Microsoft has recently announced at the AMD Fusion Developer Summit the introduction of a new technology called C++ Accelerated Massive Parallelism or shortly C++ AMP that helps C++ developers use the GPU for parallel programming. The new technology will be part of Visual C++, integrated with full support (edit, build, debug, profile) in the next version of Visual Studio. It is built in modern C++ on top of DirectX. It will provide an STL-like library, part of the Concurrency namespace, delivered in a amp.h header.

S.Somasegar (Senior Vice President of the Developer Division) said:

By building on the Windows DirectX platform, our implementation of C++ AMP allows you to target hardware from all the major hardware vendors. We expect that it will be part of the next Visual C++ compiler and fully integrated in the next release of Visual Studio experience.

You can read more about it here:

Looking forward for the first CTP and samples.

UPDATE
Herb Sutter introduces C++ AMP at the AMD Fusion Developer Summit 11

Daniel Moth digs deeper into C++ AMP with code samples and more

Herb Sutter

How long is your password? How long it will take a 100,000,000 GPU cores running at what, a million attempts per second to crack your password just by brute force? That where almost a kid can write that. An that is just a tiny example of how game changing this is.

, , , , Hits for this post: 42001 .

Channel9 recently posted a video with the Parallel Computing Concurrency Runtime team talking, mainly, about tasks and continuations, new features to the Parallel Patterns Library. These are already available through the ConcRT Extra’s Sample Pack. You can watch the half hour interview with the team here.

Besides the new stuff they shown, I particularly liked two things that Artur Laksberg said. The first was about the difference between parallelism and concurrency:

Parallelism is doing the same amount of work faster by utilizing multiple cores, multiple processors. Concurrency is understood as doing more work in the same amount of time.

The other one was about threads and tasks:

We want people to stop thinking about threads and start thinking in terms of independent, or not independent, units of work. You have one piece of work, you compose it with another piece of work and you have two tasks, you join together and what you get as a result is another task. And then, concurrency, as somebody said, just happens. It just happens if the runtime decides it’s beneficial for you, that it is safe to execute those two chunks, tasks, in parallel.

Hopefully people will start understanding that threads are obsolete and they should be thinking in tasks.

UPDATE: Microsoft Technical Computing group announced yesterday the availability of a book called Parallel Programming with Microsoft Visual C++: Design patterns for Decomposition, and Coordination on Multicore Architectures, describing six key patterns for data and task parallelism and how to implement them in VC++ using the Parallel Patterns Library and Asynchronous Agents Library, which shipped with Visual Studio 2010. There is also a printed version for the book. You can read more about it on VC++ team’s blog.

, , , , , Hits for this post: 35075 .

At the beginning of this year I have posted a short questionnaire on concurrency on my blog. I have promised to publish the results, so here they are. The number of respondents was 27 (and not everybody answered all the questions).

Q: What is your role in application development?

Q: How many cores has your primary working station (PC, laptop)?

Q: What type of application(s) are you developing?
A: Windows and web applications, ERP, banking, internal, back-end, integrations (addins: IE, Outlook, etc.), VxWorks kernel-base applications under PPC603, Point-of-Sale, internal, Web site + ERP backend + e-commerce, web applications, Windows applications, Mac applications, control software, visualizations, scada, Linux embedded application, Windows forms, Windows .Net with WPF/WCF and Silverlight, integration technology platforms for CRM and ERP, Kernel mode drivers, filesystem drivers, graphic engines, CAD.

Q: What programming languages are used for your application development?

Q: One a scale from 1 to 5 how do you rate your level of awareness on concurrency issues?

Q: On a scale from 1 to 5 how do you rate your activity in getting informed and acquiring competence on concurrency (trends, tools, etc.)?

Q: What do you expect to happen to developers that won’t learn to program concurrently?

Q: On a scale from 1 to 5, how would you grade the level of parallization of your application?

Q: Does your application performance scale proportionally with the number of cores?

