Add LJ_FR2 mode: Two-slot frame info.

src/host/buildvm_libbc.h

@@ -3,6 +3,20 @@
 static const int libbc_endian = 0;
 
 static const uint8_t libbc_code[] = {
+#if LJ_FR2
+0,1,2,0,0,1,2,24,1,0,0,76,1,2,0,241,135,158,166,3,220,203,178,130,4,0,1,2,0,
+0,1,2,24,1,0,0,76,1,2,0,243,244,148,165,20,198,190,199,252,3,0,1,2,0,0,0,3,
+16,0,5,0,21,1,0,0,76,1,2,0,0,2,10,0,0,0,15,16,0,12,0,16,1,9,0,41,2,1,0,21,3,
+0,0,41,4,1,0,77,2,8,128,18,6,1,0,18,8,5,0,59,9,5,0,66,6,3,2,10,6,0,0,88,7,1,
+128,76,6,2,0,79,2,248,127,75,0,1,0,0,2,11,0,0,0,16,16,0,12,0,16,1,9,0,43,2,
+0,0,18,3,0,0,41,4,0,0,88,5,7,128,18,7,1,0,18,9,5,0,18,10,6,0,66,7,3,2,10,7,
+0,0,88,8,1,128,76,7,2,0,70,5,3,3,82,5,247,127,75,0,1,0,0,1,2,0,0,0,3,16,0,12,
+0,21,1,0,0,76,1,2,0,0,2,10,0,0,2,30,16,0,12,0,21,2,0,0,11,1,0,0,88,3,7,128,
+8,2,0,0,88,3,23,128,59,3,2,0,43,4,0,0,64,4,2,0,76,3,2,0,88,3,18,128,16,1,14,
+0,41,3,1,0,3,3,1,0,88,3,14,128,3,1,2,0,88,3,12,128,59,3,1,0,22,4,1,1,18,5,2,
+0,41,6,1,0,77,4,4,128,23,8,1,7,59,9,7,0,64,9,8,0,79,4,252,127,43,4,0,0,64,4,
+2,0,76,3,2,0,75,0,1,0,0,2,0
+#else
 0,1,2,0,0,1,2,24,1,0,0,76,1,2,0,241,135,158,166,3,220,203,178,130,4,0,1,2,0,
 0,1,2,24,1,0,0,76,1,2,0,243,244,148,165,20,198,190,199,252,3,0,1,2,0,0,0,3,
 16,0,5,0,21,1,0,0,76,1,2,0,0,2,9,0,0,0,15,16,0,12,0,16,1,9,0,41,2,1,0,21,3,
@@ -15,6 +29,7 @@ static const uint8_t libbc_code[] = {
 0,41,3,1,0,3,3,1,0,88,3,14,128,3,1,2,0,88,3,12,128,59,3,1,0,22,4,1,1,18,5,2,
 0,41,6,1,0,77,4,4,128,23,8,1,7,59,9,7,0,64,9,8,0,79,4,252,127,43,4,0,0,64,4,
 2,0,76,3,2,0,75,0,1,0,0,2,0
+#endif
 };
 
 static const struct { const char *name; int ofs; } libbc_map[] = {

src/lib_base.c

@@ -87,10 +87,11 @@ static int ffh_pairs(lua_State *L, MMS mm)
   cTValue *mo = lj_meta_lookup(L, o, mm);
   if ((LJ_52 || tviscdata(o)) && !tvisnil(mo)) {
     L->top = o+1;  /* Only keep one argument. */
-    copyTV(L, L->base-1, mo);  /* Replace callable. */
+    copyTV(L, L->base-1-LJ_FR2, mo);  /* Replace callable. */
     return FFH_TAILCALL;
   } else {
     if (!tvistab(o)) lj_err_argt(L, 1, LUA_TTABLE);
+    if (LJ_FR2) { copyTV(L, o-1, o); o--; }
     setfuncV(L, o-1, funcV(lj_lib_upvalue(L, 1)));
     if (mm == MM_pairs) setnilV(o+1); else setintV(o+1, 0);
     return FFH_RES(3);
@@ -132,7 +133,7 @@ LJLIB_ASM(setmetatable)		LJLIB_REC(.)
     lj_err_caller(L, LJ_ERR_PROTMT);
   setgcref(t->metatable, obj2gco(mt));
   if (mt) { lj_gc_objbarriert(L, t, mt); }
-  settabV(L, L->base-1, t);
+  settabV(L, L->base-1-LJ_FR2, t);
   return FFH_RES(1);
 }
 
@@ -145,6 +146,7 @@ LJLIB_CF(getfenv)		LJLIB_REC(.)
     o = lj_debug_frame(L, level, &level);
     if (o == NULL)
       lj_err_arg(L, 1, LJ_ERR_INVLVL);
+    if (LJ_FR2) o--;
   }
   fn = &gcval(o)->fn;
   settabV(L, L->top++, isluafunc(fn) ? tabref(fn->l.env) : tabref(L->env));
@@ -166,6 +168,7 @@ LJLIB_CF(setfenv)
     o = lj_debug_frame(L, level, &level);
     if (o == NULL)
       lj_err_arg(L, 1, LJ_ERR_INVLVL);
+    if (LJ_FR2) o--;
   }
   fn = &gcval(o)->fn;
   if (!isluafunc(fn))
@@ -258,7 +261,7 @@ LJLIB_ASM(tonumber)		LJLIB_REC(.)
   if (base == 10) {
     TValue *o = lj_lib_checkany(L, 1);
     if (lj_strscan_numberobj(o)) {
-      copyTV(L, L->base-1, o);
+      copyTV(L, L->base-1-LJ_FR2, o);
       return FFH_RES(1);
     }
 #if LJ_HASFFI
@@ -271,11 +274,11 @@ LJLIB_ASM(tonumber)		LJLIB_REC(.)
 	    ct->size <= 4 && !(ct->size == 4 && (ct->info & CTF_UNSIGNED))) {
 	  int32_t i;
 	  lj_cconv_ct_tv(cts, ctype_get(cts, CTID_INT32), (uint8_t *)&i, o, 0);
-	  setintV(L->base-1, i);
+	  setintV(L->base-1-LJ_FR2, i);
 	  return FFH_RES(1);
 	}
 	lj_cconv_ct_tv(cts, ctype_get(cts, CTID_DOUBLE),
-		       (uint8_t *)&(L->base-1)->n, o, 0);
+		       (uint8_t *)&(L->base-1-LJ_FR2)->n, o, 0);
 	return FFH_RES(1);
       }
     }
@@ -291,14 +294,14 @@ LJLIB_ASM(tonumber)		LJLIB_REC(.)
       while (lj_char_isspace((unsigned char)(*ep))) ep++;
       if (*ep == '\0') {
 	if (LJ_DUALNUM && LJ_LIKELY(ul < 0x80000000u))
-	  setintV(L->base-1, (int32_t)ul);
+	  setintV(L->base-1-LJ_FR2, (int32_t)ul);
 	else
-	  setnumV(L->base-1, (lua_Number)ul);
+	  setnumV(L->base-1-LJ_FR2, (lua_Number)ul);
 	return FFH_RES(1);
       }
     }
   }
-  setnilV(L->base-1);
+  setnilV(L->base-1-LJ_FR2);
   return FFH_RES(1);
 }
 
@@ -308,11 +311,11 @@ LJLIB_ASM(tostring)		LJLIB_REC(.)
   cTValue *mo;
   L->top = o+1;  /* Only keep one argument. */
   if (!tvisnil(mo = lj_meta_lookup(L, o, MM_tostring))) {
-    copyTV(L, L->base-1, mo);  /* Replace callable. */
+    copyTV(L, L->base-1-LJ_FR2, mo);  /* Replace callable. */
     return FFH_TAILCALL;
   }
   lj_gc_check(L);
-  setstrV(L, L->base-1, lj_strfmt_obj(L, L->base));
+  setstrV(L, L->base-1-LJ_FR2, lj_strfmt_obj(L, L->base));
   return FFH_RES(1);
 }
 
