- Fractional LMUL Constraint
Re: Fractional LMUL Constraint
Robin Kruppe <hanna.kruppe@...>
On Sat, 21 Mar 2020 at 20:46, Bill Huffman <huffman@...
Maybe this has been stated already, and I just haven't seen it, but it seems like there's a constraint with fractional LMUL that wasn't there before. The compiler must ensure that: LMUL >= SEW/ELEN.
I think that my concerns for lane crossing with fractional LMUL came from mentally violating this constraint without realizing its significance.
In more detail, a number of constraints, and their origins must be:
- VLEN >= ELEN from the spec
- VLMAX = LMUL*VLEN/SEW from the spec
- VLMAX >= 1 seems obvious
- LMUL*VLEN/SEW >= 1 from #2 and #3
- LMUL >= SEW/VLEN from #4
- LMUL >= SEW/ELEN if we are not to increase minimum VLEN in #1
#6 is put into compiler visible terms (without VLEN or VLMAX). It is a new constraint, I think, since always before LMUL >= 1 meant it was forced to be satisfied because of the existing constraint that ELEN >= SEW.
Not only is it a constraint, but violating it would bring no advantage. The computation would use registers just as efficiently with a larger LMUL. For example, if there were only ELEN/2 and ELEN elements and widening/narrowing
operations between them, LMUL < 1/2 would not reduce the number of registers used, but would reduce VLMAX for longer vectors without providing any advantage for shorter vectors.
Join firstname.lastname@example.org to automatically receive all group messages.