RFR(S): 8193927: Optimize scanning code for oops.

[Doerr, Martin]

Hi,

I think, ideally, ImmediateOopsEmitted should only be true on x86 and maybe zero (where it is pointless).

If I understand it correctly, only usages of oop_Relocation::spec_for_immediate() need it which uses oop_index = 0. Other oop_Relocations use the CodeBlob's oop pool and the oops will be found there by iterating over "oop* p = oops_begin(); p < oops_end(); p++".

I think the comment should be a little more precise because every uses immediate oops. The point is that the affected ones are not part of the CodeBlob's oop pool.
+   /* Knowing whether oops are emitted to the code cache as immediates */    \
+   /* allows to skip walks of the code cache if there are none.        */    \

So I think the optimization may be nice to have for aarch/arm as well, but ARM experts should decide on that.

Best regards,
Martin


-----Original Message-----
From: hotspot-compiler-dev [mailto:hotspot-compiler-dev-bounces at openjdk.java.net] On Behalf Of Lindenmaier, Goetz
Sent: Dienstag, 2. Januar 2018 11:04
To: Andrew Haley <aph at redhat.com>; hotspot-compiler-dev at openjdk.java.net
Subject: RE: RFR(S): 8193927: Optimize scanning code for oops.

Hi Andrew, 

so I'll leave it as is.  Thanks for checking.

Best regards,
  Goetz.

> -----Original Message-----
> From: Andrew Haley [mailto:aph at redhat.com]
> Sent: Dienstag, 2. Januar 2018 10:15
> To: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>; hotspot-compiler-
> dev at openjdk.java.net
> Subject: Re: RFR(S): 8193927: Optimize scanning code for oops.
> 
> On 21/12/17 08:30, Lindenmaier, Goetz wrote:
> 
> > Some platforms don't emit immediate oops to the code. If so, scans
> > of the code for oops can be skipped.
> >
> > Add flag ImmediateOopsEmitted to each platform specifying the behavior.
> > Only search code for immediate oops if this flag is set. Make sure no
> > oops are emitted to code if the flag is not set.
> >
> > @aarch people: should the flag set to 'false' on aarch/arm?  So far it
> > is true, which defaults to the old behavior.
> 
> Sure, we emit immediate oops in generated code.
> 
> --
> Andrew Haley
> Java Platform Lead Engineer
> Red Hat UK Ltd. <https://www.redhat.com>
> EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671