@@ -535,7 +538,7 @@ LJLIB_CF(coroutine_status)
   if (co == L) s = "running";
   else if (co->status == LUA_YIELD) s = "suspended";
   else if (co->status != 0) s = "dead";
-  else if (co->base > tvref(co->stack)+1) s = "normal";
+  else if (co->base > tvref(co->stack)+1+LJ_FR2) s = "normal";
   else if (co->top == co->base) s = "dead";
   else s = "suspended";
   lua_pushstring(L, s);
@@ -577,8 +580,8 @@ static int ffh_resume(lua_State *L, lua_State *co, int wrap)
       (co->status == 0 && co->top == co->base)) {
     ErrMsg em = co->cframe ? LJ_ERR_CORUN : LJ_ERR_CODEAD;
     if (wrap) lj_err_caller(L, em);
-    setboolV(L->base-1, 0);
-    setstrV(L, L->base, lj_err_str(L, em));
+    setboolV(L->base-1-LJ_FR2, 0);
+    setstrV(L, L->base-LJ_FR2, lj_err_str(L, em));
     return FFH_RES(2);
   }
   lj_state_growstack(co, (MSize)(L->top - L->base));

src/lib_bit.c

@@ -33,7 +33,7 @@ static int bit_result64(lua_State *L, CTypeID id, uint64_t x)
 {
   GCcdata *cd = lj_cdata_new_(L, id, 8);
   *(uint64_t *)cdataptr(cd) = x;
-  setcdataV(L, L->base-1, cd);
+  setcdataV(L, L->base-1-LJ_FR2, cd);
   return FFH_RES(1);
 }
 #else
@@ -56,7 +56,7 @@ LJLIB_ASM(bit_tobit)		LJLIB_REC(bit_tobit)
 {
 #if LJ_HASFFI
   CTypeID id = 0;
-  setintV(L->base-1, (int32_t)lj_carith_check64(L, 1, &id));
+  setintV(L->base-1-LJ_FR2, (int32_t)lj_carith_check64(L, 1, &id));
   return FFH_RES(1);
 #else
   lj_lib_checknumber(L, 1);

src/lib_ffi.c

@@ -138,7 +138,7 @@ static int ffi_index_meta(lua_State *L, CTState *cts, CType *ct, MMS mm)
       }
     }
     copyTV(L, base, L->top);
-    tv = L->top-1;
+    tv = L->top-1-LJ_FR2;
   }
   return lj_meta_tailcall(L, tv);
 }

src/lib_math.c

@@ -57,7 +57,7 @@ LJLIB_ASM(math_log)		LJLIB_REC(math_log)
 #else
     x = lj_vm_log2(x); y = 1.0 / lj_vm_log2(y);
 #endif
-    setnumV(L->base-1, x*y);  /* Do NOT join the expression to x / y. */
+    setnumV(L->base-1-LJ_FR2, x*y);  /* Do NOT join the expression to x / y. */
     return FFH_RES(1);
   }
   return FFH_RETRY;

src/lib_string.c

@@ -58,7 +58,7 @@ LJLIB_ASM(string_byte)		LJLIB_REC(string_range 0)
   lj_state_checkstack(L, (MSize)n);
   p = (const unsigned char *)strdata(s) + start;
   for (i = 0; i < n; i++)
-    setintV(L->base + i-1, p[i]);
+    setintV(L->base + i-1-LJ_FR2, p[i]);
   return FFH_RES(n);
 }
 
@@ -72,7 +72,7 @@ LJLIB_ASM(string_char)		LJLIB_REC(.)
       lj_err_arg(L, i, LJ_ERR_BADVAL);
     buf[i-1] = (char)k;
   }
-  setstrV(L, L->base-1, lj_str_new(L, buf, (size_t)nargs));
+  setstrV(L, L->base-1-LJ_FR2, lj_str_new(L, buf, (size_t)nargs));
   return FFH_RES(1);
 }
 

src/lj_api.c

@@ -284,8 +284,8 @@ LUA_API int lua_equal(lua_State *L, int idx1, int idx2)
     } else {
       L->top = base+2;
       lj_vm_call(L, base, 1+1);
-      L->top -= 2;
-      return tvistruecond(L->top+1);
+      L->top -= 2+LJ_FR2;
+      return tvistruecond(L->top+1+LJ_FR2);
     }
   }
 }
@@ -307,8 +307,8 @@ LUA_API int lua_lessthan(lua_State *L, int idx1, int idx2)
     } else {
       L->top = base+2;
       lj_vm_call(L, base, 1+1);
-      L->top -= 2;
-      return tvistruecond(L->top+1);
+      L->top -= 2+LJ_FR2;
+      return tvistruecond(L->top+1+LJ_FR2);
     }
   }
 }
@@ -704,8 +704,8 @@ LUA_API void lua_concat(lua_State *L, int n)
       n -= (int)(L->top - top);
       L->top = top+2;
       lj_vm_call(L, top, 1+1);
-      L->top--;
-      copyTV(L, L->top-1, L->top);
+      L->top -= 1+LJ_FR2;
+      copyTV(L, L->top-1, L->top+LJ_FR2);
     } while (--n > 0);
   } else if (n == 0) {  /* Push empty string. */
     setstrV(L, L->top, &G(L)->strempty);
@@ -724,8 +724,8 @@ LUA_API void lua_gettable(lua_State *L, int idx)
   if (v == NULL) {
     L->top += 2;
     lj_vm_call(L, L->top-2, 1+1);
-    L->top -= 2;
-    v = L->top+1;
+    L->top -= 2+LJ_FR2;
+    v = L->top+1+LJ_FR2;
   }
   copyTV(L, L->top-1, v);
 }
@@ -740,8 +740,8 @@ LUA_API void lua_getfield(lua_State *L, int idx, const char *k)
   if (v == NULL) {
     L->top += 2;
     lj_vm_call(L, L->top-2, 1+1);
-    L->top -= 2;
-    v = L->top+1;
+    L->top -= 2+LJ_FR2;
+    v = L->top+1+LJ_FR2;
   }
   copyTV(L, L->top, v);
   incr_top(L);
@@ -882,13 +882,14 @@ LUA_API void lua_settable(lua_State *L, int idx)
   o = lj_meta_tset(L, t, L->top-2);
   if (o) {
     /* NOBARRIER: lj_meta_tset ensures the table is not black. */
-    copyTV(L, o, L->top-1);
     L->top -= 2;
+    copyTV(L, o, L->top+1);
   } else {
-    L->top += 3;
-    copyTV(L, L->top-1, L->top-6);
-    lj_vm_call(L, L->top-3, 0+1);
-    L->top -= 3;
+    TValue *base = L->top;
+    copyTV(L, base+2, base-3-LJ_FR2);
+    L->top = base+3;
+    lj_vm_call(L, base, 0+1);
+    L->top -= 3+LJ_FR2;
   }
 }
 
@@ -902,14 +903,14 @@ LUA_API void lua_setfield(lua_State *L, int idx, const char *k)
   setstrV(L, &key, lj_str_newz(L, k));
   o = lj_meta_tset(L, t, &key);
   if (o) {
-    L->top--;
     /* NOBARRIER: lj_meta_tset ensures the table is not black. */
-    copyTV(L, o, L->top);
+    copyTV(L, o, --L->top);
   } else {
-    L->top += 3;
-    copyTV(L, L->top-1, L->top-6);
-    lj_vm_call(L, L->top-3, 0+1);
-    L->top -= 2;
+    TValue *base = L->top;
+    copyTV(L, base+2, base-3-LJ_FR2);
+    L->top = base+3;
+    lj_vm_call(L, base, 0+1);
+    L->top -= 2+LJ_FR2;
   }
 }
 
@@ -1016,11 +1017,24 @@ LUA_API const char *lua_setupvalue(lua_State *L, int idx, int n)
 
 /* -- Calls --------------------------------------------------------------- */
 
+#if LJ_FR2
+static TValue *api_call_base(lua_State *L, int nargs)
+{
+  TValue *o = L->top, *base = o - nargs;
+  L->top = o+1;
+  for (; o > base; o--) copyTV(L, o, o-1);
+  setnilV(o);
+  return o+1;
+}
+#else
+#define api_call_base(L, nargs)	(L->top - (nargs))
+#endif
+
 LUA_API void lua_call(lua_State *L, int nargs, int nresults)
 {
   api_check(L, L->status == 0 || L->status == LUA_ERRERR);
   api_checknelems(L, nargs+1);
-  lj_vm_call(L, L->top - nargs, nresults+1);
+  lj_vm_call(L, api_call_base(L, nargs), nresults+1);
 }
 
