Re: Questions about modifing AST.

[Eric Tang <tytang at cse dot cuhk dot edu dot hk>]

> 
> On 27-Oct-2002, Eric Tang <tytang@cse.cuhk.edu.hk> wrote:
> > 	i want to convert floating point to fixed point arithmetics
> > 	in compiler level.
> 
> Why?
> 
> Are you talking about changing the language semantics?
> Or are you just talking about a compiler optimization?

	Since i want to remove all the floating point operations(like
	float type, and other math functions applied on floating point)
	They are time-consuming, and sometimes not necessary for
	some applications. Fixed-point can be replaced and thus
	optimize the program in this way.

	So i am wondering how to do this is better.
¡@	First, user write a normal floating point programs(even
	handling small fractional parts, C programmer also use
	float type). Then i want to anaylse each float type 
	representation needed, such that the integral and 
	fractional bits required. Finally, replace the float
	type with my fixed-point defined in another C program,
	and also replace some maths functions specially defined
	for fixed-point too. That's all i want to do.

	I know that similar researchs have been done with the
	help of SUIF. As gcc is a common tools for programmer and
	also has been ported to various platform. So i want to 
	do it with GCC.
	
	So i really need you all experts to give me advices.
	
	The language is mainly "C" only!
> 
> If you want to do this, it's better to write a pass which operates
> directly on GCC's AST representation, rather than trying to read
> in the AST from the dump file (which won't work, IIUC, because the dump
> file is deliberately incomplete).
	
	If i want to operate directly on GCC's AST, i need to modify
	the source code of gcc?? since i don't want to modify the
	gcc compiler and want to do it seperately. So it is more 
	portable?!

	It is all my idea in my mind. I am a newbie, so pls help! thx.

Regards,
Eric