/* Emit or eliminate collected PHIs. */
static void loop_emit_phi(jit_State *J, IRRef1 *subst, IRRef1 *phi, IRRef nphi)
{
int pass2 = 0;
IRRef i, nslots;
IRRef invar = J->chain[IR_LOOP];
/* Pass #1: mark redundant and potentially redundant PHIs. */
for (i = 0; i < nphi; i++) {
IRRef lref = phi[i];
IRRef rref = subst[lref];
if (lref == rref || rref == REF_DROP) { /* Invariants are redundant. */
irt_setmark(IR(lref)->t);
} else if (!(IR(rref)->op1 == lref || IR(rref)->op2 == lref)) {
/* Quick check for simple recurrences failed, need pass2. */
irt_setmark(IR(lref)->t);
pass2 = 1;
}
}
/* Pass #2: traverse variant part and clear marks of non-redundant PHIs. */
if (pass2) {
for (i = J->cur.nins-1; i > invar; i--) {
IRIns *ir = IR(i);
if (!irref_isk(ir->op1)) irt_clearmark(IR(ir->op1)->t);
if (!irref_isk(ir->op2)) irt_clearmark(IR(ir->op2)->t);
}
}
/* Pass #3: add PHIs for variant slots without a corresponding SLOAD. */
nslots = J->baseslot+J->maxslot;
for (i = 1; i < nslots; i++) {
IRRef ref = tref_ref(J->slot[i]);
while (!irref_isk(ref) && ref != subst[ref]) {
IRIns *ir = IR(ref);
irt_clearmark(ir->t); /* Unmark potential uses, too. */
if (irt_isphi(ir->t) || irt_ispri(ir->t))
break;
irt_setphi(ir->t);
if (nphi >= LJ_MAX_PHI)
lj_trace_err(J, LJ_TRERR_PHIOV);
phi[nphi++] = (IRRef1)ref;
ref = subst[ref];
if (ref > invar)
break;
}
}
/* Pass #4: emit PHI instructions or eliminate PHIs. */
for (i = 0; i < nphi; i++) {
IRRef lref = phi[i];
IRIns *ir = IR(lref);
if (!irt_ismarked(ir->t)) { /* Emit PHI if not marked. */
IRRef rref = subst[lref];
if (rref > invar)
irt_setphi(IR(rref)->t);
emitir_raw(IRT(IR_PHI, irt_type(ir->t)), lref, rref);
} else { /* Otherwise eliminate PHI. */
irt_clearmark(ir->t);
irt_clearphi(ir->t);
}
}
}
/* Emit the conversions collected during backpropagation. */
static IRRef narrow_conv_emit(jit_State *J, NarrowConv *nc)
{
/* The fins fields must be saved now -- emitir() overwrites them. */
IROpT guardot = irt_isguard(fins->t) ? IRTG(IR_ADDOV-IR_ADD, 0) : 0;
IROpT convot = fins->ot;
IRRef1 convop2 = fins->op2;
NarrowIns *next = nc->stack; /* List of instructions from backpropagation. */
NarrowIns *last = nc->sp;
NarrowIns *sp = nc->stack; /* Recycle the stack to store operands. */
while (next < last) { /* Simple stack machine to process the ins. list. */
NarrowIns ref = *next++;
IROpT op = narrow_op(ref);
if (op == NARROW_REF) {
*sp++ = ref;
} else if (op == NARROW_CONV) {
*sp++ = emitir_raw(convot, ref, convop2); /* Raw emit avoids a loop. */
} else if (op == NARROW_SEXT) {
lua_assert(sp >= nc->stack+1);
sp[-1] = emitir(IRT(IR_CONV, IRT_I64), sp[-1],
(IRT_I64<<5)|IRT_INT|IRCONV_SEXT);
} else if (op == NARROW_INT) {
lua_assert(next < last);
*sp++ = nc->t == IRT_I64 ?
