GH-113464: Remove the extra jump via `_SIDE_EXIT` in `_EXIT_TRACE` (GH-118545)

6 files changed, 31 insertions, 53 deletions

diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h
index aa39404..988464b 100644
--- a/Include/internal/pycore_uop_ids.h
+++ b/Include/internal/pycore_uop_ids.h
@@ -240,33 +240,32 @@ extern "C" {
 #define _SET_ADD SET_ADD
 #define _SET_FUNCTION_ATTRIBUTE SET_FUNCTION_ATTRIBUTE
 #define _SET_UPDATE SET_UPDATE
-#define _SIDE_EXIT 429
-#define _START_EXECUTOR 430
-#define _STORE_ATTR 431
-#define _STORE_ATTR_INSTANCE_VALUE 432
-#define _STORE_ATTR_SLOT 433
+#define _START_EXECUTOR 429
+#define _STORE_ATTR 430
+#define _STORE_ATTR_INSTANCE_VALUE 431
+#define _STORE_ATTR_SLOT 432
 #define _STORE_ATTR_WITH_HINT STORE_ATTR_WITH_HINT
 #define _STORE_DEREF STORE_DEREF
-#define _STORE_FAST 434
-#define _STORE_FAST_0 435
-#define _STORE_FAST_1 436
-#define _STORE_FAST_2 437
-#define _STORE_FAST_3 438
-#define _STORE_FAST_4 439
-#define _STORE_FAST_5 440
-#define _STORE_FAST_6 441
-#define _STORE_FAST_7 442
+#define _STORE_FAST 433
+#define _STORE_FAST_0 434
+#define _STORE_FAST_1 435
+#define _STORE_FAST_2 436
+#define _STORE_FAST_3 437
+#define _STORE_FAST_4 438
+#define _STORE_FAST_5 439
+#define _STORE_FAST_6 440
+#define _STORE_FAST_7 441
 #define _STORE_FAST_LOAD_FAST STORE_FAST_LOAD_FAST
 #define _STORE_FAST_STORE_FAST STORE_FAST_STORE_FAST
 #define _STORE_GLOBAL STORE_GLOBAL
 #define _STORE_NAME STORE_NAME
 #define _STORE_SLICE STORE_SLICE
-#define _STORE_SUBSCR 443
+#define _STORE_SUBSCR 442
 #define _STORE_SUBSCR_DICT STORE_SUBSCR_DICT
 #define _STORE_SUBSCR_LIST_INT STORE_SUBSCR_LIST_INT
 #define _SWAP SWAP
-#define _TIER2_RESUME_CHECK 444
-#define _TO_BOOL 445
+#define _TIER2_RESUME_CHECK 443
+#define _TO_BOOL 444
 #define _TO_BOOL_BOOL TO_BOOL_BOOL
 #define _TO_BOOL_INT TO_BOOL_INT
 #define _TO_BOOL_LIST TO_BOOL_LIST
@@ -276,13 +275,13 @@ extern "C" {
 #define _UNARY_NEGATIVE UNARY_NEGATIVE
 #define _UNARY_NOT UNARY_NOT
 #define _UNPACK_EX UNPACK_EX
-#define _UNPACK_SEQUENCE 446
+#define _UNPACK_SEQUENCE 445
 #define _UNPACK_SEQUENCE_LIST UNPACK_SEQUENCE_LIST
 #define _UNPACK_SEQUENCE_TUPLE UNPACK_SEQUENCE_TUPLE
 #define _UNPACK_SEQUENCE_TWO_TUPLE UNPACK_SEQUENCE_TWO_TUPLE
 #define _WITH_EXCEPT_START WITH_EXCEPT_START
 #define _YIELD_VALUE YIELD_VALUE
-#define MAX_UOP_ID 446
+#define MAX_UOP_ID 445
 
