On Sunday 14 June 2009, Øyvind Harboe wrote:
> my whole idea with using 32 bit sizes was that *no* bit array
> library would be needed....

Maybe call it a "bignum" package then?  Without things
like modular exponentiation.  Or addition.  Or ... ;)

My model is that a DR or IR scan is basically working
with a bit array ... the TDI bits were likely constructed
from a bunch of fields, and the TDO bits will likely then
need to be split into fields.  Some of those fields will
be more than 32 bits long.

So it feels to me like decent bit array tools should be
the guts of the data manipulation parts of JTAG code.
Maybe fancy tools that support fragmented arrays, or
maybe simpler ones.


> I don't think there are *any* places in OpenOCD where bit array
> code is needed or even particularly convenient if we had 32
> bit sizes.

Today that might be true.  But ... boundary scan registers
routinely take several hundred bits.  Other data registers
do too.  And there are even instruction registers that take
more than 32 bits.  (Those DSPs with 38 bit IRs, and 46 bits
in newer siblings, are examples.)

 
> My only (slight) concern with 32 bit would be that perhaps 64
> bit targets(a looong way off) would be better off w/64 bit
> sizes, though creating some JTAG helper fn's on top of a 32
> bit API would be easy peasy.

Why not just have an interface that doesn't *care* what word
size applies underneath?  Settling on 32 is better than
settling on 8 ... but specifying 32 is not necessary.  Such
things tend to cause needless trouble over time.

- Dave

_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to