Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!ucbvax!pasteur!cory.Berkeley.EDU!fadden
From: fadden@cory.Berkeley.EDU (Andy McFadden)
Newsgroups: comp.sys.apple2
Subject: Re: programming in assembly faster than HLL
Message-ID: <13656@pasteur.Berkeley.EDU>
Date: 16 May 91 22:25:36 GMT
References: <3579@kluge.fiu.edu>
Sender: news@pasteur.Berkeley.EDU
Reply-To: fadden@cory.Berkeley.EDU
Lines: 54

In article <3579@kluge.fiu.edu> acmfiu@serss0.fiu.edu (ACMFIU) writes:
>for those of you claiming that *you* can program in assembly faster than
>in *any* HLL, remember that BASIC is also an HLL. given the *same* libraries
>as BASIC, does your proposition still hold?

Yes, it does.  BASIC is interpreted.

Compiled BASIC isn't too bad though.

>                                            (personally i don't believe anyone
>who says their faster in assembly for the simple fact that HLL was created
>by assembly programmers ... chicken/egg thing doesn't apply here).

Does anybody understand what this means?

I think the point is that HLLs are every bit as fast as assembly, which is
rarely the case given competent programmers can common compiler technology.

If you require a demonstration, I can send you Bresenham's algorithm written
in APW C and in assembly.  The EXACT SAME algorithm is an order of magnitude
faster in assembly.  No special libraries; just 65816 code.

Why the difference?  APW C doesn't try to optimize its output.  It uses
constructs which work reliably in every case, but are slower than the code
I can produce.  I can use whichever addressing mode is appropriate for a
situation; a compiler has to work very hard to achieve the same results.  I
can also eliminate redundant loads and stores, I don't have to set or clear
the carry before every adc/sbc if I know the previous value, I can keep
things in the X and Y registers, etc, etc.

All things that a compiler COULD do, if it were sophisticated enough.

What a compiler *can't* do is modify the basic algorithm used.  Often times,
a fair portion of the speed increase from hand-coding a routine in assembly
comes from minor modifications of the algorithm, perhaps using special
features of an ML.  For example, a routine to move data from one region
of memory to another might copy it byte by byte; the compiler can't assume
that it's just a memory move, so it has to encode the loop.  An assembly
programmer could just use MVN.

You could also do what the FTA did to Bresenham's and expand the loop
slightly, which lopped 25% off of the execution time (I had a near-religious
experience when I saw their stuff... sigh).

To summarize, the people who claim that they can program in assembly faster
than in "*any* HLL" have yet to be proven wrong.  However, a good compiler
can narrow the gap considerably.

>albert

-- 
fadden@cory.berkeley.edu (Andy McFadden)
..!ucbvax!cory!fadden
fadden@hermes.berkeley.edu (when cory throws up)