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: [COMMITTED PATCH] Remove explicit inline on malloc perturb functions.

> On Wed, Dec 17, 2014 at 10:42:19AM -0800, Roland McGrath wrote:
> > 2014-12-17  Roland McGrath  <>
> > 
> > 	* malloc/malloc.c (alloc_perturb, free_perturb): Remove inline keyword.
> > 
> Roland, did you ran benchtest before commiting that?

No.  What I did verify is that on x86_64-linux-gnu with gcc 4.8.2 (Ubuntu
4.8.2-19ubuntu1), where the warning was not produced beforehand, it did
not change the generated code.

> You caused a 5% performance regression on malloc microbenchmark when
> using gcc version 4.4.7 (Debian 4.4.7-8)

I assume you're talking about x86_64-linux-gnu, but you should be
clear when citing such things (in fact, you should probably say
something about the hardware, since the results might be quite
different even with the same compiler on different hardware).

That is older than the minimum required to build 2.21 (which is 4.6).
Even if it were true with 4.6, I'm not convinced we should care.  It's
always going to be the case that not all compiler vintages will
produce good code from the same source.  We need to be able to move
forward with cleanups and new best practices and not worry overmuch
about building libc with compilers that are very old.  As of now, I'd
say that anything that is suboptimal with a compiler older than 4.8
but just fine with 4.8 or newer is something we should not be worrying


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