 LUA_API int lua_pcall(lua_State *L, int nargs, int nresults, int errfunc)
@@ -1038,7 +1052,7 @@ LUA_API int lua_pcall(lua_State *L, int nargs, int nresults, int errfunc)
     api_checkvalidindex(L, o);
     ef = savestack(L, o);
   }
-  status = lj_vm_pcall(L, L->top - nargs, nresults+1, ef);
+  status = lj_vm_pcall(L, api_call_base(L, nargs), nresults+1, ef);
   if (status) hook_restore(g, oldh);
   return status;
 }
@@ -1046,12 +1060,14 @@ LUA_API int lua_pcall(lua_State *L, int nargs, int nresults, int errfunc)
 static TValue *cpcall(lua_State *L, lua_CFunction func, void *ud)
 {
   GCfunc *fn = lj_func_newC(L, 0, getcurrenv(L));
+  TValue *top = L->top;
   fn->c.f = func;
-  setfuncV(L, L->top, fn);
-  setlightudV(L->top+1, checklightudptr(L, ud));
+  setfuncV(L, top++, fn);
+  if (LJ_FR2) setnilV(top++);
+  setlightudV(top++, checklightudptr(L, ud));
   cframe_nres(L->cframe) = 1+0;  /* Zero results. */
-  L->top += 2;
-  return L->top-1;  /* Now call the newly allocated C function. */
+  L->top = top;
+  return top-1;  /* Now call the newly allocated C function. */
 }
 
 LUA_API int lua_cpcall(lua_State *L, lua_CFunction func, void *ud)
@@ -1068,10 +1084,11 @@ LUA_API int lua_cpcall(lua_State *L, lua_CFunction func, void *ud)
 LUALIB_API int luaL_callmeta(lua_State *L, int idx, const char *field)
 {
   if (luaL_getmetafield(L, idx, field)) {
-    TValue *base = L->top--;
-    copyTV(L, base, index2adr(L, idx));
-    L->top = base+1;
-    lj_vm_call(L, base, 1+1);
+    TValue *top = L->top--;
+    if (LJ_FR2) setnilV(top++);
+    copyTV(L, top++, index2adr(L, idx));
+    L->top = top;
+    lj_vm_call(L, top-1, 1+1);
     return 1;
   }
   return 0;
@@ -1098,12 +1115,14 @@ LUA_API int lua_yield(lua_State *L, int nresults)
     } else {  /* Yield from hook: add a pseudo-frame. */
       TValue *top = L->top;
       hook_leave(g);
-      top->u64 = cframe_multres(cf);
-      setcont(top+1, lj_cont_hook);
-      setframe_pc(top+1, cframe_pc(cf)-1);
-      setframe_gc(top+2, obj2gco(L), LJ_TTHREAD);
-      setframe_ftsz(top+2, ((char *)(top+3)-(char *)L->base)+FRAME_CONT);
-      L->top = L->base = top+3;
+      (top++)->u64 = cframe_multres(cf);
+      setcont(top, lj_cont_hook);
+      if (LJ_FR2) top++;
+      setframe_pc(top, cframe_pc(cf)-1);
+      if (LJ_FR2) top++;
+      setframe_gc(top, obj2gco(L), LJ_TTHREAD);
+      setframe_ftsz(top, ((char *)(top+1)-(char *)L->base)+FRAME_CONT);
+      L->top = L->base = top+1;
 #if LJ_TARGET_X64
       lj_err_throw(L, LUA_YIELD);
 #else
@@ -1120,7 +1139,9 @@ LUA_API int lua_yield(lua_State *L, int nresults)
 LUA_API int lua_resume(lua_State *L, int nargs)
 {
   if (L->cframe == NULL && L->status <= LUA_YIELD)
-    return lj_vm_resume(L, L->top - nargs, 0, 0);
+    return lj_vm_resume(L,
+      L->status == 0 ? api_call_base(L, nargs) : L->top - nargs,
+      0, 0);
   L->top = L->base;
   setstrV(L, L->top, lj_err_str(L, LJ_ERR_COSUSP));
   incr_top(L);

src/lj_arch.h

@@ -365,8 +365,11 @@
 #endif
 #endif
 
+/* 2-slot frame info. */
+#define LJ_FR2			0
+
 /* Disable or enable the JIT compiler. */
-#if defined(LUAJIT_DISABLE_JIT) || defined(LJ_ARCH_NOJIT) || defined(LJ_OS_NOJIT)
+#if defined(LUAJIT_DISABLE_JIT) || defined(LJ_ARCH_NOJIT) || defined(LJ_OS_NOJIT) || LJ_FR2
 #define LJ_HASJIT		0
 #else
 #define LJ_HASJIT		1

src/lj_asm.c

@@ -1913,7 +1913,7 @@ static void asm_tail_link(ASMState *as)
     mres = (int32_t)(snap->nslots - baseslot);
     switch (bc_op(*pc)) {
     case BC_CALLM: case BC_CALLMT:
-      mres -= (int32_t)(1 + bc_a(*pc) + bc_c(*pc)); break;
+      mres -= (int32_t)(1 + LJ_FR2 + bc_a(*pc) + bc_c(*pc)); break;
     case BC_RETM: mres -= (int32_t)(bc_a(*pc) + bc_d(*pc)); break;
     case BC_TSETM: mres -= (int32_t)bc_a(*pc); break;
     default: if (bc_op(*pc) < BC_FUNCF) mres = 0; break;

src/lj_asm_arm.h

@@ -481,7 +481,7 @@ static void asm_retf(ASMState *as, IRIns *ir)
 {
   Reg base = ra_alloc1(as, REF_BASE, RSET_GPR);
   void *pc = ir_kptr(IR(ir->op2));
-  int32_t delta = 1+bc_a(*((const BCIns *)pc - 1));
+  int32_t delta = 1+LJ_FR2+bc_a(*((const BCIns *)pc - 1));
   as->topslot -= (BCReg)delta;
   if ((int32_t)as->topslot < 0) as->topslot = 0;
   irt_setmark(IR(REF_BASE)->t);  /* Children must not coalesce with BASE reg. */

src/lj_asm_mips.h

@@ -372,7 +372,7 @@ static void asm_retf(ASMState *as, IRIns *ir)
 {
   Reg base = ra_alloc1(as, REF_BASE, RSET_GPR);
   void *pc = ir_kptr(IR(ir->op2));
-  int32_t delta = 1+bc_a(*((const BCIns *)pc - 1));
+  int32_t delta = 1+LJ_FR2+bc_a(*((const BCIns *)pc - 1));
   as->topslot -= (BCReg)delta;
   if ((int32_t)as->topslot < 0) as->topslot = 0;
   irt_setmark(IR(REF_BASE)->t);  /* Children must not coalesce with BASE reg. */

src/lj_asm_ppc.h

@@ -361,7 +361,7 @@ static void asm_retf(ASMState *as, IRIns *ir)
 {
   Reg base = ra_alloc1(as, REF_BASE, RSET_GPR);
   void *pc = ir_kptr(IR(ir->op2));
-  int32_t delta = 1+bc_a(*((const BCIns *)pc - 1));
+  int32_t delta = 1+LJ_FR2+bc_a(*((const BCIns *)pc - 1));
   as->topslot -= (BCReg)delta;
   if ((int32_t)as->topslot < 0) as->topslot = 0;
   irt_setmark(IR(REF_BASE)->t);  /* Children must not coalesce with BASE reg. */

src/lj_asm_x86.h

@@ -643,7 +643,7 @@ static void asm_retf(ASMState *as, IRIns *ir)
 {
   Reg base = ra_alloc1(as, REF_BASE, RSET_GPR);
   void *pc = ir_kptr(IR(ir->op2));
-  int32_t delta = 1+bc_a(*((const BCIns *)pc - 1));
+  int32_t delta = 1+LJ_FR2+bc_a(*((const BCIns *)pc - 1));
   as->topslot -= (BCReg)delta;
   if ((int32_t)as->topslot < 0) as->topslot = 0;
   irt_setmark(IR(REF_BASE)->t);  /* Children must not coalesce with BASE reg. */

src/lj_bcdump.h

@@ -42,8 +42,9 @@
 #define BCDUMP_F_BE		0x01
 #define BCDUMP_F_STRIP		0x02
 #define BCDUMP_F_FFI		0x04
+#define BCDUMP_F_FR2		0x08
 