lj_ir_kint64(J, (int64_t)(int32_t)*next++) :
lj_ir_kint(J, *next++);
} else { /* Regular IROpT. Pops two operands and pushes one result. */
IRRef mode = nc->mode;
lua_assert(sp >= nc->stack+2);
sp--;
/* Omit some overflow checks for array indexing. See comments above. */
if ((mode & IRCONV_CONVMASK) == IRCONV_INDEX) {
if (next == last && irref_isk(narrow_ref(sp[0])) &&
(uint32_t)IR(narrow_ref(sp[0]))->i + 0x40000000u < 0x80000000u)
guardot = 0;
else /* Otherwise cache a stronger check. */
mode += IRCONV_CHECK-IRCONV_INDEX;
}
sp[-1] = emitir(op+guardot, sp[-1], sp[0]);
/* Add to cache. */
if (narrow_ref(ref))
narrow_bpc_set(J, narrow_ref(ref), narrow_ref(sp[-1]), mode);
}
}
lua_assert(sp == nc->stack+1);
return nc->stack[0];
}
/* Add or merge a snapshot. */
void lj_snap_add(jit_State *J)
{
MSize nsnap = J->cur.nsnap;
MSize nsnapmap = J->cur.nsnapmap;
/* Merge if no ins. inbetween or if requested and no guard inbetween. */
if ((nsnap > 0 && J->cur.snap[nsnap-1].ref == J->cur.nins) ||
(J->mergesnap && !irt_isguard(J->guardemit))) {
if (nsnap == 1) { /* But preserve snap #0 PC. */
emitir_raw(IRT(IR_NOP, IRT_NIL), 0, 0);
goto nomerge;
}
nsnapmap = J->cur.snap[--nsnap].mapofs;
} else {
nomerge:
lj_snap_grow_buf(J, nsnap+1);
J->cur.nsnap = (uint16_t)(nsnap+1);
}
J->mergesnap = 0;
J->guardemit.irt = 0;
snapshot_stack(J, &J->cur.snap[nsnap], nsnapmap);
}
/* Unroll loop. */
static void loop_unroll(jit_State *J)
{
IRRef1 phi[LJ_MAX_PHI];
uint32_t nphi = 0;
IRRef1 *subst;
SnapNo onsnap;
SnapShot *osnap, *loopsnap;
SnapEntry *loopmap, *psentinel;
IRRef ins, invar;
/* Use temp buffer for substitution table.
** Only non-constant refs in [REF_BIAS,invar) are valid indexes.
** Caveat: don't call into the VM or run the GC or the buffer may be gone.
*/
invar = J->cur.nins;
subst = (IRRef1 *)lj_str_needbuf(J->L, &G(J->L)->tmpbuf,
(invar-REF_BIAS)*sizeof(IRRef1)) - REF_BIAS;
subst[REF_BASE] = REF_BASE;
/* LOOP separates the pre-roll from the loop body. */
emitir_raw(IRTG(IR_LOOP, IRT_NIL), 0, 0);
/* Grow snapshot buffer and map for copy-substituted snapshots.
