RFR: 8365926: RISC-V: Performance regression in renaissance (chi-square) [v2]

[Robbin Ehn]

On Mon, 1 Sep 2025 10:10:31 GMT, Hamlin Li <mli at openjdk.org> wrote:

> ```
> JDK-23 (last version with trampoline calls)
> Mean: 3189.5827
> Standard Deviation: 284.6478
> 
> JDK-25
> Mean: 3424.8905
> Standard Deviation: 222.2208
> 
> Patch:
> Mean: 3144.8535
> Standard Deviation: 229.2577
> ```
> 
> For the performance data, do you have some data for applying this fix on top of the next commit after`JDK-23 (last version with trampoline calls)`? I think this data might be more helpful to understand the performance comparison between old trampoline, stub and this pr.

JDK-23 is last released version with trampoline calls.
JDK-24 is first released version with load calls.

What I can do is run ~jdk-24-prelease version which have both and backport to it... running that now...

> src/hotspot/cpu/riscv/nativeInst_riscv.cpp line 111:
> 
>> 109:   if (mt_safe) {
>> 110:     OrderAccess::release();
>> 111:     ICache::invalidate_range(jal_pc, NativeInstruction::instruction_size);
> 
> should `jal_pc` be `instruction_address()`?

We have:

auipc // instruction_address()                                                                                     # Never changed
ld       // instruction_address() + NativeInstruction::instruction_size                    # Never changed
jal(r)  // instruction_address() + 2 * NativeInstruction::instruction_size (jal_pc) # jal<->jalr

We only change the instruction at "instruction_address() + 2 * NativeInstruction::instruction_size".

Note that jal_pos and jal_pc means a "jump and link instruction", not specifically jal or jalr.

Make sense?

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

PR Comment: https://git.openjdk.org/jdk/pull/26944#issuecomment-3242627359
PR Review Comment: https://git.openjdk.org/jdk/pull/26944#discussion_r2314145607