RFR: 8228622: [lworld] Ineffective codegeneration for flattened arrays checks causes large performance regression on List iteration

Tobias Hartmann tobias.hartmann at oracle.com
Wed Sep 18 08:31:13 UTC 2019


Hi Roland,

great work!

All tests pass now and I've only found some minor things:
- Should TypeInstPtr::_flatten_array not be TypeInstPtr::_flat_array or _flattened_array?
- memnode.cpp:2360: Missing whitespace before {"
- parse2.cpp:85 and 277: Why are you not simply comparing 'flattened' to zero? Is it because it
might be int/long depending on compressed klass ptrs?
- phaseX.cpp:1695: Please add a comment describing why that code is needed.
- type.cpp:4127 'value' -> 'flatten_array'?
- TestLWorld.java: Shouldn't that test verify that an uncommon trap was added and that we don't have
a LOAD_UNKNOWN_VALUE match?
- ValueTypeTest.java already defines the compilation levels (see line 110)

No need to send a new webrev.

Thanks,
Tobias

On 17.09.19 13:14, Roland Westrelin wrote:
> 
> Hi Tobias,
> 
> Thanks for the testing results. Here is a new webrev.
> 
> http://cr.openjdk.java.net/~roland/8228622/webrev.02/
> 
> Roland.
> 


More information about the valhalla-dev mailing list