Remove redundancy in _PyJit_TryInitializeTracing

Author: markshannon

Include/internal/pycore_optimizer.h

@@ -87,7 +87,6 @@ typedef struct _PyJitTracerInitialState {
     PyFunctionObject *func; // Strong
     struct _PyExecutorObject *executor; // Strong
     _Py_CODEUNIT *start_instr;
-    _Py_CODEUNIT *close_loop_instr;
     _Py_CODEUNIT *jump_backward_instr;
 } _PyJitTracerInitialState;
 
@@ -470,8 +469,8 @@ PyAPI_FUNC(int) _PyJit_translate_single_bytecode_to_trace(PyThreadState *tstate,
 
 PyAPI_FUNC(int)
 _PyJit_TryInitializeTracing(PyThreadState *tstate, _PyInterpreterFrame *frame,
-    _Py_CODEUNIT *curr_instr, _Py_CODEUNIT *start_instr,
-    _Py_CODEUNIT *close_loop_instr, _PyStackRef *stack_pointer, int chain_depth, _PyExitData *exit,
+    _Py_CODEUNIT *curr_instr, _PyStackRef *stack_pointer,
+    int chain_depth, _PyExitData *exit,
     int oparg, _PyExecutorObject *current_executor);
 
 PyAPI_FUNC(void) _PyJit_FinalizeTracing(PyThreadState *tstate, int err);

Modules/_testinternalcapi/test_cases.c.h

@@ -3373,13 +3373,8 @@ dummy_func(
                 (this_instr->op.code == JUMP_BACKWARD_JIT || is_resume)) &&
                 next_instr->op.code != ENTER_EXECUTOR) {
                 /* Back up over EXTENDED_ARGs so executor is inserted at the correct place */
-                _Py_CODEUNIT *insert_exec_at = this_instr;
-                while (oparg > 255) {
-                    oparg >>= 8;
-                    insert_exec_at--;
-                }
-                int succ = _PyJit_TryInitializeTracing(tstate, frame, this_instr, insert_exec_at,
-                    is_resume ? insert_exec_at : next_instr, stack_pointer, 0, NULL, oparg, NULL);
+                int succ = _PyJit_TryInitializeTracing(
+                    tstate, frame, this_instr, stack_pointer, 0, NULL, oparg, NULL);
                 if (succ) {
                     ENTER_TRACING();
                 }
@@ -6066,7 +6061,7 @@ dummy_func(
                 // Note: it's safe to use target->op.arg here instead of the oparg given by EXTENDED_ARG.
                 // The invariant in the optimizer is the deopt target always points back to the first EXTENDED_ARG.
                 // So setting it to anything else is wrong.
-                int succ = _PyJit_TryInitializeTracing(tstate, frame, target, target, target, stack_pointer, chain_depth, exit, target->op.arg, previous_executor);
+                int succ = _PyJit_TryInitializeTracing(tstate, frame, target, stack_pointer, chain_depth, exit, target->op.arg, previous_executor);
                 exit->temperature = restart_backoff_counter(exit->temperature);
                 if (succ) {
                     GOTO_TIER_ONE_CONTINUE_TRACING(target);

Python/bytecodes.c

@@ -826,8 +826,7 @@ _PyJit_translate_single_bytecode_to_trace(
             _Py_FALLTHROUGH;
         case JUMP_BACKWARD_NO_INTERRUPT:
         {
-            if ((next_instr != tracer->initial_state.close_loop_instr) &&
-                (next_instr != tracer->initial_state.start_instr) &&
+            if ((next_instr != tracer->initial_state.start_instr) &&
                 uop_buffer_length(&tracer->code_buffer) > CODE_SIZE_NO_PROGRESS &&
                 // For side exits, we don't want to terminate them early.
                 tracer->initial_state.exit == NULL &&
@@ -840,8 +839,8 @@ _PyJit_translate_single_bytecode_to_trace(
                 OPT_STAT_INC(inner_loop);
                 ADD_TO_TRACE(_EXIT_TRACE, 0, 0, target);
                 uop_buffer_last(trace)->operand1 = true; // is_control_flow
-                DPRINTF(2, "JUMP_BACKWARD not to top ends trace %p %p %p\n", next_instr,
-                    tracer->initial_state.close_loop_instr, tracer->initial_state.start_instr);
+                DPRINTF(2, "JUMP_BACKWARD not to top ends trace %p %p\n", next_instr,
+                    tracer->initial_state.start_instr);
                 goto done;
             }
             break;
@@ -977,8 +976,7 @@ _PyJit_translate_single_bytecode_to_trace(
         }
     }
     // Loop back to the start
-    int is_first_instr = tracer->initial_state.close_loop_instr == next_instr ||
-        tracer->initial_state.start_instr == next_instr;
+    int is_first_instr = tracer->initial_state.start_instr == next_instr;
     if (is_first_instr && uop_buffer_length(trace) > CODE_SIZE_NO_PROGRESS) {
         if (needs_guard_ip) {
             ADD_TO_TRACE(_SET_IP, 0, (uintptr_t)next_instr, 0);
@@ -1002,7 +1000,7 @@ _PyJit_translate_single_bytecode_to_trace(
 Py_NO_INLINE int
 _PyJit_TryInitializeTracing(
     PyThreadState *tstate, _PyInterpreterFrame *frame, _Py_CODEUNIT *curr_instr,
-    _Py_CODEUNIT *start_instr, _Py_CODEUNIT *close_loop_instr, _PyStackRef *stack_pointer, int chain_depth,
+    _PyStackRef *stack_pointer, int chain_depth,
     _PyExitData *exit, int oparg, _PyExecutorObject *current_executor)
 {
     _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
@@ -1022,6 +1020,7 @@ _PyJit_TryInitializeTracing(
     if (oparg > 0xFFFF) {
         return 0;
     }
+    _Py_CODEUNIT *start_instr = curr_instr - (oparg > 0xFF);
     PyObject *func = PyStackRef_AsPyObjectBorrow(frame->f_funcobj);
     if (func == NULL || !PyFunction_Check(func)) {
         return 0;
@@ -1038,7 +1037,7 @@ _PyJit_TryInitializeTracing(
         PyUnicode_AsUTF8(code->co_qualname),
         PyUnicode_AsUTF8(code->co_filename),
         code->co_firstlineno,
-        2 * INSTR_IP(close_loop_instr, code),
+        2 * INSTR_IP(curr_instr, code),
         chain_depth);
 #endif
     /* Set up tracing buffer*/
@@ -1048,7 +1047,6 @@ _PyJit_TryInitializeTracing(
     ADD_TO_TRACE(_MAKE_WARM, 0, 0, 0);
 
     tracer->initial_state.start_instr = start_instr;
-    tracer->initial_state.close_loop_instr = close_loop_instr;
     tracer->initial_state.code = (PyCodeObject *)Py_NewRef(code);
     tracer->initial_state.func = (PyFunctionObject *)Py_NewRef(func);
     tracer->initial_state.executor = (_PyExecutorObject *)Py_XNewRef(current_executor);