This is the mail archive of the mailing list for the GCC 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: ia64 tls support

> On Mon, May 27, 2002 at 08:37:24AM +0200, Jan Hubicka wrote:
> > Just curious, does the RISCy machines has similar feature?  They usually
> > don't have load-modify-store isntructions, but they can have something
> > as synchronization primitive.
> The two most common are load-lock/store-conditional (ll/sc) 
> and compare-and-swap.
> Both of these will require a loop, but I recommend that the
> loop be buried inside the pattern.  Apart from not wanting
> to over-complicate the cfg, ll/sc sometimes has restrictions
> like "there can be no taken branches between the two insns"
> and "there can be no other memory references between the two
The requirement of loop looks little but unforutnate, but ...
> insns".  The later restriction means that we cannot allow
> reload to insert spill insns, which pretty much necessitates
> using a single pattern.

OK, so I will make patch for the extra pattern and add locking for i386
and lets see how popular profiling of threaded applications gets.
I tend to agree that we don't have to come with complex sollution for
problem not interesting for too many developers at the moment.
I would like the PDO to be more widely used, so thats why I think some
thread safety sollution is needed, but I am fine with the locking.

> r~

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