[0/3] use C11 memory model GCC builtin atomics

  Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
with GCC C11 memory model __atomic builtins.

This series contributes to converging on standard atomics in 23.11 but is
kept separate as there may be sensitivity to converting from __sync to the
C11 memory model builtins.

Tyler Retzlaff (3):
  bus/vmbus: use C11 memory model GCC builtin atomics
  crypto/ccp: use C11 memory model GCC builtin atomics
  eal: use C11 memory model GCC builtin atomics

 drivers/bus/vmbus/vmbus_channel.c    |  2 +-
 drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
 lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
 3 files changed, 21 insertions(+), 19 deletions(-)
  

> From: Tyler Retzlaff [mailto:roretzla@linux.microsoft.com]
> Sent: Monday, 27 March 2023 16.30
> 
> Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> with GCC C11 memory model __atomic builtins.
> 
> This series contributes to converging on standard atomics in 23.11 but is
> kept separate as there may be sensitivity to converting from __sync to the
> C11 memory model builtins.
> 
> Tyler Retzlaff (3):
>   bus/vmbus: use C11 memory model GCC builtin atomics
>   crypto/ccp: use C11 memory model GCC builtin atomics
>   eal: use C11 memory model GCC builtin atomics
> 
>  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
>  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
>  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
>  3 files changed, 21 insertions(+), 19 deletions(-)
> 
> --
> 1.8.3.1
> 

Series-reviewed-by: Morten Brørup <mb@smartsharesystems.com>
  

On Mon, Mar 27, 2023 at 4:30 PM Tyler Retzlaff
<roretzla@linux.microsoft.com> wrote:
>
> Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> with GCC C11 memory model __atomic builtins.
>
> This series contributes to converging on standard atomics in 23.11 but is
> kept separate as there may be sensitivity to converting from __sync to the
> C11 memory model builtins.

- Looking at the patches, I thought the conversion was rather straightforward.
But this mention about "sensitivity" stopped me from merging.
Did I miss some risk with the changes of this series?


>
> Tyler Retzlaff (3):
>   bus/vmbus: use C11 memory model GCC builtin atomics
>   crypto/ccp: use C11 memory model GCC builtin atomics
>   eal: use C11 memory model GCC builtin atomics
>
>  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
>  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
>  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
>  3 files changed, 21 insertions(+), 19 deletions(-)


- I noticed that the vhost library has been providing an internal
wrapper for some __sync atomic with older GCC.
Some details are in the commitlog c16915b87109 ("vhost: improve dirty
pages logging performance").

Could it affect the existing legacy API performance?
  

On Wed, May 24, 2023 at 02:51:50PM +0200, David Marchand wrote:
> On Mon, Mar 27, 2023 at 4:30 PM Tyler Retzlaff
> <roretzla@linux.microsoft.com> wrote:
> >
> > Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> > with GCC C11 memory model __atomic builtins.
> >
> > This series contributes to converging on standard atomics in 23.11 but is
> > kept separate as there may be sensitivity to converting from __sync to the
> > C11 memory model builtins.
> 
> - Looking at the patches, I thought the conversion was rather straightforward.
> But this mention about "sensitivity" stopped me from merging.
> Did I miss some risk with the changes of this series?
> 
> 
> >
> > Tyler Retzlaff (3):
> >   bus/vmbus: use C11 memory model GCC builtin atomics
> >   crypto/ccp: use C11 memory model GCC builtin atomics
> >   eal: use C11 memory model GCC builtin atomics
> >
> >  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
> >  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
> >  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
> >  3 files changed, 21 insertions(+), 19 deletions(-)
> 
> 
> - I noticed that the vhost library has been providing an internal
> wrapper for some __sync atomic with older GCC.
> Some details are in the commitlog c16915b87109 ("vhost: improve dirty
> pages logging performance").
> 
> Could it affect the existing legacy API performance?

Yes.

gcc documents that you can replace __sync_<op> with __atomic_<op> using
SEQ_CST ordering.

When the __atomic_<op> builtins were initially introduced they generated
sub-optimal (you can interpret as slower) codegen relative to the
existing __sync_<op> builtins which was fixed in later gcc releases.

I do not know the actual version of gcc, but the commit you reference
indicates GCC_VERSION < 70100 is that boundary.

I (perhaps incorrectly) assumed that if the CI performance tests didn't
indicate a regression that the replacement of the remaining and minimal
use of the legacy API would have negligable impact.

If this is a bad assumption or there are concerns, I could update the series
to do the conditional __sync vs __atomic throughout.

Let me know how you'd like to proceed.

Thanks!

