Merge branch 'master' into v2.1

src/Makefile

@@ -119,8 +119,10 @@ XCFLAGS=
 #
 # Use the system provided memory allocator (realloc) instead of the
 # bundled memory allocator. This is slower, but sometimes helpful for
-# debugging. It's helpful for Valgrind's memcheck tool, too. This option
-# cannot be enabled on x64, since the built-in allocator is mandatory.
+# debugging. This option cannot be enabled on x64, since realloc usually
+# doesn't return addresses in the right address range.
+# OTOH this option is mandatory for Valgrind's memcheck tool on x64 and
+# the only way to get useful results from it for all other architectures.
 #XCFLAGS+= -DLUAJIT_USE_SYSMALLOC
 #
 # This define is required to run LuaJIT under Valgrind. The Valgrind

src/lib_aux.c

@@ -302,7 +302,7 @@ static int panic(lua_State *L)
 
 #ifdef LUAJIT_USE_SYSMALLOC
 
-#if LJ_64
+#if LJ_64 && !defined(LUAJIT_USE_VALGRIND)
 #error "Must use builtin allocator for 64 bit target"
 #endif
 

src/lj_gc.c

@@ -625,6 +625,8 @@ static size_t gc_onestep(lua_State *L)
   case GCSsweep: {
     MSize old = g->gc.total;
     setmref(g->gc.sweep, gc_sweep(g, mref(g->gc.sweep, GCRef), GCSWEEPMAX));
+    lua_assert(old >= g->gc.total);
+    g->gc.estimate -= old - g->gc.total;
     if (gcref(*mref(g->gc.sweep, GCRef)) == NULL) {
       if (g->strnum <= (g->strmask >> 2) && g->strmask > LJ_MIN_STRTAB*2-1)
 	lj_str_resize(L, g->strmask >> 1);  /* Shrink string table. */
@@ -638,8 +640,6 @@ static size_t gc_onestep(lua_State *L)
 	g->gc.debt = 0;
       }
     }
-    lua_assert(old >= g->gc.total);
-    g->gc.estimate -= old - g->gc.total;
     return GCSWEEPMAX*GCSWEEPCOST;
     }
   case GCSfinalize:

src/lj_state.c

@@ -178,7 +178,7 @@ static void close_state(lua_State *L)
     g->allocf(g->allocd, G2GG(g), sizeof(GG_State), 0);
 }
 
-#if LJ_64
+#if LJ_64 && !(defined(LUAJIT_USE_VALGRIND) && defined(LUAJIT_USE_SYSMALLOC))
 lua_State *lj_state_newstate(lua_Alloc f, void *ud)
 #else
 LUA_API lua_State *lua_newstate(lua_Alloc f, void *ud)

src/lj_trace.c

@@ -618,6 +618,7 @@ static TValue *trace_state(lua_State *L, lua_CFunction dummy, void *ud)
       }
       lj_opt_split(J);
       lj_opt_sink(J);
+      if (!J->loopref) J->cur.snap[J->cur.nsnap-1].count = SNAPCOUNT_DONE;
       J->state = LJ_TRACE_ASM;
       break;