Feature or enhancement

Proposal:

From the discourse discussion:

Integers are ubiquitous, both in Python code and the virtual machine.
However our representation of integers is somewhat clunky and inefficient.

There is an old technique for handling ints efficiently in dynamic language VMs, called tagged pointers.
A normal object pointer always has its low bits set to zero, as objects are 8 or 16 byte aligned in memory.
We can use those low bits as “tags” to denote the meaning of the high bits.
Historically, the low bit has been set to zero to indicate that the high bits are a pointer, and to 1 to indicate that the high bits are an integer.

We already use another tagging scheme within the interpreter and in the frame stack, the tag indicating whether the reference is borrowed.

I expect to start small with this, initially just fixing #132508 then improving the performance of iteration over lists and tuples.

Has this already been discussed elsewhere?

I have already discussed this feature proposal on Discourse

Links to previous discussion of this feature:

https://discuss.python.org/t/using-tagged-pointers-to-support-efficient-integer-operations/87950

Linked PRs

• gh-132509: Speed up bytes iteration in FOR_ITER_VIRTUAL #157353