Re: [Dri-users] Radeon (R100) strange bottleneck

[Keith W. <ke...@tu...>]

> ok done that (I'm running SuSE 8.1, oprofile is only included in 8.2 but 
> it compiles easily). Looks like almost all time is spent in only two 
> functions, here is the oprofile data (not all of course) from running 
> nwn 20 seconds or so:
> 
> vma      samples  %           symbol name             image name
> 
> 000755e0 226      0.314711    memcpy                  /lib/libc.so.6
> 00072900 635      0.884253    strcmp                  /lib/libc.so.6
> 00000000 4698     6.54208     (no symbol) /usr/local/games/nwn/nwmain
> 00149280 14074    19.5984     r200WriteStencilSpan_24_8 
> /usr/X11R6/lib/modules/dri/r200_dri.so
> 00148580 48114    66.9999     r200_mba_z32 
> /usr/X11R6/lib/modules/dri/r200_dri.so
> 
> Does that help?
> (btw oprofile looks to be a really nice package)

Well it does a lot - it really looks like we're in some sort of a fallback - 
those are span functions to access the depth buffer, and shouldn't be getting 
any use outside of fallbacks.  Maybe there's something like a fallback I"m not 
catching with the DEBUG switch...

Anyway, what would be interesting now is a backtrace from those functions. 
Probably the easiest way go get this is to try & run the 'nwmain' from a 
debugger and just hit 'Ctrl-C' or 'Ctrl-|' (or maybe generate some other 
signal with kill or whatever) and see where you end up.  It won't take more 
than a couple of random goes to hit those functions.

Keith