slot based allocator for lua, usage of lua_createtable

apps/twnlua/bindgen.py

@@ -8,6 +8,9 @@ with open(sys.argv[1], 'r') if sys.argv[1] != "-" else sys.stdin as f:
 api = json.loads(api_source)
 
 
+# TODO: use lua_to--x functions for defaults, so that fewer function calls are made
+# TODO: use lua_createtable for preallocation
+
 def default(parameter):
     basetype = parameter["type"].rsplit(' *', 1)[0]
     if parameter["type"] == "float":
@@ -32,7 +35,7 @@ def default(parameter):
 
 
 def to_table(typedesc, variable, indent = 0):
-    binding = ' ' * indent + "lua_newtable(L);\n"
+    binding = ' ' * indent + "lua_createtable(L, 0, %i);\n" % len(typedesc["fields"])
     for field in typedesc["fields"]:
         if field["type"] == "float":
             binding += ' ' * indent + "lua_pushnumber(L, (double)(%s));\n" % (variable + ".%s" % field["name"])
@@ -48,7 +51,8 @@ def to_table(typedesc, variable, indent = 0):
 
 
 print('#include "twn_game_api.h"\n')
-print('#define STB_DS_IMPLEMENTATION') # TODO: reuse implementation from the engine
+# TODO: reuse implementation from the engine, this also breaks with statically compiled build
+print('#define STB_DS_IMPLEMENTATION')
 print('#include <stb_ds.h>')
 print('#include <lua.h>')
 print('#include <lualib.h>')
@@ -88,11 +92,13 @@ for procedure, procedure_desc in api["procedures"].items():
             else:
                 raise BaseException("Unhandled parameter type '%s'" % (parameter["type"]))
 
+    # binding += "    lua_pop(L, %i);\n" % (1 + len(procedure_desc["params"]) if "params" in procedure_desc else 0)
+
     if "return" in procedure_desc:
         if procedure_desc["return"] == "bool":
-            binding += "    lua_pushboolean(L, (int)%s(%s));\n" % (procedure, ", ".join(param["name"] for param in procedure_desc["params"]))
+            binding += "    lua_pushboolean(L, (int)(%s(%s)));\n" % (procedure, ", ".join(param["name"] for param in procedure_desc["params"]))
         elif procedure_desc["return"] == "float":
-            binding += "    lua_pushnumber(L, (double)%s(%s));\n" % (procedure, ", ".join(param["name"] for param in procedure_desc["params"]))
+            binding += "    lua_pushnumber(L, (double)(%s(%s)));\n" % (procedure, ", ".join(param["name"] for param in procedure_desc["params"]))
         elif procedure_desc["return"] == "char *":
             binding += "    lua_pushstring(L, %s(%s));\n" % (procedure, ", ".join(param["name"] for param in procedure_desc["params"]))
         elif type(procedure_desc["return"]) is dict or procedure_desc["return"] in api["types"]:
@@ -159,7 +165,7 @@ loader = "void bindgen_load_%s(lua_State *L) {\n" % api["name"]
 modules = set(api["procedures"][procedure]["module"] for procedure in api["procedures"])
 for module in modules:
     loader += "    bindgen_init();\n"
-    loader += "    lua_newtable(L);\n"
+    loader += "    lua_createtable(L, 0, %i);\n" % len(api["procedures"])
     for procedure, procedure_desc in api["procedures"].items():
         if procedure_desc["module"] == module:
             loader += "    lua_pushcfunction(L, binding_%s);\n" % procedure

apps/twnlua/game.c

@@ -43,6 +43,50 @@ static int physfs_loader(lua_State *L) {
 }
 
 
+/* WARN! experimental and will probably be removed */
+/* it was an attempt to ease memory usage problems posed by using lua */
+/* it saved ~100MiB, but it still hogs ~500MiB for no reason on my PC */
+static void *custom_alloc(void *ud, void *ptr, size_t osize, size_t nsize) {
+    (void)ud;
+
+    /* small allocations are placed in slots, as there's a big chance they will not need to be resized */
+    static char slots[1024][64];
+    static int16_t free_slots[1024] = { [0] = -1 };
+    static size_t free_slot_count = 1024;
+
+    if (free_slots[0] == -1)
+        for (int i = 0; i < 1024; i++)
+            free_slots[i] = (int16_t)i;
+
+    if (nsize == 0) {
+        if (ptr && (char *)ptr >= &slots[0][0] && (char *)ptr <= &slots[1024-1][64-1])
+            free_slots[free_slot_count++] = (int16_t)(((uintptr_t)ptr - (uintptr_t)slots) / 64);
+        else
+            SDL_free(ptr);
+        return NULL;
+    } else {
+        if (!ptr && nsize <= 64 && free_slot_count > 0) {
+            /* use a slot */
+            return slots[free_slots[--free_slot_count]];
+        }
+
+        if ((char *)ptr >= &slots[0][0] && (char *)ptr <= &slots[1024-1][64-1]) {
+            /* still fits */
+            if (nsize <= 64)
+                return ptr;
+
+            /* move from slot to dynamic memory */
+            void *mem = SDL_malloc(nsize);
+            SDL_memcpy(mem, ptr, osize);
+            free_slots[free_slot_count++] = (int16_t)(((uintptr_t)ptr - (uintptr_t)slots) / 64);
+            return mem;
+        }
+
+        return SDL_realloc(ptr, nsize);
+    }
+}
+
+
 void game_tick(void) {
     if (ctx.initialization_needed) {
         if (!ctx.udata)
@@ -57,6 +101,8 @@ void game_tick(void) {
         }
         state->L = luaL_newstate();
 
+        lua_setallocf(state->L, custom_alloc, NULL);
+
         /* fakey version of luaL_openlibs() that excludes file i/o and os stuff */
         {
             static const luaL_Reg loaded_libs[] = {
@@ -79,7 +125,7 @@ void game_tick(void) {
 
         /* package.searchers = { physfs_loader } */
         lua_getglobal(state->L, "package");
-        lua_newtable(state->L);
+        lua_createtable(state->L, 0, 1);
         lua_setfield(state->L, -2, "searchers");
 
         lua_getfield(state->L, -1, "searchers");

src/twn_textures.c

@@ -439,6 +439,8 @@ TextureKey textures_get_key(TextureCache *cache, const char *path) {
         }
     }
 
+    /* TODO: this will be bad when dynamic strings are involved */
+    /* to mitigate that we could free ptr_to_texture each frame */
     /* try loading */
     last_texture = textures_load(cache, path);
     hmput(ptr_to_texture, path, last_texture);