Re[2]: Suggestion: add a new api lua_pushuserdata

[Cloud Wu <cloudwu@...>]

Hello Roberto,

Wednesday, December 13, 2006, 7:17:17 PM, you wrote:

RI> By "expensive" you mean too slow? If you are going to call back Lua,
RI> it seems that two extra table accesses should make no relevant difference.

I think sometimes it seems slow, if I only call back lua for a simple operation. (IMHO)

I wrote a small program to test :

The code in C is:
-------------test.c -------------------
static void *p[1024];

static int a=0;

static int
test(lua_State *L)
{
        lua_pushlightuserdata(L,p);
        lua_rawget(L,LUA_REGISTRYINDEX);
        lua_pushlightuserdata(L,p[a]);
        lua_rawget(L,-2);
        a=(a+1)%1024;
        return 1;
}


EXPORT int
luaopen_test(lua_State *L)
{
        int i;
        lua_pushlightuserdata(L,p);
        lua_createtable(L,0,1024);
        for (i=0;i<1024;i++) {
                p[i]=lua_newuserdata(L,4);
                lua_pushlightuserdata(L,p[i]);
                lua_insert(L,-2);
                lua_rawset(L,-3);
        }
        lua_settable(L,LUA_REGISTRYINDEX);

        lua_pushcfunction(L,test);

        return 1;
}
------------------------------------

In lua, I use os.clock() to timing:
------------------------------------

local test=require"test"

local function dummy()
end

function timing(f)
        local t=os.clock()
        for i=1,1024*1024 do
                f()
        end
        print(os.clock()-t)
end

timing(test)
timing(dummy)
----------------------------------

In my PC, the result is

0.219
0.094

test() cost more time than dummy()

-- 
Best regards,
 Cloud                            mailto:cloudwu@163.com
            http://blog.codingnow.com

[学问二字, 须要拆开看, 学是学, 问是问]