A&O Jérémie Laval 2009-07-21: Improving, testing and integrating ParallelFx into Mono

[Jérémie Laval]

Ahoy,

Achievement
-----------------

- Finish ReaderWriterLockSlim. [DONE]
- Re-enable Parallel.ForEach loop. [DONE]
- Write Barrier class. [DONE]

Objectives
--------------

- Start PLinq rewrite with focus on using Partitioner internally
- Prepare and send patches to devel-list for System.Threading.Tasks namespace
- Implement the rest of the Lazy classes.

Bonus
--------

- LazyInitializer class.

- Good deal of work on speeding up Moonlight : http://garuma.wordpress.com/2009/07/16/joining-the-moonlight-fun/ . Coming is also an improved ShivaVG with PBO&VBO (data on GPU side) and accelerated math operations à la Mono.Simd

Cheers!

--
Jérémie Laval
jeremi...@gmail.com
http://garuma.wordpress.com

[Miguel de Icaza]

Hello,

- Finish ReaderWriterLockSlim. [DONE]

Mhm, we already have an implementation of this on SVN:

 /mcs/class/System.Core/System.Threading/ReaderWriterLockSlim.cs

It is based on Vance's description of what this was supposed to do.

- Good deal of work on speeding up Moonlight : http://garuma.wordpress.com/2009/07/16/joining-the-moonlight-fun/ . Coming is also an improved ShivaVG with PBO&VBO (data on GPU side) and accelerated math operations à la Mono.Simd

Oh!   Love to hear more about this!

Miguel

[Jérémie Laval]

Hi Miguel !

On Wed, Jul 22, 2009 at 12:10 AM, Miguel de Icaza <miguel....@gmail.com> wrote:

Hello,

- Finish ReaderWriterLockSlim. [DONE]

Mhm, we already have an implementation of this on SVN:

 /mcs/class/System.Core/System.Threading/ReaderWriterLockSlim.cs
 

It is based on Vance's description of what this was supposed to do.

Yup, I used a different algorithm for mine (http://research.sun.com/scalable/pubs/SPAA2009-ScalableRWLock.pdf) and we are just going to use the one which perform the best for most people. If the testing is going the same way as in the paper, the performance of this rwl should be better than the other when around 5-7 threads are using the lock depending on the scenario.

 

- Good deal of work on speeding up Moonlight : http://garuma.wordpress.com/2009/07/16/joining-the-moonlight-fun/ . Coming is also an improved ShivaVG with PBO&VBO (data on GPU side) and accelerated math operations à la Mono.Simd

Oh!   Love to hear more about this!

PBO is done albeit a small bug with one of the blending mode (but doesn't really interfere with cairo openvg backend anyway). I spend some time today speeding up the math operations with liboil (http://liboil.freedesktop.org/wiki/) and I should get VBO soon too (tough I need to test some stuff beforehand because on my system it does the bizarre thing of being slower than a traditional vertex array).
 

Miguel

Cheers !