[VectorAPI] Enhancement of floating-point math vector API implementation

[Maurizio Cimadamore]

On 09/01/2023 07:11, Xiaohong Gong wrote:
> If Foreign Linker API supports the vector calling convention and used by Vector API for such math operations, the corresponding vector intrinsics are not needed anymore. Does this mean the difference result issue from interpreter/c1 and c2 compiler will do not exist? And the fixing is not needed anymore? Thanks!

The Foreign Linker API does not support, at the time of writing, vector 
types in calling conventions. This can be added by:

* adding a memory layout for vectors (probably a new subclass of 
GroupLayout, with width, lane info etc.)
* picking a suitable Java carrier for vector objects passed to and fro 
the downcall method handles generated by the API (it is tempting to just 
use Vector here, but I'm not sure if that would be the best choice 
overall - perhaps memory segment, of suitable size? Or a java array?)
* implement the calling conventions accordingly - this involves some 
Java changes to the appropriate CallArranger, which determines the 
"bindings" for shuffling a Java call into a native call and back. I 
believe that once the correct bindings are in place (e.g. according to 
the SysV spec for Linux x64), the VM support would just follow the 
recipe in the binding and perform the correct shuffle (so that no JVM 
changes should be required - Jorn can correct me if I'm wrong in my 
assumption here)

Maurizio