hash_sparse: change length checks

dont early return on 0 len, let it be the falltrough case, this seems to help branch prediction

src/blua/lapi.c

@@ -519,6 +519,23 @@ LUA_API int lua_pushthread (lua_State *L) {
 ** get functions (Lua -> stack)
 */
 
+static void auxgetstr (lua_State *L, const TValue *t, const char *k) {
+  const TValue *slot;
+  TValue key;
+  TString *str = luaS_new(L, k);
+  api_checkvalidindex(L, t);
+  if (luaV_fastget(L, t, str, slot, luaH_getstr)) {
+    setobj2s(L, L->top, slot);
+    api_incr_top(L);
+  }
+  else {
+    setsvalue(L, &key, str);
+    luaV_finishget(L, t, L->top, L->top, slot);
+    api_incr_top(L);
+  }
+  lua_unlock(L);
+}
+
 
 LUA_API void lua_gettable (lua_State *L, int idx) {
   StkId t;
@@ -531,15 +548,8 @@ LUA_API void lua_gettable (lua_State *L, int idx) {
 
 
 LUA_API void lua_getfield (lua_State *L, int idx, const char *k) {
-  StkId t;
-  TValue key;
   lua_lock(L);
-  t = index2adr(L, idx);
-  api_checkvalidindex(L, t);
-  setsvalue(L, &key, luaS_new(L, k));
-  luaV_gettable(L, t, &key, L->top);
-  api_incr_top(L);
-  lua_unlock(L);
+  auxgetstr(L, index2adr(L, idx), k);
 }
 
 
@@ -630,6 +640,26 @@ LUA_API void lua_getfenv (lua_State *L, int idx) {
 ** set functions (stack -> Lua)
 */
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wcast-qual"
+static void auxsetstr (lua_State *L, const TValue *t, const char *k) {
+  TValue key;
+  const TValue *slot;
+  TString *str = luaS_new(L, k);
+  api_checknelems(L, 1);
+  api_checkvalidindex(L, t);
+  if (luaV_fastset(L, t, str, slot, luaH_setstr, L->top)) {
+    setobj2t(L, cast(TValue *, slot), L->top - 1);
+    L->top--;  /* pop value */
+  }
+  else {
+    setsvalue2s(L, &key, str);
+    luaV_finishset(L, t, &key, L->top - 1, slot);
+    L->top--;  /* pop value */
+  }
+  lua_unlock(L);
+}
+
 
 LUA_API void lua_settable (lua_State *L, int idx) {
   StkId t;
@@ -642,18 +672,12 @@ LUA_API void lua_settable (lua_State *L, int idx) {
   lua_unlock(L);
 }
 
+#pragma GCC diagnostic pop
+
 
 LUA_API void lua_setfield (lua_State *L, int idx, const char *k) {
-  StkId t;
-  TValue key;
   lua_lock(L);
-  api_checknelems(L, 1);
-  t = index2adr(L, idx);
-  api_checkvalidindex(L, t);
-  setsvalue(L, &key, luaS_new(L, k));
-  luaV_settable(L, t, &key, L->top - 1);
-  L->top--;  /* pop value */
-  lua_unlock(L);
+  auxsetstr(L, index2adr(L, idx), k);
 }
 
 

src/blua/lauxlib.c

@@ -20,6 +20,7 @@
 #define lauxlib_c
 #define LUA_LIB
 
+#include "ldebug.h"
 #include "lua.h"
 
 #include "lauxlib.h"
@@ -468,8 +469,19 @@ LUALIB_API char *luaL_prepbuffer (luaL_Buffer *B) {
 
 
 LUALIB_API void luaL_addlstring (luaL_Buffer *B, const char *s, size_t l) {
-  while (l--)
-    luaL_addchar(B, *s++);
+ while (l) {
+    size_t space = bufffree(B);
+    if (space == 0) {
+      luaL_prepbuffer(B);
+      lua_assert(bufffree(B) == LUAL_BUFFERSIZE);
+      space = LUAL_BUFFERSIZE;
+    }
+    if (space > l) space = l;
+    memcpy(B->p, s, space);
+    B->p += space;
+    s += space;
+    l -= space;
+  }
 }
 
 
@@ -686,6 +698,6 @@ static int panic (lua_State *L) {
 
 LUALIB_API lua_State *luaL_newstate (void) {
   lua_State *L = lua_newstate(l_alloc, NULL);
-  if (L) lua_atpanic(L, &panic);
+  if (l_likely(L)) lua_atpanic(L, &panic);
   return L;
 }

src/blua/lcode.c

@@ -626,8 +626,20 @@ static void codenot (FuncState *fs, expdesc *e) {
   removevalues(fs, e->t);
 }
 
+/*
+** Check whether expression 'e' is a literal string
+*/
+static int isKstr (FuncState *fs, expdesc *e) {
+  return (e->k == VK && ttisstring(&fs->f->k[e->u.s.info]));
+}
 
