Øyvind Harboe wrote: > On Wed, May 6, 2009 at 11:50 AM, Laurent Gauch > <laurent.ga...@amontec.com> wrote: > >> Ųyvind Harboe wrote: >> >>> On Wed, May 6, 2009 at 11:10 AM, Laurent Gauch >>> <laurent.ga...@amontec.com> wrote: >>> >>> >>>> Yes it could. But giving this out_mask info to the low layer JTAG API can >>>> really help to accelerate the JTAG interface itself and accelerating the >>>> OpenOCD by the way. >>>> >>>> >>> This is an extraordinary claim that I do not accept at face value. >>> >>> >> extraordinary ? >> When we created the OpenOCD JTAG API (Dominic and I at Amontec in 2005 >> before any SVN server, any forum ...), we was thinking to provide an >> extensible JTAG layer to OpenOCD allowing to accelerate the JTAG Layer >> Interface! The _mask could be a very important information to have in the >> Low Layer JTAG interface. >> > > There is not a lot of techincal specifics in your post for me to comment on. > > I claim to be pretty much up to speed on the original design ideas. > > I also get the impression that you didn't look much at the patch: > > Clearly we can reintroduce these constructs without complicating the > normal method invocations by creating helper functions on top of some > future lower level drivers. > > What the patch fixes is that 99% of the invocation sites are more > straightforward. > > If these sort of optimisations were attempted in the drivers, then there > *would* be changes done to the JTAG API and driver APIs. I even > believe that we are *better* positioned now to attack such optimisation > problems if we don't have lots of old cruft to releate to. > > > " technical specifics " ... it is concept !
We are working on layers like an ISO-OSI 7 layers. Actually, we do not have so much layers in our OpenOCD. But the JTAG API interface is certainly the more important layer in the OpenOCD concept. Actually not all features from JTAG API are used by upper layers (calling functions). On the Transport layer ISO-OSI there are a lot of features not used by all upper layers. Does they simplify Transport Layer because a large part of upper layers do not use some of the Transport Layer features? Yes, you could re-implement the CHECK structure in upper Layer by an helper, but we know that checking at very low layer can accelerate future JTAG interface. Anyway, you are free to simplify JTAG API interface by deleting the "_mask"s, but this will be a very **BAD REGRESSION** for OpenOCD. There are a lot of other things to do on OpenOCD that removing good features from a defined layer. Also, do you think adding some NULL pointers in the stack will really change the final datarate of OpenOCD ? I do not think ? but I think the OpenOCD can be a lot faster, and without having an embedded ARM9 or ARM11 Linux running OPENOCD. Laurent http://www.amontec.com _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