-#define BCDUMP_F_KNOWN		(BCDUMP_F_FFI*2-1)
+#define BCDUMP_F_KNOWN		(BCDUMP_F_FR2*2-1)
 
 /* Type codes for the GC constants of a prototype. Plus length for strings. */
 enum {

src/lj_bcread.c

@@ -394,6 +394,7 @@ static int bcread_header(LexState *ls)
       bcread_byte(ls) != BCDUMP_VERSION) return 0;
   bcread_flags(ls) = flags = bcread_uleb128(ls);
   if ((flags & ~(BCDUMP_F_KNOWN)) != 0) return 0;
+  if ((flags & BCDUMP_F_FR2) != LJ_FR2*BCDUMP_F_FR2) return 0;
   if ((flags & BCDUMP_F_FFI)) {
 #if LJ_HASFFI
     lua_State *L = ls->L;

src/lj_bcwrite.c

@@ -309,8 +309,9 @@ static void bcwrite_header(BCWriteCtx *ctx)
   *p++ = BCDUMP_HEAD3;
   *p++ = BCDUMP_VERSION;
   *p++ = (ctx->strip ? BCDUMP_F_STRIP : 0) +
-		       (LJ_BE ? BCDUMP_F_BE : 0) +
-		       ((ctx->pt->flags & PROTO_FFI) ? BCDUMP_F_FFI : 0);
+	 LJ_BE*BCDUMP_F_BE +
+	 ((ctx->pt->flags & PROTO_FFI) ? BCDUMP_F_FFI : 0) +
+	 LJ_FR2*BCDUMP_F_FR2;
   if (!ctx->strip) {
     p = lj_strfmt_wuleb128(p, len);
     p = lj_buf_wmem(p, name, len);

src/lj_ccallback.c

@@ -422,7 +422,7 @@ static void callback_conv_args(CTState *cts, lua_State *L)
 
   if (slot < cts->cb.sizeid && (id = cts->cb.cbid[slot]) != 0) {
     ct = ctype_get(cts, id);
-    rid = ctype_cid(ct->info);
+    rid = ctype_cid(ct->info);  /* Return type. x86: +(spadj<<16). */
     fn = funcV(lj_tab_getint(cts->miscmap, (int32_t)slot));
     fntp = LJ_TFUNC;
   } else {  /* Must set up frame first, before throwing the error. */
@@ -431,9 +431,16 @@ static void callback_conv_args(CTState *cts, lua_State *L)
     fn = (GCfunc *)L;
     fntp = LJ_TTHREAD;
   }
-  o->u32.lo = LJ_CONT_FFI_CALLBACK;  /* Continuation returns from callback. */
-  o->u32.hi = rid;  /* Return type. x86: +(spadj<<16). */
-  o++;
+  /* Continuation returns from callback. */
+  if (LJ_FR2) {
+    (o++)->u64 = LJ_CONT_FFI_CALLBACK;
+    (o++)->u64 = rid;
+    o++;
+  } else {
+    o->u32.lo = LJ_CONT_FFI_CALLBACK;
+    o->u32.hi = rid;
+    o++;
+  }
   setframe_gc(o, obj2gco(fn), fntp);
   setframe_ftsz(o, ((char *)(o+1) - (char *)L->base) + FRAME_CONT);
   L->top = L->base = ++o;
@@ -486,8 +493,13 @@ static void callback_conv_args(CTState *cts, lua_State *L)
   L->top = o;
 #if LJ_TARGET_X86
   /* Store stack adjustment for returns from non-cdecl callbacks. */
-  if (ctype_cconv(ct->info) != CTCC_CDECL)
+  if (ctype_cconv(ct->info) != CTCC_CDECL) {
+#if LJ_FR2
+    (L->base-3)->u64 |= (nsp << (16+2));
+#else
     (L->base-2)->u32.hi |= (nsp << (16+2));
+#endif
+  }
 #endif
   while (gcsteps-- > 0)
     lj_gc_check(L);
@@ -496,7 +508,11 @@ static void callback_conv_args(CTState *cts, lua_State *L)
 /* Convert Lua object to callback result. */
 static void callback_conv_result(CTState *cts, lua_State *L, TValue *o)
 {
+#if LJ_FR2
+  CType *ctr = ctype_raw(cts, (uint16_t)(L->base-3)->u64);
+#else
   CType *ctr = ctype_raw(cts, (uint16_t)(L->base-2)->u32.hi);
+#endif
 #if LJ_TARGET_X86
   cts->cb.gpr[2] = 0;
 #endif
@@ -565,7 +581,7 @@ void LJ_FASTCALL lj_ccallback_leave(CTState *cts, TValue *o)
   }
   callback_conv_result(cts, L, o);
   /* Finally drop C frame and continuation frame. */
-  L->top -= 2;
+  L->top -= 2+2*LJ_FR2;
   L->base = obase;
   L->cframe = cframe_prev(L->cframe);
   cts->cb.slot = 0;  /* Blacklist C function that called the callback. */

src/lj_crecord.c

@@ -1123,7 +1123,7 @@ static void crec_snap_caller(jit_State *J)
   ptrdiff_t delta;
   if (!frame_islua(base-1) || J->framedepth <= 0)
     lj_trace_err(J, LJ_TRERR_NYICALL);
-  J->pc = frame_pc(base-1); delta = 1+bc_a(J->pc[-1]);
+  J->pc = frame_pc(base-1); delta = 1+LJ_FR2+bc_a(J->pc[-1]);
   L->top = base; L->base = base - delta;
   J->base[-1] = TREF_FALSE;
   J->base -= delta; J->baseslot -= (BCReg)delta;

src/lj_debug.c

@@ -24,7 +24,7 @@
 /* Get frame corresponding to a level. */
 cTValue *lj_debug_frame(lua_State *L, int level, int *size)
 {
-  cTValue *frame, *nextframe, *bot = tvref(L->stack);
+  cTValue *frame, *nextframe, *bot = tvref(L->stack)+LJ_FR2;
   /* Traverse frames backwards. */
   for (nextframe = frame = L->base-1; frame > bot; ) {
     if (frame_gc(frame) == obj2gco(L))
@@ -184,7 +184,7 @@ static TValue *debug_localname(lua_State *L, const lua_Debug *ar,
   TValue *nextframe = size ? frame + size : NULL;
   GCfunc *fn = frame_func(frame);
   BCPos pc = debug_framepc(L, fn, nextframe);
-  if (!nextframe) nextframe = L->top;
+  if (!nextframe) nextframe = L->top+LJ_FR2;
   if ((int)slot1 < 0) {  /* Negative slot number is for varargs. */
     if (pc != NO_BCPOS) {
       GCproto *pt = funcproto(fn);
@@ -194,7 +194,7 @@ static TValue *debug_localname(lua_State *L, const lua_Debug *ar,
 	  nextframe = frame;
 	  frame = frame_prevd(frame);
 	}
-	if (frame + slot1 < nextframe) {
+	if (frame + slot1+LJ_FR2 < nextframe) {
 	  *name = "(*vararg)";
 	  return frame+slot1;
 	}
@@ -205,7 +205,7 @@ static TValue *debug_localname(lua_State *L, const lua_Debug *ar,
   if (pc != NO_BCPOS &&
       (*name = debug_varname(funcproto(fn), pc, slot1-1)) != NULL)
     ;
-  else if (slot1 > 0 && frame + slot1 < nextframe)
+  else if (slot1 > 0 && frame + slot1+LJ_FR2 < nextframe)
     *name = "(*temporary)";
   return frame+slot1;
 }
@@ -268,7 +268,7 @@ restart:
 	*name = strdata(gco2str(proto_kgc(pt, ~(ptrdiff_t)bc_c(ins))));
 	if (ip > proto_bc(pt)) {
 	  BCIns insp = ip[-1];
-	  if (bc_op(insp) == BC_MOV && bc_a(insp) == ra+1 &&
+	  if (bc_op(insp) == BC_MOV && bc_a(insp) == ra+1+LJ_FR2 &&
 	      bc_d(insp) == bc_b(ins))
 	    return "method";
 	}
@@ -290,7 +290,7 @@ const char *lj_debug_funcname(lua_State *L, cTValue *frame, const char **name)
   cTValue *pframe;
   GCfunc *fn;
   BCPos pc;
-  if (frame <= tvref(L->stack))
+  if (frame <= tvref(L->stack)+LJ_FR2)
     return NULL;
   if (frame_isvarg(frame))
     frame = frame_prevd(frame);

