On 12/04/2014 01:00 PM, Mark Wielaard wrote:
This solves a couple of crashers reported by Alexander.
This will probably have some performance impact, but I haven't measured
it yet. It would be good to have some performance tests. We also need
some overflow check for leb128 reading.
Josh, which tests did you use last time when you did the performance
improvements of this code? I am afraid I won't be able to keep it as fast,
but it would be good to know how much this slows things down.
I couldn't remember, but I found my original intro:
So the numbers I reported were with "tests/varlocs -k >/dev/null".
The good news is that your patch is barely slower, 81.64s -> 82.34s.
That's almost noise, but it's consistently a little slower over several
attempts. That's to be expected with more checks though.
The bad news is that these numbers in the 80s are back in the
neighborhood of where I was before those optimization patches. That's
not an apples-to-apples comparison, since I'm now on F21 gcc and kernel,
but that's still a lot to slip back. :(
I'll see if I can grab that old kernel debuginfo to do a more direct