commit 87cfb3dd2eb8268e9e0b5b1a55c0d021cdec56f8
parent 2aea575f1267f801875df749f0826914c1e1c36d
Author: Quentin Carbonneaux <quentin.carbonneaux@yale.edu>
Date: Fri, 25 Dec 2015 17:49:09 -0500
wip fp lifting in spill.c, pheeew
Diffstat:
| M | lisc/spill.c | | | 159 | +++++++++++++++++++++++++++++++++++++++++++------------------------------------ |
1 file changed, 87 insertions(+), 72 deletions(-)
diff --git a/lisc/spill.c b/lisc/spill.c
@@ -22,8 +22,10 @@ loopmark(Blk *hd, Blk *b, Phi *p)
* loop headers */
for (z=0; z<BITS; z++)
hd->gen.t[z] |= b->gen.t[z];
- if (b->nlive > hd->nlive)
- hd->nlive = b->nlive;
+ if (b->nlive[0] > hd->nlive[0])
+ hd->nlive[0] = b->nlive[0];
+ if (b->nlive[1] > hd->nlive[1])
+ hd->nlive[1] = b->nlive[1];
for (n=0; n<b->npred; n++)
loopmark(hd, b->pred[n], b->phi);
}
@@ -77,7 +79,8 @@ fillcost(Fn *fn)
}
if (hd && debug['S']) {
fprintf(stderr, "\t%-10s", b->name);
- fprintf(stderr, " (% 3d) ", b->nlive);
+ fprintf(stderr, " (% 3d ", b->nlive[0]);
+ fprintf(stderr, "% 3d) ", b->nlive[1]);
dumpts(&b->gen, fn->tmp, stderr);
}
}
@@ -155,7 +158,7 @@ slot(int t)
*
* invariant: slot4 <= slot8
*/
- if (tmp[t].wide) {
+ if (KWIDE(tmp[t].cls)) {
s = slot8;
if (slot4 == slot8)
slot4 += 2;
@@ -175,15 +178,6 @@ slot(int t)
}
static void
-store(Ref r, int s)
-{
- if (tmp[r.val].wide)
- emit(OStorel, 0, R, r, SLOT(s));
- else
- emit(OStorew, 0, R, r, SLOT(s));
-}
-
-static void
limit(Bits *b, int k, Bits *fst)
{
static int *tarr, maxt;
@@ -229,23 +223,48 @@ sethint(Bits *u, ulong r)
static void
reloads(Bits *u, Bits *v)
{
- int t;
+ /* fixme, oooh really... */
+ static int kload[] = {
+ [Kw] = OLoadsw, [Kl] = OLoadl,
+ [Ks] = OLoads, [Kd] = OLoadd
+ };
+ int t, k;
for (t=Tmp0; t<ntmp; t++)
- if (BGET(*u, t) && !BGET(*v, t))
- emit(OLoad, tmp[t].wide, TMP(t), slot(t), R);
+ if (BGET(*u, t) && !BGET(*v, t)) {
+ k = tmp[t].cls;
+ emit(kload[k], k, TMP(t), slot(t), R);
+ }
+}
+
+static void
+store(Ref r, int s)
+{
+ static int kstore[] = {
+ [Kw] = OStorew, [Kl] = OStorel,
+ [Ks] = OStores, [Kd] = OStored,
+ };
+
+ if (s != -1)
+ emit(kstore[tmp[r.val].cls], 0, R, r, SLOT(s));
+}
+
+static int
+regcpy(Ins *i)
+{
+ return i->op == OCopy && isreg(i->arg[0]);
}
static Ins *
dopm(Blk *b, Ins *i, Bits *v)
{
- int n, s, t;
+ int n, t;
Bits u;
Ins *i1;
ulong r;
/* consecutive copies from
- * registers need to handled
+ * registers need to be handled
* as one large instruction
*
* fixme: there is an assumption
@@ -261,14 +280,10 @@ dopm(Blk *b, Ins *i, Bits *v)
if (!req(i->to, R))
if (BGET(*v, t)) {
BCLR(*v, t);
- s = tmp[t].slot;
- if (s != -1)
- store(i->to, s);
+ store(i->to, tmp[t].slot);
}
BSET(*v, i->arg[0].val);
- } while (i != b->ins &&
- (i-1)->op == OCopy &&
- isreg((i-1)->arg[0]));
+ } while (i != b->ins && regcpy(i-1));
u = *v;
if (i != b->ins && (i-1)->op == OCall) {
v->t[0] &= ~calldef(*(i-1), 0);
@@ -305,8 +320,8 @@ void
spill(Fn *fn)
{
Blk *b, *s1, *s2, *hd;
- int n, m, z, l, t, lvarg[2];
- Bits u, v, w;
