Comments on: Minimizing memory use in embedded systems tip#2 – Be completely consistent in your coding style

By: Nigel Jones

Nigel Jones — Mon, 16 Jan 2012 13:36:40 +0000

Sure. One compiler I use all the time that does it is IAR’s AVR compiler. IAR calls it ‘Cross call’ optimization. You can specify the number of passes the compiler will use. For example, one project of mine compiles down to 42156 bytes with full size optimization except for cross call optimization. If I turn on unlimited cross call optimization, then the size reduces to 37006 – an impressive reduction of 12.2% . The compile time does of course go up slightly.

By: Garry

Garry — Mon, 16 Jan 2012 01:09:41 +0000

Would you please give a reference to a HLL language *compiler* that does the “subroutines of subroutines” code compression?

I have used more than 20 different programming languages (FORTRAN, BASIC, Pascal, C, LISP, Ada, Postscript, Objective-C, C++, Scheme, Java, Erlang, Ruby, Python, JavaScript, several assemblers, several “4GLs”, etc.), and numerous implementations of them, I’ve read and read about about many more programming systems.

The ONLY language system I have ever seen, or read about, that use the mechanism the article describes are “threaded” or “threaded and knotted” systems, which I believe are rare and unusual. Forth and its progeny is one branch, and postscript another. All that I know of are interpreters, and not compilers.

By: Will

Will — Sun, 15 Jan 2012 13:53:43 +0000

Threaded is the term you are looking for.

By: Nigel Jones

Nigel Jones — Wed, 09 Sep 2009 10:54:16 +0000

Thanks for the information Anders – if I'd thought about it I'd have asked you first! Regarding the effectiveness on different architectures. If you are programming an 8051 then pretty much regardless of what you write in a HLL the object code comes out the same – for the simple reason that there really is only one way of doing many things in the 8051. Conversely CPUs that are register rich with orthogonal instruction can do a task in dozens of different ways. As a result I'd expect that being consistent in your coding style would actually be more beneficial with these types of CPUs. Having said that, I've never run a test to support my hypothesis

By: Anders

Anders — Tue, 08 Sep 2009 06:44:34 +0000

I forgot to say that this kind of optimization is generally most efficient for 8-bit CPU's and accumulator based architectures. This is because there will often be a lot of common sequences to work with, even if the programmer is inconsistent in his coding style.(The paper I mentioned is named "Analyzing and compressing assembly code" by Fraser et al. from 1984)

By: Anders

Anders — Mon, 07 Sep 2009 16:10:22 +0000

Well, There are several names but the most common are probably cross-jump and cross-call. They differ in what scope they have for the optimization.) One of the first papers on this method as applied to file scope called it a combination of cross-jump and procedural abstraction.

By: Nigel Jones

Nigel Jones — Sat, 05 Sep 2009 10:41:47 +0000

That was my first thought. However upon investigation, CSE seems to apply at the source code level – i.e. identifying common sub expressions in the source code. This technique applies at the object code level. As such it appears to be a combination of peep hole optimization + CSE.

By: darkwzrd

darkwzrd — Sat, 05 Sep 2009 08:29:51 +0000

"As a final note, does anyone know the formal name for this type of optimization?"I'm guessing "Common Subexpression Elimination", or CSE?