Date:    19971010
From:    Josef Hanzal
To:      piclist@mitvma.mit.edu
Subject: C versus assembler

>> We regularily win with compiled C against hand written assembler. The
>> difference is the compiler doesn't forget a programmers trick.
>>
>> Walter Banks
>
>Yes, but after looking at your output and remembering the trick I can
>write assembler code that will at least tie it and probably beat it.
>(After all, I know more info than the compiler.)

Never ending C versus assembler battle... Despite of all the counter-examples
anyone has or will post, I agree with following text (found at the Fast
Forward Engineering answer page)

5a. An excellent C compiler can generate better code than an excellent
assembly-language programmer, because the compiler can safely generate
incomprehensible, non-maintainable, "brittle" code that the programmer
wouldn't dare write.

5b. An excellent assembly-language programmer can write better code than an
excellent C compiler, because the programmer can see a much larger picture
of the problem than the compiler can.

And maybe one more point - if any of the two languages was better in all
cases, then how would you explain, that the other one still survives ?

Regards

Josef


Date:    19971010
From:    John Payson
Subject: Re: C versus assembler
To:      piclist@mitvma.mit.edu

> Never ending C versus assembler battle... Despite of all the counter-examples
> anyone has or will post, I agree with following text (found at the Fast
> Forward Engineering answer page)
>
> 5a. An excellent C compiler can generate better code than an excellent
> assembly-language programmer, because the compiler can safely generate
> incomprehensible, non-maintainable, "brittle" code that the programmer
> wouldn't dare write.
>
> 5b. An excellent assembly-language programmer can write better code than an
> excellent C compiler, because the programmer can see a much larger picture
> of the problem than the compiler can.

I would rephrase these slightly:

[1] Given sufficient effort, a proficient assembly language program who
    puts code size or execution speed above all other concerns (such as
    maintainability) can produce smaller or faster code than a compiler
    can write.  Such code may require major reworking for even "minor"
    changes, but in applications where a 10% reduction in code size can
    save a few cents each on 100,000 units such code may be worth the
    time and hassle required to produce it.

[2] In many embedded applications, there is a level of code size and speed
    which is "good enough"; any improvements beyond this point serve no
    useful purpose.  A programmer writing in C and using a decent compiler
    can often produce "good enough" code with less effort than an assembly
    language programmer.

[3] Many programs spend 90% or more of their time executing a very small part
    their code (5-10% or less).  Even small improvements in this part can
    yield big improvements in performance.  Fortuitously, it is often such
    small pieces where hand-assembly significantly beats compiled code; one
    game I wrote for a PC had about 100 lines of assembly (about 60 real
    instructions) and a couple thousand lines of Pascal.  The assembly code
    took about 80% of the program's execution time, and was about 30 times
    as fast as what the compiler would have produced.  As a result, a game
    that would have run about 2 frames/sec if it were all in Pascal instead
    ran at 60.


Date:    19971010
From:    William Chops Westfield
To:      piclist@mitvma.mit.edu
Subject: Re: C versus assembler

    And maybe one more point - if any of the two languages was better in all
    cases, then how would you explain, that the other one still survives ?

Please repeat to yourself, over and over again until you are convinced,
"having the fastest possible most efficient code is not always important."

As an extreme example, I think I can state "assembler is always faster than
interpreted basic" without much fear of contradiction.  That doesn't mean
that there is no reason for the basic stamp to exist...

Sometime, I shall have to tell you all my views on why Cisco had to abandon
horizontally microprogrammed bit-slice microcontrollers as their preferred
engine for network controllers.  If the reasons aren't obvious :-(

BillW