ref: d552fed38514dc799c7fea95dfb632c8816c3f60
dir: /sys/src/cmd/kc/reg.c/
#include "gc.h" Reg* rega(void) { Reg *r; r = freer; if(r == R) { r = alloc(sizeof(*r)); } else freer = r->link; *r = zreg; return r; } int rcmp(const void *a1, const void *a2) { Rgn *p1, *p2; int c1, c2; p1 = (Rgn*)a1; p2 = (Rgn*)a2; c1 = p2->cost; c2 = p1->cost; if(c1 -= c2) return c1; return p2->varno - p1->varno; } void regopt(Prog *p) { Reg *r, *r1, *r2; Prog *p1; int i, z; long initpc, val, npc; ulong vreg; Bits bit; struct { long m; long c; Reg* p; } log5[6], *lp; firstr = R; lastr = R; nvar = 0; regbits = 0; for(z=0; z<BITS; z++) { externs.b[z] = 0; params.b[z] = 0; consts.b[z] = 0; addrs.b[z] = 0; } /* * pass 1 * build aux data structure * allocate pcs * find use and set of variables */ val = 5L * 5L * 5L * 5L * 5L; lp = log5; for(i=0; i<5; i++) { lp->m = val; lp->c = 0; lp->p = R; val /= 5L; lp++; } val = 0; for(; p != P; p = p->link) { switch(p->as) { case ADATA: case AGLOBL: case ANAME: case ASIGNAME: continue; } r = rega(); if(firstr == R) { firstr = r; lastr = r; } else { lastr->link = r; r->p1 = lastr; lastr->s1 = r; lastr = r; } r->prog = p; r->pc = val; val++; lp = log5; for(i=0; i<5; i++) { lp->c--; if(lp->c <= 0) { lp->c = lp->m; if(lp->p != R) lp->p->log5 = r; lp->p = r; (lp+1)->c = 0; break; } lp++; } r1 = r->p1; if(r1 != R) switch(r1->prog->as) { case ARETURN: case AJMP: case ARETT: r->p1 = R; r1->s1 = R; } /* * left side always read */ bit = mkvar(&p->from, p->as==AMOVW); for(z=0; z<BITS; z++) r->use1.b[z] |= bit.b[z]; /* * right side depends on opcode */ bit = mkvar(&p->to, 0); if(bany(&bit)) switch(p->as) { default: diag(Z, "reg: unknown asop: %A", p->as); break; /* * right side write */ case ANOP: case AMOVB: case AMOVBU: case AMOVH: case AMOVHU: case AMOVW: case AFMOVF: case AFMOVD: for(z=0; z<BITS; z++) r->set.b[z] |= bit.b[z]; break; /* * funny */ case AJMPL: for(z=0; z<BITS; z++) addrs.b[z] |= bit.b[z]; break; } } if(firstr == R) return; initpc = pc - val; npc = val; /* * pass 2 * turn branch references to pointers * build back pointers */ for(r = firstr; r != R; r = r->link) { p = r->prog; if(p->to.type == D_BRANCH) { val = p->to.offset - initpc; r1 = firstr; while(r1 != R) { r2 = r1->log5; if(r2 != R && val >= r2->pc) { r1 = r2; continue; } if(r1->pc == val) break; r1 = r1->link; } if(r1 == R) { nearln = p->lineno; diag(Z, "ref not found\n%P", p); continue; } if(r1 == r) { nearln = p->lineno; diag(Z, "ref to self\n%P", p); continue; } r->s2 = r1; r->p2link = r1->p2; r1->p2 = r; } } if(debug['R']) { p = firstr->prog; print("\n%L %D\n", p->lineno, &p->from); } /* * pass 2.5 * find looping structure */ for(r = firstr; r != R; r = r->link) r->active = 0; change = 0; loopit(firstr, npc); if(debug['R'] && debug['v']) { print("\nlooping structure:\n"); for(r = firstr; r != R; r = r->link) { print("%ld:%P", r->loop, r->prog); for(z=0; z<BITS; z++) bit.b[z] = r->use1.b[z] | r->use2.b[z] | r->set.b[z]; if(bany(&bit)) { print("\t"); if(bany(&r->use1)) print(" u1=%B", r->use1); if(bany(&r->use2)) print(" u2=%B", r->use2); if(bany(&r->set)) print(" st=%B", r->set); } print("\n"); } } /* * pass 3 * iterate propagating usage * back until flow graph is complete */ loop1: change = 0; for(r = firstr; r != R; r = r->link) r->active = 0; for(r = firstr; r != R; r = r->link) if(r->prog->as == ARETURN) prop(r, zbits, zbits); loop11: /* pick up unreachable code */ i = 0; for(r = firstr; r != R; r = r1) { r1 = r->link; if(r1 && r1->active && !r->active) { prop(r, zbits, zbits); i = 1; } } if(i) goto loop11; if(change) goto loop1; /* * pass 4 * iterate propagating register/variable synchrony * forward until graph is complete */ loop2: change = 0; for(r = firstr; r != R; r = r->link) r->active = 0; synch(firstr, zbits); if(change) goto loop2; /* * pass 5 * isolate regions * calculate costs (paint1) */ r = firstr; if(r) { for(z=0; z<BITS; z++) bit.b[z] = (r->refahead.b[z] | r->calahead.b[z]) & ~(externs.b[z] | params.b[z] | addrs.b[z] | consts.b[z]); if(bany(&bit)) { nearln = r->prog->lineno; warn(Z, "used and not set: %B", bit); if(debug['R'] && !debug['w']) print("used and not set: %B\n", bit); } } if(debug['R'] && debug['v']) print("\nprop structure:\n"); for(r = firstr; r != R; r = r->link) r->act = zbits; rgp = region; nregion = 0; for(r = firstr; r != R; r = r->link) { if(debug['R'] && debug['v']) print("%P\n set = %B; rah = %B; cal = %B\n", r->prog, r->set, r->refahead, r->calahead); for(z=0; z<BITS; z++) bit.b[z] = r->set.b[z] & ~(r->refahead.b[z] | r->calahead.b[z] | addrs.b[z]); if(bany(&bit)) { nearln = r->prog->lineno; warn(Z, "set and not used: %B", bit); if(debug['R']) print("set an not used: %B\n", bit); excise(r); } for(z=0; z<BITS; z++) bit.b[z] = LOAD(r) & ~(r->act.b[z] | addrs.b[z]); while(bany(&bit)) { i = bnum(bit); rgp->enter = r; rgp->varno = i; change = 0; if(debug['R'] && debug['v']) print("\n"); paint1(r, i); bit.b[i/32] &= ~(1L<<(i%32)); if(change <= 0) { if(debug['R']) print("%L$%d: %B\n", r->prog->lineno, change, blsh(i)); continue; } rgp->cost = change; nregion++; if(nregion >= NRGN) { warn(Z, "too many regions"); goto brk; } rgp++; } } brk: qsort(region, nregion, sizeof(region[0]), rcmp); /* * pass 6 * determine used registers (paint2) * replace code (paint3) */ rgp = region; for(i=0; i<nregion; i++) { bit = blsh(rgp->varno); vreg = paint2(rgp->enter, rgp->varno); vreg = allreg(vreg, rgp); if(debug['R']) { if(rgp->regno >= NREG) print("%L$%d F%d: %B\n", rgp->enter->prog->lineno, rgp->cost, rgp->regno-NREG, bit); else print("%L$%d R%d: %B\n", rgp->enter->prog->lineno, rgp->cost, rgp->regno, bit); } if(rgp->regno != 