

In this case you need to give the type of i because there are several possible overloads for ParallelArrays.Shift(inputArray, [| -i |]) and the algorithm used for type inference can’t pick a specific one automatically.

re: GPGPU and x64 Multicore Programming with Accelerator from F#

Jurgen — Thu, 17 Dec 2009 15:06:07 GMT

Hi Satnam,

Very cool blog post. I was playing around with the code sample in VS 2010 but ran into some type inference issues. For the 1D sample on the line

kernel.[i] * ParallelArrays.Shift(inputArray, -i )

F# didn't seem to recognize the -i argument properly ([| -i |] does work) and also it didn't recognize the multiplication operator. I was able to get it to work using

let rec expr (i : int) =

let e = kernel.[i] * (ParallelArrays.Shift(inputArray, [| -i |]) : FloatParallelArray )

if i = 0 then e else e + expr (i-1)

but surely there must be a better way?!? I look forward to play with this more!

Jurgen

re: GPGPU and x64 Multicore Programming with Accelerator from F#

billh — Thu, 17 Dec 2009 08:44:59 GMT

FYI: Although this feed is visible in Google Reader, the post is not. I've not tried other readers, so I don't know whether it affects them as well or not.

re: GPGPU and x64 Multicore Programming with Accelerator from F#

honu — Wed, 16 Dec 2009 11:15:55 GMT

Thanks Satnam.

If you read this excellant blog, you'll probably also enjoy " ...Can non-numerical application developers take advantage from the new LRBni instructions?, by Daniele Paolo Scarpazza

— a member of the Multicore Computing Department at the IBM T. J. Watson Research Center. He

is the author of articles and scholarly papers on parallel algorithms and performance optimization on emerging multicore architectures

http://i.cmpnet.com/ddj/digest/2009/DDD_1209.pdf