[dpdk-dev] [PATCH v2 0/8] use GCC's C11 atomic builtins for test

Tyler Retzlaff roretzla at linux.microsoft.com
Wed Jun 30 23:49:37 CEST 2021


On Wed, Jun 30, 2021 at 08:25:44PM +0000, Honnappa Nagarahalli wrote:
> <snip>
> 
> > >
> > > As I mentioned earlier in this thread, GCC supports 2 types of
> > > atomics. "Use GCC atomic builtins" does not help distinguish between
> > > them. In "GCC's C11 atomic builtins" - "C11" indicates which atomics
> > > we are using, "atomic builtins" indicates that we are NOT using APIs
> > > from stdatomic.h
> > 
> > if you need a term to distinguish the two sets of atomics in gcc you can qualify
> > it with "Memory Model Aware" which is straight from the gcc manual.
> "Memory model aware" sounds too generic. The same page [1] also makes it clear that the built-in functions match the requirements for the C11 memory model.

allow me to put your interpretation of the manual that you linked side
by side with what the manual text actually says verbatim.

your text from above
  "built-in functions match the requirements for the C11 memory model."

the actual text from your link
  "built-in functions approximately match the requirements for the C++11 memory model."

* you've chosen to drop approximately from the wording to try and make
  your argument.

* you've also chosen to substitute C11 in place of C++11. again
  presumably for the same reason.

in fact the entire page does not mention C11 even once, it also goes on
to highlight a specific deviation from C++11 with this excerpt "because
of a deficiency in C++11's semantics for memory_order_consume"

> There are also several patches merged in the past which do not use the term "memory model aware". I would prefer to be consistent.

i prefer the history represent the change. that previous submitters and
reviewers lacked precision is not my concern nor is consistency a reason
to continue documenting history incorrectly.

i'm waiting to ack the change, it's up to you. you've already spent more
time arguing than it would have taken to submit a v2 correcting the
problem.

> 
> [1] https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html


More information about the dev mailing list