RFR (S) 8141044: C1 should fold (this == null) to false

[Aleksey Shipilev]

On 11/01/2015 10:25 PM, Vitaly Davidovich wrote:
> Doesn't your patch only work for the implicit 'this' argument to
> instance methods? It seems like only local at slot 0 is marked as
> receiver - what if 'this' is passed as one of the remaining args?

See e.g. the targeted benchmark from
  https://bugs.openjdk.java.net/browse/JDK-8141044


    @Benchmark
    @CompilerControl(CompilerControl.Mode.DONT_INLINE)
    public void test() {
        m(this);
    }

    private void m(Object o) {
        if (o == null) {
            throw new IllegalArgumentException();
        }
    }

...where "this" from the caller is passed through the slot 1 (slot 0 is
taken by the receiver for m()):

  private void m(java.lang.Object);
     0: aload_1
     1: ifnonnull     12
     4: new           #3   // class java/lang/IAE
     7: dup
     8: invokespecial #4   // Method java/lang/IAE."<init>":()V
     11: athrow
     12: return


If you are compiling m() alone, there is no chance in statically
figuring out slot 1 contains "this". C2 is also unable to figure this
out, if you, say, break the inlining for m(). So yes, you might say it
only works for the implicit "this", but...

If you inline m() into test(), then "this"-ness would propagate from the
test() to the m(). This happens even with the trivial dataflow that C1
apparently does (by figuring out the "defining" instruction), and that
will naturally conclude "o" is actually "this".

This is why the benchmark improves in C1 with a suggested patch.

Thanks,
-Aleksey

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20151102/47808565/signature-0001.asc>