Re: PATCH libatomic

[maya%NetBSD.org@localhost]

On Fri, May 08, 2020 at 06:23:01PM -0400, Thor Lancelot Simon wrote:
> On Fri, May 08, 2020 at 10:03:06AM +0200, Kamil Rytarowski wrote:
> > On 08.05.2020 02:14, Thor Lancelot Simon wrote:
> > > 
> > > Not without performance penalty for every atomic operation, unless you propose
> > > to do this by binary patch as is done in the kernel.
> > 
> > There is atomic penalty, but it is the contract and design of this (C
> > and C++) feature. Atomics can be legitimately lock-free or non-lock-free
> > and this is a feature.
> 
> Adding an extra test or, worse, an indirect branch before every atomic
> operation makes it far worse than it has to be.  An uncontended locked
> transaction on the bus may cost nothing.  An indirect branch followed
> by the same transaction consumes all sorts of microarchitectural resources
> will not just be slower but also impact the performance of _other_ code
> too.  That's why the kernel binary-patches out lock prefixes instead of
> using indirection for atomics.

The indirection only applies to the first call. The magic is within
rtld.