Re: [UrJTAG-dev] FT2232 Cable Driver

[Waschk,Kolja <ur...@ix...>]

Hi Arnim,

> Some parts of existing code were modified.

So far it looks very good, I endorse merging it into trunk. I made some
quick tests with an USB-Blaster attached, all fine.

> The cable is still no speed demon :-( Guess we need a different concept
> for the cable API to get a better performance.

We could introduce buffering replacements for "chain_clock", "transfer",
even "get_tdo" etc. Such replacement would just schedule the actual
operation for later in some static buffer. Execution of the operations
collected in the buffer would happen at a later time; either whenever
the caller asks for results, or with some kind of flush() function.

Instead of:

...other operations
chain_clock()
chain_clock()
...other operations
chain_clock()
chain_clock()

We should do

...other operations
chain_clock_defer()
chain_clock_defer()
...other operations
chain_clock_defer()
chain_clock()


For functions like "get_tdo" and "transfer" which normally return
results immediately, all calling code would have to be looked at -
whether it uses the result at all and _when_ it uses the result. If the
result isn't required immediately, the "get_tdo" request could be
buffered and the data could retrieved later with another call:

Instead of:

... other operations
x = get_tdo
... other operations>

New version:

... other operations 
(void)get_tdo_defer
... other operations
x = get_tdo_result

Tracking this as an Enhancement request:
 [ 1832990 ] Speed improvements for USB cables