Q: What libraries and tools are used in your application for concurrency?
A: System.Threading, The built-in threading classes, OS support (CRT, Win32 API), VS2010, .Net Framework – Thread, ThreadPool, Task, Monitor, lock() and many own written libraries (synchronized objects, asynchron calls and functions), MFC framework, .NET framework, Qt framework, Java framework, My own using RAII, nothing special: self made priority queues and thread pools, Win32 API’s, Windows threads.

Q: Give examples of routines in your application that are executed in worker threads.
A: loading of grids which care displayed at startup in the main application window, logging, tracing, loading data (in some cases), long running tasks (computations, report generation also in some cases), HTTP lenghtly posts, Outlook inbox scan for text, a job/task manager for things that are triggered by web UI but are not required to output any result to the client, data processing, database interactions, database data loading/uploading, looped data acquisition presented in chart monitors, sattelite data scaning and sectioned for DVB-S systems, any operations where I need to start/stop, all long running operations, application request processing, connection management, I/O, Open large files, communication, file uploads, update checking.

Q: Give examples of routines in your application that execute in the main (UI) thread and can be refactored to execute in worker threads.
A: loading all application modules, SQL queries, draw rows of panels – maybe even threading each individual block per row, all data processing which take more than a few seconds and which are not dependent or influence other events on UI, I’ve been multi-threaded programming for so long, I usually design the app for multi-theading up front, long (minutes) calculations, user triggered opening of multiple files, opening of a single file, certain post-processing of data, database stuff.

Q: Give examples of routines in your application that are parallelized.
A: logging and Tracing with multiple outputs (File System, Debug Window, Activity Logs written to database), UI, background processes, sending emails, invokes an event multithreaded in an own ThreadPool, all data processing in an application running in a sattelite set top box, zapit-thread, epg-thread, timer thread, other threads controlling whichever driver involved: front panel display driver, sound driver etc., remote connection handling, data processing, 3d transformations.

Q: Give examples of routines in your application that could be parallelized to improve performance.
A: cost computation and distribution for productions chains, the rest, all timers routines should turn to threads, all routines that take more than a few seconds and locks the UI, send-request-and-wait-response logic (actually I was thinking of SEDA kind of processing), importing models from project files.

, , , Hits for this post: 26737 .

I think application development faces two challenges nowadays: 64-bit and multi-core/many-core hardware. Switching from 32 to 64 bit is just another step in the evolution of processors. There was a time when we switched from 8 to 16, and then from 16 to 32. There are problems that arise every time, but probably in 10-20 years we will have 128 bit platforms. On the other hand, multi-core/many-core is a different shift not only in development but also thinking. We either run one core or multiple core processors. My working stations has 8 cores (4 physical and 4 virtualized) and so does my laptop. Still, I don’t see an 8 times improvement of the applications running on these machines; yet I know 8 times I not what I should expect. But not even 4 times. For instance, the time for building from scratch the application I’ve working on with Visual Studio 2008 dropped from 20 minutes to 8 minutes; that’s a 2.5 improvement. I know that having N cores doesn’t mean that applications can run N times faster, because not everything can run in several threads, and then we have problems with resource access, synchronization and others. All these prevent applications run N times faster. But the problem is that we are not thinking in parallel. We are still used to program in a single thread; and when I say that I mean not only most developers are not used to do parallelization for boosting performance of some routines, but also many operations are run in the main (UI) thread, making that GUI freeze.

Therefore I would like to get some feedback from people working in applications development to get an idea about awareness, issues, solutions regarding concurrency. Please take several minutes to answer the questions in this survey.

I’d like to quote James Reinders, lead evangelist and Director of Marketing and Business Development for Intel Software Development Products, who said that:

I am still confident that software development in 2016 will not be kind to programmers who have not learned to “Think Parallel.”

The “not parallel” era we are now exiting will appear to be a very primitive time in the history of computers when people look back in a hundred years. The world works in parallel, and it is time for computer programs to do the same.

Doing one thing at a time is “so yesterday.”

The questionnaire bellow is also available here.

Thank you for answering the questionnaire.

, , , Hits for this post: 27978 .