On Thu, Dec 10, 2009 at 9:50 PM, David Brownell <davi...@pacbell.net> wrote: > On Thursday 10 December 2009, Ųyvind Harboe wrote: >> GDB load is at 600kBytes/s @ 16MHz for arm7 now with zy1000 revc. > > Where previously it was ... ?
It went from ~580-90 to 600-15'ish. > The thing about micro-optimization is that it doesn't > necessarily affect anything of significance. You can > shuffle cycles around easily enough, even reduce them, > but when the bottleneck is I/O it just doesn't matter. > > And in most systems I've worked with, I/O is indeed > the bottleneck. Agreed that on smaller micros, there > are cases where it's othewise; is this such a case? Not really. The performance is CPU bound here. There is a 100mBit connection on ethernet and the data going out of the jtag chain could easily be doubled. I've worked quite a bit on OpenOCD optimizations and there is plenty of evidence that the gdb packet processing is a bottleneck on embedded hosts. This particular optimization was partly a refactoring to allow further work to be done. Micro optimizations are great when they just get rid of silly overhead that shouldn't play any part in design decisions. -- Øyvind Harboe US toll free 1-866-980-3434 / International +47 51 63 25 00 http://www.zylin.com/zy1000.html ARM7 ARM9 ARM11 XScale Cortex JTAG debugger and flash programmer _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