+/*
+** Create expression 't[k]'. 't' must have its final result already in a
+** register or upvalue. Upvalues can only be indexed by literal strings.
+*/
 void luaK_indexed (FuncState *fs, expdesc *t, expdesc *k) {
+  if (t->k == VUPVAL && !isKstr(fs, k))  /* upvalue indexed by non string? */
+    luaK_exp2anyreg(fs, t);  /* put it in a register */
   t->u.s.aux = luaK_exp2RK(fs, k);
   t->k = VINDEXED;
 }
@@ -790,18 +802,19 @@ void luaK_posfix (FuncState *fs, BinOpr op, expdesc *e1, expdesc *e2) {
     case OPR_BXOR: codearith(fs, OP_BXOR, e1, e2); break;
     case OPR_BSHL: codearith(fs, OP_BSHL, e1, e2); break;
     case OPR_BSHR: codearith(fs, OP_BSHR, e1, e2); break;
-    case OPR_ADD: codearith(fs, OP_ADD, e1, e2); break;
-    case OPR_SUB: codearith(fs, OP_SUB, e1, e2); break;
-    case OPR_MUL: codearith(fs, OP_MUL, e1, e2); break;
-    case OPR_DIV: codearith(fs, OP_DIV, e1, e2); break;
-    case OPR_MOD: codearith(fs, OP_MOD, e1, e2); break;
-    case OPR_POW: codearith(fs, OP_POW, e1, e2); break;
-    case OPR_EQ: codecomp(fs, OP_EQ, 1, e1, e2); break;
-    case OPR_NE: codecomp(fs, OP_EQ, 0, e1, e2); break;
-    case OPR_LT: codecomp(fs, OP_LT, 1, e1, e2); break;
-    case OPR_LE: codecomp(fs, OP_LE, 1, e1, e2); break;
-    case OPR_GT: codecomp(fs, OP_LT, 0, e1, e2); break;
-    case OPR_GE: codecomp(fs, OP_LE, 0, e1, e2); break;
+    case OPR_ADD: case OPR_SUB: case OPR_MUL: case OPR_DIV:
+    case OPR_MOD: case OPR_POW: {
+      codearith(fs, op - OPR_ADD + OP_ADD, e1, e2);
+      break;
+    }
+    case OPR_EQ: case OPR_LT: case OPR_LE: {
+      codecomp(fs, op - OPR_EQ + OP_EQ, 1, e1, e2);
+      break;
+    }
+    case OPR_NE: case OPR_GT: case OPR_GE: {
+      codecomp(fs, op - OPR_NE + OP_EQ, 0, e1, e2);
+      break;
+    }
     default: lua_assert(0);
   }
 }

src/blua/lcode.h

@@ -21,13 +21,13 @@
 
 
 /*
-** grep "ORDER OPR" if you change these enums
+** grep "ORDER OPR" if you change these enums  (ORDER OP)
 */
 typedef enum BinOpr {
   OPR_ADD, OPR_SUB, OPR_MUL, OPR_DIV, OPR_MOD, OPR_POW,
   OPR_CONCAT,
-  OPR_NE, OPR_EQ,
-  OPR_LT, OPR_LE, OPR_GT, OPR_GE,
+  OPR_EQ, OPR_LT, OPR_LE,
+  OPR_NE, OPR_GT, OPR_GE,
   OPR_AND, OPR_OR,
   OPR_BAND, OPR_BOR, OPR_BXOR,
   OPR_BSHL, OPR_BSHR,

src/blua/ldebug.c

@@ -542,16 +542,39 @@ static const char *getobjname (lua_State *L, CallInfo *ci, int stackpos,
 
 
 static const char *getfuncname (lua_State *L, CallInfo *ci, const char **name) {
+  TMS tm = 0;
   Instruction i;
   if ((isLua(ci) && ci->tailcalls > 0) || !isLua(ci - 1))
     return NULL;  /* calling function is not Lua (or is unknown) */
   ci--;  /* calling function */
   i = ci_func(ci)->l.p->code[currentpc(L, ci)];
-  if (GET_OPCODE(i) == OP_CALL || GET_OPCODE(i) == OP_TAILCALL ||
-      GET_OPCODE(i) == OP_TFORLOOP)
-    return getobjname(L, ci, GETARG_A(i), name);
-  else
-    return NULL;  /* no useful name can be found */
+  switch (GET_OPCODE(i)) {
+    case OP_CALL:
+    case OP_TAILCALL:
+    case OP_TFORLOOP:
+      return getobjname(L, ci, GETARG_A(i), name);
+    case OP_GETGLOBAL:
+    case OP_SELF:
+    case OP_GETTABLE: tm = TM_INDEX; break;
+    case OP_SETGLOBAL:
+    case OP_SETTABLE: tm = TM_NEWINDEX; break;
+    case OP_EQ: tm = TM_EQ; break;
+    case OP_ADD: tm = TM_ADD; break;
+    case OP_SUB: tm = TM_SUB; break;
+    case OP_MUL: tm = TM_MUL; break;
+    case OP_DIV: tm = TM_DIV; break;
+    case OP_MOD: tm = TM_MOD; break;
+    case OP_POW: tm = TM_POW; break;
+    case OP_UNM: tm = TM_UNM; break;
+    case OP_LEN: tm = TM_LEN; break;
+    case OP_LT: tm = TM_LT; break;
+    case OP_LE: tm = TM_LE; break;
+    case OP_CONCAT: tm = TM_CONCAT; break;
+    default:
+      return NULL;  /* else no useful name can be found */
+  }
+  *name = getstr(G(L)->tmname[tm]);
+  return "metamethod";
 }
 
