query reg. gcc

Mike Stump mrs@apple.com
Thu Dec 23 20:33:00 GMT 2004


On Thursday, December 23, 2004, at 10:39  AM, Ian Lance Taylor wrote:
> In general, I agree that that would be a useful feature if it is in 
> fact possible.

See the compile server branch for a partial existence proof it is 
possible.  :-)

It caches compilation fragments, when the cache it valid, things 
compile about as they can be validated, printf can spew out code and 
bump a couple of pointers.  About a 90x speedup when it hits, and 
that's untuned, tuned, I'd expect it to be better than that.

If people use precompiled headers perfectly, then the compile server is 
the same speed as a normal precompiled header compile (or so).  When 
precompiled headers aren't used, the speed approaches the speed of the 
perfect precompiled headers.  I say approaches, as it does lots of 
validation and symbol management.



More information about the Gcc mailing list