Xref: utzoo comp.arch:10558 comp.lang.misc:3057
Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!apple!ames!xanth!mcnc!decvax!ima!haddock!suitti
From: suitti@haddock.ima.isc.com (Stephen Uitti)
Newsgroups: comp.arch,comp.lang.misc
Subject: Re: Programming and Machine Operations
Message-ID: <13979@haddock.ima.isc.com>
Date: 9 Jul 89 04:35:25 GMT
References: <57125@linus.UUCP> <1989Jun24.230056.27774@utzoo.uucp> <13970@haddock.ima.isc.com> <1398@l.cc.purdue.edu>
Reply-To: suitti@haddock.ima.isc.com (Stephen Uitti)
Organization: Interactive Systems, Boston
Lines: 75

In various articles, cik & suitti argue...
>>suitti
>cik

>> The code sequence is: ...
>> 
>> void lvecadd(a, b, c, s) /* a = b + c, length s */
>> long *a; long *b; long *c; long s;
>> {
>> 	do {
>> 		*a++ = *b++ + *c++;
>> 	} while (--s != 0);
>> }
For the VAX:
L18:addl3	(r9)+,(r10)+,r0
movl	r0,(r11)+
decl	r8
jneq	L18

>
>I am afraid I will have to give you a D- on this.  Most of the time I would
>not even bother with a call, considering the code length.  But the code is
>bad for vectors of length >2.  How about
>{
>	end = a + s;
> 	do {
> 		*a++ = *b++ + *c++;
> 	} while (a < end);
>}

L26:addl3	(r9)+,(r10)+,r0
movl	r0,(r11)+
cmpl	r11,r7
jlss	L26

These are the loops as coded by the VAX PCC based C compiler.
I've no idea why it felt the movl was needed.  In any case both
code fragments compile to the same number of instructions.  My
decl has fewer arguments than your cmpl.  Without running the
test, there is no way to know which will run faster.  It depends
on which VAX.  A microVAX might (and probably does) have
different relative times than a 780.  Still, i expected the
compiler to use some sort of subtract one and branch if not zero
instruction instead of a decrement, and a branch.  Unfortunately,
i didn't check my VAX architecture handbook in advance, and i'd
forgotten (through disuse) that the VAX has an SOBGTR rather than
an SOBNE.  I should have used
	} while (s > 0);
On a PDP-11, the "SOB" instruction uses a test for not equal to
zero.  Further, on a PDP-11 (and many other machines) an extra
variable may strain the register resources of the machine
requiring some high use variable be put into slower storage.  A
really good compiler should have been able to convert the end
condition and use the right instruction.  The VAX PCC based
compiler is not a "really good compiler".

Also, i was considering inlined code, even if originally
presented as a function.  If the vector lengths are large (and
experience shows that they never are), it doesn't matter if it is
a subroutine or not, just as the computation for 'end' in your
example is immaterial.

Yeah, i know, facts in an argument.

Fortunately, i went to a school that didn't have letter grades.
It was pass/fail.  Many schools, including Purdue, often treat
their grad students with some dignity and respect.  I had the
good fortune to find an undergrad school that did so.  I never
had to argue with instructors.  It was always real clear if i'd
passed or failed (did i do the work? yes: pass, no: fail).  Thus,
i never got a D- for code that worked.  Once, however, in high
school, i managed to get an A (best you could get), with a U
(unsatisfactory) for effort.

Stephen.