On Thu, 23 Jun 2005, Mario Mendes Jr. wrote:

> OK, maybe not necessarily limited to pics, but....
>
> So, I remember when I started reading about cpus/mcus a long time agon
> about why banking and paging came about.
>
> Being a beginer, I do find banking and paging to be a nusiance,
> nevermind making code a little more complex to deal with.
>
> But rather than trash the idea, what I wanted to know from the more
> experienced guys in the list is: are there any actual benefits to
> banking and paging?  Is there ever a situation where banking and paging
> actuall makes things easier/better?

In theory banked and paged address spaces make for more compact program 
code since the addresses used for any indirect access will be shorter. 
But that is history (once upon a time code storage was extremely 
expensive and nobody had optimising compilers). On most 8 bit micros 
there were addressing modes for 'short' and 'long' distance of indirect 
objects. The 8080, Z80, MCS51 and most Motorola 8 bit CPUs all have this 
feature. When storage became cheap the banking (and shadowing which is a 
form of banking) became obsolete and they are still used only for 
acceleration (f.ex. the BIOS in a pc is copied to RAM and then mapped 
out during boot - code runs up to 100 times faster from ram than from 
flash BIOS), and in 'legacy' architectures (like mask programmed 4 bit 
cpus).

I do not know why Microchip chose to go with banking and paging instead 
of making wider instruction words (to make room for the extra address 
bits). It may have saved some silicon real estate (in the form of less 
code storage bits required for the same task). Others went with wide 
words all the way. The ARM is a good example, its words encode a lot of 
information: example (from 'Introduction to the LPC2000'):

 	if(Z==1) R1 = R2+(R3*4)

encodes as:

 	EQADDS R1,R2,R3,LSL #2

which is a single word on 32 bits. On a pic one would do:

 	btfss	STATUS,Z
 	goto	noz

 	rlf	R3,w	; w  = r3 * 2
 	movwf	R1	; r1 = r3 * 2
 	addwf	R2,w    ; w  = r3 * 2 + r2
 	addwf	R1,f	; r1 = 2 * r3 + 2 * r2 + r2 = r2 + r3 * 4 
noz:

which takes 6 x 12 or 6 x 14 bits and only operates on 8 bit numbers in 
a 32-register file. So the pic consumes at least 72 bits of code to do 
1/4 (8 bit registers instead of 32) of the equivalent single 32 bit ARM 
instruction word, and runs in 6 T cycles (the ARM takes 1 T cycle to do 
the job).

On the other hand, a simple bit banging code can consume less space on a 
pic than on an ARM. E.g. toggling an output can be done as:

 	movlw	0xFF,w
 	mofvwf	RB
again:
 	xorwf	RB,f
 	goto	again

the loop uses 2 x 12 bit words and runs in 3T per loop. An equivalent 
ARM loop will use 2 x 32 bit words and run in 2 or 3T (depending on MAM 
settings). so the PIC saves 60% code space because it uses narrower code 
words (which is possible due to banking).

Peter
-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist