This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: ppc64 vDSO in mainline

Steve Munroe wrote:
So the additional overhead (at least 9 cycles with the NULL pointer check)
does matter. In the new memcmp I can compare 8 bytes per cycle (8 x 9 ==
72 bytes) so this overhead is significant.

You ignore completely real-world issues like loading the cache. Sure, in your synthesized test cases you might see effects for very short strings or memory blocks. But these are not that important. Once you work with bigger data sets they don't fit in the caches and then a single miss costs you so much more.

So, get this all set up as not only I proposed, then do some
measurements.  If it should turn out to be a real issue in real life, we
can think about some way to fix it.  That way in no case must involve
writing to the vdso or exposing it directly to user code.  prelink, for
instance, would be at least hampered in its work.  If no function in the
vdso needs to reference any PC relative data and we can get by with a
simple asm call, there are some quite dirty ways to avoid the extra
indirection, but I'm not willing to discuss any of this until the simple
solution is implemented, tested, and profiled first.

And anyway, if you think this work should really be your highest
priority as PPC maintainer, you're wrong.  There is a *MUCH* more
important issue waiting to be fixed.  If you don't know what I mean,
contact me directly.

â Ulrich Drepper â Red Hat, Inc. â 444 Castro St â Mountain View, CA â

Attachment: signature.asc
Description: OpenPGP digital signature

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]