On 5/28/2014 12:35, Drago, William @ MWG - NARDAEAST wrote:
Bigger code gets kicked out of the processor cache faster, so the
processor has to go back to main memory more often.
Don't modern compilers consider what effects the speed optimizations will have
on the pipeline and the cache and optimize accordingly?
The compiler probably won't be looking at things like cache pressure.
All the compiler will care about is that this function now executes
twice as fast, yay! It won't consider that it had to make it twice as
large as the -Os version, which will therefore stay in cache 4x as long,
so that cache thrashing will throw away the 2x benefit.
As you say, it probably won't make a function so large that it never
fits into L1 in the first place.
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users