> 
> -- 
> David Marchand
  

On Wed, May 24, 2023 at 09:05:08AM -0700, Tyler Retzlaff wrote:
> On Wed, May 24, 2023 at 02:51:50PM +0200, David Marchand wrote:
> > On Mon, Mar 27, 2023 at 4:30 PM Tyler Retzlaff
> > <roretzla@linux.microsoft.com> wrote:
> > >
> > > Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> > > with GCC C11 memory model __atomic builtins.
> > >
> > > This series contributes to converging on standard atomics in 23.11 but is
> > > kept separate as there may be sensitivity to converting from __sync to the
> > > C11 memory model builtins.
> > 
> > - Looking at the patches, I thought the conversion was rather straightforward.
> > But this mention about "sensitivity" stopped me from merging.
> > Did I miss some risk with the changes of this series?
> > 
> > 
> > >
> > > Tyler Retzlaff (3):
> > >   bus/vmbus: use C11 memory model GCC builtin atomics
> > >   crypto/ccp: use C11 memory model GCC builtin atomics
> > >   eal: use C11 memory model GCC builtin atomics
> > >
> > >  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
> > >  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
> > >  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
> > >  3 files changed, 21 insertions(+), 19 deletions(-)
> > 
> > 
> > - I noticed that the vhost library has been providing an internal
> > wrapper for some __sync atomic with older GCC.
> > Some details are in the commitlog c16915b87109 ("vhost: improve dirty
> > pages logging performance").
> > 
> > Could it affect the existing legacy API performance?
> 
> Yes.
> 
> gcc documents that you can replace __sync_<op> with __atomic_<op> using
> SEQ_CST ordering.
> 
> When the __atomic_<op> builtins were initially introduced they generated
> sub-optimal (you can interpret as slower) codegen relative to the
> existing __sync_<op> builtins which was fixed in later gcc releases.
> 
> I do not know the actual version of gcc, but the commit you reference
> indicates GCC_VERSION < 70100 is that boundary.
> 
> I (perhaps incorrectly) assumed that if the CI performance tests didn't
> indicate a regression that the replacement of the remaining and minimal
> use of the legacy API would have negligable impact.
> 
> If this is a bad assumption or there are concerns, I could update the series
> to do the conditional __sync vs __atomic throughout.
> 
> Let me know how you'd like to proceed.

Anything further here or want to keep it as is?

> 
> Thanks!
> 
> > 
> > -- 
> > David Marchand
  

On Fri, Jun 2, 2023 at 6:18 AM Tyler Retzlaff
<roretzla@linux.microsoft.com> wrote:
>
> On Wed, May 24, 2023 at 09:05:08AM -0700, Tyler Retzlaff wrote:
> > On Wed, May 24, 2023 at 02:51:50PM +0200, David Marchand wrote:
> > > On Mon, Mar 27, 2023 at 4:30 PM Tyler Retzlaff
> > > <roretzla@linux.microsoft.com> wrote:
> > > >
> > > > Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> > > > with GCC C11 memory model __atomic builtins.
> > > >
> > > > This series contributes to converging on standard atomics in 23.11 but is
> > > > kept separate as there may be sensitivity to converting from __sync to the
> > > > C11 memory model builtins.
> > >
> > > - Looking at the patches, I thought the conversion was rather straightforward.
> > > But this mention about "sensitivity" stopped me from merging.
> > > Did I miss some risk with the changes of this series?
> > >
> > >
> > > >
> > > > Tyler Retzlaff (3):
> > > >   bus/vmbus: use C11 memory model GCC builtin atomics
> > > >   crypto/ccp: use C11 memory model GCC builtin atomics
> > > >   eal: use C11 memory model GCC builtin atomics
> > > >
> > > >  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
> > > >  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
> > > >  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
> > > >  3 files changed, 21 insertions(+), 19 deletions(-)
> > >
> > >
> > > - I noticed that the vhost library has been providing an internal
> > > wrapper for some __sync atomic with older GCC.
> > > Some details are in the commitlog c16915b87109 ("vhost: improve dirty
> > > pages logging performance").
> > >
> > > Could it affect the existing legacy API performance?
> >
> > Yes.
> >
> > gcc documents that you can replace __sync_<op> with __atomic_<op> using
> > SEQ_CST ordering.
> >
> > When the __atomic_<op> builtins were initially introduced they generated
> > sub-optimal (you can interpret as slower) codegen relative to the
> > existing __sync_<op> builtins which was fixed in later gcc releases.
> >
> > I do not know the actual version of gcc, but the commit you reference
> > indicates GCC_VERSION < 70100 is that boundary.
> >
> > I (perhaps incorrectly) assumed that if the CI performance tests didn't
> > indicate a regression that the replacement of the remaining and minimal
> > use of the legacy API would have negligable impact.

I don't think all the targets/tests are run in the CI (when compared
to what is done during a -rcX validation).
But I can understand this assumption.


> >
> > If this is a bad assumption or there are concerns, I could update the series
> > to do the conditional __sync vs __atomic throughout.

Nobody else complained/reacted on the topic.
The conditional would have to be duplicated in various places.

So ok, let's go with this approach and see if the rc1 validation
(which may run more tests) catches something.


> >
> > Let me know how you'd like to proceed.
>
> Anything further here or want to keep it as is?

Nop, thanks.
  

On Mon, Mar 27, 2023 at 4:30 PM Tyler Retzlaff
<roretzla@linux.microsoft.com> wrote:
>
> Replace the use of __sync_<op>_and_fetch and __sync_fetch_and_<op> atomics
> with GCC C11 memory model __atomic builtins.
>
> This series contributes to converging on standard atomics in 23.11 but is
> kept separate as there may be sensitivity to converting from __sync to the
> C11 memory model builtins.
>
> Tyler Retzlaff (3):
>   bus/vmbus: use C11 memory model GCC builtin atomics
>   crypto/ccp: use C11 memory model GCC builtin atomics
>   eal: use C11 memory model GCC builtin atomics
>
>  drivers/bus/vmbus/vmbus_channel.c    |  2 +-
>  drivers/crypto/ccp/ccp_dev.c         |  6 ++++--
>  lib/eal/include/generic/rte_atomic.h | 32 ++++++++++++++++----------------
>  3 files changed, 21 insertions(+), 19 deletions(-)

Please use get-maintainers.sh when submitting patches.

The change looks ok to me and rc1 validation will confirm if there is
an impact on performance.
Series applied, thanks.