Help me implement high accuracy system with XUP boards

Hello everyone,

I have got two Xilinx boards: XUPV2-Pro and XUPV5-LX110T. And I want to implement an optimized system which needs calculations as for instance: sine, cosine, square root, div, mul and matrix inversion. It also requires a high accuracy in the data, I mean, in some cases it needs computing with ten decimals.

I have a bit experience in VHDL and PPC (Adding customs IP's to the PLB and OPB) with the XUPV2P. I would say that I am more familiar using the XUPV2P than XUPV5L. However, I have not done systems which require high accuracy using FPGA (should I use Floating-point or Fixed point?).

I have idea about how could be the hardware design of the system (creating customs IP's), but I have many doubts regarding the platform (XUPV2 or XUPV5?), the implementation because of the high accuracy and the tools that I must use. I would like to make a compact, efficient and high accuracy system based on FPGA. So I would be grateful if anyone could give any guidance or advice.

Thanks in advance,
Alex

Re: Help with XUP boards

I am not sure about Virtex-II, but afaik it is not supported by the newest version of ISE, so it would be better to use Virtex-5, because the newest ISE should support VHDL-2008 and new math libs. Usually, for simple maths numeric_std lib is used, but there are more libs now, check them **broken link removed**.

I suppose ISE doesn't support VHDL2k8. Definitely XST doesn't support it. So for it's only Riviera-PRO/Active and Synplify I think.

PowerPC looks nice considering floating point operations. However you do not specify how fast you want to calculate.

I cant tell you which board you should choose but if you choose XUPV5 check FPGA Developer

Thanks for the answers.

it would be better to use Virtex-5, because the newest ISE should support VHDL-2008 and new math libs for simple maths numeric_std lib is used, but there are more libs now, check them HERE

Click to expand...

I also think that Virtex-5 would be a good choise, and although I haven't had many experience with this board it seems similar to the Virtex-2P in some things. Regarding the link, I didn't see any source code debugged for the tool that I have Xilinx 10.1. Anyway, I think it could be helpful for my system.

Definitely XST doesn't support it. So for it's only Riviera-PRO/Active and Synplify I think

Click to expand...

You mean that I can't create a VHDL design with VHDL2k8 and add it as peripheral in the PPC or microblaze, don't you?

you do not specify how fast you want to calculate

Click to expand...

The data that I obtain has 1KHz (sampling frequency) but the signal processing has a high computational cost. So I think that the PPC would be suited, however I want a flexible system which can be adapted to different architectures.

I read about the System Generator for digital signal processing, what do you think about this tool?

Thanks!!
Alex

Alex02 said:

You mean that I can't create a VHDL design with VHDL2k8 and add it as peripheral in the PPC or microblaze, don't you?Alex

Click to expand...

Yes. Ive just tried to use simple 'process (all)' in ISE 12.3 but it didnt synthesize because of the error. However you can add new fixed_pkg or float_pkg packages to your designs and synthesize it. More you can read here:
**broken link removed**

Alex02 said:

I read about the System Generator for digital signal processing, what do you think about this tool?

Click to expand...

Its a pretty nice tool. There are a lot of examples/workshops on Xilinx sites considering SG and XUPV5 even with HIL - hardware in the loop but using only JTAG (still nice).

I think I will use XUPV5. Regarding VHDL-2008 and the tools: System Generator, Riviera-PRO/Active and Synplify, I will see how I can use them in order to implement the system. Is there another tool or thing that I should consider to do sine, x^(3/2), div, mul with floating point operators using this board?

Thanks!
Alex

Hi, guys ,i have just got the XUPv5 board , and i used to use the Virtex 4 FPGA and ISE 9.1. I like ISE 9.1, but there's some error inferred while i mapped the program in Virtex 5 device, so i would like to change to a latest one. In my opinion , ISE 12.3/ISE 12.4/ISE 13.1/ISE 13.2 would be good. By the way ,i implemented the maths calculation in Virtex 4 device several months ago , to finish my wave control project.And i used the Cordic IP core supplied by ISE (core generator). You could get your accuracy, cause the IP core could support up to 32 bits. Before implementing the maths algorithm with the Cordic IP core , just take a look at the PDF from Xilinx, called "Xilinx logic core, Cordic v3.0", and you could find the way to use it. The data format the core used is just quite special. If you could not find the PDF, just write a mail to rocky.rockychan@163.com, and would send you the document. Also, i used the fpu IP by ISE, and it's quite useful. Just have a try.

Hi rockybc,

Thanks for the comment, I'll see what you say. It is worth saying that I've found tools like flopoco which is interesting to do floating point cores and there is also a FPU core in opencores.