src/lj_dispatch.c

@@ -393,7 +393,7 @@ static BCReg cur_topslot(GCproto *pt, const BCIns *pc, uint32_t nres)
   if (bc_op(ins) == BC_UCLO)
     ins = pc[bc_j(ins)];
   switch (bc_op(ins)) {
-  case BC_CALLM: case BC_CALLMT: return bc_a(ins) + bc_c(ins) + nres-1+1;
+  case BC_CALLM: case BC_CALLMT: return bc_a(ins) + bc_c(ins) + nres-1+1+LJ_FR2;
   case BC_RETM: return bc_a(ins) + bc_d(ins) + nres-1;
   case BC_TSETM: return bc_a(ins) + nres-1;
   default: return pt->framesize;

src/lj_err.c

@@ -106,7 +106,7 @@ static void *err_unwind(lua_State *L, void *stopcf, int errcode)
 	return cf;
       }
     }
-    if (frame <= tvref(L->stack))
+    if (frame <= tvref(L->stack)+LJ_FR2)
       break;
     switch (frame_typep(frame)) {
     case FRAME_LUA:  /* Lua frame. */
@@ -171,7 +171,7 @@ static void *err_unwind(lua_State *L, void *stopcf, int errcode)
   }
   /* No C frame. */
   if (errcode) {
-    L->base = tvref(L->stack)+1;
+    L->base = tvref(L->stack)+1+LJ_FR2;
     L->cframe = NULL;
     unwindstack(L, L->base);
     if (G(L)->panic)
@@ -494,7 +494,7 @@ LJ_NOINLINE void lj_err_mem(lua_State *L)
 /* Find error function for runtime errors. Requires an extra stack traversal. */
 static ptrdiff_t finderrfunc(lua_State *L)
 {
-  cTValue *frame = L->base-1, *bot = tvref(L->stack);
+  cTValue *frame = L->base-1, *bot = tvref(L->stack)+LJ_FR2;
   void *cf = L->cframe;
   while (frame > bot && cf) {
     while (cframe_nres(cframe_raw(cf)) < 0) {  /* cframe without frame? */
@@ -530,8 +530,8 @@ static ptrdiff_t finderrfunc(lua_State *L)
       break;
     case FRAME_PCALL:
     case FRAME_PCALLH:
-      if (frame_ftsz(frame) >= (ptrdiff_t)(2*sizeof(TValue)))  /* xpcall? */
-	return savestack(L, frame-1);  /* Point to xpcall's errorfunc. */
+      if (frame_func(frame_prevd(frame))->c.ffid == FF_xpcall)
+	return savestack(L, frame_prevd(frame)+1);  /* xpcall's errorfunc. */
       return 0;
     default:
       lua_assert(0);
@@ -554,8 +554,9 @@ LJ_NOINLINE void lj_err_run(lua_State *L)
       lj_err_throw(L, LUA_ERRERR);
     }
     L->status = LUA_ERRERR;
-    copyTV(L, top, top-1);
+    copyTV(L, top+LJ_FR2, top-1);
     copyTV(L, top-1, errfunc);
+    if (LJ_FR2) setnilV(top++);
     L->top = top+1;
     lj_vm_call(L, top, 1+1);  /* Stack: |errfunc|msg| -> |msg| */
   }
@@ -630,6 +631,7 @@ LJ_NOINLINE void lj_err_optype_call(lua_State *L, TValue *o)
   const BCIns *pc = cframe_Lpc(L);
   if (((ptrdiff_t)pc & FRAME_TYPE) != FRAME_LUA) {
     const char *tname = lj_typename(o);
+    if (LJ_FR2) o++;
     setframe_pc(o, pc);
     setframe_gc(o, obj2gco(L), LJ_TTHREAD);
     L->top = L->base = o+1;

src/lj_ffrecord.c

@@ -107,6 +107,7 @@ static void recff_stitch(jit_State *J)
   TValue *pframe = frame_prevl(base-1);
   TRef trcont;
 
+  lua_assert(!LJ_FR2);  /* TODO_FR2: handle frame shift. */
   /* Move func + args up in Lua stack and insert continuation. */
   memmove(&base[1], &base[-1], sizeof(TValue)*(J->maxslot+1));
   setframe_ftsz(base+1, ((char *)(base+1) - (char *)pframe) + FRAME_CONT);
@@ -466,6 +467,7 @@ static void LJ_FASTCALL recff_xpcall(jit_State *J, RecordFFData *rd)
     TValue argv0, argv1;
     TRef tmp;
     int errcode;
+    lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
     /* Swap function and traceback. */
     tmp = J->base[0]; J->base[0] = J->base[1]; J->base[1] = tmp;
     copyTV(J->L, &argv0, &rd->argv[0]);

src/lj_frame.h

@@ -11,7 +11,16 @@
 
 /* -- Lua stack frame ----------------------------------------------------- */
 
-/* Frame type markers in callee function slot (callee base-1). */
+/* Frame type markers in LSB of PC (4-byte aligned) or delta (8-byte aligned:
+**
+**    PC  00  Lua frame
+** delta 001  C frame
+** delta 010  Continuation frame
+** delta 011  Lua vararg frame
+** delta 101  cpcall() frame
+** delta 110  ff pcall() frame
+** delta 111  ff pcall() frame with active hook
+*/
 enum {
   FRAME_LUA, FRAME_C, FRAME_CONT, FRAME_VARG,
   FRAME_LUAP, FRAME_CP, FRAME_PCALL, FRAME_PCALLH
@@ -21,12 +30,47 @@ enum {
 #define FRAME_TYPEP		(FRAME_TYPE|FRAME_P)
 
 /* Macros to access and modify Lua frames. */
+#if LJ_FR2
+/* Two-slot frame info, required for 64 bit PC/GCRef:
+**
+**                   base-2  base-1      |  base  base+1 ...
+**                  [func   PC/delta/ft] | [slots ...]
+**                  ^-- frame            | ^-- base   ^-- top
+**
+** Continuation frames:
+**
+**   base-4  base-3  base-2  base-1      |  base  base+1 ...
+**  [cont      PC ] [func   PC/delta/ft] | [slots ...]
+**                  ^-- frame            | ^-- base   ^-- top
+*/
+#define frame_gc(f)		(gcval((f)-1))
+#define frame_ftsz(f)		((ptrdiff_t)(f)->ftsz)
+#define frame_pc(f)		((const BCIns *)frame_ftsz(f))
+#define setframe_gc(f, p, tp)	(setgcVraw((f)-1, (p), (tp)))
+#define setframe_ftsz(f, sz)	((f)->ftsz = (sz))
+#define setframe_pc(f, pc)	((f)->ftsz = (int64_t)(intptr_t)(pc))
+#else
+/* One-slot frame info, sufficient for 32 bit PC/GCRef:
+**
+**              base-1              |  base  base+1 ...
+**              lo     hi           |
+**             [func | PC/delta/ft] | [slots ...]
+**             ^-- frame            | ^-- base   ^-- top
+**
+** Continuation frames:
+**
+**  base-2      base-1              |  base  base+1 ...
+**  lo     hi   lo     hi           |
+** [cont | PC] [func | PC/delta/ft] | [slots ...]
+**             ^-- frame            | ^-- base   ^-- top
+*/
 #define frame_gc(f)		(gcref((f)->fr.func))
 #define frame_ftsz(f)		((ptrdiff_t)(f)->fr.tp.ftsz)
 #define frame_pc(f)		(mref((f)->fr.tp.pcr, const BCIns))
 #define setframe_gc(f, p, tp)	(setgcref((f)->fr.func, (p)), UNUSED(tp))
 #define setframe_ftsz(f, sz)	((f)->fr.tp.ftsz = (int32_t)(sz))
 #define setframe_pc(f, pc)	(setmref((f)->fr.tp.pcr, (pc)))
+#endif
 