+ int n, m, z, l, t, k, nr, lvarg[2];
+ Bits u, v[2], w, mask[2];
Ins *i;
int j, s;
Phi *p;
@@ -320,18 +335,14 @@ spill(Fn *fn)
slot8 = 0;
assert(ntmp < NBit*BITS);
- for (b=fn->start; b; b=b->link) {
- for (p=b->phi; p; p=p->link)
- tmp[p->to.val].wide = p->wide;
- for (i=b->ins; i-b->ins < b->nins; i++)
- if (rtype(i->to) == RTmp)
- /* Note: this only works because
- * all OCmp operations were eliminated
- * indeed, the wide bit of those refer
- * to the size of the operands
- */
- tmp[i->to.val].wide = i->wide;
- }
+ BZERO(mask[0]);
+ BZERO(mask[1]);
+ for (t=Tmp0; t<ntmp; t++)
+ BSET(mask[KBASE(tmp[t].cls)], t);
+ for (t=0; t < NIReg; t++) /* could use the .cls of regs */
+ BSET(mask[0], RAX + t);
+ for (t=0; t < NFReg; t++)
+ BSET(mask[1], XMM0 + t);
for (n=fn->nblk-1; n>=0; n--) {
/* invariant: m>n => in,out of m updated */
@@ -342,44 +353,51 @@ spill(Fn *fn)
curi = 0;
s1 = b->s1;
s2 = b->s2;
- BZERO(v);
hd = 0;
if (s1 && s1->id <= n)
hd = s1;
if (s2 && s2->id <= n)
if (!hd || s2->id >= hd->id)
hd = s2;
- if (hd) {
- /* back-edge */
- l = hd->nlive;
- for (z=0; z<BITS; z++)
- v.t[z] = hd->gen.t[z] & b->out.t[z];
- j = bcnt(&v);
- if (j < NReg) {
- w = b->out;
- for (z=0; z<BITS; z++)
- w.t[z] &= ~v.t[z];
- j = bcnt(&w); /* live through */
- limit(&w, NReg - (l - j), 0);
- for (z=0; z<BITS; z++)
- v.t[z] |= w.t[z];
- } else if (j > NReg)
- limit(&v, NReg, 0);
- } else if (s1) {
- v = liveon(b, s1);
- w = v;
- if (s2) {
- u = liveon(b, s2);
+ for (k=0; k<2; k++) {
+ nr = k == 0 ? NIReg : NFReg;
+ if (hd) {
+ /* back-edge */
for (z=0; z<BITS; z++) {
- v.t[z] |= u.t[z];
- w.t[z] &= u.t[z];
+ v[k].t[z] = b->out.t[z]
+ & hd->gen.t[z]
+ & mask[k].t[z];
+ w.t[z] = b->out.t[z]
+ & ~hd->gen.t[z]
+ & mask[k].t[z];
+ }
+ j = bcnt(&v[k]);
+ if (j < nr) {
+ j = bcnt(&w); /* live through */
+ l = hd->nlive[k];
+ limit(&w, nr - (l - j), 0);
+ for (z=0; z<BITS; z++)
+ v[k].t[z] |= w.t[z];
+ } else
+ limit(&v[k], nr, 0);
+ } else if (s1) {
+ w = liveon(b, s1);
+ v[k] = w;
+ if (s2) {
+ u = liveon(b, s2);
+ for (z=0; z<BITS; z++) {
+ v[k].t[z] |= u.t[z];
+ v[k].t[z] &= mask[k].t[z];
+ w.t[z] &= u.t[z];
+ }
}
+ limit(&v[k], nr, &w);
}
- assert(bcnt(&w) <= NReg);
- limit(&v, NReg, &w);
}
- b->out = v;
- assert(bcnt(&v) <= NReg);
+ BZERO(b->out);
+ for (z=0; z<BITS; z++)
+ for (k=0; k<2; k++)
+ b->out.t[z] |= v[k].t[z];
/* 2. process the block instructions */
curi = &insb[NIns];
@@ -387,7 +405,7 @@ spill(Fn *fn)
for (i=&b->ins[b->nins]; i!=b->ins;) {
assert(bcnt(&v) <= NReg);
i--;
- if (i->op == OCopy && isreg(i->arg[0])) {
+ if (regcpy(i)) {
i = dopm(b, i, &v);
continue;
}
@@ -448,8 +466,7 @@ spill(Fn *fn)
if (r)
sethint(&v, r);
reloads(&u, &v);
- if (s != -1)
- store(i->to, s);
+ store(i->to, s);
emiti(*i);
}
assert(!r || b==fn->start);
@@ -459,9 +476,7 @@ spill(Fn *fn)
t = p->to.val;
if (BGET(v, t)) {
BCLR(v, t);
- s = tmp[t].slot;
- if (s != -1)
- store(p->to, s);
+ store(p->to, tmp[t].slot);
} else
p->to = slot(p->to.val);
}
