Keep maximum frame extent in snap->topslot.

diff --git a/src/lj_asm.c b/src/lj_asm.c
index be6a11ff56cb03032901b14487bd6407b0763594..16671a8a0ad644a6ae7deccff49ba91f95cd743d 100644 (file)
--- a/src/lj_asm.c
+++ b/src/lj_asm.c
@@ -903,30 +903,6 @@ static uint32_t asm_callx_flags(ASMState *as, IRIns *ir)
   return (nargs | (ir->t.irt << CCI_OTSHIFT));
 }
 
-/* Get extent of the stack for a snapshot. */
-static BCReg asm_stack_extent(ASMState *as, SnapShot *snap, BCReg *ptopslot)
-{
-  SnapEntry *map = &as->T->snapmap[snap->mapofs];
-  MSize n, nent = snap->nent;
-  BCReg baseslot = 0, topslot = 0;
-  /* Must check all frames to find topslot (outer can be larger than inner). */
-  for (n = 0; n < nent; n++) {
-    SnapEntry sn = map[n];
-    if ((sn & SNAP_FRAME)) {
-      IRIns *ir = IR(snap_ref(sn));
-      GCfunc *fn = ir_kfunc(ir);
-      if (isluafunc(fn)) {
-       BCReg s = snap_slot(sn);
-       BCReg fs = s + funcproto(fn)->framesize;
-       if (fs > topslot) topslot = fs;
-       baseslot = s;
-      }
-    }
-  }
-  *ptopslot = topslot;
-  return baseslot;
-}
-
 /* Calculate stack adjustment. */
 static int32_t asm_stack_adjust(ASMState *as)
 {
@@ -1415,13 +1391,30 @@ static void asm_head_side(ASMState *as)
 
 /* -- Tail of trace ------------------------------------------------------- */
 
+/* Get base slot for a snapshot. */
+static BCReg asm_baseslot(ASMState *as, SnapShot *snap, int *gotframe)
+{
+  SnapEntry *map = &as->T->snapmap[snap->mapofs];
+  MSize n;
+  for (n = snap->nent; n > 0; n--) {
+    SnapEntry sn = map[n-1];
+    if ((sn & SNAP_FRAME)) {
+      *gotframe = 1;
+      return snap_slot(sn);
+    }
+  }
+  return 0;
+}
+
 /* Link to another trace. */
 static void asm_tail_link(ASMState *as)
 {
   SnapNo snapno = as->T->nsnap-1;  /* Last snapshot. */
   SnapShot *snap = &as->T->snap[snapno];
-  BCReg baseslot = asm_stack_extent(as, snap, &as->topslot);
+  int gotframe = 0;
+  BCReg baseslot = asm_baseslot(as, snap, &gotframe);
 
+  as->topslot = snap->topslot;
   checkmclim(as);
   ra_allocref(as, REF_BASE, RID2RSET(RID_BASE));
 
@@ -1454,8 +1447,8 @@ static void asm_tail_link(ASMState *as)
   /* Sync the interpreter state with the on-trace state. */
   asm_stack_restore(as, snap);
 
-  /* Root traces that grow the stack need to check the stack at the end. */
-  if (!as->parent && as->topslot)
+  /* Root traces that add frames need to check the stack at the end. */
+  if (!as->parent && gotframe)
     asm_stack_check(as, as->topslot, NULL, as->freeset & RSET_GPR, snapno);
 }
 

diff --git a/src/lj_jit.h b/src/lj_jit.h
index e80547ab547d416c56d4289a1fe2d721b46c4641..11dc973731df8f5fb9562eb993d211a2bdc6f10b 100644 (file)
--- a/src/lj_jit.h
+++ b/src/lj_jit.h
@@ -138,9 +138,9 @@ typedef struct SnapShot {
   uint16_t mapofs;     /* Offset into snapshot map. */
   IRRef1 ref;          /* First IR ref for this snapshot. */
   uint8_t nslots;      /* Number of valid slots. */
+  uint8_t topslot;     /* Maximum frame extent. */
   uint8_t nent;                /* Number of compressed entries. */
   uint8_t count;       /* Count of taken exits for this snapshot. */
-  uint8_t unused;
 } SnapShot;
 
 #define SNAPCOUNT_DONE 255     /* Already compiled and linked a side trace. */

diff --git a/src/lj_opt_loop.c b/src/lj_opt_loop.c
index 8d2935f635636358cd479d192318480802711b70..c3d115b2560b256b105c6d8b0608f2dd2cb29768 100644 (file)
--- a/src/lj_opt_loop.c
+++ b/src/lj_opt_loop.c
@@ -199,6 +199,7 @@ static void loop_subst_snap(jit_State *J, SnapShot *osnap,
   snap->mapofs = (uint16_t)nmapofs;
   snap->ref = (IRRef1)J->cur.nins;
   snap->nslots = nslots;
+  snap->topslot = osnap->topslot;
   snap->count = 0;
   nmap = &J->cur.snapmap[nmapofs];
   /* Substitute snapshot slots. */

diff --git a/src/lj_snap.c b/src/lj_snap.c
index a2025d8816d158bec7ed2d38674bcca2252f33f6..89f739825d1d830dd1d0fd1892eb80f1fbdd1da2 100644 (file)
--- a/src/lj_snap.c
+++ b/src/lj_snap.c
@@ -63,7 +63,8 @@ static MSize snapshot_slots(jit_State *J, SnapEntry *map, BCReg nslots)
     if (ref) {
       SnapEntry sn = SNAP_TR(s, tr);
       IRIns *ir = IR(ref);
-      if (ir->o == IR_SLOAD && ir->op1 == s && ref > retf) {
+      if (!(sn & (SNAP_CONT|SNAP_FRAME)) &&
+         ir->o == IR_SLOAD && ir->op1 == s && ref > retf) {
        /* No need to snapshot unmodified non-inherited slots. */
        if (!(ir->op2 & IRSLOAD_INHERIT))
          continue;
@@ -81,16 +82,19 @@ static MSize snapshot_slots(jit_State *J, SnapEntry *map, BCReg nslots)
 }
 
