Previous | Next --- Slide 31 of 35
Back to Lecture Thumbnails
Kapteyn

I'm guessing that offsets are calculated by subtracting the base from each 32 byte value in the cache line? (Also the offsets must be signed because we might have negative or positive offsets.)

If this is the case, to make this compression fast and not linear in the number of chunks in the cache line, we would want to perform this subtraction in parallel. Does this mean the cache has access to SIMD units to perform the compression?

admintio42

It would be interesting to know which computational resources the cache would be allowed to use to compress the data.

bwf

@admintio42 I don't think it would be the case that existing resources would be used to compress data(since that would just add extra memory costs), instead, I'd image that specific resources would be given solely for this purpose.