Offset filed of buffer dose not work

[Gxiandy]

Hi,

This is excellent work! We try to use it to solve some memory allocation problems. We found that when we set a value to the offset filed of a buffer (which will be regard as a fixed position buffer according to the annotation)， the solver will re-allocate this buffer instead of using the offet we provided.

Did I use it wrong? Or is this still a todo feature?

[mmoffitt]

Ah, yes that optional<Offset> field is confusing ... when reading solution files, we use it to store the final (fixed) offset assignment, but it's not a property that is honored by the solver. It would take some extra work if you want the solver to recognize such fixed objects and place buffers around them.

[Gxiandy]

Thank you for your reply. Now I understand the meaning of offset filed. Do you have any suggestions If I want to add the feature I mentioned above？

[mmoffitt]

I've just pushed a change that should add solver support for fixed offsets. Would you mind giving it a try?

[Gxiandy]

That's great! I‘d like to !

[Gxiandy]

Thanks for your work. I've tried this fix in our project and it works. However, we found that for some cases, minimalloc can not found a solution within the required time (which is 30s for 200 buffers)， but when we use first-fit method for memory allocation, we will get right solution in very short time. Seems that the partial-allocation problem is not friendly to minimalloc? btw, I suggest that use number of backtrack times to determine whether timeout occurs because it is more stable.

[mmoffitt]

There's one more change I ought to make for the "fixed offset" feature to be fully performant -- I'll let you know once it's checked in, might take me a couple days.

[mmoffitt]

Just added one more check that should allow for earlier termination -- let me know if this fixes the issue (if not, I can try a few more things).