Here is some statistics without likely and with likely functions. It seems that using likely() gives better performance with Phenom I too. % Plain knl % No likely % With likely % dvb_ringbuffer_write 5,9 62,8 8,7 81,3 5,7 79,2 dvb_dmx_swfilter_packet 1,2 12,8 0,7 6,5 0,8 11,1 dvb_dmx_swfilter_204 2,3 24,5 1,3 12,1 0,7 9,7 Here "Plain knl %" is "perf top -d 30" percentage. 24,5 12,1 and 9,7 are percentages without a patch, with basic patch and last is with "likely" functions using patch. Regards, Marko Ristola 08.04.2011 18:40, Marko Ristola kirjoitti: > Avoid unnecessary DVB TS 188 sized packet copying from DMA buffer into stack. > Backtrack one 188 sized packet just after some garbage bytes when possible. > This obsoletes patch https://patchwork.kernel.org/patch/118147/ > > Signed-off-by: Marko Ristola marko.ristola@xxxxxxxxxxx > diff --git a/drivers/media/dvb/dvb-core/dvb_demux.c b/drivers/media/dvb/dvb-core/dvb_demux.c > index 4a88a3e..faa3671 100644 > --- a/drivers/media/dvb/dvb-core/dvb_demux.c > +++ b/drivers/media/dvb/dvb-core/dvb_demux.c > @@ -478,97 +478,94 @@ void dvb_dmx_swfilter_packets(struct dvb_demux *demux, const u8 *buf, > > EXPORT_SYMBOL(dvb_dmx_swfilter_packets); > -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
