RFR: 8259822: [PPC64] Support the prefixed instruction format added in POWER10 [v2]
Kazunori Ogata
ogatak at openjdk.java.net
Tue Jan 26 19:38:59 UTC 2021
On Thu, 21 Jan 2021 01:38:17 GMT, Corey Ashford <github.com+51754783+CoreyAshford at openjdk.org> wrote:
>> Kazunori Ogata has updated the pull request incrementally with one additional commit since the last revision:
>>
>> Update based on review comments
>
> src/hotspot/cpu/ppc/assembler_ppc.inline.hpp line 157:
>
>> 155: }
>> 156:
>> 157: inline void Assembler::paddi_or_addi(Register d, Register a, long si34) {
>
> defined but not used.
It is intended to show an example of an efficient set of macros/functions for a prefixed instruction. Since a non-prefixed (32-bit) instruction is more efficient if the argument fits within the operand fields because of smaller I-cache usage and no need for the padding nop, it is convenient to have a macro/function that select better format. I intended to show an example of recommended set of useful macros/functions when adding support for a prefixed instruction. I'm planning to use this function in my next patch. Is it better to move this to the next patch?
> src/hotspot/cpu/ppc/assembler_ppc.inline.hpp line 209:
>
>> 207: inline void Assembler::psubi(Register d, Register a, long si34) { Assembler::paddi( d, a, -si34, false); }
>> 208:
>> 209: inline void Assembler::pli_or_li(Register d, long si34) {
>
> Defined but not used.
Same reason as the next comment (paddi_or_addi), but necessity of this one may be weaker.
-------------
PR: https://git.openjdk.java.net/jdk/pull/2095
More information about the hotspot-compiler-dev
mailing list