KeyWorld Posted April 12, 2013 Share Posted April 12, 2013 Hi. If I remember correctly I wrote some things related to this post on the forum, at the beginning of rathena, I'm not able to find it ever, so maybe it was in the staff section. It was talking about script optimization, more specially, how some fake commands (if/else/while/for) are converting to label in the source and can be improved. Since I don't have the motivation (and the time) to work on a patch, I let you the result of my investigations here (because it was waiting from one year on my ftp). So first of all, this modifications required two commands : jump_nonzero, goto_ifexist. There will be used to rewrite all the for() while() code and optimize the code. // jump_nonzero() - inverse of jump_zero. BUILDIN_DEF(jump_nonzero,"il") BUILDIN_FUNC(jump_nonzero) { if( script_getnum(st,2) ) { if( !data_islabel(script_getdata(st,3)) ){ ShowError("script: jump_zero: not label !\n"); st->state=END; return 1; } st->pos = script_getnum(st,3); st->state = GOTO; } return 0; } // goto_ifexist() - Jump to a label if it exist (required a string !) BUILDIN_DEF(goto_ifexist,"s") BUILDIN_FUNC(goto_ifexist) { int n = search_str( script_getstr(st,2) ); if( n == -1 || str_data[n].type != C_POS ) {// label not found return 0; } st->pos = str_data[n].label; st->state = GOTO; return 0; } ---------------------------------------------------- Now, let's have fun: For NPC Script: for ( <a>; <b>; <c> ) { <d>; }Currently compiled to:<a>; __FR%x_J: jump_zero( <b>, __FR%x_FIN ); goto __FR%x_BGN; __FR%x_NXT: <c>; goto __FR%x_J; __FR%x_BGN: <d>; goto __FR%x_NXT; __FR%x_FIN:My proposal:<a>; jump_zero( <b>, __FR%x_FIN ); __FR%x_J: <d>; __FR%x_NXT: <c>; jump_nonzero( <b>, __FR%x_J ); __FR%x_FIN: While NPC Script: while ( <a> ) { <b> } Currently compiled to: __WL%x_NXT: jump_zero( <a>, __WL%x_FIN ); <b> goto __WL%x_NXT; __WL%x_FIN: My proposal: goto __WL%x_NXT; __WL%x_BGN: <b>; __WL%x_NXT: jump_nonzero( <a>, __WL%x_BGN ); __WL%x_FIN: Do-While NPC Script: do { <a> } while( <b> ); Currently compiled to: __DO%x_BGN: <a>; __DO%x_NXT: jump_zero( <b>, __DO%x_FIN ); goto __DO%x_BGN; __DO%x_FIN: My proposal: __DO%x_BGN: <a>; __DO%x_NXT: jump_nonzero( <b>, __DO%x_BGN ); __DO%x_FIN: Switch NPC Script: switch(<a>) { case 1: <b> break; default: <c> } Currently compiled code (simplify, the original is more obscure): set [email protected]__SW%x_VAL, <a>; goto __SW%x_%xJ; __SW%x_%xJ: __SW%x_%x: if(%d != [email protected]__SW%x_VAL) goto __SW%x_%x; <b>; goto __SW%x_FIN; // break __SW%x_%x: goto __SW%x_DEF; __SW%x_DEF: <c>; set [email protected]__SW%x_VAL,0; goto __SW%x_FIN; __SW%x_FIN: My proposal: goto_ifexist("__SW%x_" + <a>); goto_ifexist("__SW%x_DEF"); goto __SW%x_FIN; __SW%x_1: <b> goto __SW%x_FIN; // break; __SW%x_DEF: <d> __SW%x_FIN: 7 Quote Link to comment Share on other sites More sharing options...
nanakiwurtz Posted April 17, 2013 Share Posted April 17, 2013 Are you sure that the original code for 'for' isn't like this? <a>; __FR%x_J: jump_zero( <b>, __FR%x_FIN ); goto __FR%x_BGN; __FR%x_NXT: <c>; goto __FR%x_J; // <---- __FR%x_BGN: <d>; goto __FR%x_NXT; __FR%x_FIN: My suggestion is to use a forced jump instead using the jump_nonzero: <a>; __FR%x_J: jump_zero( <b>, __FR%x_FIN ); <d>; __FR%x_NXT: <c>; goto __FR%x_J; __FR%x_FIN: And in your proposal for the 'while', why are you checking the <b>? Isn't the check routine is in <a>? Sorry to bother you with these newbie questions (=_=') Thanks KeyWorld 1 Quote Link to comment Share on other sites More sharing options...
KeyWorld Posted April 17, 2013 Author Share Posted April 17, 2013 Yeah, I was a little confused, corrected thanks In your for() version, you are using a goto + jump_zero, when I just use a jump_nonzero. So no it's not better. 1 Quote Link to comment Share on other sites More sharing options...
Euro Posted May 6, 2013 Share Posted May 6, 2013 If this could optimize the script commands, I hope this gets implemented soon. Quote Link to comment Share on other sites More sharing options...
Joseph Posted May 29, 2013 Share Posted May 29, 2013 Improvements, why not? Quote Link to comment Share on other sites More sharing options...