Re: State of support for the ISO C++ Transactional Memory TS and remanining work

[Torvald Riegel <triegel@redhat.com>]

On Wed, 2015-11-11 at 15:04 +0000, Szabolcs Nagy wrote:
> On 10/11/15 18:29, Torvald Riegel wrote:
> > On Tue, 2015-11-10 at 17:26 +0000, Szabolcs Nagy wrote:
> >> On 09/11/15 00:19, Torvald Riegel wrote:
> >>> I've not yet created tests for the full list of functions specified as
> >>> transaction-safe in the TS, but my understanding is that this list was
> >>> created after someone from the ISO C++ TM study group looked at libstdc
> >>> ++'s implementation and investigated which functions might be feasible
> >>> to be declared transaction-safe in it.
> >>>
> >>
> >> is that list available somewhere?
> >
> > See the TM TS, N4514.
> >
> 
> i was looking at an older version,
> things make more sense now.
> 
> i think system() should not be transaction safe..
> 
> i wonder what's the plan for getting libc functions
> instrumented (i assume that is needed unless hw
> support is used).

No specific plans so far.  We'll wait and see, I guess.  TM is still in
a chicken-and-egg situation.

> >> xmalloc
> >>     the runtime exits on memory allocation failure,
> >>     so it is not possible to use it safely.
> >>     (i think it should be possible to roll back the
> >>     transaction in case of internal allocation failure
> >>     and retry with a strategy that does not need dynamic
> >>     allocation).
> >
> > Not sure what you mean by "safely".  Hardening against out-of-memory
> > situations hasn't been considered to be of high priority so far, but I'd
> > accept patches for that that don't increase complexity signifantly and
> > don't hamper performance.
> >
> 
> i consider this a library safety issue.
> 
> (a library or runtime is not safe to use if it may terminate
> the process in case of internal failures.)

If it is truly a purely internal failure, then aborting might be the
best thing one can do if there is no sensible way to try to recover from
the error (ie, take a fail-fast approach).
Out-of-memory errors are not purely internal failures.  I agree that it
would be nice to have a fallback, but for some features there simply is
none (eg, the program can't require rollback to be possible and yet not
provide enough memory for this to be achievable).  Given that this
transactions have to be used from C programs too, there's not much
libitm can do except perhaps call user-supplied handlers.

> >> uint64_t GTM::gtm_spin_count_var = 1000;
> >>     i guess this was supposed to be tunable.
> >>     it seems libitm needs some knobs (strategy, retries,
> >>     spin counts), but there is no easy way to adapt these
> >>     for a target/runtime environment.
> >
> > Sure, more performance tuning knobs would be nice.
> >
> 
> my problem was with getting the knobs right at runtime.
> 
> (i think this will need a solution to make tm practically
> useful, there are settings that seem to be sensitive to
> the properties of the underlying hw.. this also seems
> to be a problem for glibc lock elision retry policies.)

Yes, that applies to many tuning settings in lots of places.  And
certainly to TM implementations too :)

> >> sys_futex0
> >>     i'm not sure why this has arch specific implementations
> >>     for some targets but not others. (syscall is not in the
> >>     implementation reserved namespace).
> >
> > Are there archs that support libitm but don't have a definition of this
> > one?
> >
> 
> i thought all targets were supported on linux
> (the global lock based strategies should work)
> i can prepare a sys_futex0 for arm and aarch64.

arm and aarch64 should be supported according to configure.tgt.  Also
see the comment in config/linux/futex_bits.h if you want to change
something there.  I haven't tried arm at all so far.