Linux/Windows bitfile equivalence?

--- Quote Start ---

The Linux and Windows tools don't produce the same bitfile (either sof or jic) - is this to be expected?

The tools I'm using are:

XP SP3, 32-bit: 12.1/SP1 build 243

Linux x86_64: 12.1/SP1 build 243, 32-bit

There are lots of differences in the output - it's not just a date, for example.

--- Quote End ---

From memory some of the tools include the tool version in the sof/pof so differing output files isn't surprising.

(I think if you should get an identical file out of each tool when it's run if all inputs are the same).

Nial.

--- Quote Start ---

From memory some of the tools include the tool version in the sof/pof so differing output files isn't surprising.

Nial.

--- Quote End ---

No, the tool versions are the same - build 243, both 32-bit, downloaded together. There are lots of differences - about 10% of all bytes for the SOF file, and a bit less than 2% for the JIC file, so it's not a date/version number/etc thing. It's a bit surprising - maybe the code calls a platform-specific random number generator (if it does, it shouldn't, in my opinion).

--- Quote Start ---

It doesn't use any random generator, all the "randomness" is generated from the seed.

--- Quote End ---

In normal usage, you might expect a 'seed' to seed a random number generator, implying the existence of an actual generator. A single seed by itself isn't of much value.

--- Quote Start ---

The real question is why is it so important? Different seeds will also give you different results, but if your project is constrained correctly it should just work as well.

--- Quote End ---

Primarily as a sanity check. A colleague uses a DOS batch file to implement a project, and I use Linux makefile. If the two produce different results with the same seed, then the obvious conclusion is that one of us has got our setup wrong or, at the least, that the two flows aren't equivalent. Secondly, I have some experience of developing translators that run on both Linux and Windows, and differing output on the two platforms is a good indicator of a bug somewhere, which might be the result of differing word widths, different and incompatible implementations of STL or the C library, and so on. At the least, you have to know why they're different, and let your customers know that it doesn't matter.

I think the seed issue is a different matter. If you can't route under one set of circumstances - the seed - you try it differently. It's important that the Linux and Windows versions route identically with the same seed, otherwise there's no specific correlation between the seed value and the ability, or otherwise, to route.