 

src/blua/ldo.c

@@ -164,7 +164,7 @@ void luaD_reallocCI (lua_State *L, int newsize) {
 
 
 void luaD_growstack (lua_State *L, int n) {
-  if (l_likely(n <= L->stacksize))  /* double size is enough? */
+  if (n <= L->stacksize)  /* double size is enough? */
     luaD_reallocstack(L, 2*L->stacksize);
   else
     luaD_reallocstack(L, L->stacksize + n);
@@ -318,7 +318,7 @@ int luaD_precall (lua_State *L, StkId func, int nresults) {
     ci->top = L->top + LUA_MINSTACK;
     lua_assert(ci->top <= L->stack_last);
     ci->nresults = nresults;
-    if (L->hookmask & LUA_MASKCALL)
+    if (l_unlikely(L->hookmask & LUA_MASKCALL))
       luaD_callhook(L, LUA_HOOKCALL, -1);
     lua_unlock(L);
     n = (*curr_func(L)->c.f)(L);  /* do the actual call */

src/blua/liolib.c

@@ -356,20 +356,21 @@ static int read_line (lua_State *L, FILE *f) {
 
 
 static int read_chars (lua_State *L, FILE *f, size_t n) {
-  size_t rlen;  /* how much to read */
-  size_t nr;  /* number of chars actually read */
+  size_t tbr = n;  /* number of chars to be read */
+  size_t rlen;  /* how much to read in each cycle */
+  size_t nr;  /* number of chars actually read in each cycle */
   luaL_Buffer b;
   luaL_buffinit(L, &b);
   rlen = LUAL_BUFFERSIZE;  /* try to read that much each time */
   do {
     char *p = luaL_prepbuffer(&b);
-    if (rlen > n) rlen = n;  /* cannot read more than asked */
+    if (rlen > tbr) rlen = tbr;  /* cannot read more than asked */
     nr = fread(p, sizeof(char), rlen, f);
     luaL_addsize(&b, nr);
-    n -= nr;  /* still have to read `n' chars */
-  } while (n > 0 && nr == rlen);  /* until end of count or eof */
+    tbr -= nr;  /* still have to read 'tbr' chars */
+  } while (tbr > 0 && nr == rlen);  /* until end of count or eof */
   luaL_pushresult(&b);  /* close buffer */
-  return (n == 0 || lua_objlen(L, -1) > 0);
+  return (tbr == 0 || tbr < n);  /* true iff read something */
 }
 
 
@@ -432,13 +433,13 @@ static int f_read (lua_State *L) {
 
 static int io_readline (lua_State *L) {
   FILE *f = *(FILE **)lua_touserdata(L, lua_upvalueindex(1));
-  int sucess;
+  int success;
   if (f == NULL)  /* file is already closed? */
     luaL_error(L, "file is already closed");
-  sucess = read_line(L, f);
+  success = read_line(L, f);
   if (ferror(f))
     return luaL_error(L, "%s", strerror(errno));
-  if (sucess) return 1;
+  if (success) return 1;
   else {  /* EOF */
     if (lua_toboolean(L, lua_upvalueindex(2))) {  /* generator created file? */
       lua_settop(L, 0);

src/blua/lopcodes.h

@@ -232,6 +232,8 @@ OP_VARARG/*	A B	R(A), R(A+1), ..., R(A+B-1) = vararg		*/
   (*) In OP_SETLIST, if (B == 0) then B = `top';
       if (C == 0) then next `instruction' is real C
 
+  (*) In OP_SELF, the index must be a string.
+
   (*) For comparisons, A specifies what condition the test should accept
       (true or false).
 

src/blua/lparser.c

@@ -881,10 +881,10 @@ static const struct {
   lu_byte right; /* right priority */
 } priority[] = {  /* ORDER OPR */
    {6, 6}, {6, 6}, {7, 7}, {7, 7}, {7, 7},  /* `+' `-' `/' `%' */
-   {10, 9}, {5, 4},                 /* power and concat (right associative) */
-   {3, 3}, {3, 3},                  /* equality and inequality */
-   {3, 3}, {3, 3}, {3, 3}, {3, 3},  /* order */
-   {2, 2}, {1, 1}                   /* logical (and/or) */
+   {10, 9}, {5, 4},                 /* ^, .. (right associative) */
+   {3, 3}, {3, 3}, {3, 3},          /* ==, <, <= */
+   {3, 3}, {3, 3}, {3, 3},          /* ~=, >, >= */
+   {2, 2}, {1, 1}                   /* and, or */
    ,{6, 6}, {6, 6}, {6, 6},           /* bit-wise (band/bor/bxor) */
    {7, 7}, {7, 7}		    /* shl/shr */
 };

src/blua/lstring.c

@@ -51,7 +51,7 @@ static TString *newlstr (lua_State *L, const char *str, size_t l,
                                        unsigned int h) {
   TString *ts;
   stringtable *tb;
-  if (l+1 > (MAX_SIZET - sizeof(TString))/sizeof(char))
+  if (l_unlikely(l+1 > (MAX_SIZET - sizeof(TString))/sizeof(char)))
     luaM_toobig(L);
   ts = cast(TString *, luaM_malloc(L, (l+1)*sizeof(char)+sizeof(TString)));
   ts->tsv.len = l;
@@ -72,18 +72,72 @@ static TString *newlstr (lua_State *L, const char *str, size_t l,
 }
 
