GH-113710: Tier 2 optimizer: check the function instead of checking globals. (GH-116410)

4 files changed, 27 insertions, 36 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 15794cb..ad4ea4e 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -4058,12 +4058,8 @@ dummy_func(
             null = NULL;
         }
 
-        tier2 op(_CHECK_GLOBALS, (dict/4 -- )) {
-            DEOPT_IF(GLOBALS() != dict);
-        }
-
-        tier2 op(_CHECK_BUILTINS, (dict/4 -- )) {
-            DEOPT_IF(BUILTINS() != dict);
+        tier2 op(_CHECK_FUNCTION, (func/4 -- )) {
+            DEOPT_IF(frame->f_funcobj != func);
         }
 
         /* Internal -- for testing executors */
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 806f748..a057466 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -3789,15 +3789,9 @@
             break;
         }
 
-        case _CHECK_GLOBALS: {
-            PyObject *dict = (PyObject *)CURRENT_OPERAND();
-            if (GLOBALS() != dict) goto deoptimize;
-            break;
-        }
-
-        case _CHECK_BUILTINS: {
-            PyObject *dict = (PyObject *)CURRENT_OPERAND();
-            if (BUILTINS() != dict) goto deoptimize;
+        case _CHECK_FUNCTION: {
+            PyObject *func = (PyObject *)CURRENT_OPERAND();
+            if (frame->f_funcobj != func) goto deoptimize;
             break;
         }
 
diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c
index 18e3382..0f25527 100644
--- a/Python/optimizer_analysis.c
+++ b/Python/optimizer_analysis.c
@@ -154,10 +154,10 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer,
 
     /* These values represent stacks of booleans (one bool per bit).
      * Pushing a frame shifts left, popping a frame shifts right. */
-    uint32_t builtins_checked = 0;
+    uint32_t function_checked = 0;
     uint32_t builtins_watched = 0;
-    uint32_t globals_checked = 0;
     uint32_t globals_watched = 0;
+    uint32_t prechecked_function_version = 0;
     if (interp->dict_state.watchers[GLOBALS_WATCHER_ID] == NULL) {
         interp->dict_state.watchers[GLOBALS_WATCHER_ID] = globals_watcher_callback;
     }
@@ -176,13 +176,13 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer,
                     PyDict_Watch(BUILTINS_WATCHER_ID, builtins);
                     builtins_watched |= 1;
                 }
-                if (builtins_checked & 1) {
+                if (function_checked & 1) {
                     buffer[pc].opcode = NOP;
                 }
                 else {
-                    buffer[pc].opcode = _CHECK_BUILTINS;
+                    buffer[pc].opcode = _CHECK_FUNCTION;
                     buffer[pc].operand = (uintptr_t)builtins;
-                    builtins_checked |= 1;
+                    function_checked |= 1;
                 }
                 break;
             case _GUARD_GLOBALS_VERSION:
@@ -198,36 +198,39 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer,
                     _Py_BloomFilter_Add(dependencies, globals);
                     globals_watched |= 1;
                 }
-                if (globals_checked & 1) {
+                if (function_checked & 1) {
                     buffer[pc].opcode = NOP;
                 }
                 else {
-                    buffer[pc].opcode = _CHECK_GLOBALS;
+                    buffer[pc].opcode = _CHECK_FUNCTION;
                     buffer[pc].operand = (uintptr_t)globals;
-                    globals_checked |= 1;
+                    function_checked |= 1;
                 }
                 break;
             case _LOAD_GLOBAL_BUILTINS:
-                if (globals_checked & builtins_checked & globals_watched & builtins_watched & 1) {
+                if (function_checked & globals_watched & builtins_watched & 1) {
                     convert_global_to_const(inst, builtins);
                 }
                 break;
             case _LOAD_GLOBAL_MODULE:
-                if (globals_checked & globals_watched & 1) {
+                if (function_checked & globals_watched & 1) {
                     convert_global_to_const(inst, globals);
                 }
                 break;
             case _PUSH_FRAME:
             {
-                globals_checked <<= 1;
-                globals_watched <<= 1;
-                builtins_checked <<= 1;
                 builtins_watched <<= 1;
+                globals_watched <<= 1;
+                function_checked <<= 1;
                 PyFunctionObject *func = (PyFunctionObject *)buffer[pc].operand;
                 if (func == NULL) {
                     return 1;
                 }
                 assert(PyFunction_Check(func));
+                if (prechecked_function_version == func->func_version) {
+                    function_checked |= 1;
+                }
+                prechecked_function_version = 0;
                 globals = func->func_globals;
                 builtins = func->func_builtins;
                 if (builtins != interp->builtins) {
@@ -237,16 +240,18 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer,
             }
             case _POP_FRAME:
             {
-                globals_checked >>= 1;
-                globals_watched >>= 1;
-                builtins_checked >>= 1;
                 builtins_watched >>= 1;
+                globals_watched >>= 1;
+                function_checked >>= 1;
                 PyFunctionObject *func = (PyFunctionObject *)buffer[pc].operand;
                 assert(PyFunction_Check(func));
                 globals = func->func_globals;
                 builtins = func->func_builtins;
                 break;
             }
+            case _CHECK_FUNCTION_EXACT_ARGS:
+                prechecked_function_version = (uint32_t)buffer[pc].operand;
+                break;
             default:
                 if (op_is_end(opcode)) {
                     return 1;
diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index 32dfca9..a114b03 100644
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -1951,11 +1951,7 @@
             break;
         }
 
-        case _CHECK_GLOBALS: {
-            break;
-        }
-
-        case _CHECK_BUILTINS: {
+        case _CHECK_FUNCTION: {
             break;
         }