qbe

commit 73add5daaef5f6b764ac2ddf260b27f86d12e1af
parent 32cb8d315afae78dfe35ce57512b7015a0cb0fb4
Author: Quentin Carbonneaux <quentin.carbonneaux@yale.edu>
Date:   Mon, 20 Jul 2015 17:53:58 -0400

start work on spiller

Diffstat:
M
lisc/lisc.h
 | 
17
+++++++++++++----
A
lisc/spill.c
 | 
228
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

2 files changed, 241 insertions(+), 4 deletions(-)
diff --git a/lisc/lisc.h b/lisc/lisc.h
@@ -32,11 +32,11 @@ enum {
 	R13,
 	R14,
 	R15,
+	NReg = R15 - RAX + 1
 };
 
 enum {
-	NReg    = 32,
-	Tmp0    = NReg+1,
+	Tmp0    = 33,
 
 	NString = 32,
 	NPred   = 15,
@@ -44,11 +44,11 @@ enum {
 	NIns    = 256,
 
 	BITS    = 4,
-	NBit    = 8 * sizeof(unsigned long long),
+	NBit    = 8 * sizeof(uint64_t),
 };
 
 struct Bits {
-	unsigned long long t[BITS];
+	uint64_t t[BITS];
 };
 
 #define BGET(b, n) (1&((b).t[n/NBit]>>(n%NBit)))
@@ -131,6 +131,7 @@ struct Blk {
 	Blk **pred;
 	uint npred;
 	Bits in, out;
+	int loop;
 	char name[NString];
 	int id;
 };
@@ -143,6 +144,7 @@ struct Sym {
 	} type;
 	char name[NString];
 	uint ndef, nuse;
+	int cost;
 };
 
 struct Fn {
@@ -168,3 +170,10 @@ void ssafix(Fn *, int);
 
