hmm, well, let me put a disclaimer on that.. I'd be interested to check with someone at ARM who knows the pipeline well, to see if something like this could be implemented without having to write a full cycle accurate simulation of the ARM.. it's possible that a tool like I describe might be more difficult than it's worth if it requires an elaborate simulation of the inner workings of the ARM. But it was a feasible thing to write such a tool, it would be very useful for optimizing neon code
possibly on the topic of "NEON Optimization Forum Collaboration tools" (or possibly off-topic and more of a general tools thing)..
I've always wanted a static code analyzer for neon that given a specified cortex-a processor, could tell me which instructions would stall, and for how many cycles and why. It would be a helpful thing for optimizing NEON code. I guess not a hard thing to write for someone who has a deep understanding of cortex-a* pipeline.. (ie. ARM)
BR,-ROn Mon, Apr 25, 2011 at 12:54 PM, Kurt Taylor <kurt.taylor@linaro.org> wrote:
_______________________________________________At the last TSC meeting we refined the list of TRs for the Multimedia team. https://wiki.linaro.org/Cycles/1111/TechnicalTopics/Multimedia
I have taken the liberty to put these into a wiki page based on the feedback received during the meeting and some new items added after. Unless told otherwise, I will be creating the Multimedia WG blueprints from this list. Here they are sorted by priority:
https://wiki.linaro.org/WorkingGroups/Middleware/Multimedia/BlueprintsFor1111
Please review and be thinking about any questions you have, as well as a rough sizing for the amount of work required.
Comments?
Kurt Taylor (irc krtaylor)
linaro-multimedia mailing list
linaro-multimedia@lists.linaro.org
http://lists.linaro.org/mailman/listinfo/linaro-multimedia