Felipe Balbi wrote: > Hi Troy, > > On Mon, Dec 15, 2008 at 01:15:02PM -0700, Troy Kisky wrote: >> You need to post performance tests pre and post patch to >> let us know how much this degrades performance. Remember >> to turn off dma before the test. I personally don't >> like this patch. But I'm biased, as I added the assembly >> fifo access in the first place. > > personally I think using assembly to optimize pio transfers is not a > great deal. I'd rather spend time on dma and have pio written in C so > anyone can easily follow what's going on there. > > Anyways, I did some really quick and simple measurements (timed printk before > and after calling read/write fifo) and we spend on average about 10ns > more with the C implementation. > > I'd say it's something we can live with, what say you ?
Yeah, I expected a much greater difference. 10ns for every 32 bytes is trivial. > >> The description should also be improved. More like, "replace >> assembly language with similar C code" > > to me it says the same, with the difference that I'm saying the > reimplementation is for fifo handling. Sorry, about that comment. I ignored the subject line of your patch and was looking at only the description when I wrote that. _______________________________________________ Davinci-linux-open-source mailing list [email protected] http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
