[Bf-committers] inlining tests redux.

John K. Walton bf-committers@blender.org
Sun, 30 Nov 2003 23:38:10 -0500 (EST)

On Sun, 30 Nov 2003, Robert Wenzlaff wrote:

> UnNamed and I discovered that -O3 implies inlining small functions, and since 
> my build flags are set to -O3 the test I reported at Sunday's meeting were 
> bogus.
> Still no major gains in rendertime, (18.55 with out  inlining vs 18.02 with, 
> for a 350kb binary size increase)  but it occured to me that the matrix math 
> in arithb.c is probably only a small part of rendering. 
> What other test could I perform that would be a good indicator of arithb 
> speed?  
i'm not sure what conclusions you're drawing. however, generally inline
optimization trys to reduce context swaps, or subroutine stack manipulation
by bringing the 'potentially executable code into I cache'. the downside 
is that you maybe flushing cache more that you need to. i expect this 
level of optimization if very processor dependant, and won't yield a 
'general performance improvement'. well, that's my 2 cents.