 
-TString *luaS_newlstr (lua_State *L, const char *str, size_t l) {
-  GCObject *o;
+// string hash taken from luaJIT
+
+/* Unaligned load of uint32_t. */
+static inline __attribute__((always_inline)) uint32_t getu32(const void* ptr)
+{
+  uint32_t result;
+  memcpy(&result, ptr, 4);
+  return result;
+}
+
+// smol macro to look a little less messy lul
+#define getu8(p) *cast(const uint8_t *, p)
+
+/* Keyed sparse ARX string hash. Constant time. */
+static unsigned hash_sparse(const char *str, size_t len)
+{
+  /* Constants taken from lookup3 hash by Bob Jenkins. */
+  unsigned int a, b, h = cast(unsigned int, len);
+#define rol(x, n) (((x)<<(n)) | ((x)>>(-cast(int, n)&(8*sizeof(x)-1))))
+  if (len >= 4) {  /* Caveat: unaligned access! */
+    a  = getu32(str);
+    h ^= getu32(str+len-4);
+    b  = getu32(str+(len>>1)-2);
+    h ^= b; h -= rol(b, 14);
+    b += getu32(str+(len>>2)-1);
+  } else if (len > 0) {
+    a  = getu8(str);
+    h ^= getu8(str+len-1);
+    b  = getu8(str+(len>>1));
+    h ^= b; h -= rol(b, 14);
+  } else {
+    return 0;
+  }
+  a ^= h; a -= rol(h, 11);
+  b ^= a; b -= rol(a, 25);
+  h ^= b; h -= rol(b, 16);
+#undef rol
+  return h;
+}
+
+#undef getu8
+
+
+static inline __attribute__((always_inline)) unsigned luaS_hash (const char *str, size_t l) {
+#if 0
   unsigned int h = cast(unsigned int, l);  /* seed */
   size_t step = (l>>5)+1;  /* if string is too long, don't hash all its chars */
   size_t l1;
   for (l1=l; l1>=step; l1-=step)  /* compute hash */
     h = h ^ ((h<<5)+(h>>2)+cast(unsigned char, str[l1-1]));
+  return h;
+#else
+  return hash_sparse(str, l);
+#endif
+}
+
+TString *luaS_newlstr (lua_State *L, const char *str, size_t l) {
+  GCObject *o;
+  unsigned int h = luaS_hash(str, l);
   for (o = G(L)->strt.hash[lmod(h, G(L)->strt.size)];
        o != NULL;
        o = o->gch.next) {
     TString *ts = rawgco2ts(o);
-    if (l_unlikely(ts->tsv.len == l && (memcmp(str, getstr(ts), l) == 0))) {
+    if (ts->tsv.hash == h &&
+        ts->tsv.len == l &&
+       (memcmp(str, getstr(ts), l) == 0)) {
       /* string may be dead */
       if (isdead(G(L), o)) changewhite(o);
       return ts;

src/blua/ltm.c

@@ -61,7 +61,7 @@ TValue *luaT_gettm (Table *events, TMS event, TString *ename) {
 }
 
 
-TValue *luaT_gettmbyobj (lua_State *L, TValue *o, TMS event) {
+TValue *luaT_gettmbyobj (lua_State *L, const TValue *o, TMS event) {
   Table *mt;
   switch (ttype(o)) {
     case LUA_TTABLE:

src/blua/ltm.h

@@ -55,7 +55,7 @@ LUAI_DATA const char *const luaT_typenames[];
 
 
 LUAI_FUNC TValue *luaT_gettm (Table *events, TMS event, TString *ename);
-LUAI_FUNC TValue *luaT_gettmbyobj (lua_State *L, TValue *o, TMS event);
+LUAI_FUNC TValue *luaT_gettmbyobj (lua_State *L, const TValue *o, TMS event);
 LUAI_FUNC void luaT_init (lua_State *L);
 
 #endif

src/blua/lvm.c

@@ -28,7 +28,7 @@
 
 
 
-/* limit for table tag-method chains (to avoid loops) */
+/* limit for table tag-method chains (to avoid infinite loops) */
 #define MAXTAGLOOP	100
 
 
@@ -105,61 +105,94 @@ static void callTM (lua_State *L, const TValue *f, const TValue *p1,
 }
 
