[En-Nut-Discussion] Not fixed!!!: Re: confirmed!!! Re: NutOS 4.4.0 on ARM7: possibly bug in NutEnterCritical / NutExitCritical

[duane ellis]

Ole Reinhardt wrote:
>> I had to admit, that this LR used for ARM's call/ret confuses me every 
>> time again. Anyone else here with a more ARM-compatible brain structure?
>>     
>
> Sorry me either :(
>
>   

I am very familiar with ARM.
What you are speaking about is the "frame pointer" - and 'argument 
pointer', the "got" pointer, and so forth.

Just remember this:
    Most other CPUs - push the return address on the stack.
    Most other CPUs - pop the return address from the stack.
    The same can be done in the IRQ handler.

Those are TWO memory references. The question is: Do you really need them?
Can  your program be faster if you can remove those two memory references?

Think of your program execution as a "call-graph-tree" - you have LEAF 
nodes at the end.
LEAF nodes *never* call any other function.
If the CPU has enough "registers" for the "leaf" function - and it often 
does.
One can optimize away the PUSH/POP return address.
How much faster would your program be?

Also - how often does a "tail-call" occur?
ie: Function (A) calls (B) - then B returns and A immediately returns.
This technique, if handled well, lets the program (1) reduce stack space
and (2) not have to execute a series of pop/rts/pop/rts/pop/rts - to dig
out of some deep nested set of functions. Some of the RTS instructions 
go away.

Hence, ARM does not automatically PUSH things onto the stack.

Also - when you write hardware state machines in assembler, you learn 
amazing tricks using LR/R14 in amazing ways.

-Duane.