 #define frame_type(f)		(frame_ftsz(f) & FRAME_TYPE)
 #define frame_typep(f)		(frame_ftsz(f) & FRAME_TYPEP)
@@ -42,9 +86,16 @@ enum {
 
 enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
 
+#if LJ_FR2
+#define frame_contpc(f)		(frame_pc((f)-2))
+#define frame_contv(f)		(((f)-3)->u64)
+#else
 #define frame_contpc(f)		(frame_pc((f)-1))
 #define frame_contv(f)		(((f)-1)->u32.lo)
-#if LJ_64
+#endif
+#if LJ_FR2
+#define frame_contf(f)		((ASMFunction)(uintptr_t)((f)-3)->u64)
+#elif LJ_64
 #define frame_contf(f) \
   ((ASMFunction)(void *)((intptr_t)lj_vm_asm_begin + \
 			 (intptr_t)(int32_t)((f)-1)->u32.lo))
@@ -54,7 +105,7 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
 #define frame_iscont_fficb(f) \
   (LJ_HASFFI && frame_contv(f) == LJ_CONT_FFI_CALLBACK)
 
-#define frame_prevl(f)		((f) - (1+bc_a(frame_pc(f)[-1])))
+#define frame_prevl(f)		((f) - (1+LJ_FR2+bc_a(frame_pc(f)[-1])))
 #define frame_prevd(f)		((TValue *)((char *)(f) - frame_sized(f)))
 #define frame_prev(f)		(frame_islua(f)?frame_prevl(f):frame_prevd(f))
 /* Note: this macro does not skip over FRAME_VARG. */

src/lj_gc.c

@@ -268,12 +268,12 @@ static MSize gc_traverse_frames(global_State *g, lua_State *th)
 {
   TValue *frame, *top = th->top-1, *bot = tvref(th->stack);
   /* Note: extra vararg frame not skipped, marks function twice (harmless). */
-  for (frame = th->base-1; frame > bot; frame = frame_prev(frame)) {
+  for (frame = th->base-1; frame > bot+LJ_FR2; frame = frame_prev(frame)) {
     GCfunc *fn = frame_func(frame);
     TValue *ftop = frame;
     if (isluafunc(fn)) ftop += funcproto(fn)->framesize;
     if (ftop > top) top = ftop;
-    gc_markobj(g, fn);  /* Need to mark hidden function (or L). */
+    if (!LJ_FR2) gc_markobj(g, fn);  /* Need to mark hidden function (or L). */
   }
   top++;  /* Correct bias of -1 (frame == base-1). */
   if (top > tvref(th->maxstack)) top = tvref(th->maxstack);
@@ -284,7 +284,7 @@ static MSize gc_traverse_frames(global_State *g, lua_State *th)
 static void gc_traverse_thread(global_State *g, lua_State *th)
 {
   TValue *o, *top = th->top;
-  for (o = tvref(th->stack)+1; o < top; o++)
+  for (o = tvref(th->stack)+1+LJ_FR2; o < top; o++)
     gc_marktv(g, o);
   if (g->gc.state == GCSatomic) {
     top = tvref(th->stack) + th->stacksize;
@@ -456,13 +456,14 @@ static void gc_call_finalizer(global_State *g, lua_State *L,
   int errcode;
   TValue *top;
   lj_trace_abort(g);
-  top = L->top;
-  L->top = top+2;
   hook_entergc(g);  /* Disable hooks and new traces during __gc. */
   g->gc.threshold = LJ_MAX_MEM;  /* Prevent GC steps. */
-  copyTV(L, top, mo);
-  setgcV(L, top+1, o, ~o->gch.gct);
-  errcode = lj_vm_pcall(L, top+1, 1+0, -1);  /* Stack: |mo|o| -> | */
+  top = L->top;
+  copyTV(L, top++, mo);
+  if (LJ_FR2) setnilV(top++);
+  setgcV(L, top, o, ~o->gch.gct);
+  L->top = top+1;
+  errcode = lj_vm_pcall(L, top, 1+0, -1);  /* Stack: |mo|o| -> | */
   hook_restore(g, oldh);
   g->gc.threshold = oldt;  /* Restore GC threshold. */
   if (errcode)

src/lj_lib.h

@@ -47,8 +47,13 @@ LJ_FUNC GCtab *lj_lib_checktabornil(lua_State *L, int narg);
 LJ_FUNC int lj_lib_checkopt(lua_State *L, int narg, int def, const char *lst);
 
 /* Avoid including lj_frame.h. */
+#if LJ_FR2
+#define lj_lib_upvalue(L, n) \
+  (&gcref((L->base-2)->gcr)->fn.c.upvalue[(n)-1])
+#else
 #define lj_lib_upvalue(L, n) \
   (&gcref((L->base-1)->fr.func)->fn.c.upvalue[(n)-1])
+#endif
 
 #if LJ_TARGET_WINDOWS
 #define lj_lib_checkfpu(L) \

src/lj_meta.c

@@ -80,12 +80,16 @@ int lj_meta_tailcall(lua_State *L, cTValue *tv)
   TValue *base = L->base;
   TValue *top = L->top;
   const BCIns *pc = frame_pc(base-1);  /* Preserve old PC from frame. */
-  copyTV(L, base-1, tv);  /* Replace frame with new object. */
-  top->u32.lo = LJ_CONT_TAILCALL;
-  setframe_pc(top, pc);
-  setframe_gc(top+1, obj2gco(L), LJ_TTHREAD);  /* Dummy frame object. */
-  setframe_ftsz(top+1, ((char *)(top+2) - (char *)base) + FRAME_CONT);
-  L->base = L->top = top+2;
+  copyTV(L, base-1-LJ_FR2, tv);  /* Replace frame with new object. */
+  if (LJ_FR2)
+    (top++)->u64 = LJ_CONT_TAILCALL;
+  else
+    top->u32.lo = LJ_CONT_TAILCALL;
+  setframe_pc(top++, pc);
+  if (LJ_FR2) top++;
+  setframe_gc(top, obj2gco(L), LJ_TTHREAD);  /* Dummy frame object. */
+  setframe_ftsz(top, ((char *)(top+1) - (char *)base) + FRAME_CONT);
+  L->base = L->top = top+1;
   /*
   ** before:   [old_mo|PC]    [... ...]
   **                         ^base     ^top
@@ -116,11 +120,13 @@ static TValue *mmcall(lua_State *L, ASMFunction cont, cTValue *mo,
   */
   TValue *top = L->top;
   if (curr_funcisL(L)) top = curr_topL(L);
-  setcont(top, cont);  /* Assembler VM stores PC in upper word. */
-  copyTV(L, top+1, mo);  /* Store metamethod and two arguments. */
-  copyTV(L, top+2, a);
-  copyTV(L, top+3, b);
-  return top+2;  /* Return new base. */
+  setcont(top++, cont);  /* Assembler VM stores PC in upper word or FR2. */
+  if (LJ_FR2) setnilV(top++);
+  copyTV(L, top++, mo);  /* Store metamethod and two arguments. */
+  if (LJ_FR2) setnilV(top++);
+  copyTV(L, top, a);
+  copyTV(L, top+1, b);
+  return top;  /* Return new base. */
 }
 
 /* -- C helpers for some instructions, called from assembler VM ----------- */
@@ -256,10 +262,11 @@ TValue *lj_meta_cat(lua_State *L, TValue *top, int left)
       ** after mm:  [...][CAT stack ...] <--push-- [result]
       ** next step: [...][CAT stack .............]
       */
-      copyTV(L, top+2, top);  /* Careful with the order of stack copies! */
-      copyTV(L, top+1, top-1);
-      copyTV(L, top, mo);
+      copyTV(L, top+2*LJ_FR2+2, top);  /* Carefully ordered stack copies! */
+      copyTV(L, top+2*LJ_FR2+1, top-1);
+      copyTV(L, top+LJ_FR2, mo);
       setcont(top-1, lj_cont_cat);
+      if (LJ_FR2) { setnilV(top); setnilV(top+2); top += 2; }
       return top+1;  /* Trigger metamethod call. */
     } else {
       /* Pick as many strings as possible from the top and concatenate them:
@@ -327,12 +334,14 @@ TValue *lj_meta_equal(lua_State *L, GCobj *o1, GCobj *o2, int ne)
 	return (TValue *)(intptr_t)ne;
     }
     top = curr_top(L);
-    setcont(top, ne ? lj_cont_condf : lj_cont_condt);
-    copyTV(L, top+1, mo);
+    setcont(top++, ne ? lj_cont_condf : lj_cont_condt);
+    if (LJ_FR2) setnilV(top++);
+    copyTV(L, top++, mo);
+    if (LJ_FR2) setnilV(top++);
     it = ~(uint32_t)o1->gch.gct;
-    setgcV(L, top+2, o1, it);
-    setgcV(L, top+3, o2, it);
-    return top+2;  /* Trigger metamethod call. */
+    setgcV(L, top, o1, it);
+    setgcV(L, top+1, o2, it);
+    return top;  /* Trigger metamethod call. */
   }
   return (TValue *)(intptr_t)ne;
 }
@@ -431,7 +440,8 @@ void lj_meta_call(lua_State *L, TValue *func, TValue *top)
   TValue *p;
   if (!tvisfunc(mo))
     lj_err_optype_call(L, func);
-  for (p = top; p > func; p--) copyTV(L, p, p-1);
+  for (p = top; p > func+2*LJ_FR2; p--) copyTV(L, p, p-1);
+  if (LJ_FR2) copyTV(L, func+2, func);
   copyTV(L, func, mo);
 }
 

