René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Use tagged pointers on the stack in the default build.","articleBody":"Currently all references to objects in frameobjects use `_PyStackRef` instead of `PyObject *`.\r\nThis is necessary for the free-threaded build to support deferred references.\r\n\r\nFor the default build `_PyStackRef` is just an alias for `PyObject *`.\r\nWe should change `_PyStackRef` to use proper tagged pointers in the default build for two important reasons:\r\n\r\n* It will reduce the maintenance burden of using tagged pointers if they were the same in both builds\r\n* It offers a lot of optimization potential. The overhead of reference counting operations is large, and tagged pointers will allow us to reduce that overhead considerably.\r\n\r\nMy initial implementation is [0.8% slower](https://github.com/faster-cpython/benchmarking-public/tree/main/results/bm-20241206-3.14.0a2+-0c20416), although I'd like to get that closer to 0 before merging anything. There is some speedup in the GC due to streamlined immortality checks, and some slowdown due to increased overhead of turning new `PyObject *` references into `_PyStackRef`s.\r\n\r\nThis small slowdown will allow us a large speedup (maybe more than 5%) as we can do the following:\r\n* Reduce the overhead of refcount operations by using tagged references for the majority of `LOAD_` instructions in the interpreter.\r\n* Completely eliminate many decref operations by tracking which references are tagged in the JIT.\r\n\r\nThe tagging scheme:\r\n Tag | Meaning\r\n--- | --- \r\n00 | Normal pointers\r\n01 | Pointers with embedded reference count\r\n10 | Unused\r\n11 | Pointer to immortal object\u003csup\u003e1\u003c/sup\u003e (including NULL)\r\n\r\nThis tagging scheme is chosen as it provides the best performance for the most common operations:\r\n* PyStackRef_DUP: Can check to see if the object's reference count needs updating with a single check and no memory read: `ptr \u0026 1`\r\n* PyStackRef_CLOSE: As for PyStackRef_DUP, only a single bit check is needed\r\n* PyStackRef_XCLOSE: Since `NULL` is treated as immortal and tagged, this is the same as PyStackRef_CLOSE.\r\n\r\nMaintaining the invariant that tag `11` is used for all immortal objects is a bit expensive, but can be mitigated by pushing the conversion from `PyObject *` to `_PyStackRef` down to a point where it is known whether an object is newly created or not.\r\nFor newly created objects `PyStackRef_FromPyObjectStealMortal` can be used which performs no immortality check.\r\n\r\n --------------\r\n\r\n1. Actually, any object that was immortal when the reference was created. References to objects that are made immortal after the reference is created would have the low bits set to `00`, or `01`.   This is OK as immortal refcounts have a huge margin of error and the number of possible references to one of these immortal objects is very small.\n\n\u003c!-- gh-linked-prs --\u003e\n### Linked PRs\n* gh-127875\n* gh-128121\n* gh-130785\n* gh-131072\n* gh-131140\n* gh-131198\n* gh-131365\n* gh-131500\n* gh-131508\n* gh-136178\n* gh-136206\n\u003c!-- /gh-linked-prs --\u003e\n","author":{"url":"https://github.com/markshannon","@type":"Person","name":"markshannon"},"datePublished":"2024-12-06T18:31:10.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":14},"url":"https://github.com/127705/cpython/issues/127705"}