RFR: 8255553: [PPC64] Introduce and use setbc and setnbc P10 instructions [v6]

[Martin Doerr]

On Wed, 4 Nov 2020 23:04:10 GMT, Ziviani <github.com+670087+jrziviani at openjdk.org> wrote:

>> - setbc RT,BI: sets RT to 1 if CR(BI) is 1, otherwise 0.
>> - setnbc RT,BI: sets RT to -1 if CR(BI) is 1, otherwise 0.
>> Ref: PowerISA 3.1, page 129.
>> 
>> These instructions are particularly interesting to improve the following
>> pattern `(src1<src2)? -1: ((src1>src2)? 1: 0)`, which can be found in
>> `instruct cmpL3_reg_reg_ExEx()@ppc.ad`, by removing its branches.
>> 
>> Long.toString, that generate such pattern in getChars, has showed a
>> good performance gain by using these new instructions.
>> 
>> Example:
>> for (int i = 0; i < 200_000; i++)
>>   res = Long.toString((long)i);
>> 
>> java -Xcomp -XX:CompileThreshold=1 -XX:-TieredCompilation TestToString
>> 
>> Without setbc (average): 0.1178 seconds
>> With setbc (average): 0.0396 seconds
>
> Ziviani has refreshed the contents of this pull request, and previous commits have been removed. The incremental views will show differences compared to the previous content of the PR.

Very nice! Unfortunately, I had forgotten one detail (see inline).
I'll try to find a 2nd reviewer in our team.

src/hotspot/cpu/ppc/ppc.ad line 11422:

> 11420: 
> 11421: // Manifest a CmpL3 result in an integer register.
> 11422: instruct cmpL3_reg_reg(iRegIdst dst, iRegLsrc src1, iRegLsrc src2) %{

I had forgotten one detail in my previous review. Sorry for that. We need to model the CR0 effect:
instruct cmpL3_reg_reg(iRegIdst dst, iRegLsrc src1, iRegLsrc src2, flagsRegCR0 cr0) %{
  match(Set dst (CmpL3 src1 src2));
  effect(KILL cr0);
(Same for other nodes.)

src/hotspot/cpu/ppc/ppc.ad line 11766:

> 11764:     __ fcmpu(CCR0, $src1$$FloatRegister, $src2$$FloatRegister);
> 11765:     __ cror(CCR0, Assembler::less, CCR0, Assembler::summary_overflow);
> 11766:     __ set_cmp3($dst$$Register);

Why not
set_cmpu3($dst$$Register, true); // C2 requires unordered to get treated like less
?
(same for CmpD3)

src/hotspot/cpu/ppc/templateTable_ppc_64.cpp line 1611:

> 1609: 
> 1610:   __ fcmpu(CCR0, Rfirst, Rsecond); // compare
> 1611:   // if unordered_result is 1, treat unordered_result like 'greater than'

Please add
assert(unordered_result == 1 || unordered_result == -1, "only supported");

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

Changes requested by mdoerr (Reviewer).

PR: https://git.openjdk.java.net/jdk/pull/907