inlining with local offset calculation

Fidget-Spinner · Dec 10, 2024 · 7644afe · 7644afe
1 parent e79ad55
commit 7644afe
Show file tree

Hide file tree

Showing 14 changed files with 318 additions and 231 deletions.
diff --git a/Include/internal/pycore_ceval.h b/Include/internal/pycore_ceval.h
@@ -276,6 +276,7 @@ PyAPI_FUNC(PyObject *)_PyEval_MatchKeys(PyThreadState *tstate, PyObject *map, Py
 PyAPI_FUNC(void) _PyEval_MonitorRaise(PyThreadState *tstate, _PyInterpreterFrame *frame, _Py_CODEUNIT *instr);
 PyAPI_FUNC(int) _PyEval_UnpackIterableStackRef(PyThreadState *tstate, _PyStackRef v, int argcnt, int argcntafter, _PyStackRef *sp);
 PyAPI_FUNC(void) _PyEval_FrameClearAndPop(PyThreadState *tstate, _PyInterpreterFrame *frame);
+PyAPI_FUNC(void) _PyEval_InlinedFrameClear(_PyStackRef *start, int to_pop);
 PyAPI_FUNC(PyObject **) _PyObjectArray_FromStackRefArray(_PyStackRef *input, Py_ssize_t nargs, PyObject **scratch);
 
 PyAPI_FUNC(void) _PyObjectArray_Free(PyObject **array, PyObject **scratch);

diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
@@ -330,6 +330,9 @@ struct _Py_UOpsPEAbstractFrame {
     // Usually _INIT_CALL_PY_EXACT_ARGS
     // If it's not NULL that indicates it's virtual.
     _PyUOpInstruction *init_frame_inst;
+
+    // If inlined, where the localsplus starts relative to caller's localsplus.
+    int inline_localsplus_offset_from_caller;
 };
 
 typedef struct _Py_UOpsPEAbstractFrame _Py_UOpsPEAbstractFrame;

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
@@ -5065,6 +5065,22 @@ dummy_func(
             GOTO_UNWIND();
         }
 
+        tier2 op(_GROW_STACK, (unused/4, null_out_count/2 --)) {
+            for (int i = 0; i < null_out_count; i++) {
+                stack_pointer[i] = PyStackRef_NULL;
+            }
+            stack_pointer += null_out_count;
+        }
+
+        tier2 op(_SHRINK_STACK, (unused/4, shrink_count/2 --)) {
+            _PyStackRef tos = stack_pointer[-1];
+            stack_pointer--;
+            _PyEval_InlinedFrameClear(stack_pointer, shrink_count);
+            stack_pointer -= shrink_count;
+            stack_pointer++;
+            stack_pointer[-1] = tos;
+        }
+
         /* Progress is guaranteed if we DEOPT on the eval breaker, because
          * ENTER_EXECUTOR will not re-enter tier 2 with the eval breaker set. */
         tier2 op(_TIER2_RESUME_CHECK, (--)) {