[vector] Fix reduce_add2F/reduce_add2D for incorrect associativity

Qi Feng Qi.Feng at arm.com
Wed Mar 4 07:02:57 UTC 2020


Hi,

Please help me review this patch.

Webrev: http://cr.openjdk.java.net/~yzhang/vectorapi/vectorapi.fadd/webrev.00/

The implementation of these two instructs on AArch64 accumulate it's arguments in the wrong order, since the addition of floating point values doesn't obey associative law.

Thanks,
Qi



More information about the panama-dev mailing list