 /* Add frame links at the end of the snapshot. */
-static void snapshot_framelinks(jit_State *J, SnapEntry *map)
+static BCReg snapshot_framelinks(jit_State *J, SnapEntry *map)
 {
   cTValue *frame = J->L->base - 1;
   cTValue *lim = J->L->base - J->baseslot;
+  cTValue *ftop = frame + funcproto(frame_func(frame))->framesize;
   MSize f = 0;
   map[f++] = SNAP_MKPC(J->pc);  /* The current PC is always the first entry. */
   while (frame > lim) {  /* Backwards traversal of all frames above base. */
     if (frame_islua(frame)) {
       map[f++] = SNAP_MKPC(frame_pc(frame));
       frame = frame_prevl(frame);
+      if (frame + funcproto(frame_func(frame))->framesize > ftop)
+       ftop = frame + funcproto(frame_func(frame))->framesize;
     } else if (frame_iscont(frame)) {
       map[f++] = SNAP_MKFTSZ(frame_ftsz(frame));
       map[f++] = SNAP_MKPC(frame_contpc(frame));
@@ -102,6 +106,7 @@ static void snapshot_framelinks(jit_State *J, SnapEntry *map)
     }
   }
   lua_assert(f == (MSize)(1 + J->framedepth));
+  return (BCReg)(ftop - lim);
 }
 
 /* Take a snapshot of the current stack. */
@@ -114,7 +119,7 @@ static void snapshot_stack(jit_State *J, SnapShot *snap, MSize nsnapmap)
   lj_snap_grow_map(J, nsnapmap + nslots + (MSize)J->framedepth+1);
   p = &J->cur.snapmap[nsnapmap];
   nent = snapshot_slots(J, p, nslots);
-  snapshot_framelinks(J, p + nent);
+  snap->topslot = (uint8_t)snapshot_framelinks(J, p + nent);
   snap->mapofs = (uint16_t)nsnapmap;
   snap->ref = (IRRef1)J->cur.nins;
   snap->nent = (uint8_t)nent;
@@ -338,7 +343,6 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
   SnapEntry *map = &T->snapmap[snap->mapofs];
   SnapEntry *flinks = &T->snapmap[snap_nextofs(T, snap)-1];
   int32_t ftsz0;
-  BCReg nslots = snap->nslots;
   TValue *frame;
   BloomFilter rfilt = snap_renamefilter(T, snapno);
   const BCIns *pc = snap_pc(map[nent]);
@@ -348,9 +352,9 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
   setcframe_pc(cframe_raw(L->cframe), pc+1);
 
   /* Make sure the stack is big enough for the slots from the snapshot. */
-  if (LJ_UNLIKELY(L->base + nslots > tvref(L->maxstack))) {
+  if (LJ_UNLIKELY(L->base + snap->topslot > tvref(L->maxstack))) {
     L->top = curr_topL(L);
-    lj_state_growstack(L, nslots - curr_proto(L)->framesize);
+    lj_state_growstack(L, snap->topslot - curr_proto(L)->framesize);
   }
 
   /* Fill stack slots with data from the registers and spill slots. */
@@ -364,27 +368,9 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
     IRIns *ir = &T->ir[ref];
     if (irref_isk(ref)) {  /* Restore constant slot. */
       lj_ir_kvalue(L, o, ir);
-      if ((sn & (SNAP_CONT|SNAP_FRAME))) {
-       /* Overwrite tag with frame link. */
-       o->fr.tp.ftsz = s != 0 ? (int32_t)*flinks-- : ftsz0;
-       if ((sn & SNAP_FRAME)) {
-         GCfunc *fn = ir_kfunc(ir);
-         if (isluafunc(fn)) {
-           MSize framesize = funcproto(fn)->framesize;
-           L->base = ++o;
-           if (LJ_UNLIKELY(o + framesize > tvref(L->maxstack))) {
-             ptrdiff_t fsave = savestack(L, frame);
-             L->top = o;
-             lj_state_growstack(L, framesize);  /* Grow again. */
-             frame = restorestack(L, fsave);
-           }
-         }
-       }
-      }
     } else if (!(sn & SNAP_NORESTORE)) {
       IRType1 t = ir->t;
       RegSP rs = ir->prev;
-      lua_assert(!(sn & (SNAP_CONT|SNAP_FRAME)));
       if (LJ_UNLIKELY(bloomtest(rfilt, ref)))
        rs = snap_renameref(T, snapno, ref, rs);
       if (ra_hasspill(regsp_spill(rs))) {  /* Restore from spill slot. */
@@ -438,10 +424,14 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
            (uint32_t)ex->gpr[regsp_reg(rs)-RID_MIN_GPR];
       }
     }
+    if ((sn & (SNAP_CONT|SNAP_FRAME))) {  /* Overwrite tag with frame link. */
+      o->fr.tp.ftsz = s != 0 ? (int32_t)*flinks-- : ftsz0;
+      L->base = o+1;
+    }
   }
   switch (bc_op(*pc)) {
   case BC_CALLM: case BC_CALLMT: case BC_RETM: case BC_TSETM:
-    L->top = frame + nslots;
+    L->top = frame + snap->nslots;
     break;
   default:
     L->top = curr_topL(L);