[LLVMdev] LLVM Concurrency and Undef

[Eli Friedman]

On Sat, Sep 10, 2011 at 7:13 PM, Peter Lawrence
<peterl95124 at sbcglobal.net> wrote:
>
> On Aug 22, 2011, at 5:29 PM, Eli Friedman wrote:
>
> LLVM Atomics and Concurrency guide also states the following
>
> "Note that speculative loads are allowed; a load which is part of a
>
> race returns undef, but does not have undefined behavior"
>
> If the speculative loads returns an undef and the returned value is
>
> used, then it results in an undefined behavior. Am I correct?
>
> It behaves like any other undef value... which do often lead to
>
> undefined behavior.
>
> Eli,
>       Java by necessity has to be well defined in spite of data-races
> (because, IIUC, detecting
> data-races in just about anything and java-byte-codes in particular is an
> N-P-complete problem,
> so cannot be part of the byte-code validation process, therefore any
> byte-code which does have
> a data-race and does pass validation must still execute with defined
> behavior).

Yes.

> so, again IIUC, it appears that LLVM-IR could not be used for Java if we
> import C/C++'s
> strict "data-race implies undefined" semantics into our IR.

You can't use regular load and store instructions, but I have recently
added load and store instructions with the appropriate semantics for
Java to the LLVM IR.  See
http://llvm.org/docs/Atomics.html#o_unordered .

-Eli