 
-void luaV_gettable (lua_State *L, TValue *t, TValue *key, StkId val) {
-  int loop;
+/*
+** Finish the table access 'val = t[key]'.
+** if 'slot' is NULL, 't' is not a table; otherwise, 'slot' points to
+** t[k] entry (which must be nil).
+*/
+void luaV_finishget (lua_State *L, const TValue *t, TValue *key, StkId val,
+                      const TValue *slot) {
+  int loop;  /* counter to avoid infinite loops */
+  const TValue *tm;  /* metamethod */
   for (loop = 0; loop < MAXTAGLOOP; loop++) {
-    TValue *tm;
-    if (ttistable(t)) {  /* `t' is a table? */
-      Table *h = hvalue(t);
-      const TValue *res = luaH_get(h, key); /* do a primitive get */
-      if (!ttisnil(res) ||  /* result is no nil? */
-          (tm = fasttm(L, h->metatable, TM_INDEX)) == NULL) { /* or no TM? */
-        setobj2s(L, val, res);
+    if (slot == NULL) {  /* 't' is not a table? */
+      lua_assert(!ttistable(t));
+      tm = luaT_gettmbyobj(L, t, TM_INDEX);
+      if (l_unlikely(ttisnil(tm)))
+        luaG_typeerror(L, t, "index");  /* no metamethod */
+      /* else will try the metamethod */
+    }
+    else {  /* 't' is a table */
+      lua_assert(ttisnil(slot));
+      tm = fasttm(L, hvalue(t)->metatable, TM_INDEX);  /* table's metamethod */
+      if (tm == NULL) {  /* no metamethod? */
+        setnilvalue(val);  /* result is nil */
         return;
       }
-      /* else will try the tag method */
+      /* else will try the metamethod */
     }
-    else if (l_unlikely(ttisnil(tm = luaT_gettmbyobj(L, t, TM_INDEX))))
-      luaG_typeerror(L, t, "index");
-    if (ttisfunction(tm)) {
-      callTMres(L, val, tm, t, key);
+    if (ttisfunction(tm)) {  /* is metamethod a function? */
+      callTMres(L, val, tm, t, key);  /* call it */
       return;
     }
-    t = tm;  /* else repeat with `tm' */
+    t = tm;  /* else try to access 'tm[key]' */
+    if (luaV_fastget(L,t,key,slot,luaH_get)) {  /* fast track? */
+      setobj2s(L, val, slot);  /* done */
+      return;
+    }
+    /* else repeat (tail call 'luaV_finishget') */
   }
-  luaG_runerror(L, "loop in gettable");
+  luaG_runerror(L, "'__index' chain too long; possible loop");
 }
 
-
-void luaV_settable (lua_State *L, TValue *t, TValue *key, StkId val) {
-  int loop;
+/*
+** Finish a table assignment 't[key] = val'.
+** If 'slot' is NULL, 't' is not a table.  Otherwise, 'slot' points
+** to the entry 't[key]', or to 'luaO_nilobject' if there is no such
+** entry.  (The value at 'slot' must be nil, otherwise 'luaV_fastset'
+** would have done the job.)
+*/
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wcast-qual"
+void luaV_finishset (lua_State *L, const TValue *t, TValue *key,
+                     StkId val, const TValue *slot) {
+  int loop;  /* counter to avoid infinite loops */
   for (loop = 0; loop < MAXTAGLOOP; loop++) {
-    TValue *tm = NULL;
-    if (ttistable(t)) {  /* `t' is a table? */
+    const TValue *tm;
+    if (slot != NULL) {
+      lua_assert(ttistable(t) && ttisnil(slot));
       Table *h = hvalue(t);
-      TValue *oldval = luaH_set(L, h, key); /* do a primitive set */
-      /* oldval is nil=> look for newindex, oldval is not nil => look for usedindex */
-      if (!((ttisnil(oldval) && ((tm = fasttm(L, h->metatable, TM_NEWINDEX)) != NULL)) ||
-         ((!ttisnil(oldval)) && ((tm = fasttm(L, h->metatable, TM_USEDINDEX)) != NULL)))) {
-        setobj2t(L, oldval, val);
+      /* must check the metamethod */
+      /* slot is nil=> look for newindex, slot is not nil => look for usedindex */
+      if (!((ttisnil(slot) && ((tm = fasttm(L, h->metatable, TM_NEWINDEX)) != NULL)) ||
+         ((!ttisnil(slot)) && ((tm = fasttm(L, h->metatable, TM_USEDINDEX)) != NULL)))) {
+        /* no metamethod and (now) there is an entry with given key */
+        setobj2t(L, cast(TValue *, slot), val);
         luaC_barriert(L, h, val);
         return;
       }
-      /* else will try the tag method */
+      /* else will try the metamethod */
     }
-    else if (l_unlikely(ttisnil(tm = luaT_gettmbyobj(L, t, TM_NEWINDEX))))
-      luaG_typeerror(L, t, "index");
+    else {  /* not a table; check metamethod */
+      if (l_unlikely(ttisnil(tm = luaT_gettmbyobj(L, t, TM_NEWINDEX))))
+        luaG_typeerror(L, t, "index");
+    }
+    /* try the metamethod */
     if (ttisfunction(tm)) {
       callTM(L, tm, t, key, val);
       return;
     }
-    t = tm;  /* else repeat with `tm' */
+    t = tm;  /* else repeat assignment over 'tm' */
+    if (luaV_fastset(L, t, key, slot, luaH_set, val))
+      return;  /* done */
+    /* else loop */
   }
-  luaG_runerror(L, "loop in settable");
+  luaG_runerror(L, "settable chain too long; possible loop");
 }
+#pragma GCC diagnostic pop
 
