Note: The information below is taken from the Readme-Src.txt file found in the source release of Rose Shank.
Source release and the IL4 compiler by:
Rose Shank was an experimental 4k intro. The experiment was to see if it was feasible to build a small virtual machine with a custom bytecode to run the intro, and if it was easier/smaller. It was a given that faster it is not. :)
This language was named IL4 Lisp-ahtava (the compiler is called il4c). As the name says, the language uses the lisp syntax as its syntax. That’s about all there’s common to it. The reason the lisp parantheses syntax was chosen, was to facilitate fast turnabout while developing the language further (adding new constructs was a lot easier this way, as the lexer/parser stayed the same).
Il4c reads and parses all the given source files given to it at once. The files are interpreted as if they were joined together, and all symbols are globally visible. IL4 then performs a number of simple optimizations over the code. This includes stripping functions that are never called. Not much is done to the actual code. This gives greater control to the IL4 programmer (for example in C a small change may make a huge effect on the inlining and code size in general). All inlining and such should be done by hand. Il4c then generates a suitable bytecode for the remaining code. Only the operations that are needed are assigned a bytecode. As a last step, il4c generates one assembly file “out.asm” which contains both the generated bytecode interpreter and the actual bytecode. This asm file is designed to be compiled by nasm. The linker used may be the microsoft linker, but Crinkler is the only linker ever tested with il4c.
To compile Rose Shank, run compile.sh with cygwin. You may need to modify it to set the correct library paths (uses the win32 libraries from Visual Studio).
A few words about the IL4 language (as there is virtually no ducmentation about it) to get the prominent reader started.
Start from “test.il4” (the startup function “main” is there) and “core.il4” (most of the basic math functions and such are there).
Comments start with a # char and go up to the end of the line.
The language is totally untyped (well, strictly speaking the type is “32-bits” :). That means you have to be careful to call the right arithmetic and comparison functions (as well as others). The values may very well be pointers, integers, floats or whatever.
On the top level an IL4 file may contain roughly the following elements:
(const pi 3.1415926535897932384626433832795)
(var room_floor_lights) (var cam_ang 0.0)
# Random function returning a float with an integer interval [mini,maxi[. (fun rand_rangef (mini maxi) i2f (rand_range mini maxi))
Notes on example: “i2f” is a function that takes one argument, and it changes an integer to a float. “rand_range” is another function that takes 2 arguments.
(asmfun +i (a b) "pop eax" "add [esp], eax")
Most of the things inside function bodies are:
(fun glVertex3fv (arr) (stdcall (external_symbol "_glVertex3fv@4") arr)) (fun glViewport (x y w h) (stdcall (external_symbol "_glViewport@16") x y w h))
There are different forms for functions which return a floating point value, namely “stdcall-fp” and “cdecl-fp”.
Modify the compile.sh script and change the parameters a little. There are a few optimizations which really made the packed code bigger (even though the raw object file was smaller). Also the “-save-debug” switch is useful as it saves the middle forms (tree forms before/after optimization) for the program. “il4c -help” gives the list of command line options.
Read & try to understand the generated “out.asm” file. Some quick notes about it: - Every assembly function end with a “jmp ebx” - Every opcode has a 16bit loopup index to it’s code - Every bytecode function has a 16bit lookup index to it’s code - Every bytecode function has 16bits of function header and one return bytecode at the end (8bit); total overhead 3 bytes - Constants currently take 4 bytes each. - Every global that has a preset value takes 4 bytes (even if the same value is used as a constant elsewhere). - Globals that have no preset values take no space.
Doubles were a bit hacked in (as the general type is “32-bits”). The only functions that required doubles in Rose Shank were OpenGL calls which follow the stdcall calling convention. What we do, is use a asmfun which converts a float to a double on the stack. Because the stdcall convention requires for the function to clean it’s own arguments, this hack works (as the correct amount of items are on the stack after the call).
To compile the HD version, modify compile.sh to use “resolution_hd.il4” instead of “resolution.il4”.
These sources and the IL4 compiler are released under the GPLv2 license (see LICENSE.txt).