Structure of the HotSpot Interpreter

[Andrew Haley]

On 5/30/24 14:47, Julian Waters wrote:
> I've recently been trying to learn more about HotSpot and studying its
> internals, but the structure of the Interpreter seems to elude me
> still. I'm aware that HotSpot doesn't use a traditional switch case
> (Well, at least not usually, looking at you Zero Port), but how it
> functions is more or less still a black box to me. What kind of
> dispatch mechanism does it use, for instance? Is it Direct Threaded,
> Indirect Threaded, Token Threaded, or something else entirely? Is
> there somewhere I can learn about how everything connects together?
> I've tried reading the HotSpot documentation online but there doesn't
> seem to be an in-depth explanation in them for how it all fits
> together, I'd greatly appreciate if someone points me in the right
> direction

It's a bytecode interpreter, so token threaded.

Print it out with -XX:+PrintInterpreter. Here's lmul (for Arm).

lmul  105 lmul  [0x0000ffff785428c0, 0x0000ffff785428e0]  32 bytes

--------------------------------------------------------------------------------
   0x0000ffff785428c0:   ldr		x0, [x20], #0x10    ;;@FILE: /home/aph/theRealAph-jdk/src/hotspot/share/interpreter/templateInterpreterGenerator.cpp
                                                             ;;  359:     case ltos: vep = __ pc(); __ pop(ltos); lep = __ pc(); generate_and_dispatch(t); break;
   0x0000ffff785428c4:   ldr		x1, [x20], #0x10    ;;  359:     case ltos: vep = __ pc(); __ pop(ltos); lep = __ pc(); generate_and_dispatch(t); break;
                                                             ;;  378:   __ verify_FPU(1, t->tos_in());
                                                             ;;  391:     __ dispatch_prolog(tos_out, step);
   0x0000ffff785428c8:   mul		x0, x0, x1

Fetch the next bytecode. x22 is the bytecode pointer:

   0x0000ffff785428cc:   ldrb		w8, [x22, #1]!      ;;  403:     __ dispatch_epilog(tos_out, step);

Offset to the dispatch table:

   0x0000ffff785428d0:   add		w9, w8, #0x500

Load the address of the next action, and jump to it:

   0x0000ffff785428d4:   ldr		x9, [x21, w9, uxtw #3]
   0x0000ffff785428d8:   br		x9

The interesting source is in the various templateInterpreter and templateTable files.

-- 
Andrew Haley  (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671