5
\$\begingroup\$

Most FPGA developement boards have a 50 MHz clock source onboard. However, the FPGAs are typically able to work faster than this. For multiplying the clock speed it seems to be needed to use a custom vendor specific IP of the FPGA, that must be configured and placed in the design.

Is there a way around this? Can I program/define my own PLL in VHDL, that is preferably synthesized using the internal hard coded PLL blocks of the FPGA and works for Altera and Xilinx without having to manually insert the PLL to a project via MegaWizard & Cons?

\$\endgroup\$
4
  • 2
    \$\begingroup\$ clock generation and distribution is, if you like, a 'level below' the standard logic fabric. There is so much riding on the cross chip skew, and power dissipation associated with huge nets, that it's optimised to the Nth for any particular device, family, vendor etc etc. You could make a generic multiplier, but you'd have to route it through the dedicated clock buffers into the clocking infrastructure, and then it wouldn't perform anything like the dedicated resources. With clocking, you just have to suck up what you're given on a part by part basis. \$\endgroup\$
    – Neil_UK
    Commented Jan 21, 2019 at 9:22
  • 1
    \$\begingroup\$ Besides the input @Neil_UK gave, why is the the IP a problem for you? You could just generate the IP than look into the generated files and if you are lucky the part of the code you are looking for isn't enrypted. Also, you can just write a code wrapper, like with constants and generate statements f.E. which instantiates the component depending on your hardware. \$\endgroup\$
    – Eggi
    Commented Jan 21, 2019 at 10:01
  • \$\begingroup\$ Yeah, that would be a possibility. However, is there any way to avoid those IP-Managers? Something like "#ifdef XILINX" constructs for actually using the one or the other generated IP dependent? --- Main issue is: I want to avoid using any GUI wizzards, instead have VHDL and other source files that might be generated by some external tool. \$\endgroup\$
    – SDwarfs
    Commented Jan 21, 2019 at 10:04
  • \$\begingroup\$ If you want ifdef's you should switch to Verilog. And I agree the Xilinx GUI was great until some @~$£%$ [censored] decided to go back to schematics. \$\endgroup\$
    – Oldfart
    Commented Jan 21, 2019 at 10:22

1 Answer 1

4
\$\begingroup\$

You can do this by including vendor specific primitives in your code.

Both Xilinx and Altera have PLL primitives that can be instantiated in the source code with no gui.

Xilinx: PLL2_ADV and PLL_BASE (from page 351 onward)

Altera: altpll User guide

While the Altera AltPLL IP does have a gui (and they highly recommend you use it, due to the wide range of PLL configuration parameters they have available), you don't actually have to use the GUI. Or, you can use the GUI once and then just keep the vhdl file it generates. I tested this by removing the IP gui file (.qip) from the project and added the source only, and it synthesized fine.

In both cases, you'll need to include the libraries at the top of your code.

-- Altera
LIBRARY altera_mf;
USE altera_mf.all;


-- Xilinx
LIBRARY UNISIM;
USE UNISIM.vcomponents.all;

The biggest issue is differentiating between vendors. The --pragma directive doesn't have a way to distinguish between vendors, so you'll need to use generate statements to do so. See this answer for more info about that: what is #define equivalent in VHDL

\$\endgroup\$
4
  • \$\begingroup\$ Thanks! This should work for me. Also thanks for the pointers to the #define-equivalents. \$\endgroup\$
    – SDwarfs
    Commented Jan 21, 2019 at 12:34
  • 1
    \$\begingroup\$ One global generic definition of "IS_XILINX" or "IS_ALTERA" and a if-generate at the target system specific locations (as suggested by the linked Q&A) should do the trick. A global quick to find configuration point is acceptable in my setup. \$\endgroup\$
    – SDwarfs
    Commented Jan 21, 2019 at 12:38
  • \$\begingroup\$ @SDwarfs Just to be clear, the answer to the question in the title you wrote is no. Different "platforms", even those from the same vendor, may have different ways of specifying and configuring their PLLs. The approach suggested here works as long as you provide options for all "platforms". \$\endgroup\$ Commented Jan 21, 2019 at 14:37
  • \$\begingroup\$ @Elliot: Sure, that's surely why they provide the IP Wizards instead of giving us just the VHDL libraries. But, in a way this could be used to write a library that supports multiple FPGA variants having a common interface, where you only need to take care to change some generic constants to the right settings. It's not platform independent, but allows for better compatibility. \$\endgroup\$
    – SDwarfs
    Commented Jan 21, 2019 at 16:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.