src/lj_obj.h

@@ -131,7 +131,7 @@ typedef struct SBuf {
 /* Frame link. */
 typedef union {
   int32_t ftsz;		/* Frame type and size of previous frame. */
-  MRef pcr;		/* Overlaps PC for Lua frames. */
+  MRef pcr;		/* Or PC for Lua frames. */
 } FrameLink;
 
 /* Tagged value. */
@@ -147,12 +147,16 @@ typedef LJ_ALIGN(8) union TValue {
     , uint32_t it;	/* Internal object tag. Must overlap MSW of number. */
     )
   };
+#if LJ_FR2
+  int64_t ftsz;		/* Frame type and size of previous frame, or PC. */
+#else
   struct {
     LJ_ENDIAN_LOHI(
       GCRef func;	/* Function for next frame (or dummy L). */
     , FrameLink tp;	/* Link to previous frame. */
     )
   } fr;
+#endif
   struct {
     LJ_ENDIAN_LOHI(
       uint32_t lo;	/* Lower 32 bits of number. */
@@ -584,7 +588,11 @@ struct lua_State {
 #define registry(L)		(&G(L)->registrytv)
 
 /* Macros to access the currently executing (Lua) function. */
+#if LJ_FR2
+#define curr_func(L)		(&gcref((L->base-2)->gcr)->fn)
+#else
 #define curr_func(L)		(&gcref((L->base-1)->fr.func)->fn)
+#endif
 #define curr_funcisL(L)		(isluafunc(curr_func(L)))
 #define curr_proto(L)		(funcproto(curr_func(L)))
 #define curr_topL(L)		(L->base + curr_proto(L)->framesize)
@@ -732,10 +740,16 @@ static LJ_AINLINE void setlightudV(TValue *o, void *p)
 #if LJ_64
 #define checklightudptr(L, p) \
   (((uint64_t)(p) >> 47) ? (lj_err_msg(L, LJ_ERR_BADLU), NULL) : (p))
+#else
+#define checklightudptr(L, p)	(p)
+#endif
+
+#if LJ_FR2
+#define setcont(o, f)		((o)->u64 = (uint64_t)(uintptr_t)(void *)(f))
+#elif LJ_64
 #define setcont(o, f) \
   ((o)->u64 = (uint64_t)(void *)(f) - (uint64_t)lj_vm_asm_begin)
 #else
-#define checklightudptr(L, p)	(p)
 #define setcont(o, f)		setlightudV((o), (void *)(f))
 #endif
 

src/lj_parse.c

@@ -662,14 +662,14 @@ static void bcemit_method(FuncState *fs, ExpDesc *e, ExpDesc *key)
   BCReg idx, func, obj = expr_toanyreg(fs, e);
   expr_free(fs, e);
   func = fs->freereg;
-  bcemit_AD(fs, BC_MOV, func+1, obj);  /* Copy object to first argument. */
+  bcemit_AD(fs, BC_MOV, func+1+LJ_FR2, obj);  /* Copy object to 1st argument. */
   lua_assert(expr_isstrk(key));
   idx = const_str(fs, key);
   if (idx <= BCMAX_C) {
-    bcreg_reserve(fs, 2);
+    bcreg_reserve(fs, 2+LJ_FR2);
     bcemit_ABC(fs, BC_TGETS, func, obj, idx);
   } else {
-    bcreg_reserve(fs, 3);
+    bcreg_reserve(fs, 3+LJ_FR2);
     bcemit_AD(fs, BC_KSTR, func+2, idx);
     bcemit_ABC(fs, BC_TGETV, func, obj, func+2);
     fs->freereg--;
@@ -1913,11 +1913,11 @@ static void parse_args(LexState *ls, ExpDesc *e)
   lua_assert(e->k == VNONRELOC);
   base = e->u.s.info;  /* Base register for call. */
   if (args.k == VCALL) {
-    ins = BCINS_ABC(BC_CALLM, base, 2, args.u.s.aux - base - 1);
+    ins = BCINS_ABC(BC_CALLM, base, 2, args.u.s.aux - base - 1 - LJ_FR2);
   } else {
     if (args.k != VVOID)
       expr_tonextreg(fs, &args);
-    ins = BCINS_ABC(BC_CALL, base, 2, fs->freereg - base);
+    ins = BCINS_ABC(BC_CALL, base, 2, fs->freereg - base - LJ_FR2);
   }
   expr_init(e, VCALL, bcemit_INS(fs, ins));
   e->u.s.aux = base;
@@ -1957,6 +1957,7 @@ static void expr_primary(LexState *ls, ExpDesc *v)
       parse_args(ls, v);
     } else if (ls->tok == '(' || ls->tok == TK_string || ls->tok == '{') {
       expr_tonextreg(fs, v);
+      if (LJ_FR2) bcreg_reserve(fs, 1);
       parse_args(ls, v);
     } else {
       break;
@@ -2537,7 +2538,8 @@ static void parse_for_iter(LexState *ls, GCstr *indexname)
   lex_check(ls, TK_in);
   line = ls->linenumber;
   assign_adjust(ls, 3, expr_list(ls, &e), &e);
-  bcreg_bump(fs, 3);  /* The iterator needs another 3 slots (func + 2 args). */
+  /* The iterator needs another 3 [4] slots (func [pc] | state ctl). */
+  bcreg_bump(fs, 3+LJ_FR2);
   isnext = (nvars <= 5 && predict_next(ls, fs, exprpc));
   var_add(ls, 3);  /* Hidden control variables. */
   lex_check(ls, TK_do);

