FreeLists / openbeos / [openbeos] Re: Kernel status (POSIX)

["Michael Phipps" <mphipps1@xxxxxxxxxxxxxxxx>]

>> yet not impact portability whatsoever, since there's still the C 
>> functions available
>> for new ports until whatever time they choose to write the optimized 
>> version, or not. :)
>
>And this is the key. Write whatever you please in assembly, *as long as
>* there is a C fallback. Of course, some things have to be done in 
>assembler in the kernel, and use it for that, but always provide a 
>fallback everywhere else.
>-Nathan

Let me put it this way - 

OTHER THAN stuff from NewOS (which is Travis' domain), anyone who wants
to use asm at all will have to justify it. There is almost never a good reason.
I would want things like a code profile showing that the code is the performance
issue, real evidence that the code CAUSES a performance issue, and to have 
looked
at it as a team and decided that there is *no* other way. 

Premature optimization is the root of bad coding. Moving to asm is the most 
common
(that I can think of) premature optimization that people make. I use profilers 
all the time 
(whether -pg or some other methodology) on code that is performance crititical. 
I am winding
up such a project at my full time job. Once the code was 100% complete AND 
tested AND
regession testable AND documented, we work on getting it faster. Using 
profiling, I have fixed 
a couple, maybe as many as 5 functions and multiplied the speed of the process 
by 1.5. 
That, not guessing, not using folk stories and wives tales, is the way to 
optimize. :-)

Travis writes most of his code in C. Only the things that need to be in asm are.

Does that make everyone comfortable? Except those who want to write asm? :-)