 
-static int call_binTM (lua_State *L, TValue *p1, TValue *p2,
+static int call_binTM (lua_State *L, const TValue *p1, const TValue *p2,
                        StkId res, TMS event) {
   const TValue *tm = luaT_gettmbyobj(L, p1, event);  /* try first operand */
   if (ttisnil(tm))
@@ -184,7 +217,7 @@ static const TValue *get_compTM (lua_State *L, Table *mt1, Table *mt2,
 }
 
 
-static int call_orderTM (lua_State *L, TValue *p1, TValue *p2,
+static int call_orderTM (lua_State *L, const TValue *p1, const TValue *p2,
                          TMS event) {
   const TValue *tm1 = luaT_gettmbyobj(L, p1, event);
   const TValue *tm2;
@@ -366,6 +399,7 @@ static void Arith (lua_State *L, StkId ra, TValue *rb,
 
 
 #define arith_op(op,tm) { \
+        StkId ra = RA(i); \
         TValue *rb = RKB(i); \
         TValue *rc = RKC(i); \
         if (ttisnumber(rb) && ttisnumber(rc)) { \
@@ -377,6 +411,24 @@ static void Arith (lua_State *L, StkId ra, TValue *rb,
       }
 
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wcast-qual"
+
+/*
+** copy of 'luaV_gettable', but protecting call to potential metamethod
+** (which can reallocate the stack)
+*/
+#define gettableProtected(L,t,k,v)  { const TValue *slot; \
+  if (luaV_fastget(L,t,k,slot,luaH_get)) { setobj2s(L, v, slot); } \
+  else Protect(luaV_finishget(L,t,k,v,slot)); }
+
+
+/* same for 'luaV_settable' */
+#define settableProtected(L,t,k,v) { const TValue *slot; \
+  if (!luaV_fastset(L,t,k,slot,luaH_set,v)) \
+    Protect(luaV_finishset(L,t,k,v,slot)); }
+
+
 