** Need up to twice the number of snapshots minus #0 and loop snapshot.
** Need up to twice the number of entries plus fallback substitutions
** from the loop snapshot entries for each new snapshot.
** Caveat: both calls may reallocate J->cur.snap and J->cur.snapmap!
*/
onsnap = J->cur.nsnap;
lj_snap_grow_buf(J, 2*onsnap-2);
lj_snap_grow_map(J, J->cur.nsnapmap*2+(onsnap-2)*J->cur.snap[onsnap-1].nent);
/* The loop snapshot is used for fallback substitutions. */
loopsnap = &J->cur.snap[onsnap-1];
loopmap = &J->cur.snapmap[loopsnap->mapofs];
/* The PC of snapshot #0 and the loop snapshot must match. */
psentinel = &loopmap[loopsnap->nent];
lua_assert(*psentinel == J->cur.snapmap[J->cur.snap[0].nent]);
*psentinel = SNAP(255, 0, 0); /* Replace PC with temporary sentinel. */
/* Start substitution with snapshot #1 (#0 is empty for root traces). */
osnap = &J->cur.snap[1];
/* Copy and substitute all recorded instructions and snapshots. */
for (ins = REF_FIRST; ins < invar; ins++) {
IRIns *ir;
IRRef op1, op2;
if (ins >= osnap->ref) /* Instruction belongs to next snapshot? */
loop_subst_snap(J, osnap++, loopmap, subst); /* Copy-substitute it. */
/* Substitute instruction operands. */
ir = IR(ins);
op1 = ir->op1;
if (!irref_isk(op1)) op1 = subst[op1];
op2 = ir->op2;
if (!irref_isk(op2)) op2 = subst[op2];
if (irm_kind(lj_ir_mode[ir->o]) == IRM_N &&
op1 == ir->op1 && op2 == ir->op2) { /* Regular invariant ins? */
subst[ins] = (IRRef1)ins; /* Shortcut. */
} else {
/* Re-emit substituted instruction to the FOLD/CSE/etc. pipeline. */
IRType1 t = ir->t; /* Get this first, since emitir may invalidate ir. */
IRRef ref = tref_ref(emitir(ir->ot & ~IRT_ISPHI, op1, op2));
subst[ins] = (IRRef1)ref;
if (ref != ins) {
IRIns *irr = IR(ref);
if (ref < invar) { /* Loop-carried dependency? */
/* Potential PHI? */
if (!irref_isk(ref) && !irt_isphi(irr->t) && !irt_ispri(irr->t)) {
irt_setphi(irr->t);
if (nphi >= LJ_MAX_PHI)
lj_trace_err(J, LJ_TRERR_PHIOV);
phi[nphi++] = (IRRef1)ref;
}
/* Check all loop-carried dependencies for type instability. */
if (!irt_sametype(t, irr->t)) {
if (irt_isinteger(t) && irt_isinteger(irr->t))
continue;
else if (irt_isnum(t) && irt_isinteger(irr->t)) /* Fix int->num. */
ref = tref_ref(emitir(IRTN(IR_CONV), ref, IRCONV_NUM_INT));
else if (irt_isnum(irr->t) && irt_isinteger(t)) /* Fix num->int. */
ref = tref_ref(emitir(IRTGI(IR_CONV), ref,
IRCONV_INT_NUM|IRCONV_CHECK));
else
lj_trace_err(J, LJ_TRERR_TYPEINS);
subst[ins] = (IRRef1)ref;
irr = IR(ref);
goto phiconv;
}
} else if (ref != REF_DROP && irr->o == IR_CONV &&
ref > invar && irr->op1 < invar) {
/* May need an extra PHI for a CONV. */
ref = irr->op1;
irr = IR(ref);
phiconv:
if (ref < invar && !irref_isk(ref) && !irt_isphi(irr->t)) {
