gh-115685: Type/values propagate for TO_BOOL in tier 2 #115686
Conversation
|
When you're done making the requested changes, leave the comment: And if you don't make the requested changes, you will be poked with soft cushions! |
|
Hey @Fidget-Spinner did you make the requested changes? You need to do the dance to make Bedevere happy. :-) |
|
I have made the requested changes; please review again. |
|
Odd. That test is still failing?! |
| assert(PyLong_CheckExact(get_const(right))); | ||
| PyObject *temp = _PyLong_Add((PyLongObject *)get_const(left), | ||
| (PyLongObject *)get_const(right)); |
There was a problem hiding this comment.
Maybe there should be s get_long_const() helper which returns (PyLongObject *)get_const(x) and also includes the assert(PyLong_CheckExact()) call, to make code like this more compact. (Same for float.)
Python/bytecodes.c
Outdated
| pure op (_POP_TOP_LOAD_CONST_INLINE_BORROW, (ptr/4, pop -- value)) { | ||
| TIER_TWO_ONLY; | ||
| DECREF_INPUTS(); | ||
| value = ptr; | ||
| } |
There was a problem hiding this comment.
Maybe there could be an even more specialized opcode that doesn't DECREF its input? That would be handy for the case where you introduced this -- the input is known to be None, so we don't really need to bother with the DECREF of the immortal value.
There was a problem hiding this comment.
The expectation is that the _POP_TOP part will be removed in a later pass.
LOAD_CONST ; _POP_TOP_LOAD_CONST_INLINE_BORROW -> _LOAD_CONST_INLINE_BORROW
Python/optimizer_analysis.c
Outdated
| { | ||
| sym_set_type(sym, Py_TYPE(const_val)); | ||
| sym_set_flag(sym, TRUE_CONST | KNOWN | NOT_NULL); | ||
| Py_XSETREF(sym->const_val, Py_NewRef(const_val)); |
There was a problem hiding this comment.
This is temporary, just ignore this for now. I will remove it in #115817.
| @@ -220,7 +220,7 @@ dummy_func(void) { | |||
| } | |||
|
|
|||
| op(_TO_BOOL_NONE, (value -- res)) { | |||
| if (sym_get_const(value) == Py_None) { | |||
| if (sym_is_const(value) && (sym_get_const(value) == Py_None)) { | |||
There was a problem hiding this comment.
Why is this necessary? Shouldn't sym_get_const return NULL for anything that isn't a constant?
It does seem rather verbose to have to prefix every sys_get_const with a sys_is_const.
There was a problem hiding this comment.
In the code:
static inline PyObject *
sym_get_const(_Py_UOpsSymType *sym)
{
assert(sym_is_const(sym));
assert(sym->const_val);
return sym->const_val;
}So we assume there is always something there when there's sym_get_const. I think I can just change the contract to return NULL on non constants though.
There was a problem hiding this comment.
The new sym_get_const() indeed returns NULL in this case.
|
@Fidget-Spinner: It looks like this PR is totally out of date. It adds a new ERROR_IF() macro that should probably be done in a separate cleanup pass. I can do that next (once gh-116062 is merged). The key part of this PR is to add optimizer specializations for |
|
Since you've addressed Mark's question I don't see a reason to wait for his explicit approval. |
TO_BOOL#115685