🚀 go-pugleaf

In article <lj2h6hINNikj@exodus.Eng.Sun.COM>, nand@not-and.Eng.Sun.COM (Nand Mulchandani) writes:
|> A couple of things make it a *little* harder :
|> 
|> 1) Branches with delay slots (you have to be able to represent them and
|>    deal with them).

Well, actually, the assembler doesn't *have* to know about delay slots.
It's a nice frill if it knows about them well enough to warn you when you
do something unusual in a delay slot, such as a multi-instruction macro.

In general, I don't like "smart" assemblers, trying to fill delay slots
for me.  If I want "smart", I'll use a compiler, thank you.

|> 2) In general, chip workarounds are done in the assembler, and this code
|>    is just yuckier to deal with

This has always really ticked me off.  The SPARC assembler, for instance,
has workarounds (read KLUGEs) for all sorts of stupid things.  When I try
to code a test for a processor that doesn't need the damned kluges, I get
them anyway.  In the sort of tests that I write, I want only the instruction
I'm asking for, no more, and no less.  With the SPARC assembler, so far as
I know, I can't even disable the damned kluges.  So, I have to regress to
the level of machine language programming, and fool the assembler by
inserting an instruction as if it were a data word (can you say manual
assembly?).  Ick!

|> 3) branches have limited scope (on SPARC, you get only 22 bits in the branch
|>    instruction). It is reasonable to leave a check in the assembler to detect
|>    branches to targets that are larger than the displacement will allow.

Yes, but this is true of nearly all microprocessors.  Nothing special about
RISC here.

|> 4) Code to check architectural correctness (try doing something illegal
|>    with the SPARC assembler - if it sqawks, then someone had to write the code 
|>    to do the check). This is very important, because the machine will not always
|>    execute "arbitrary assembly language". 

Again, I consider this sort of thing to be a frill.  



In general, an assembler is an assembler is an assembler.  They're pretty
easy to do, unless you are planning to get too fancy.  I like plain, vanilla
assemblers, perhaps with a simple macro preprocessor.  
-- 

...Russ Schnapp
BIX: rschnapp           Email: netcom!metaflow!rschnapp or rschnapp@metaflow.com
Metaflow Technologies   Voice: 619/452-6608x230;  FAX: 619/452-0401
La Jolla, California    Unless otw specified, I`m speaking only for myself!