irt_setphi(irr->t);
if (nphi >= LJ_MAX_PHI)
lj_trace_err(J, LJ_TRERR_PHIOV);
phi[nphi++] = (IRRef1)ref;
}
}
}
}
}
if (!irt_isguard(J->guardemit)) /* Drop redundant snapshot. */
J->cur.nsnapmap = (uint16_t)J->cur.snap[--J->cur.nsnap].mapofs;
lua_assert(J->cur.nsnapmap <= J->sizesnapmap);
*psentinel = J->cur.snapmap[J->cur.snap[0].nent]; /* Restore PC. */
loop_emit_phi(J, subst, phi, nphi, onsnap);
}
/* Emit or eliminate collected PHIs. */
static void loop_emit_phi(jit_State *J, IRRef1 *subst, IRRef1 *phi, IRRef nphi,
SnapNo onsnap)
{
int passx = 0;
IRRef i, j, nslots;
IRRef invar = J->chain[IR_LOOP];
/* Pass #1: mark redundant and potentially redundant PHIs. */
for (i = 0, j = 0; i < nphi; i++) {
IRRef lref = phi[i];
IRRef rref = subst[lref];
if (lref == rref || rref == REF_DROP) { /* Invariants are redundant. */
irt_clearphi(IR(lref)->t);
} else {
phi[j++] = (IRRef1)lref;
if (!(IR(rref)->op1 == lref || IR(rref)->op2 == lref)) {
/* Quick check for simple recurrences failed, need pass2. */
irt_setmark(IR(lref)->t);
passx = 1;
}
}
}
nphi = j;
/* Pass #2: traverse variant part and clear marks of non-redundant PHIs. */
if (passx) {
SnapNo s;
for (i = J->cur.nins-1; i > invar; i--) {
IRIns *ir = IR(i);
if (!irref_isk(ir->op2)) irt_clearmark(IR(ir->op2)->t);
if (!irref_isk(ir->op1)) {
irt_clearmark(IR(ir->op1)->t);
if (ir->op1 < invar &&
ir->o >= IR_CALLN && ir->o <= IR_CARG) { /* ORDER IR */
ir = IR(ir->op1);
while (ir->o == IR_CARG) {
if (!irref_isk(ir->op2)) irt_clearmark(IR(ir->op2)->t);
if (irref_isk(ir->op1)) break;
ir = IR(ir->op1);
irt_clearmark(ir->t);
}
}
}
}
for (s = J->cur.nsnap-1; s >= onsnap; s--) {
SnapShot *snap = &J->cur.snap[s];
SnapEntry *map = &J->cur.snapmap[snap->mapofs];
MSize n, nent = snap->nent;
for (n = 0; n < nent; n++) {
IRRef ref = snap_ref(map[n]);
if (!irref_isk(ref)) irt_clearmark(IR(ref)->t);
}
}
}
/* Pass #3: add PHIs for variant slots without a corresponding SLOAD. */
nslots = J->baseslot+J->maxslot;
for (i = 1; i < nslots; i++) {
IRRef ref = tref_ref(J->slot[i]);
while (!irref_isk(ref) && ref != subst[ref]) {
IRIns *ir = IR(ref);
irt_clearmark(ir->t); /* Unmark potential uses, too. */
if (irt_isphi(ir->t) || irt_ispri(ir->t))
break;
irt_setphi(ir->t);
if (nphi >= LJ_MAX_PHI)
lj_trace_err(J, LJ_TRERR_PHIOV);
phi[nphi++] = (IRRef1)ref;
ref = subst[ref];
if (ref > invar)
break;
}
}
/* Pass #4: propagate non-redundant PHIs. */
while (passx) {
passx = 0;
for (i = 0; i < nphi; i++) {
IRRef lref = phi[i];
IRIns *ir = IR(lref);
if (!irt_ismarked(ir->t)) { /* Propagate only from unmarked PHIs. */
IRIns *irr = IR(subst[lref]);
if (irt_ismarked(irr->t)) { /* Right ref points to other PHI? */
