vwredsum.vs

{
  let SEW      = get_sew();
  let LMUL_pow = get_lmul_pow();
  let SEW_widen      = SEW * 2;
  let LMUL_pow_widen = LMUL_pow + 1;
  let num_elem_vs = get_num_elem(LMUL_pow, SEW);
  let num_elem_vd = get_num_elem(0, SEW_widen); /* vd regardless of LMUL setting */

  if illegal_reduction_widen(SEW_widen, LMUL_pow_widen) then { handle_illegal(); return RETIRE_FAIL };

  if unsigned(vl) == 0 then return RETIRE_SUCCESS; /* if vl=0, no operation is performed */

  let 'n = num_elem_vs;
  let 'd = num_elem_vd;
  let 'm = SEW;
  let 'o = SEW_widen;

  let vm_val  : vector('n, dec, bool)     = read_vmask(num_elem_vs, vm, 0b00000);
  let vd_val  : vector('d, dec, bits('o)) = read_vreg(num_elem_vd, SEW_widen, 0, vd);
  let vs2_val : vector('n, dec, bits('m)) = read_vreg(num_elem_vs, SEW, LMUL_pow, vs2);
  let mask    : vector('n, dec, bool)     = init_masked_source(num_elem_vs, LMUL_pow, vm_val);

  sum : bits('o) = read_single_element(SEW_widen, 0, vs1); /* vs1 regardless of LMUL setting */
  foreach (i from 0 to (num_elem_vs - 1)) {
    if mask[i] then {
      let elem : bits('o) = match funct6 {
        IVV_VWREDSUMU  => to_bits(SEW_widen, unsigned(vs2_val[i])),
        IVV_VWREDSUM   => to_bits(SEW_widen, signed(vs2_val[i]))
      };
      sum = sum + elem
    }
  };

  write_single_element(SEW_widen, 0, vd, sum);
  /* other elements in vd are treated as tail elements, currently remain unchanged */
  /* TODO: configuration support for agnostic behavior */
  vstart = zeros();
  RETIRE_SUCCESS
}