Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
9 years ago

Windows vs Linux tool set

I haven't found a discussion of this anywhere. Please point me to the thread if one exists.

I want to know whether the Windows based tools and the Linux based tools will produce equivalent (not necessarily identical)

bitstreams. This goes for HDL and OpenCL inputs.

In the case of HDL, are the input files the same? I am assuming that the GUI for Quartus Prime Pro would actually

present the same display. (Valid assumption?) But does code underneath produce the same results?

Thanks,

Dan Poznanovic

Cray Inc.

7 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    I only use HDL - these are identical toolsets. You can even use both together when doing seed sweeps (set up windows and linux machines as possible build nodes from a single windows/linux master.).

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    They do not produce identical results. The difference comes from a minor difference in floating point results that affects the fitter. Across a bunch of seeds, the averages will be identical, and for any given seed it will be random which one is better or another, but if you get a result on Linux and for some reason need to get identical results on Windows, you won't.

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    You will not even get the some results on different Windows machines. There is something in the fitter that is machine depended.

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    You should get the same results on different windows, at least that used to be the case and I'm fairly certain I've seen it recently. I have heard that changing the number of processors used can change the result, but not positive. (It was kind of a weird thing, whereby if it was set to maximum allowed, you're supposed to get the same result even if the machines have different# of procs, but if you manually set it to different numbers, you get different results.) I never looked into it enough to be sure.

    Most of the time, I don't think it's that important. The difference is just a seed variation. If there is a build that is being released to production, I personally recommend zipping it up, along with the /db and /incremental_db. It may be large, but it completely removes any issue of "is this reproducible". Even if it is reproducible, it's a lot faster to unzip it and open it rather than running a whole compile.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    As for equivalence, it depends on how good you set your timing requirements. As Quartus will produce a different placement and routing, the resulting configurations will have different timing. If you have completely defined your timing requirements and Timequest says in both cases that they are met, then you can say that both bitsreams are equivalent. But if you forgot some requirements, or if some have a wrong value, then you could see a difference in timing between a configuration generated from Windows and one from Linux.

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    9 years ago

    Daixiwen,

    Good point. A correctly constrained design that meets timing should be equivalent even with different place-and-routes. If two designs meet timing but behave differently, it's more important to debug what is wrong than trying to get the two compiles to have identical results. (I'm not sure if this is the original poster's issue, just bringing it up. There are reasons that being able to get the same results is important, but there have been many times I've seen user's more concerned about this than they probably should be...)