irt_clearmark(irr->t); /* Mark that PHI as non-redundant. */
passx = 1; /* Retry. */
}
}
}
}
/* Pass #5: emit PHI instructions or eliminate PHIs. */
for (i = 0; i < nphi; i++) {
IRRef lref = phi[i];
IRIns *ir = IR(lref);
if (!irt_ismarked(ir->t)) { /* Emit PHI if not marked. */
IRRef rref = subst[lref];
if (rref > invar)
irt_setphi(IR(rref)->t);
emitir_raw(IRT(IR_PHI, irt_type(ir->t)), lref, rref);
} else { /* Otherwise eliminate PHI. */
irt_clearmark(ir->t);
irt_clearphi(ir->t);
}
}
}
/* Replay snapshot state to setup side trace. */
void lj_snap_replay(jit_State *J, GCtrace *T)
{
SnapShot *snap = &T->snap[J->exitno];
SnapEntry *map = &T->snapmap[snap->mapofs];
MSize n, nent = snap->nent;
BloomFilter seen = 0;
int pass23 = 0;
J->framedepth = 0;
/* Emit IR for slots inherited from parent snapshot. */
for (n = 0; n < nent; n++) {
SnapEntry sn = map[n];
BCReg s = snap_slot(sn);
IRRef ref = snap_ref(sn);
IRIns *ir = &T->ir[ref];
TRef tr;
/* The bloom filter avoids O(nent^2) overhead for de-duping slots. */
if (bloomtest(seen, ref) && (tr = snap_dedup(J, map, n, ref)) != 0)
goto setslot;
bloomset(seen, ref);
if (irref_isk(ref)) {
tr = snap_replay_const(J, ir);
} else if (!regsp_used(ir->prev)) {
pass23 = 1;
lua_assert(s != 0);
tr = s;
} else {
IRType t = irt_type(ir->t);
uint32_t mode = IRSLOAD_INHERIT|IRSLOAD_PARENT;
if (LJ_SOFTFP && (sn & SNAP_SOFTFPNUM)) t = IRT_NUM;
if (ir->o == IR_SLOAD) mode |= (ir->op2 & IRSLOAD_READONLY);
tr = emitir_raw(IRT(IR_SLOAD, t), s, mode);
}
setslot:
J->slot[s] = tr | (sn&(SNAP_CONT|SNAP_FRAME)); /* Same as TREF_* flags. */
J->framedepth += ((sn & (SNAP_CONT|SNAP_FRAME)) && s);
if ((sn & SNAP_FRAME))
J->baseslot = s+1;
}
if (pass23) {
IRIns *irlast = &T->ir[snap->ref];
pass23 = 0;
/* Emit dependent PVALs. */
for (n = 0; n < nent; n++) {
SnapEntry sn = map[n];
IRRef refp = snap_ref(sn);
IRIns *ir = &T->ir[refp];
if (regsp_reg(ir->r) == RID_SUNK) {
if (J->slot[snap_slot(sn)] != snap_slot(sn)) continue;
pass23 = 1;
lua_assert(ir->o == IR_TNEW || ir->o == IR_TDUP ||
ir->o == IR_CNEW || ir->o == IR_CNEWI);
if (ir->op1 >= T->nk) snap_pref(J, T, map, nent, seen, ir->op1);
if (ir->op2 >= T->nk) snap_pref(J, T, map, nent, seen, ir->op2);
if (LJ_HASFFI && ir->o == IR_CNEWI) {
if (LJ_32 && refp+1 < T->nins && (ir+1)->o == IR_HIOP)
snap_pref(J, T, map, nent, seen, (ir+1)->op2);
} else {
IRIns *irs;
for (irs = ir+1; irs < irlast; irs++)
if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
if (snap_pref(J, T, map, nent, seen, irs->op2) == 0)
snap_pref(J, T, map, nent, seen, T->ir[irs->op2].op1);
else if ((LJ_SOFTFP || (LJ_32 && LJ_HASFFI)) &&
irs+1 < irlast && (irs+1)->o == IR_HIOP)
snap_pref(J, T, map, nent, seen, (irs+1)->op2);
}
}
} else if (!irref_isk(refp) && !regsp_used(ir->prev)) {
