sync sp of _DYNAMIC_EXIT

Author: Fidget-Spinner

Include/internal/pycore_uop_metadata.h

@@ -5307,7 +5307,9 @@ dummy_func(
     #endif
             // Disabled for now (gh-139109) as it slows down dynamic code tremendously.
             // Compile and jump to the cold dynamic executors in the future.
+            SYNC_SP();
             GOTO_TIER_ONE(frame->instr_ptr);
+            Py_UNREACHABLE();
         }
 
         tier2 op(_CHECK_VALIDITY, (--)) {
@@ -5470,14 +5472,19 @@ dummy_func(
                 assert(tstate->jit_exit == exit);
                 exit->executor = executor;
                 TIER2_TO_TIER2(exit->executor);
+                Py_UNREACHABLE();
             }
             else {
                 if (frame->owner >= FRAME_OWNED_BY_INTERPRETER) {
+                    SYNC_SP();
                     GOTO_TIER_ONE(target);
+                    Py_UNREACHABLE();
                 }
                 if (!backoff_counter_triggers(temperature)) {
                     exit->temperature = advance_backoff_counter(temperature);
+                    SYNC_SP();
                     GOTO_TIER_ONE(target);
+                    Py_UNREACHABLE();
                 }
                 _PyExecutorObject *previous_executor = _PyExecutor_FromExit(exit);
                 assert(tstate->current_executor == (PyObject *)previous_executor);

Python/bytecodes.c

@@ -560,7 +560,7 @@ _PyJIT_Compile(_PyExecutorObject *executor, const _PyUOpInstruction trace[], siz
     unsigned char *code = memory;
     state.trampolines.mem = memory + code_size;
     unsigned char *data = memory + code_size + state.trampolines.size + code_padding;
-    assert(trace[0].opcode == _START_EXECUTOR_r00 || trace[0].opcode == _COLD_EXIT_r00 ||  trace[0].opcode == _COLD_DYNAMIC_EXIT);
+    assert(trace[0].opcode == _START_EXECUTOR_r00 || trace[0].opcode == _COLD_EXIT_r00 ||  trace[0].opcode == _COLD_DYNAMIC_EXIT_r00);
     for (size_t i = 0; i < length; i++) {
         const _PyUOpInstruction *instruction = &trace[i];
         group = &stencil_groups[instruction->opcode];

Python/executor_cases.c.h

@@ -1655,14 +1655,14 @@ _PyExecutor_GetColdDynamicExecutor(void)
 {
     PyInterpreterState *interp = _PyInterpreterState_GET();
     if (interp->cold_dynamic_executor != NULL) {
-        assert(interp->cold_dynamic_executor->trace[0].opcode == _COLD_DYNAMIC_EXIT);
+        assert(interp->cold_dynamic_executor->trace[0].opcode == _COLD_DYNAMIC_EXIT_r00);
         return interp->cold_dynamic_executor;
     }
     _PyExecutorObject *cold = allocate_executor(0, 1);
     if (cold == NULL) {
         Py_FatalError("Cannot allocate core JIT code");
     }
-    ((_PyUOpInstruction *)cold->trace)->opcode = _COLD_DYNAMIC_EXIT;
+    ((_PyUOpInstruction *)cold->trace)->opcode = _COLD_DYNAMIC_EXIT_r00;
 #ifdef _Py_JIT
     cold->jit_code = NULL;
     cold->jit_size = 0;