次の方法で共有

Yet Another Technical Support’s added value: Engagement of Dev Teams

  • [アーティクル]
  • Some bloggers within Technical Support
  • Example of added value given by Technical Support: Devs’ engagement 
  • Comments about NETCF on ARMv6 and ARMv7

I often have to do some marketing for Microsoft Technical Support , because it’s something everyone with questions or problems around Microsoft products must test out to understand its added value… Apart from myself smile_regular we’re plenty of great people on all technologies: to name just a few, the following are representatives of the Technical Support – watch out their blogs, as they’re so plenty of “knowledge”! smile_nerd

There are many others I simply hadn’t yet the honor to work with, I simply couldn’t mention all of them!

So, what is the added value I want to present today? This time I’d like to demonstrate that one of benefit of using Technical Support is that we may have a preferential engagement with the Product Groups, when necessary. This may differ in forms depending on the technology, but no doubt that we can engage the people who actually wrote the software. And an example of this happened to me some weeks ago…

In that particular case who opened a Service Request was the OEM of a Windows CE 6.0-based device: you can say that I help ISVs, but this particular OEM was posing questions about developing a NETCF v3.5 application and wanted to receive support about Application Development rather than OS Customization through Platform Builder, therefore it fell in area of expertise. The device was based on a ARMv7 processor and was used as navigation system, hence it was heavily dependent on mathematical functions. On this processor, the OEM noticed that calling the same mathematical functions in a NETCF application required much more time than in a native application. The impression was that the native application leverages on the mathematical co-processor and asked if there’s any way to configure the NETCF Runtime to do the same either at application-layer or system-wide through an OS Customization.

To make it short, in this case I had the pleasure to work directly with Abhinaba Basu, member of the NETCF Dev Team – see his blog posts as a result of this co-operation:

Apart from the deep explanations given by Abhinaba in his blog, questions and answers were:

1- Is this an expected result?
NetCF 3.5 lacks Armv6/v7 specific optimizations as it is based on Armv4i. In particular we don’t leverage the FPU (Floating Point Unit), as NETCF does only floating point emulation. Hence NetCF math functions will be much slower than doing the same in a native C++ program, which in contrast could be compiled to use the hardware FPU.

2- Is ARMv7 a fully-supported architecture by the NETCF Runtime?
NetCF 3.5 is based on Armv4i and Armv7 is backward compatible with Armv4i. Hence all of the functionality of NetCF3.5 should work fine though may not yield optimal performance.

3- Is there any possible way to achieve similar results with NETCF?
Math routines could be written in native and the same could be PInvoked from managed code. Obviously you need to ensure that the cost of P/Invoke marshalling is not offsetting the saving in computation perf. A good idea to do that is to reduce the number of P/Invoke calls by bulking the computation.

In particular, in that specific case the solution proposed was to develop a native DLL that internally invoked native math functions and exposes some “math expressions” to a .NET client. I mean a native DLL exposing f(a,b,c) which internally is like {[sin(a) x sin(b)] / cos(c)}. Moreover, if for example you know that you need to do 3 of these operations then create a buffer of inputs and outputs and push that to the native dll so that it can do all 3 in one go.

 

Cheers,
~raffaele

Comments