[Discuss-gnuradio] fusb_darwin / osx code

[Michael Dickens]

On Jan 1, 2006, at 5:43 PM, Eric Blossom wrote:
> I am almost certain that the problem is that you are not submitting
> multiple asynchonous requests to the USB subsystem.  This 4MB/sec
> "limit" is what we are also seeing on NetBSD, which is also (at this
> time) lacking a method to stream multiple bulk requests.
>
> Since the existing libusb darwin/osx implementation issues non- 
> blocking
> bulk reads and writes, but then immediately blocks waiting for them to
> complete, it does seem that we have all the bits and pieces that we
> need to construct a fusb_osx.

Ah ... I was wondering what the linux code did, and that would make  
sense; I had thought about that possibility, but didn't want to go  
there yet as coding will take a bit of effort (as opposed to the  
quick efforts thus far: async w/ feedback, async w/o feedback, sync,  
async w/ isoch; all with the same results except for no feedback,  
which just didn't work at that really wasn't a surprise).

As OSX's IOKit provides an async callback with optional user defined  
arguments (e.g. used by libusb to get the actual # of Bytes  
transmitted), that mechanism can be used to determine when data is  
successfully transmitted or received, and then removed from an  
appropriate queue.  I'll do it in USRP's fusb_darwin (similarly to  
the linux code), since it'll be easier to do the list buffering in C+ 
+.  I will work on that no later than Tuesday, hopefully have a  
concept by the end of the week. - MLD