From: Ø. H. <oyv...@zy...> - 2008-07-29 21:42:09
|
On Tue, Jul 29, 2008 at 8:31 PM, Spen <sp...@sp...> wrote: >> >> Why is this an advantage? >> > > currently we access gdb using tcp, a work in progress uses pipes instead of > tcp, eg. > (gdb) target remote | openocd --pipe is RLE required to make pipes work? > > This is much faster than tcp and cutting down the traffic is a step. > >> >> I'd want to see some measurements showing that this is an improvement. >> > > from the tests i have done it is either the same or slightly quicker. > It largely depends on the data being read from the target. Any numbers? I'd think this would adversely affect single stepping performance and slightly help larger memory reads on low bandwidth channels. >> It's a performance regression for embedded hosts as it >> currently stands. >> > > without testing we will never know for sure, but you may be correct. > The problem is only you have access to a supported emebedded host, so i > cannot test. I know for sure the code would have to be optimized/tinkered to avoid regressions and I'm skeptical that it will, overall, be measurable difference. Since I'm not a believer, I'm not very keen on working on this... got other things on OpenOCD where I think my effort could make a bigger splash... -- Øyvind Harboe http://www.zylin.com/zy1000.html ARM7 ARM9 XScale Cortex JTAG debugger and flash programmer |