Computed branch

laser_scientist · Joined: 26 Jun 2013 Posts: 7

Hello,

I am using a PIC24EP512GP806, trying to do an efficient computed branch using the BRA Wd instruction via some assembly code (below). However, CCS chokes on this instruction, giving an "expecting an opcode mnemonic" error. This is a valid opcode for this device, and CCS help even has this instruction in its list for inline assembly. I'm using version 5.018 of the compiler. Does anyone know why I might be getting this error? (Note that the actual application has many more entries in the jump table -- this is just a test case)

Thanks,
Dan

temtronic · Posted: Sun Dec 21, 2014 7:52 pm

I don't have a real answer for you as it 'looks OK...' but the Switch statement is pretty clever. Maybe cut a small example then dump the listing to see how CCS does it ?

just an idea...

Jay

Ttelmah · Joined: 11 Mar 2010 Posts: 19520

Get the Microchip assembly reference
(70157F.PDF).

Look at page128 (the page covering the 'computed branch' instruction).

At the top, there is a table of processors that support this. Note that the PIC24E, is _not_ ticked.....

Remember also, the address is relative to the PC. Much easier for a table based approach to use an absolute address. The easy way to do this, is to calculate the address required, push it onto the stack, and then execute a return.

laser_scientist · Joined: 26 Jun 2013 Posts: 7

Ttelmah, you're absolutely right about the reference. I was going by the device datasheet (70616g.pdf), where on page 488, they do have the BRA Wn instruction listed. I'm guessing that the assembly reference is correct though. And thanks for the alternative suggestion. That's what I'll do.

temtronic, thanks for the suggestion. That's actually how I decided to go with a table approach. From the code that CCS generates, I could see that they were effectively just doing a sequential if-then:

temtronic · Posted: Mon Dec 22, 2014 7:40 am

re: switch()...

it's been said that CCS changes the way it code switch() based on the number of entries. Maybe after 16(?) it happens??
Might be worth a coffee to code and see what they do !!

cheers
Jay

Ttelmah · Joined: 11 Mar 2010 Posts: 19520

I think it is very high on the PIC24.
Reason is that 'straight through' path on these chips only takes one instruction time, so it is quicker for a very large number of tests to use this approach.
A quick play, shows it refusing to switch, even well past the point where switching would be more economical....
However a simple array of function names works. So something like: