private static unsafe void Add(Vector256 <UInt64>[] s, Vector256 <UInt32>[] v, int shift)
{
int shift_sets = shift / Vector256 <UInt64> .Count;
int shift_rems = shift % Vector256 <UInt64> .Count;
if (shift_rems == 0)
{
#if DEBUG
Debug <OverflowException> .Assert(checked (shift_sets + v.Length) <= s.Length);
#endif
fixed(Vector256 <UInt64> *ps = s)
{
fixed(Vector256 <UInt32> *pv = v)
{
for (int i = 0, store_idx = shift_sets; i < v.Length; i++, store_idx++)
{
ps[store_idx] = Avx2.Add(ps[store_idx], pv[i].AsUInt64());
}
}
}
}
else
{
#if DEBUG
Debug <OverflowException> .Assert(checked (shift_sets + v.Length) < s.Length);
#endif
Vector256 <UInt64> ml = Mask256.MSV(checked ((uint)(shift_rems * 2))).AsUInt64();
Vector256 <UInt64> mh = Mask256.LSV(checked ((uint)(shift_rems * 2))).AsUInt64();
byte mm_perm = shift_rems switch {
1 => MM_PERM_CBAD,
2 => MM_PERM_BADC,
3 => MM_PERM_ADCB,
_ => throw new ArgumentException(nameof(shift_rems))
};
int store_idx = shift_sets;
Vector256 <UInt64> uh, ul, u;
fixed(Vector256 <UInt64> *ps = s)
{
fixed(Vector256 <UInt32> *pv = v)
{
u = Avx2.Permute4x64(pv[0].AsUInt64(), mm_perm);
ul = Avx2.And(u, ml);
ps[store_idx] = Avx2.Add(ps[store_idx], ul);
store_idx++;
for (int i = 1; i < v.Length; i++)
{
uh = Avx2.And(u, mh);
u = Avx2.Permute4x64(pv[i].AsUInt64(), mm_perm);
ul = Avx2.And(u, ml);
ps[store_idx] = Avx2.Add(ps[store_idx], Avx2.Or(uh, ul));
store_idx++;
}
uh = Avx2.And(u, mh);
ps[store_idx] = Avx2.Add(ps[store_idx], uh);
}
}
}
}