lua_assert(ir->o == IR_CONV && ir->op2 == IRCONV_NUM_INT);
J->slot[snap_slot(sn)] = snap_pref(J, T, map, nent, seen, ir->op1);
}
}
/* Replay sunk instructions. */
for (n = 0; pass23 && n < nent; n++) {
SnapEntry sn = map[n];
IRRef refp = snap_ref(sn);
IRIns *ir = &T->ir[refp];
if (regsp_reg(ir->r) == RID_SUNK) {
TRef op1, op2;
if (J->slot[snap_slot(sn)] != snap_slot(sn)) { /* De-dup allocs. */
J->slot[snap_slot(sn)] = J->slot[J->slot[snap_slot(sn)]];
continue;
}
op1 = ir->op1;
if (op1 >= T->nk) op1 = snap_pref(J, T, map, nent, seen, op1);
op2 = ir->op2;
if (op2 >= T->nk) op2 = snap_pref(J, T, map, nent, seen, op2);
if (LJ_HASFFI && ir->o == IR_CNEWI) {
if (LJ_32 && refp+1 < T->nins && (ir+1)->o == IR_HIOP) {
lj_needsplit(J); /* Emit joining HIOP. */
op2 = emitir_raw(IRT(IR_HIOP, IRT_I64), op2,
snap_pref(J, T, map, nent, seen, (ir+1)->op2));
}
J->slot[snap_slot(sn)] = emitir(ir->ot & ~(IRT_MARK|IRT_ISPHI), op1, op2);
} else {
IRIns *irs;
TRef tr = emitir(ir->ot, op1, op2);
J->slot[snap_slot(sn)] = tr;
for (irs = ir+1; irs < irlast; irs++)
if (irs->r == RID_SINK && snap_sunk_store(T, ir, irs)) {
IRIns *irr = &T->ir[irs->op1];
TRef val, key = irr->op2, tmp = tr;
if (irr->o != IR_FREF) {
IRIns *irk = &T->ir[key];
if (irr->o == IR_HREFK)
key = lj_ir_kslot(J, snap_replay_const(J, &T->ir[irk->op1]),
irk->op2);
else
key = snap_replay_const(J, irk);
if (irr->o == IR_HREFK || irr->o == IR_AREF) {
IRIns *irf = &T->ir[irr->op1];
tmp = emitir(irf->ot, tmp, irf->op2);
}
}
tmp = emitir(irr->ot, tmp, key);
val = snap_pref(J, T, map, nent, seen, irs->op2);
if (val == 0) {
IRIns *irc = &T->ir[irs->op2];
lua_assert(irc->o == IR_CONV && irc->op2 == IRCONV_NUM_INT);
val = snap_pref(J, T, map, nent, seen, irc->op1);
val = emitir(IRTN(IR_CONV), val, IRCONV_NUM_INT);
} else if ((LJ_SOFTFP || (LJ_32 && LJ_HASFFI)) &&
irs+1 < irlast && (irs+1)->o == IR_HIOP) {
IRType t = IRT_I64;
if (LJ_SOFTFP && irt_type((irs+1)->t) == IRT_SOFTFP)
t = IRT_NUM;
lj_needsplit(J);
if (irref_isk(irs->op2) && irref_isk((irs+1)->op2)) {
uint64_t k = (uint32_t)T->ir[irs->op2].i +
((uint64_t)T->ir[(irs+1)->op2].i << 32);
val = lj_ir_k64(J, t == IRT_I64 ? IR_KINT64 : IR_KNUM,
lj_ir_k64_find(J, k));
} else {
val = emitir_raw(IRT(IR_HIOP, t), val,
snap_pref(J, T, map, nent, seen, (irs+1)->op2));
}
tmp = emitir(IRT(irs->o, t), tmp, val);
continue;
}
tmp = emitir(irs->ot, tmp, val);
} else if (LJ_HASFFI && irs->o == IR_XBAR && ir->o == IR_CNEW) {
emitir(IRT(IR_XBAR, IRT_NIL), 0, 0);
}
}
}
}
}
J->base = J->slot + J->baseslot;
J->maxslot = snap->nslots - J->baseslot;
lj_snap_add(J);
if (pass23) /* Need explicit GC step _after_ initial snapshot. */
emitir_raw(IRTG(IR_GCSTEP, IRT_NIL), 0, 0);
}
