[Libre-soc-isa] [Bug 1071] add parallel prefix sum remap mode

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Mon May 1 09:30:00 BST 2023


--- Comment #20 from Jacob Lifshay <programmerjake at gmail.com> ---
(In reply to Luke Kenneth Casson Leighton from comment #19)
> (In reply to Luke Kenneth Casson Leighton from comment #15)
> > (In reply to Jacob Lifshay from comment #14)
> > > for i in range(0, length, chunk_size):
> > >     sum += reduce_chunk(VL=chunk_size)
> > > sum += reduce_tail(VL=length % chunk_size)  # dynamic VL here
> > 
> > drat, you're right.
> > 
> > of course. okaaay sigh so this involves creating a new Form, with Parallel
> > Reduction similar to svshape2 "carving out" its own niche... urrr...
> ah. realised there's a complication. VL and MAXVL determines the number
> of operations carried out, not the width of the reduction.
> for now the tail will have to be done by recursive macros using
> immediates manually.

that's why I'm proposing a svshape with GPR/VL input, so it can calculate the
proper VL (which is a little less than 2x bigger but doesn't follow any simple
formula I could find).

MAXVL would be adjusted following the same algorithm but using MAXVL as the
input rather than VL or a GPR.

You are receiving this mail because:
You are on the CC list for the bug.

More information about the Libre-SOC-ISA mailing list