softfloat: Specialize udiv_qrnnd for ppc64

The ISA has a 128/64-bit division instruction, though it assumes the
low 64-bits of the numerator are 0, and so requires a bit more fixup
than a full 128-bit division insn.

Backports commit 27ae5109a2ba8b6b679cce3e03e16570a34390a0 from qemu
This commit is contained in:
Richard Henderson 2018-10-08 11:08:05 -04:00 committed by Lioncash
parent 0fd568871e
commit e54a2b65c0
No known key found for this signature in database
GPG key ID: 4E3C3CC1031BA9C7

View file

@ -647,6 +647,22 @@ static inline uint64_t udiv_qrnnd(uint64_t *r, uint64_t n1,
asm("dlgr %0, %1" : "+r"(n) : "r"(d)); asm("dlgr %0, %1" : "+r"(n) : "r"(d));
*r = n >> 64; *r = n >> 64;
return n; return n;
#elif defined(_ARCH_PPC64)
/* From Power ISA 3.0B, programming note for divdeu. */
uint64_t q1, q2, Q, r1, r2, R;
asm("divdeu %0,%2,%4; divdu %1,%3,%4"
: "=&r"(q1), "=r"(q2)
: "r"(n1), "r"(n0), "r"(d));
r1 = -(q1 * d); /* low part of (n1<<64) - (q1 * d) */
r2 = n0 - (q2 * d);
Q = q1 + q2;
R = r1 + r2;
if (R >= d || R < r2) { /* overflow implies R > d */
Q += 1;
R -= d;
}
*r = R;
return Q;
#else #else
uint64_t d0, d1, q0, q1, r1, r0, m; uint64_t d0, d1, q0, q1, r1, r0, m;