RFR(M): 8213528: fp registers should not need to be saved around a CallLeafNoFP
vladimir.kozlov at oracle.com
Wed Nov 14 17:16:12 UTC 2018
On 11/14/18 12:29 AM, Roland Westrelin wrote:
>> Save/restore XMM registers in a wrapper stub is perfectly fine with me. May be we should name such calls differently -
>> NO_FP is misleading.
> Rather than a different node type, what about a field in CallLeafNode
> (bool _preserve_fp_registers;)? I would then leave both CallLeafNode and
> CallLeafNoFpNode alone for now and the patch would simply test for
> _preserve_fp_registers in lcm.cpp?
Sounds good. Is lcm.cpp only place where we do such check (in addition to code in .ad files)?
More information about the hotspot-compiler-dev