Add POP_DEAD_INPUTS

Author: mpage

Lib/test/test_generated_cases.py

@@ -1639,6 +1639,80 @@ def test_escaping_call_next_to_cmacro(self):
         """
         self.run_cases_test(input, output)
 
+    def test_pop_dead_inputs_all_live(self):
+        input = """
+        inst(OP, (a, b --)) {
+            POP_DEAD_INPUTS();
+            HAM(a, b);
+            INPUTS_DEAD();
+        }
+        """
+        output = """
+        TARGET(OP) {
+            frame->instr_ptr = next_instr;
+            next_instr += 1;
+            INSTRUCTION_STATS(OP);
+            _PyStackRef a;
+            _PyStackRef b;
+            b = stack_pointer[-1];
+            a = stack_pointer[-2];
+            HAM(a, b);
+            stack_pointer += -2;
+            assert(WITHIN_STACK_BOUNDS());
+            DISPATCH();
+        }
+        """
+        self.run_cases_test(input, output)
+
+    def test_pop_dead_inputs_some_live(self):
+        input = """
+        inst(OP, (a, b, c --)) {
+            POP_DEAD_INPUTS();
+            HAM(a);
+            INPUTS_DEAD();
+        }
+        """
+        output = """
+        TARGET(OP) {
+            frame->instr_ptr = next_instr;
+            next_instr += 1;
+            INSTRUCTION_STATS(OP);
+            _PyStackRef a;
+            a = stack_pointer[-3];
+            stack_pointer += -2;
+            assert(WITHIN_STACK_BOUNDS());
+            HAM(a);
+            stack_pointer += -1;
+            assert(WITHIN_STACK_BOUNDS());
+            DISPATCH();
+        }
+        """
+        self.run_cases_test(input, output)
+
+    def test_pop_dead_inputs_with_output(self):
+        input = """
+        inst(OP, (a, b -- c)) {
+            POP_DEAD_INPUTS();
+            c = SPAM();
+        }
+        """
+        output = """
+        TARGET(OP) {
+            frame->instr_ptr = next_instr;
+            next_instr += 1;
+            INSTRUCTION_STATS(OP);
+            _PyStackRef c;
+            stack_pointer += -2;
+            assert(WITHIN_STACK_BOUNDS());
+            c = SPAM();
+            stack_pointer[0] = c;
+            stack_pointer += 1;
+            assert(WITHIN_STACK_BOUNDS());
+            DISPATCH();
+        }
+        """
+        self.run_cases_test(input, output)
+
 
 class TestGeneratedAbstractCases(unittest.TestCase):
     def setUp(self) -> None:

Python/bytecodes.c

@@ -2174,8 +2174,7 @@ dummy_func(
             PyObject *attr_o = FT_ATOMIC_LOAD_PTR_RELAXED(ep->me_value);
             DEAD(mod_keys);
             // Clear mod_keys from stack in case we need to deopt
-            SAVE_STACK();
-            RELOAD_STACK();
+            POP_DEAD_INPUTS();
             DEOPT_IF(attr_o == NULL);
             #ifdef Py_GIL_DISABLED
             int increfed = _Py_TryIncrefCompareStackRef(&ep->me_value, attr_o, &attr);

Python/executor_cases.c.h

@@ -120,6 +120,7 @@ def __init__(self, out: CWriter):
             "PyStackRef_AsPyObjectSteal": self.stackref_steal,
             "DISPATCH": self.dispatch,
             "INSTRUCTION_SIZE": self.instruction_size,
+            "POP_DEAD_INPUTS": self.pop_dead_inputs,
         }
         self.out = out
 
@@ -348,6 +349,19 @@ def save_stack(
         self.emit_save(storage)
         return True
 
+    def pop_dead_inputs(
+        self,
+        tkn: Token,
+        tkn_iter: TokenIterator,
+        uop: Uop,
+        storage: Storage,
+        inst: Instruction | None,
+    ) -> None:
+        next(tkn_iter)
+        next(tkn_iter)
+        next(tkn_iter)
+        storage.pop_dead_inputs(self.out)
+
     def emit_reload(self, storage: Storage) -> None:
         storage.reload(self.out)
         self._print_storage(storage)

Python/generated_cases.c.h

@@ -512,6 +512,10 @@ def flush(self, out: CWriter, cast_type: str = "uintptr_t", extract_bits: bool =
         self._push_defined_outputs()
         self.stack.flush(out, cast_type, extract_bits)
 
+    def pop_dead_inputs(self, out: CWriter, cast_type: str = "uintptr_t", extract_bits: bool = True) -> None:
+        self.clear_dead_inputs()
+        self.stack.flush(out, cast_type, extract_bits)
+
     def save(self, out: CWriter) -> None:
         assert self.spilled >= 0
         if self.spilled == 0: