|
| 1 | +# The Frame Stack |
| 2 | + |
| 3 | +Each call to a Python function has an activation record, |
| 4 | +commonly known as a "frame". |
| 5 | +Python semantics allows frames to outlive the activation, |
| 6 | +so they have (before 3.11) been allocated on the heap. |
| 7 | +This is expensive as it requires many allocations and |
| 8 | +results in poor locality of reference. |
| 9 | + |
| 10 | +In 3.11, rather than have these frames scattered about memory, |
| 11 | +as happens for heap-allocated objects, frames are allocated |
| 12 | +contiguously in a per-thread stack. |
| 13 | +This improves performance significantly for two reasons: |
| 14 | +* It reduces allocation overhead to a pointer comparison and increment. |
| 15 | +* Stack allocated data has the best possible locality and will always be in |
| 16 | + CPU cache. |
| 17 | + |
| 18 | +Generator and coroutines still need heap allocated activation records, but |
| 19 | +can be linked into the per-thread stack so as to not impact performance too much. |
| 20 | + |
| 21 | +## Layout |
| 22 | + |
| 23 | +Each activation record consists of four conceptual sections: |
| 24 | + |
| 25 | +* Local variables (including arguments, cells and free variables) |
| 26 | +* Evaluation stack |
| 27 | +* Specials: The per-frame object references needed by the VM: globals dict, |
| 28 | + code object, etc. |
| 29 | +* Linkage: Pointer to the previous activation record, stack depth, etc. |
| 30 | + |
| 31 | +### Layout |
| 32 | + |
| 33 | +The specials and linkage sections are a fixed size, so are grouped together. |
| 34 | + |
| 35 | +Each activation record is laid out as: |
| 36 | +* Specials and linkage |
| 37 | +* Locals |
| 38 | +* Stack |
| 39 | + |
| 40 | +This seems to provide the best performance without excessive complexity. |
| 41 | +It needs the interpreter to hold two pointers, a frame pointer and a stack pointer. |
| 42 | + |
| 43 | +#### Alternative layout |
| 44 | + |
| 45 | +An alternative layout that was used for part of 3.11 alpha was: |
| 46 | + |
| 47 | +* Locals |
| 48 | +* Specials and linkage |
| 49 | +* Stack |
| 50 | + |
| 51 | +This has the advantage that no copying is required when making a call, |
| 52 | +as the arguments on the stack are (usually) already in the correct |
| 53 | +location for the parameters. However, it requires the VM to maintain |
| 54 | +an extra pointer for the locals, which can hurt performance. |
| 55 | + |
| 56 | +A variant that only needs the need two pointers is to reverse the numbering |
| 57 | +of the locals, so that the last one is numbered `0`, and the first in memory |
| 58 | +is numbered `N-1`. |
| 59 | +This allows the locals, specials and linkage to accessed from the frame pointer. |
| 60 | +We may implement this in the future. |
| 61 | + |
| 62 | +#### Note: |
| 63 | + |
| 64 | +> In a contiguous stack, we would need to save one fewer registers, as the |
| 65 | +> top of the caller's activation record would be the same at the base of the |
| 66 | +> callee's. However, since some activation records are kept on the heap we |
| 67 | +> cannot do this. |
| 68 | +
|
| 69 | +### Generators and Coroutines |
| 70 | + |
| 71 | +Generators and coroutines contain a `_PyInterpreterFrame` |
| 72 | +The specials sections contains the following pointers: |
| 73 | + |
| 74 | +* Globals dict |
| 75 | +* Builtins dict |
| 76 | +* Locals dict (not the "fast" locals, but the locals for eval and class creation) |
| 77 | +* Code object |
| 78 | +* Heap allocated `PyFrameObject` for this activation record, if any. |
| 79 | +* The function. |
| 80 | + |
| 81 | +The pointer to the function is not strictly required, but it is cheaper to |
| 82 | +store a strong reference to the function and borrowed references to the globals |
| 83 | +and builtins, than strong references to both globals and builtins. |
| 84 | + |
| 85 | +### Frame objects |
| 86 | + |
| 87 | +When creating a backtrace or when calling `sys._getframe()` the frame becomes |
| 88 | +visible to Python code. When this happens a new `PyFrameObject` is created |
| 89 | +and a strong reference to it placed in the `frame_obj` field of the specials |
| 90 | +section. The `frame_obj` field is initially `NULL`. |
| 91 | + |
| 92 | +The `PyFrameObject` may outlive a stack-allocated `_PyInterpreterFrame`. |
| 93 | +If it does then `_PyInterpreterFrame` is copied into the `PyFrameObject`, |
| 94 | +except the evaluation stack which must be empty at this point. |
| 95 | +The linkage section is updated to reflect the new location of the frame. |
| 96 | + |
| 97 | +This mechanism provides the appearance of persistent, heap-allocated |
| 98 | +frames for each activation, but with low runtime overhead. |
| 99 | + |
| 100 | +### Generators and Coroutines |
| 101 | + |
| 102 | + |
| 103 | +Generator objects have a `_PyInterpreterFrame` embedded in them. |
| 104 | +This means that creating a generator requires only a single allocation, |
| 105 | +reducing allocation overhead and improving locality of reference. |
| 106 | +The embedded frame is linked into the per-thread frame when iterated or |
| 107 | +awaited. |
| 108 | + |
| 109 | +If a frame object associated with a generator outlives the generator, then |
| 110 | +the embedded `_PyInterpreterFrame` is copied into the frame object. |
| 111 | + |
| 112 | + |
| 113 | +All the above applies to coroutines and async generators as well. |
| 114 | + |
| 115 | +### Field names |
| 116 | + |
| 117 | +Many of the fields in `_PyInterpreterFrame` were copied from the 3.10 `PyFrameObject`. |
| 118 | +Thus, some of the field names may be a bit misleading. |
| 119 | + |
| 120 | +For example the `f_globals` field has a `f_` prefix implying it belongs to the |
| 121 | +`PyFrameObject` struct, although it belongs to the `_PyInterpreterFrame` struct. |
| 122 | +We may rationalize this naming scheme for 3.12. |
0 commit comments