 void luaV_execute (lua_State *L, int nexeccalls) {
   LClosure *cl;
@@ -386,13 +438,12 @@ void luaV_execute (lua_State *L, int nexeccalls) {
  reentry:  /* entry point */
   lua_assert(isLua(L->ci));
   pc = L->savedpc;
-  cl = &clvalue(L->ci->func)->l;
-  base = L->base;
-  k = cl->p->k;
+  cl = &clvalue(L->ci->func)->l;  /* local reference to function's closure */
+  k = cl->p->k;  /* local reference to function's constant table */
+  base = L->base;  /* local copy of function's base */
   /* main loop of interpreter */
   for (;;) {
     const Instruction i = *pc++;
-    StkId ra;
     if ((L->hookmask & (LUA_MASKLINE | LUA_MASKCOUNT)) &&
         (--L->hookcount == 0 || L->hookmask & LUA_MASKLINE)) {
       traceexec(L, pc);
@@ -402,26 +453,28 @@ void luaV_execute (lua_State *L, int nexeccalls) {
       }
       base = L->base;
     }
-    /* warning!! several calls may realloc the stack and invalidate `ra' */
-    ra = RA(i);
     lua_assert(base == L->base && L->base == L->ci->base);
     lua_assert(base <= L->top && L->top <= L->stack + L->stacksize);
     lua_assert(L->top == L->ci->top || luaG_checkopenop(i));
     switch (GET_OPCODE(i)) {
       case OP_MOVE: {
+        StkId ra = RA(i);
         setobjs2s(L, ra, RB(i));
         continue;
       }
       case OP_LOADK: {
+        StkId ra = RA(i);
         setobj2s(L, ra, KBx(i));
         continue;
       }
       case OP_LOADBOOL: {
+        StkId ra = RA(i);
         setbvalue(ra, GETARG_B(i));
         if (GETARG_C(i)) pc++;  /* skip next instruction (if C) */
         continue;
       }
       case OP_LOADNIL: {
+        StkId ra = RA(i);
         TValue *rb = RB(i);
         do {
           setnilvalue(rb--);
@@ -429,11 +482,13 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_GETUPVAL: {
+        StkId ra = RA(i);
         int b = GETARG_B(i);
         setobj2s(L, ra, cl->upvals[b]->v);
         continue;
       }
       case OP_GETGLOBAL: {
+        StkId ra = RA(i);
         TValue g;
         TValue *rb = KBx(i);
         sethvalue(L, &g, cl->env);
@@ -442,10 +497,14 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_GETTABLE: {
-        Protect(luaV_gettable(L, RB(i), RKC(i), ra));
+        StkId ra = RA(i);
+        StkId rb = RB(i);
+        TValue *rc = RKC(i);
+        gettableProtected(L, rb, rc, ra);
         continue;
       }
       case OP_SETGLOBAL: {
+        StkId ra = RA(i);
         TValue g;
         sethvalue(L, &g, cl->env);
         lua_assert(ttisstring(KBx(i)));
@@ -453,16 +512,21 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_SETUPVAL: {
+        StkId ra = RA(i);
         UpVal *uv = cl->upvals[GETARG_B(i)];
         setobj(L, uv->v, ra);
         luaC_barrier(L, uv, ra);
         continue;
       }
       case OP_SETTABLE: {
-        Protect(luaV_settable(L, ra, RKB(i), RKC(i)));
+        StkId ra = RA(i);
+        TValue *rb = RKB(i);
+        TValue *rc = RKC(i);
+        settableProtected(L, ra, rb, rc);
         continue;
       }
       case OP_NEWTABLE: {
+        StkId ra = RA(i);
         int b = GETARG_B(i);
         int c = GETARG_C(i);
         sethvalue(L, ra, luaH_new(L, luaO_fb2int(b), luaO_fb2int(c)));
@@ -470,9 +534,16 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_SELF: {
+        StkId ra = RA(i);
+        const TValue *slot;
         StkId rb = RB(i);
-        setobjs2s(L, ra+1, rb);
-        Protect(luaV_gettable(L, rb, RKC(i), ra));
+        TValue *rc = RKC(i);
+        TString *key = rawtsvalue(rc);  /* key must be a string */
+        setobjs2s(L, ra + 1, rb);
+        if (luaV_fastget(L, rb, key, slot, luaH_getstr)) {
+          setobj2s(L, ra, slot);
+        }
+        else Protect(luaV_finishget(L, rb, rc, ra, slot));
         continue;
       }
       case OP_ADD: {
@@ -488,6 +559,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_DIV: {
+        StkId ra = RA(i);
         TValue *rb = RKB(i);
         TValue *rc = RKC(i);
         if (ttisnumber(rb) && ttisnumber(rc)) {
@@ -503,6 +575,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_MOD: {
+        StkId ra = RA(i);
         TValue *rb = RKB(i);
         TValue *rc = RKC(i);
         if (ttisnumber(rb) && ttisnumber(rc)) {
@@ -522,26 +595,27 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_BAND: {
-	arith_op(luai_numand, TM_AND);
-	continue;
+        arith_op(luai_numand, TM_AND);
+        continue;
       }
       case OP_BOR: {
-	arith_op(luai_numor, TM_OR);
-	continue;
+        arith_op(luai_numor, TM_OR);
+        continue;
       }
       case OP_BXOR: {
-	arith_op(luai_numxor, TM_XOR);
-	continue;
+        arith_op(luai_numxor, TM_XOR);
+        continue;
       }
       case OP_BSHL: {
-	arith_op(luai_numshl, TM_SHL);
-	continue;
+        arith_op(luai_numshl, TM_SHL);
+        continue;
       }
       case OP_BSHR: {
-	arith_op(luai_numshr, TM_SHR);
-	continue;
+        arith_op(luai_numshr, TM_SHR);
+        continue;
       }
       case OP_BNOT: {
+        StkId ra = RA(i);
         TValue *rb = RB(i);
         if (ttisnumber(rb)) {
           lua_Number nb = nvalue(rb);
@@ -553,6 +627,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_UNM: {
+        StkId ra = RA(i);
         TValue *rb = RB(i);
         if (ttisnumber(rb)) {
           lua_Number nb = nvalue(rb);
@@ -564,11 +639,13 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_NOT: {
+        StkId ra = RA(i);
         int res = l_isfalse(RB(i));  /* next assignment may change this value */
         setbvalue(ra, res);
         continue;
       }
       case OP_LEN: {
+        StkId ra = RA(i);
         TValue *rb = RB(i);
         switch (ttype(rb)) {
           case LUA_TTABLE: {
@@ -626,12 +703,14 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_TEST: {
+        StkId ra = RA(i);
         if (l_isfalse(ra) != GETARG_C(i))
           dojump(L, pc, GETARG_sBx(*pc));
         pc++;
         continue;
       }
       case OP_TESTSET: {
+        StkId ra = RA(i);
         TValue *rb = RB(i);
         if (l_isfalse(rb) != GETARG_C(i)) {
           setobjs2s(L, ra, rb);
@@ -641,6 +720,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_CALL: {
+        StkId ra = RA(i);
         int b = GETARG_B(i);
         int nresults = GETARG_C(i) - 1;
         if (b != 0) L->top = ra+b;  /* else previous instruction set top */
@@ -662,6 +742,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         }
       }
       case OP_TAILCALL: {
+        StkId ra = RA(i);
         int b = GETARG_B(i);
         if (b != 0) L->top = ra+b;  /* else previous instruction set top */
         L->savedpc = pc;
@@ -694,6 +775,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         }
       }
       case OP_RETURN: {
+        StkId ra = RA(i);
         int b = GETARG_B(i);
         if (b != 0) L->top = ra+b-1;
         if (L->openupval) luaF_close(L, base);
@@ -709,6 +791,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         }
       }
       case OP_FORLOOP: {
+        StkId ra = RA(i);
         lua_Number step = nvalue(ra+2);
         lua_Number idx = luai_numadd(nvalue(ra), step); /* increment index */
         lua_Number limit = nvalue(ra+1);
@@ -721,6 +804,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_FORPREP: {
+        StkId ra = RA(i);
         const TValue *init = ra;
         const TValue *plimit = ra+1;
         const TValue *pstep = ra+2;
@@ -737,6 +821,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_TFORLOOP: {
+        StkId ra = RA(i);
         StkId cb = ra + 3;  /* call base */
         setobjs2s(L, cb+2, ra+2);
         setobjs2s(L, cb+1, ra+1);
@@ -753,6 +838,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_SETLIST: {
+        StkId ra = RA(i);
         int n = GETARG_B(i);
         int c = GETARG_C(i);
         int last;
@@ -775,10 +861,12 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_CLOSE: {
+        StkId ra = RA(i);
         luaF_close(L, ra);
         continue;
       }
       case OP_CLOSURE: {
+        StkId ra = RA(i);
         Proto *p;
         Closure *ncl;
         int nup, j;
@@ -799,6 +887,7 @@ void luaV_execute (lua_State *L, int nexeccalls) {
         continue;
       }
       case OP_VARARG: {
+        StkId ra = RA(i);
         int b = GETARG_B(i) - 1;
         int j;
         CallInfo *ci = L->ci;
@@ -822,3 +911,5 @@ void luaV_execute (lua_State *L, int nexeccalls) {
     }
   }
 }
+
+#pragma GCC diagnostic pop