 /* live.c */
 void filllive(Fn *);
+
+/* isel.c */
+void isel(Fn *);
+
+/* spill.c */
+void fillcost(Fn *);
+void spill(Fn *);
diff --git a/lisc/spill.c b/lisc/spill.c
@@ -0,0 +1,228 @@
+#include "lisc.h"
+
+
+static void
+symuse(Ref r, int use, int loop, Fn *fn)
+{
+	Sym *s;
+
+	if (rtype(r) != RSym)
+		return;
+	s = &fn->sym[r.val];
+	if (s->type != STmp)
+		return;
+	if (use)
+		s->nuse++;
+	else
+		s->ndef++;
+	s->cost += loop;
+}
+
+/* evaluate spill costs of temporaries,
+ * this also fills usage information
+ * requires rpo
+ */
+void
+fillcost(Fn *fn)
+{
+	int n, m;
+	uint a;
+	Blk *b;
+	Ins *i;
+	Sym *s;
+	Phi *p;
+
+	/* todo, have somewhat proper loop
+	 * detection for example on this
+	 * cfg, it's bogus:
+	 * +> loop <+
+	 * |   /\   |
+	 * +- a  b -+
+	 */
+	for (b=fn->start; b; b=b->link)
+		b->loop = 1;
+	for (n=fn->nblk-1; n>=0; n--) {
+		b = fn->rpo[n];
+		m = n;
+		if (b->s1 && b->s1->id < m)
+			m = b->s1->id;
+		if (b->s2 && b->s2->id < m)
+			m = b->s2->id;
+		for (; m<n; m++)
+			fn->rpo[m]->loop *= 10;
+	}
+	for (s=fn->sym; s-fn->sym < fn->ntmp; s++) {
+		s->cost = 0;
+		s->nuse = 0;
+		s->ndef = 0;
+		if (s->type == SReg)
+			s->cost = 100000;
+	}
+	for (b=fn->start; b; b=b->link) {
+		for (p=b->phi; p; p=p->link) {
+			/* zero cost for temporaries used
+			 * in phi instructions */
+			assert(rtype(p->to) == RSym);
+			assert(fn->sym[p->to.val].type == STmp);
+			symuse(p->to, 0, 0, fn);
+			for (a=0; a<p->narg; a++)
+				symuse(p->arg[a], 1, 0, fn);
+		}
+		n = b->loop;
+		for (i=b->ins; i-b->ins < b->nins; i++) {
+			symuse(i->to, 0, n, fn);
+			symuse(i->arg[0], 1, n, fn);
+			symuse(i->arg[0], 1, n, fn);
+		}
+		symuse(b->jmp.arg, 1, n, fn);
+	}
+}
+
+static int
+bcnt(Bits *b) /* glad I can pull it :) */
+{
+	const uint64_t m1 = 0x5555555555555555;
+	const uint64_t m2 = 0x3333333333333333;
+	const uint64_t m3 = 0x0f0f0f0f0f0f0f0f;
+	const uint64_t m4 = 0x00ff00ff00ff00ff;
+	const uint64_t m5 = 0x0000ffff0000ffff;
+	const uint64_t m6 = 0x00000000ffffffff;
+	uint64_t tmp;
+	int z, i;
+
+	i = 0;
+	for (z=0; z<BITS; z++) {
+		tmp = b->t[z];
+		if (!tmp)
+			continue;
+		tmp = (tmp&m1) + (tmp>> 1&m1);
+		tmp = (tmp&m2) + (tmp>> 2&m2);
+		tmp = (tmp&m3) + (tmp>> 4&m3);
+		tmp = (tmp&m4) + (tmp>> 8&m4);
+		tmp = (tmp&m5) + (tmp>>16&m5);
+		tmp = (tmp&m6) + (tmp>>32&m6);
+		i += tmp;
+	}
+	return i;
+}
+
+extern Ins insb[NIns], *curi; /* shared work buffer */
+static Bits w;
+static Sym *sym;
+
+#if 0
+static void
+emit(short op, Ref to, Ref arg0, Ref arg1)
+{
+	if (curi == insb)
+		diag("spill: too many instructions");
+	*curi-- = (Ins){op, to, {arg0, arg1}};
+}
+#endif
+
+static int
+tcmp(const void *pa, const void *pb)
+{
+	int a, b;
+
+	/* so here, we make sure temporaries
+	 * in w are sorted first in the list
+	 * no matter their cost
+	 */
+	a = *(int *)pa;
+	b = *(int *)pb;
+	assert(sym[a].type==STmp && sym[b].type==STmp);
+	if (BGET(w, a))
+		return BGET(w, b) ? sym[b].cost-sym[a].cost : -1;
+	else
+		return BGET(w, b) ? +1 : sym[b].cost-sym[a].cost;
+}
+
+/* spill code insertion
+ * requires spill costs and rpo
+ *
+ * Note: this will replace liveness
+ * information (in, out) with temporaries
+ * that must be in registers at block
+ * borders
+ *
+ * Be careful with:
+ * - OCopy instructions to ensure register
+ *   constraints
+ * - Branching variables *must* be in
+ *   register
+ */
+void
+spill(Fn *fn)
+{
+	Bits v;
+	int n, z, j, k, l;
+	Blk *b, *s1, *s2;
+	Ins *i;
+	int *tmp, maxt, nt;
+
+	sym = fn->sym;
+	tmp = 0;
+	maxt = 0;
+	for (n=fn->nblk-1; n>=0; n--) {
+		/* invariant: m>n => in,out of m updated */
+		b = fn->rpo[n];
+		curi = &insb[NIns-1];
+
+		/* 1. find temporaries in registers at
+		 * the end of the block (put them in v) */
+		s1 = b->s1;
+		s2 = b->s2;
+		k = NReg - !req(b->jmp.arg, R);
+
+		if ((s1 && s1->id <= n) || (s2 && s2->id <= n)) {
+			/* potential back-edge */
+		} else {
+			if (s1) {
+				v = s1->in; /* could be in reg */
+				w = s1->in; /* will be in reg */
+			} else {
+				w = (Bits){{0}};
+				v = (Bits){{0}};
+			}
+			if (s2) {
+				for (z=0; z<BITS; z++) {
+					v.t[z] |= s2->in.t[z];
+					w.t[z] &= s2->in.t[z];
+				}
+			}
+/* #if 0 */
+			for (z=0; z<BITS; z++) {
+				assert(!(v.t[z] & ~b->out.t[z]));
+				assert(!(w.t[z] & ~b->out.t[z]));
+			}
+			assert(bcnt(&w) <= NReg);
+			assert(bcnt(&w) <= bcnt(&v));
+/* #endif */
+			nt = bcnt(&v);
+			if (nt > maxt) {
+				free(tmp);
+				tmp = alloc(nt * sizeof tmp[0]);
+				maxt = nt;
+			}
+			j=0;
+			for (l=0/*Tmp0*/; l<fn->ntmp; l++)
+				if (BGET(v, l)) {
+					assert(fn->sym[l].type == STmp);
+					tmp[j++] = l;
+				}
+			assert(j == nt);
+			qsort(tmp, nt, sizeof tmp[0], tcmp);
+			v = (Bits){{0}};
+			for (j=0; j<k; j++)
+				BSET(v, j);
+		}
+
+		/* 2. process the block instructions */
+		for (i=&b->ins[b->nins]; i!=b->ins;) {
+			i--;
+			;
+		}
+	}
+	free(tmp);
+}