gh-145742: Pre-tag operands for _LOAD_CONST_INLINE_BORROW variants

[corona10]

• Issue: Some (mostly) easy ways to reduce the size of JIT generated code #145742

[corona10]

@Fidget-Spinner @markshannon PTAL when you have times :)

[corona10]

I’ve just resolved all the conflicts. Gentle ping to @markshannon and @Fidget-Spinner.

[markshannon]

Thanks for doing this.

There is work underway to remove the many variants of _LOAD_CONST_INLINE_BORROW that also pop, swap or rotate.
Once that work is done, this PR should become much simpler.
I'd recommend leaving this for a week or two before attempting to resolve conflicts.

[corona10]

Got it

[corona10]

@markshannon I've resolved conflicts with the latest main and simplified the pre-tagging/untagging to be path-based (auto-tagging in add_op() when the opcode is a BORROW variant). PTAL.

[markshannon]

Have you tested this with Py_STACKREF_DEBUG set?

[markshannon]

Doesn't it need defining for Py_STACKREF_DEBUG as well?
Can you make this an inline function, for type safety?

[markshannon]

You need to whitelist PyStackRef_FromPreTagged as non-escaping

[corona10]

Have you tested this with Py_STACKREF_DEBUG set?

FYI, it's already broken see: #148718

[corona10]

@markshannon gentle ping?

[markshannon]

This looks fine, once updated to main.

The problem I have with it is the lying to the type system and tagged pointers.
The fix would be to differentiate between object pointers, other pointers, stackrefs and 64 bit integers in the DSL. But that's a lot of work for this small change.

So, can you either:

• Make a new issue make those changes to the DSL in the future, and then we can merge this, or
• Make the DSL changes in this PR

[markshannon]

The problem is that we always cast /4 values to PyObject * even when they aren't pointers.

[bedevere-app]

When you're done making the requested changes, leave the comment: I have made the requested changes; please review again.

[corona10]

Make the DSL changes in this PR

I prefer this one, since this PR is not urgent thing :)

[corona10]

@markshannon
I’ve added a ^ marker to annotate the pretagged cache effect.

I’d like to know whether this is the kind of DSL you had in mind, or if you would prefer a different grammar. I would be happy to follow your suggestion if you have one.

PTAL.

    op ( LOAD_CONST_INLINE_BORROW, (ptr/4^ -- value) ) {
        value = PyStackRef_FromPreTagged(ptr);
    }

[corona10]

FYI, the current Ubuntu CI failure is not related to this change; it's a Canonical infra issue that they are handling at this moment. https://x.com/ubuntu/status/2050112955132297652

[corona10]

DSL Summary

• The ^ marker on cache slots (e.g. ptr/4^) tells the DSL the operand is pre-tagged, so consumers declare uintptr_t + read_u64 instead of lying with PyObject *.
• Producers (add_op, convert_global_to_const) now call a generated _PyUop_PrepareOperand0 dispatch keyed on the same marker, so no more hardcoded _LOAD_CONST_INLINE_BORROW checks, future ^ ops are wired in by DSL alone.

[corona10]

Gentle ping @markshannon

[github-actions]

This PR is stale because it has been open for 90 days with no activity.