0) paint3(rgp->enter, rgp->varno, vreg, rgp->regno); rgp++; } /* * pass 7 * peep-hole on basic block */ if(!debug['R'] || debug['P']) peep(); /* * pass 8 * recalculate pc */ val = initpc; for(r = firstr; r != R; r = r1) { r->pc = val; p = r->prog; p1 = P; r1 = r->link; if(r1 != R) p1 = r1->prog; for(; p != p1; p = p->link) { switch(p->as) { default: val++; break; case ANOP: case ADATA: case AGLOBL: case ANAME: case ASIGNAME: break; } } } pc = val; /* * fix up branches */ if(debug['R']) if(bany(&addrs)) print("addrs: %B\n", addrs); r1 = 0; /* set */ for(r = firstr; r != R; r = r->link) { p = r->prog; if(p->to.type == D_BRANCH) p->to.offset = r->s2->pc; r1 = r; } /* * last pass * eliminate nops * free aux structures */ for(p = firstr->prog; p != P; p = p->link){ while(p->link && p->link->as == ANOP) p->link = p->link->link; } if(r1 != R) { r1->link = freer; freer = firstr; } } /* * add mov b,rn * just after r */ void addmove(Reg *r, int bn, int rn, int f) { Prog *p, *p1; Adr *a; Var *v; p1 = alloc(sizeof(*p1)); *p1 = zprog; p = r->prog; p1->link = p->link; p->link = p1; p1->lineno = p->lineno; v = var + bn; a = &p1->to; a->sym = v->sym; a->name = v->name; a->offset = v->offset; a->etype = v->etype; a->type = D_OREG; if(a->etype == TARRAY || a->sym == S) a->type = D_CONST; p1->as = AMOVW; if(v->etype == TCHAR || v->etype == TUCHAR) p1->as = AMOVB; if(v->etype == TSHORT || v->etype == TUSHORT) p1->as = AMOVH; if(v->etype == TFLOAT) p1->as = AFMOVF; if(v->etype == TDOUBLE) p1->as = AFMOVD; p1->from.type = D_REG; p1->from.reg = rn; if(rn >= NREG) { p1->from.type = D_FREG; p1->from.reg = rn-NREG; } if(!f) { p1->from = *a; *a = zprog.from; a->type = D_REG; a->reg = rn; if(rn >= NREG) { a->type = D_FREG; a->reg = rn-NREG; } if(v->etype == TUCHAR) p1->as = AMOVBU; if(v->etype == TUSHORT) p1->as = AMOVHU; } if(debug['R']) print("%P\t.a%P\n", p, p1); } Bits mkvar(Adr *a, int docon) { Var *v; int i, t, n, et, z; long o; Bits bit; Sym *s; t = a->type; if(t == D_REG && a->reg != NREG) regbits |= RtoB(a->reg); if(t == D_FREG && a->reg != NREG) regbits |= FtoB(a->reg); s = a->sym; o = a->offset; et = a->etype; if(s == S) { if(t != D_CONST || !docon || a->reg != NREG) goto none; et = TLONG; } if(t == D_CONST) { if(s == S && sval(o)) goto none; } n = a->name; v = var; for(i=0; i<nvar; i++) { if(s == v->sym) if(n == v->name) if(o == v->offset) goto out; v++; } if(s) if(s->name[0] == '.') goto none; if(nvar >= NVAR) { if(debug['w'] > 1 && s) warn(Z, "variable not optimized: %s", s->name); goto none; } i = nvar; nvar++; v = &var[i]; v->sym = s; v->offset = o; v->etype = et; v->name = n; if(debug['R']) print("bit=%2d et=%2d %D\n", i, et, a); out: bit = blsh(i); if(n == D_EXTERN || n == D_STATIC) for(z=0; z<BITS; z++) externs.b[z] |= bit.b[z]; if(n == D_PARAM) for(z=0; z<BITS; z++) params.b[z] |= bit.b[z]; if(v->etype != et || !typechlpfd[et]) /* funny punning */ for(z=0; z<BITS; z++) addrs.b[z] |= bit.b[z]; if(t == D_CONST) { if(s == S) { for(z=0; z<BITS; z++) consts.b[z] |= bit.b[z]; return bit; } if(et != TARRAY) for(z=0; z<BITS; z++) addrs.b[z] |= bit.b[z]; for(z=0; z<BITS; z++) params.b[z] |= bit.b[z]; return bit; } if(t == D_OREG) return bit; none: return zbits; } void prop(Reg *r, Bits ref, Bits cal) { Reg *r1, *r2; int z; for(r1 = r; r1 != R; r1 = r1->p1) { for(z=0; z<BITS; z++) { ref.b[z] |= r1->refahead.b[z]; if(ref.b[z] != r1->refahead.b[z]) { r1->refahead.b[z] = ref.b[z]; change++; } cal.b[z] |= r1->calahead.b[z]; if(cal.b[z] != r1->calahead.b[z]) { r1->calahead.b[z] = cal.b[z]; change++; } } switch(r1->prog->as) { case AJMPL: for(z=0; z<BITS; z++) { cal.b[z] |= ref.b[z] | externs.b[z]; ref.b[z] = 0; } break; case ATEXT: for(z=0; z<BITS; z++) { cal.b[z] = 0; ref.b[z] = 0; } break; case ARETURN: for(z=0; z<BITS; z++) { cal.b[z] = externs.b[z]; ref.b[z] = 0; } } for(z=0; z<BITS; z++) { ref.b[z] = (ref.b[z] & ~r1->set.b[z]) | r1->use1.b[z] | r1->use2.b[z]; cal.b[z] &= ~(r1->set.b[z] | r1->use1.b[z] | r1->use2.b[z]); r1->refbehind.b[z] = ref.b[z]; r1->calbehind.b[z] = cal.b[z]; } if(r1->active) break; r1->active = 1; } for(; r != r1; r = r->p1) for(r2 = r->p2; r2 != R; r2 = r2->p2link) prop(r2, r->refbehind, r->calbehind); } /* * find looping structure * * 1) find reverse postordering * 2) find approximate dominators, * the actual dominators if the flow graph is reducible * otherwise, dominators plus some other non-dominators. * See Matthew S. Hecht and Jeffrey D. Ullman, * "Analysis of a Simple Algorithm for Global Data Flow Problems", * Conf. Record of ACM Symp. on Principles of Prog. Langs, Boston, Massachusetts, * Oct. 1-3, 1973, pp. 207-217. * 3) find all nodes with a predecessor dominated by the current node. * such a node is a loop head. * recursively, all preds with a greater rpo number are in the loop */ long postorder(Reg *r, Reg **rpo2r, long n) { Reg *r1; r->rpo = 1; r1 = r->s1; if(r1 && !r1->rpo) n = postorder(r1, rpo2r, n); r1 = r->s2; if(r1 && !r1->rpo) n = postorder(r1, rpo2r, n); rpo2r[n] = r; n++; return n; } long rpolca(long *idom, long rpo1, long rpo2) { long t; if(rpo1 == -1) return rpo2; while(rpo1 != rpo2){ if(rpo1 > rpo2){ t = rpo2; rpo2 = rpo1; rpo1 = t; } while(rpo1 < rpo2){ t = idom[rpo2]; if(t >= rpo2) fatal(Z, "bad idom"); rpo2 = t; } } return rpo1; } int doms(long *idom, long r, long s) { while(s > r) s = idom[s]; return s == r; } int loophead(long *idom, Reg *r) { long src; src = r->rpo; if(r->p1 != R && doms(idom, src, r->p1->rpo)) return 1; for(r = r->p2; r != R; r = r->p2link) if(doms(idom, src, r->rpo)) return 1; return 0; } void loopmark(Reg **rpo2r, long head, Reg *r) { if(r->rpo < head || r->active == head) return; r->active = head; r->loop += LOOP; if(r->p1 != R) loopmark(rpo2r, head, r->p1); for(r = r->p2; r != R; r = r->p2link) loopmark(rpo2r, head, r); } void loopit(Reg *r, long nr) { Reg *r1; long i, d, me; if(nr > maxnr) { rpo2r = alloc(nr * sizeof(Reg*)); idom = alloc(nr * sizeof(long)); maxnr = nr; } d = postorder(r, rpo2r, 0); if(d > nr) fatal(Z, "too many reg nodes"); nr = d; for(i = 0; i < nr / 2; i++){ r1 = rpo2r[i]; rpo2r[i] = rpo2r[nr - 1 - i]; rpo2r[nr - 1 - i] = r1; } for(i = 0; i < nr; i++) rpo2r[i]->rpo = i; idom[0] = 0; for(i = 0; i < nr; i++){ r1 = rpo2r[i]; me = r1->rpo; d = -1; if(r1->p1 != R && r1->p1->rpo < me) d = r1->p1->rpo; for(r1 = r1->p2; r1 != nil; r1 = r1->p2link) if(r1->rpo < me) d = rpolca(idom, d, r1->rpo); idom[i] = d; } for(i = 0; i < nr; i++){ r1 = rpo2r[i]; r1->loop++; if(r1->p2 != R && loophead(idom, r1)) loopmark(rpo2r, i, r1); } } void synch(Reg *r, Bits dif) { Reg *r1; int z; for(r1 = r; r1 != R; r1 = r1->s1) { for(z=0; z<BITS; z++) { dif.b[z] = (dif.b[z] & ~(~r1->refbehind.b[z] & r1->refahead.b[z])) | r1->set.b[z] | r1->regdiff.b[z]; if(dif.b[z] != r1->regdiff.b[z]) { r1->regdiff.b[z] = dif.b[z]; change++; } } if(r1->active) break; r1->active = 1; for(z=0; z<BITS; z++) dif.b[z] &= ~(~r1->calbehind.b[z] & r1->calahead.b[z]); if(r1->s2 != R) synch(r1->s2, dif); } } ulong allreg(ulong b, Rgn *r) { Var *v; int i; v = var + r->varno; r->regno = 0; switch(v->etype) { default: diag(Z, "unknown etype %d/%d", bitno(b), v->etype); break; case TCHAR: case TUCHAR: case TSHORT: case TUSHORT: case TINT: case TUINT: case TLONG: case TULONG: case TIND: case TARRAY: i = BtoR(~b); if(i && r->cost > 0) { r->regno = i; return RtoB(i); } break; case TDOUBLE: case TFLOAT: i = BtoF(~b); if(i && r->cost > 0) { r->regno = i+NREG; return FtoB(i); } break; } return 0; } void paint1(Reg *r, int bn) { Reg *r1; Prog *p; int z; ulong bb; z = bn/32; bb = 1L<<(bn%32); if(r->act.b[z] & bb) return; for(;;) { if(!(r->refbehind.b[z] & bb)) break; r1 = r->p1; if(r1 == R) break; if(!(r1->refahead.b[z] & bb)) break; if(r1->act.b[z] & bb) break; r = r1; } if(LOAD(r) & ~(r->set.b[z]&~(r->use1.b[z]|r->use2.b[z])) & bb) { change -= CLOAD * r->loop; if(debug['R'] && debug['v']) print("%ld%P\tld %B $%d\n", r->loop, r->prog, blsh(bn), change); } for(;;) { r->act.b[z] |= bb; p = r->prog; if(r->use1.b[z] & bb) { change += CREF * r->loop; if(p->to.type == D_FREG && p->as == AMOVW) change = -CINF; /* cant go Rreg to Freg */ if(debug['R'] && debug['v']) print("%ld%P\tu1 %B $%d\n", r->loop, p, blsh(bn), change); } if((r->use2.b[z]|r->set.b[z]) & bb) { change += CREF * r->loop; if(p->from.type == D_FREG && p->as == AMOVW) change = -CINF; /* cant go Rreg to Freg */ if(debug['R'] && debug['v']) print("%ld%P\tu2 %B $%d\n", r->loop, p, blsh(bn), change); } if(STORE(r) & r->regdiff.b[z] & bb) { change -= CLOAD * r->loop; if(debug['R'] && debug['v']) print("%ld%P\tst %B $%d\n", r->loop, p, blsh(bn), change); } if(r->refbehind.b[z] & bb) for(r1 = r->p2; r1 != R; r1 = r1->p2link) if(r1->refahead.b[z] & bb) paint1(r1, bn); if(!