conferences/fosdem2024/fosdem2024_ddffirst/pospopcount.s

   1
   2         mtspr 9, 3                 # move r3 to CTR
   3         setvl 3,0,8,0,1,1          # set MVL=8, VL=r3=MIN(MVL,CTR)
   4         # load VL bytes (update r4 addr) but compressed (dw=8)
   5         addi 6, 0, 0               # initialise all 64-bits of r6 to zero
   6         sv.lbzu/pi/dw=8 *6, 1(4)   # should be /lf here as well
   7         # gather performs the transpose (which gets us to positional..)
   8         gbbd 8,6
   9         # now those bits have been turned around, popcount and sum them
  10         setvl 0,0,8,0,1,1          # set MVL=VL=8
  11         sv.popcntd/sw=8 *24,*8     # do the (now transposed) popcount
  12         sv.add *16,*16,*24         # and accumulate in results
  13         # branch back if CTR still non-zero. works even though VL=8
  14         sv.bc/all 16, *0, -0x28   # reduce CTR by VL and stop if -ve