| 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. |