(r->refahead.b[z] & bb)) break; r1 = r->s2; if(r1 != R) if(r1->refbehind.b[z] & bb) paint1(r1, bn); r = r->s1; if(r == R) break; if(r->act.b[z] & bb) break; if(!(r->refbehind.b[z] & bb)) break; } } ulong paint2(Reg *r, int bn) { Reg *r1; int z; ulong bb, vreg; z = bn/32; bb = 1L << (bn%32); vreg = regbits; if(!(r->act.b[z] & bb)) return vreg; for(;;) { if(!(r->refbehind.b[z] & bb)) break; r1 = r->p1; if(r1 == R) break; if(!(r1->refahead.b[z] & bb)) break; if(!(r1->act.b[z] & bb)) break; r = r1; } for(;;) { r->act.b[z] &= ~bb; vreg |= r->regu; if(r->refbehind.b[z] & bb) for(r1 = r->p2; r1 != R; r1 = r1->p2link) if(r1->refahead.b[z] & bb) vreg |= paint2(r1, bn); if(!(r->refahead.b[z] & bb)) break; r1 = r->s2; if(r1 != R) if(r1->refbehind.b[z] & bb) vreg |= paint2(r1, bn); r = r->s1; if(r == R) break; if(!(r->act.b[z] & bb)) break; if(!(r->refbehind.b[z] & bb)) break; } return vreg; } void paint3(Reg *r, int bn, long rb, int rn) { Reg *r1; Prog *p; int z; ulong bb; z = bn/32; bb = 1L << (bn%32); if(r->act.b[z] & bb) return; for(;;) { if(!(r->refbehind.b[z] & bb)) break; r1 = r->p1; if(r1 == R) break; if(!(r1->refahead.b[z] & bb)) break; if(r1->act.b[z] & bb) break; r = r1; } if(LOAD(r) & ~(r->set.b[z] & ~(r->use1.b[z]|r->use2.b[z])) & bb) addmove(r, bn, rn, 0); for(;;) { r->act.b[z] |= bb; p = r->prog; if(r->use1.b[z] & bb) { if(debug['R']) print("%P", p); addreg(&p->from, rn); if(debug['R']) print("\t.c%P\n", p); } if((r->use2.b[z]|r->set.b[z]) & bb) { if(debug['R']) print("%P", p); addreg(&p->to, rn); if(debug['R']) print("\t.c%P\n", p); } if(STORE(r) & r->regdiff.b[z] & bb) addmove(r, bn, rn, 1); r->regu |= rb; if(r->refbehind.b[z] & bb) for(r1 = r->p2; r1 != R; r1 = r1->p2link) if(r1->refahead.b[z] & bb) paint3(r1, bn, rb, rn); if(!(r->refahead.b[z] & bb)) break; r1 = r->s2; if(r1 != R) if(r1->refbehind.b[z] & bb) paint3(r1, bn, rb, rn); r = r->s1; if(r == R) break; if(r->act.b[z] & bb) break; if(!(r->refbehind.b[z] & bb)) break; } } void addreg(Adr *a, int rn) { a->sym = 0; a->name = D_NONE; a->type = D_REG; a->reg = rn; if(rn >= NREG) { a->type = D_FREG; a->reg = rn-NREG; } } /* * bit reg * 0 R9 * 1 R10 * ... ... * 4 R13 * 5 R16 * ... ... * 20 R31 */ long RtoB(int r) { if(r >= 9 && r <= 13) return 1L << (r-9); if(r >= 16 && r <= 31) return 1L << (r-11); return 0; } int BtoR(long b) { int r; b &= 0x001fffffL; if(b == 0) return 0; r = bitno(b) + 9; if(r >= 14) r += 2; return r; } /* * bit reg * 22 F4 * 23 F6 * ... ... * 31 F22 */ long FtoB(int f) { if(f < 4 || f > 22 || (f&1)) return 0; return 1L << (f/2 + 20); } BtoF(long b) { b &= 0xffc00000L; if(b == 0) return 0; return bitno(b)*2 - 40; }