Return to README CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / s390

Annotation of OpenXM_contrib/gmp/mpn/s390/README, Revision 1.1.1.1

1.1       ohara       1: All current (2001) S/390 and z/Architecture machines are single-issue,
                      2: but some newer machines have a deep pipeline.  Software-pipelining is
                      3: therefore beneficial.
                      4:
                      5: * mpn_add_n, mpn_sub_n: Use code along the lines below.  Two-way unrolling
                      6:   would be adequate.
                      7:
                      8:   mp_limb_t
                      9:   mpn_add_n (mp_ptr rp, mp_srcptr up, mp_srcptr vp, mp_size_t n)
                     10:   {
                     11:     mp_limb_t a, b, r, cy;
                     12:     mp_size_t i;
                     13:     mp_limb_t mm = -1;
                     14:
                     15:     cy = 0;
                     16:     up += n;
                     17:     vp += n;
                     18:     rp += n;
                     19:     i = -n;
                     20:     do
                     21:       {
                     22:        a = up[i];
                     23:        b = vp[i];
                     24:        r = a + b + cy;
                     25:        rp[i] = r;
                     26:        cy = (((a & b) | ((a | b) & (r ^ mm)))) >> 31;
                     27:        i++;
                     28:       }
                     29:     while (i < 0);
                     30:     return cy;
                     31:   }
                     32:
                     33: * mpn_lshift, mpn_rshift: Use SLDL/SRDL, and two-way unrolling.
                     34:
                     35: * mpn_mul_1, mpn_addmul_1, mpn_submul_1: For machines with just signed
                     36:   multiply (MR), use two loops, similar to the corresponding VAX or
                     37:   POWER functions.  Handle carry like for mpn_add_n.