src/lj_record.c

@@ -502,6 +502,7 @@ static LoopEvent rec_for(jit_State *J, const BCIns *fori, int isforl)
 static LoopEvent rec_iterl(jit_State *J, const BCIns iterins)
 {
   BCReg ra = bc_a(iterins);
+  lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
   if (!tref_isnil(getslot(J, ra))) {  /* Looping back? */
     J->base[ra-1] = J->base[ra];  /* Copy result of ITERC to control var. */
     J->maxslot = ra-1+bc_b(J->pc[-1]);
@@ -672,6 +673,7 @@ static void rec_call_setup(jit_State *J, BCReg func, ptrdiff_t nargs)
   TValue *functv = &J->L->base[func];
   TRef *fbase = &J->base[func];
   ptrdiff_t i;
+  lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
   for (i = 0; i <= nargs; i++)
     (void)getslot(J, func+i);  /* Ensure func and all args have a reference. */
   if (!tref_isfunc(fbase[0])) {  /* Resolve __call metamethod. */
@@ -788,7 +790,8 @@ void lj_record_ret(jit_State *J, BCReg rbase, ptrdiff_t gotresults)
     BCIns callins = *(frame_pc(frame)-1);
     ptrdiff_t nresults = bc_b(callins) ? (ptrdiff_t)bc_b(callins)-1 :gotresults;
     BCReg cbase = bc_a(callins);
-    GCproto *pt = funcproto(frame_func(frame - (cbase+1)));
+    GCproto *pt = funcproto(frame_func(frame - (cbase+1-LJ_FR2)));
+    lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame teardown. */
     if ((pt->flags & PROTO_NOJIT))
       lj_trace_err(J, LJ_TRERR_CJITOFF);
     if (J->framedepth == 0 && J->pt && frame == J->L->base - 1) {
@@ -973,6 +976,7 @@ static TRef rec_mm_arith(jit_State *J, RecordIndex *ix, MMS mm)
     lj_trace_err(J, LJ_TRERR_NOMM);
   }
 ok:
+  lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
   base[0] = ix->mobj;
   copyTV(J->L, basev+0, &ix->mobjv);
   lj_record_call(J, func, 2);
@@ -989,6 +993,7 @@ static TRef rec_mm_len(jit_State *J, TRef tr, TValue *tv)
     BCReg func = rec_mm_prep(J, lj_cont_ra);
     TRef *base = J->base + func;
     TValue *basev = J->L->base + func;
+    lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
     base[0] = ix.mobj; copyTV(J->L, basev+0, &ix.mobjv);
     base[1] = tr; copyTV(J->L, basev+1, tv);
 #if LJ_52
@@ -1011,6 +1016,7 @@ static void rec_mm_callcomp(jit_State *J, RecordIndex *ix, int op)
   BCReg func = rec_mm_prep(J, (op&1) ? lj_cont_condf : lj_cont_condt);
   TRef *base = J->base + func;
   TValue *tv = J->L->base + func;
+  lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
   base[0] = ix->mobj; base[1] = ix->val; base[2] = ix->key;
   copyTV(J->L, tv+0, &ix->mobjv);
   copyTV(J->L, tv+1, &ix->valv);
@@ -1261,6 +1267,7 @@ TRef lj_record_idx(jit_State *J, RecordIndex *ix)
       BCReg func = rec_mm_prep(J, ix->val ? lj_cont_nop : lj_cont_ra);
       TRef *base = J->base + func;
       TValue *tv = J->L->base + func;
+      lua_assert(!LJ_FR2);  /* TODO_FR2: handle different frame setup. */
       base[0] = ix->mobj; base[1] = ix->tab; base[2] = ix->key;
       setfuncV(J->L, tv+0, funcV(&ix->mobjv));
       copyTV(J->L, tv+1, &ix->tabv);
@@ -2135,28 +2142,28 @@ void lj_record_ins(jit_State *J)
   /* -- Calls and vararg handling ----------------------------------------- */
 
   case BC_ITERC:
-    J->base[ra] = getslot(J, ra-3);
-    J->base[ra+1] = getslot(J, ra-2);
-    J->base[ra+2] = getslot(J, ra-1);
+    J->base[ra] = getslot(J, ra-3-LJ_FR2);
+    J->base[ra+1] = getslot(J, ra-2-LJ_FR2);
+    J->base[ra+2] = getslot(J, ra-1-LJ_FR2);
     { /* Do the actual copy now because lj_record_call needs the values. */
       TValue *b = &J->L->base[ra];
-      copyTV(J->L, b, b-3);
-      copyTV(J->L, b+1, b-2);
-      copyTV(J->L, b+2, b-1);
+      copyTV(J->L, b, b-3-LJ_FR2);
+      copyTV(J->L, b+1, b-2-LJ_FR2);
+      copyTV(J->L, b+2, b-1-LJ_FR2);
     }
     lj_record_call(J, ra, (ptrdiff_t)rc-1);
     break;
 
   /* L->top is set to L->base+ra+rc+NARGS-1+1. See lj_dispatch_ins(). */
   case BC_CALLM:
-    rc = (BCReg)(J->L->top - J->L->base) - ra;
+    rc = (BCReg)(J->L->top - J->L->base) - ra - LJ_FR2;
     /* fallthrough */
   case BC_CALL:
     lj_record_call(J, ra, (ptrdiff_t)rc-1);
     break;
 
   case BC_CALLMT:
-    rc = (BCReg)(J->L->top - J->L->base) - ra;
+    rc = (BCReg)(J->L->top - J->L->base) - ra - LJ_FR2;
     /* fallthrough */
   case BC_CALLT:
     lj_record_tailcall(J, ra, (ptrdiff_t)rc-1);

src/lj_snap.c

@@ -100,6 +100,7 @@ static BCReg snapshot_framelinks(jit_State *J, SnapEntry *map)
   GCfunc *fn = frame_func(frame);
   cTValue *ftop = isluafunc(fn) ? (frame+funcproto(fn)->framesize) : J->L->top;
   MSize f = 0;
+  lua_assert(!LJ_FR2);  /* TODO_FR2: store 64 bit PCs. */
   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)) {
@@ -241,7 +242,8 @@ static BCReg snap_usedef(jit_State *J, uint8_t *udf,
     case BCMbase:
       if (op >= BC_CALLM && op <= BC_VARG) {
 	BCReg top = (op == BC_CALLM || op == BC_CALLMT || bc_c(ins) == 0) ?
-		    maxslot : (bc_a(ins) + bc_c(ins));
+		    maxslot : (bc_a(ins) + bc_c(ins)+LJ_FR2);
+	if (LJ_FR2) DEF_SLOT(bc_a(ins)+1);
 	s = bc_a(ins) - ((op == BC_ITERC || op == BC_ITERN) ? 3 : 0);
 	for (; s < top; s++) USE_SLOT(s);
 	for (; s < maxslot; s++) DEF_SLOT(s);
@@ -836,6 +838,7 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
 	snap_restoreval(J, T, ex, snapno, rfilt, ref+1, &tmp);
 	o->u32.hi = tmp.u32.lo;
       } else if ((sn & (SNAP_CONT|SNAP_FRAME))) {
+	lua_assert(!LJ_FR2);  /* TODO_FR2: store 64 bit PCs. */
 	/* Overwrite tag with frame link. */
 	setframe_ftsz(o, snap_slot(sn) != 0 ? (int32_t)*flinks-- : ftsz0);
 	L->base = o+1;

src/lj_state.c

@@ -49,6 +49,7 @@
 ** one extra slot if mobj is not a function. Only lj_meta_tset needs 5
 ** slots above top, but then mobj is always a function. So we can get by
 ** with 5 extra slots.
+** LJ_FR2: We need 2 more slots for the frame PC and the continuation PC.
 */
 
 /* Resize stack slots and adjust pointers in state. */
@@ -128,8 +129,9 @@ static void stack_init(lua_State *L1, lua_State *L)
   L1->stacksize = LJ_STACK_START + LJ_STACK_EXTRA;
   stend = st + L1->stacksize;
   setmref(L1->maxstack, stend - LJ_STACK_EXTRA - 1);
-  L1->base = L1->top = st+1;
-  setthreadV(L1, st, L1);  /* Needed for curr_funcisL() on empty stack. */
+  setthreadV(L1, st++, L1);  /* Needed for curr_funcisL() on empty stack. */
+  if (LJ_FR2) setnilV(st++);
+  L1->base = L1->top = st;
   while (st < stend)  /* Clear new slots. */
     setnilV(st++);
 }
@@ -253,7 +255,7 @@ LUA_API void lua_close(lua_State *L)
   for (i = 0;;) {
     hook_enter(g);
     L->status = 0;
-    L->base = L->top = tvref(L->stack) + 1;
+    L->base = L->top = tvref(L->stack) + 1 + LJ_FR2;
     L->cframe = NULL;
     if (lj_vm_cpcall(L, NULL, NULL, cpfinalize) == 0) {
       if (++i >= 10) break;

src/lj_trace.c

@@ -830,7 +830,7 @@ int LJ_FASTCALL lj_trace_exit(jit_State *J, void *exptr)
   ERRNO_RESTORE
   switch (bc_op(*pc)) {
   case BC_CALLM: case BC_CALLMT:
-    return (int)((BCReg)(L->top - L->base) - bc_a(*pc) - bc_c(*pc));
+    return (int)((BCReg)(L->top - L->base) - bc_a(*pc) - bc_c(*pc) + LJ_FR2);
   case BC_RETM:
     return (int)((BCReg)(L->top - L->base) + 1 - bc_a(*pc) - bc_d(*pc));
   case BC_TSETM:

src/lj_vmevent.c

@@ -27,6 +27,7 @@ ptrdiff_t lj_vmevent_prepare(lua_State *L, VMEvent ev)
     if (tv && tvisfunc(tv)) {
       lj_state_checkstack(L, LUA_MINSTACK);
       setfuncV(L, L->top++, funcV(tv));
+      if (LJ_FR2) setnilV(L->top++);
       return savestack(L, L->top);
     }
   }