src/blua/lvm.h

@@ -22,14 +22,61 @@
 	(ttype(o1) == ttype(o2) && luaV_equalval(L, o1, o2))
 
 
+/*
+** fast track for 'gettable': if 't' is a table and 't[k]' is not nil,
+** return 1 with 'slot' pointing to 't[k]' (final result).  Otherwise,
+** return 0 (meaning it will have to check metamethod) with 'slot'
+** pointing to a nil 't[k]' (if 't' is a table) or NULL (otherwise).
+** 'f' is the raw get function to use.
+*/
+#define luaV_fastget(L,t,k,slot,f) \
+  (!ttistable(t)  \
+   ? (slot = NULL, 0)  /* not a table; 'slot' is NULL and result is 0 */  \
+   : (slot = f(hvalue(t), k),  /* else, do raw access */  \
+      !ttisnil(slot)))  /* result not nil? */
+
+/*
+** standard implementation for 'gettable'
+*/
+#define luaV_gettable(L,t,k,v) { const TValue *slot; \
+  if (luaV_fastget(L,t,k,slot,luaH_get)) { setobj2s(L, v, slot); } \
+  else luaV_finishget(L,t,k,v,slot); }
+
+
+/*
+** Fast track for set table. If 't' is a table and 't[k]' is not nil,
+** call GC barrier, do a raw 't[k]=v', and return true; otherwise,
+** return false with 'slot' equal to NULL (if 't' is not a table) or
+** 'nil'. (This is needed by 'luaV_finishget'.) Note that, if the macro
+** returns true, there is no need to 'invalidateTMcache', because the
+** call is not creating a new entry.
+*/
+#define luaV_fastset(L,t,k,slot,f,v) \
+  (!ttistable(t) \
+   ? (slot = NULL, 0) \
+   : (slot = f(L, hvalue(t), k), \
+     ttisnil(slot) ? 0 \
+     : (__extension__ ({ \
+          if (valiswhite(v) && isblack(obj2gco(hvalue(t)))) \
+            luaC_barrierback(L, hvalue(t)); \
+          setobj2t(L, cast(TValue *,slot), v); \
+          1; \
+        }))))
+
+#define luaV_settable(L,t,k,v) { const TValue *slot; \
+  if (!luaV_fastset(L,t,k,slot,luaH_set,v)) \
+    luaV_finishset(L,t,k,v,slot); }
+
+
+
 LUAI_FUNC int luaV_lessthan (lua_State *L, TValue *l, TValue *r);
 LUAI_FUNC int luaV_equalval (lua_State *L, const TValue *t1, const TValue *t2);
 LUAI_FUNC const TValue *luaV_tonumber (const TValue *obj, TValue *n);
 LUAI_FUNC int luaV_tostring (lua_State *L, StkId obj);
-LUAI_FUNC void luaV_gettable (lua_State *L, TValue *t, TValue *key,
-                                            StkId val);
-LUAI_FUNC void luaV_settable (lua_State *L, TValue *t, TValue *key,
-                                            StkId val);
+LUAI_FUNC void luaV_finishget (lua_State *L, const TValue *t, TValue *key,
+                               StkId val, const TValue *tm);
+LUAI_FUNC void luaV_finishset (lua_State *L, const TValue *t, TValue *key,
+                               StkId val, const TValue *oldval);
 LUAI_FUNC void luaV_execute (lua_State *L, int nexeccalls);
 LUAI_FUNC void luaV_concat (lua_State *L, int total, int last);