> From: Högander Jouni [mailto:[EMAIL PROTECTED] > Sent: Wednesday, October 15, 2008 8:04 AM
> L3 and L4 are anyway on when not in wfi, because of sleep/wkdep, so no > power save. This is the case at least with OMAP3 not sure about OMAP2. Probably some savings inside of individual domains when they are in idle like camera. Defiantly internal autogating in CortexA8 saves power if MPU domain is in idle. ... Recall if DSP is up and running L3/L4 might be up, but MPU could be in WFI with its domain gated off (also in INACTIVE/RET/OFF). A 1st order power waster is fan out from L3/L4 as you say. Future OMAPs add more domains to reduce fan out effect even if L3/L4 is on. > In many modules there is also possibility to gate ick by HW when not > in wfi (see AUTOIDLE in SYCONFIG). I'm not sure about this but > extremely also fcks could be taken care by HW (see CLOCKACTIVITY in > SYSCONFIG). Module AUTOIDLE can save power by selective gating inside the module. An easy experiment is to measure power while stopped in the debugger and enable/disable individual auto idles. You'll be surprised how much you get back on something like security accelerators (all I-clock, so gating inside of modules helps, even if l3/l4 is up). Clock activity in some modules does do extra gating at segment idle time. Regards, Richard W.
