C1: fix "assert(has_printable_bci()) failed: _printable_bci should have been set" when using -XX:+PrintCanonicalization
christian.thalinger at oracle.com
Tue May 29 17:16:18 PDT 2012
Looks good. Thanks for fixing this. It hit me quite often while working on C1 and I never had time to fix it.
On May 29, 2012, at 12:10 PM, Krystal Mok wrote:
> Hi all,
> Could I have a couple of review for this patch, please:
> I hit the "assert(has_printable_bci()) failed: _printable_bci should have been
> set" assertion when I was doing some experiment on C1, and needed to use
> -XX:+PrintCanonicalization for verification. It turns out that this flag is
> pretty broken, that quite a few places didn't set printable_bci appropriately.
> This patch tries to fix the missing printable_bcis.
> The link also includes a simple example before and after applying this patch.
> Added set_printable_bci() to Local and Phi's constructor.
> A "Local" instruction models an incoming argument, which gets its value before
> method entry, so I'm setting all Local's printable_bci to -1.
> A reasonable printable_bci for a "Phi" instruction is the same as the start bci
> of the basic block to which it belongs. I had to use a weird cast to get rid of
> a "invalid use of incomplete type 'struct BlockBegin' error from GCC.
> Added a default x->set_printable_bci(bci()) to Canonicalizer::set_canonical().
> There are quite a few place in Canonicalizer that doesn't specify the
> printable_bci for the newly created substitution instruction. It's reasonable
> to just set that to the "current" bci, which is the bci of the instruction to
> be substituted.
> Also adjusted the order of a set_bci() call in Canonicalizer::do_If, so that
> the new code above could pick up the modified bci in set_canonical().
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the hotspot-compiler-dev