Ultrawide Core Fixes

[bkorpan]

Add configs with 256-bit fetch and >2 memory pipes.

[jerryz123]

Does this actually work? Its pretty cool if it does.

I suspect there may be some bug in the global history maintenance logic that reduces bpred performance, I assume its 2x 128b banks?
Also, I'm a bit concerned about moving the store pipe to the last pipe in the cache, that was extremely finicky to get right, but I can't think of a case of the top of my head where it might break something.

[bkorpan]

Does this actually work? Its pretty cool if it does.

I suspect there may be some bug in the global history maintenance logic that reduces bpred performance, I assume its 2x 128b banks?
Also, I'm a bit concerned about moving the store pipe to the last pipe in the cache, that was extremely finicky to get right, but I can't think of a case of the top of my head where it might break something.

I was able to run Dhrystone before doing the store pipe thing - got a score of 11500 Dhrystones/MHz. The store pipe thing doesn't work yet.