[lworld+vector] RFR: 8307715: Integrate VectorMask/Shuffle with value/primitive classes [v6]

Jatin Bhateja jbhateja at openjdk.org
Thu May 25 08:08:21 UTC 2023


On Thu, 25 May 2023 07:13:44 GMT, Xiaohong Gong <xgong at openjdk.org> wrote:

>> Make sense to me. But if  the constant oop is `gvn.zerocon(T_PRIMITIVE_OBJECT)`, it's more reasonable that accessing the multi-fields is invalid in runtime, right? I think compiler should exclude such cases before calling `InlineTypeNode::load`.  WDYT?
>
> Do you have a test when the constant oop's type is NULL_PTR? I can prepare a fixing and testing. Thanks!

I encountered this issue during my implementation when we attempt to create a VectorBoxNode while ciInlineKlass was not initialized.
https://github.com/openjdk/valhalla/blob/lworld%2Bvector/src/hotspot/share/opto/vectornode.hpp#L1692
You can make the change in above line to always create gvn.zerocon(T_PRIMITIVE_OBJECT) oop and remove special multifield handlings for constant case in InlineTypeNode::load, it will result into creation of imbalanced PhiNodes.

test Float256VectorTests.blendFloat256VectorTestsBroadcastLongSmokeTest(float[-i * 5], float[cornerCaseValue(i)], mask[false]): success

Error mixing types: vectory[8]:{float} and ftcon:0.000000

-------------

PR Review Comment: https://git.openjdk.org/valhalla/pull/845#discussion_r1205147334



More information about the valhalla-dev mailing list