RFR: 8332689: RISC-V: Use load instead of trampolines [v4]

[Fei Yang]

On Tue, 4 Jun 2024 16:34:34 GMT, Robbin Ehn <rehn at openjdk.org> wrote:

>> Hi all, please consider!
>> 
>> Today we do JAL to **dest** if **dest** is in reach (+/- 1 MB).
>> Using a very small application or running very short time we have fast patchable calls.
>> But any normal application running longer will increase the code size and code chrun/fragmentation.
>> So whatever or not you get hot fast calls rely on luck.
>> 
>> To be patchable and get code cache reach we also emit a stub trampoline which we can point the JAL to.
>> This would be the common case for a patchable call.
>> 
>> Code stream:
>> JAL <trampo>
>> Stubs:
>> AUIPC
>> LD
>> JALR
>> <DEST>
>> 
>> 
>> On some CPUs L1D and L1I can't contain the same cache line, which means the tramopline stub can bounce from L1I->L1D->L1I, which is expensive.
>> Even if you don't have that problem having a call to a jump is not the fastest way.
>> Loading the address avoids the pitsfalls of cmodx.
>> 
>> This patch suggest to solve the problems with trampolines, we take small penalty in the naive case of JAL to **dest**,
>> and instead do by default:
>> 
>> Code stream:
>> AUIPC
>> LD
>> JALR
>> Stubs:
>> <DEST>
>> 
>> An experimental option for turning trampolines back on exists.
>> 
>> It should be possible to enhanced this with the WIP [Zjid](https://github.com/riscv/riscv-j-extension) by changing the JALR to JAL and nop out the auipc+ld (as the current proposal of Zjid forces the I-fetcher to fetch instruction in order (meaning we will avoid a lot issues which arm has)) when in reach and vice-versa.
>> 
>> Numbers from VF2 (I have done them a few times, they are always overall in favor of this patch):
>> 
>> fop                                        (msec)    2239       |  2128       =  0.950424
>> h2                                         (msec)    18660      |  16594      =  0.889282
>> jython                                     (msec)    22022      |  21925      =  0.995595
>> luindex                                    (msec)    2866       |  2842       =  0.991626
>> lusearch                                   (msec)    4108       |  4311       =  1.04942
>> lusearch-fix                               (msec)    4406       |  4116       =  0.934181
>> pmd                                        (msec)    5976       |  5897       =  0.98678
>> jython                                     (msec)    22022      |  21925      =  0.995595
>> Avg:                                       0.974112                              
>> fop(xcomp)                                 (msec)    2721       |  2714       =  0.997427
>> h2(xcomp) ...
>
> Robbin Ehn has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Move shart/far code to cpp

Hi, I only have several minor comments for now. Still looking at the code.

src/hotspot/cpu/riscv/nativeInst_riscv.cpp line 330:

> 328: 
> 329:   enum RISCV_specific_constants {
> 330:     return_address_offset       =    3 * NativeInstruction::instruction_size, // ld auipc jalr

Suggestion about the code comment: `// auipc + ld + jalr`

src/hotspot/cpu/riscv/nativeInst_riscv.cpp line 636:

> 634:     call = (NativeCall*)(return_address - NativeFarCall::return_address_offset);
> 635:   } else {
> 636:     call = (NativeCall*)(return_address - NativeShortCall::instruction_size);

Maybe it's better to have a `return_address_offset` (which equals `NativeShortCall::instruction_size`) const for `NativeShortCall` too?

src/hotspot/cpu/riscv/nativeInst_riscv.hpp line 118:

> 116:  public:
> 117:   enum {
> 118:     instruction_size = 3 * Assembler::instruction_size,

It looks odd for this `instruction_size` `NativeCall` to have a size of 12 here. It should depends on whether it is a `NativeShortCall` for `NativeFarCall` at the bottom, right?

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

PR Review: https://git.openjdk.org/jdk/pull/19453#pullrequestreview-2098062531
PR Review Comment: https://git.openjdk.org/jdk/pull/19453#discussion_r1626976092
PR Review Comment: https://git.openjdk.org/jdk/pull/19453#discussion_r1627098898
PR Review Comment: https://git.openjdk.org/jdk/pull/19453#discussion_r1627095666