[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc
Subject:    Re: cache misses in gcc 3.3
From:       Mike Stump <mstump () apple ! com>
Date:       2003-02-13 22:54:09
[Download RAW message or body]

On Thursday, February 13, 2003, at 11:26 AM, Tim Josling wrote:
> So would I - but there would be hundreds of thousands of calls out of 
> the
> parse; to wrap these in timvar would be prohibitive.

Not necessarily.  At Apple, we use a magic instruction, akin to the 
read timestamp instruction on the x86.  On the x86, it is 9 ticks of 
the processor clock on an AMD processor, and that is farily low, given 
how slow compiles actually are.  One can inline and if the difference 
is < limit, just reuse the last gettimeofday time, and if >, it can 
call a real function to do it the slow way.

>  One approach I thought of could be to do sampling by setting regular 
> interrupts based on timeouts and check if the IP pointed into bison 
> (i.e. a kind of poor man's gprof), however it would be difficult to do 
> this portably.

Ack, please, no.

[prev in list] [next in list] [prev in thread] [next in thread]