 #ifdef __cplusplus
 }
diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
index d086002..475924d 100644
--- a/Include/internal/pycore_uop_metadata.h
+++ b/Include/internal/pycore_uop_metadata.h
@@ -237,7 +237,7 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_SET_IP] = 0,
     [_CHECK_STACK_SPACE_OPERAND] = HAS_DEOPT_FLAG,
     [_SAVE_RETURN_OFFSET] = HAS_ARG_FLAG,
-    [_EXIT_TRACE] = HAS_EXIT_FLAG,
+    [_EXIT_TRACE] = 0,
     [_CHECK_VALIDITY] = HAS_DEOPT_FLAG,
     [_LOAD_CONST_INLINE] = HAS_PURE_FLAG,
     [_LOAD_CONST_INLINE_BORROW] = HAS_PURE_FLAG,
@@ -252,7 +252,6 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_FATAL_ERROR] = 0,
     [_CHECK_VALIDITY_AND_SET_IP] = HAS_DEOPT_FLAG,
     [_DEOPT] = 0,
-    [_SIDE_EXIT] = 0,
     [_ERROR_POP_N] = HAS_ARG_FLAG,
     [_TIER2_RESUME_CHECK] = HAS_DEOPT_FLAG,
 };
@@ -459,7 +458,6 @@ const char *const _PyOpcode_uop_name[MAX_UOP_ID+1] = {
     [_SET_FUNCTION_ATTRIBUTE] = "_SET_FUNCTION_ATTRIBUTE",
     [_SET_IP] = "_SET_IP",
     [_SET_UPDATE] = "_SET_UPDATE",
-    [_SIDE_EXIT] = "_SIDE_EXIT",
     [_START_EXECUTOR] = "_START_EXECUTOR",
     [_STORE_ATTR] = "_STORE_ATTR",
     [_STORE_ATTR_INSTANCE_VALUE] = "_STORE_ATTR_INSTANCE_VALUE",
@@ -970,8 +968,6 @@ int _PyUop_num_popped(int opcode, int oparg)
             return 0;
         case _DEOPT:
             return 0;
-        case _SIDE_EXIT:
-            return 0;
         case _ERROR_POP_N:
             return oparg;
         case _TIER2_RESUME_CHECK:
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index e8383ed..ddada96 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -4133,7 +4133,7 @@ dummy_func(
         }
 
         tier2 op(_EXIT_TRACE, (--)) {
-            EXIT_IF(1);
+            EXIT_TO_TRACE();
         }
 
         tier2 op(_CHECK_VALIDITY, (--)) {
@@ -4266,10 +4266,6 @@ dummy_func(
             EXIT_TO_TIER1();
         }
 
-        tier2 op(_SIDE_EXIT, (--)) {
-            EXIT_TO_TRACE();
-        }
-
         tier2 op(_ERROR_POP_N, (target/2, unused[oparg] --)) {
             frame->instr_ptr = ((_Py_CODEUNIT *)_PyFrame_GetCode(frame)->co_code_adaptive) + target;
             SYNC_SP();
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index c3ee6e9..d0b794c 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -4127,10 +4127,7 @@
         }
 
         case _EXIT_TRACE: {
-            if (1) {
-                UOP_STAT_INC(uopcode, miss);
-                JUMP_TO_JUMP_TARGET();
-            }
+            EXIT_TO_TRACE();
             break;
         }
 
@@ -4319,11 +4316,6 @@
             break;
         }
 
-        case _SIDE_EXIT: {
-            EXIT_TO_TRACE();
-            break;
-        }
-
         case _ERROR_POP_N: {
             oparg = CURRENT_OPARG();
             uint32_t target = (uint32_t)CURRENT_OPERAND();
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 56768ae..c0e1be9 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -976,7 +976,7 @@ count_exits(_PyUOpInstruction *buffer, int length)
     int exit_count = 0;
     for (int i = 0; i < length; i++) {
         int opcode = buffer[i].opcode;
-        if (opcode == _SIDE_EXIT || opcode == _DYNAMIC_EXIT) {
+        if (opcode == _EXIT_TRACE || opcode == _DYNAMIC_EXIT) {
             exit_count++;
         }
     }
@@ -1021,7 +1021,7 @@ prepare_for_execution(_PyUOpInstruction *buffer, int length)
         int32_t target = (int32_t)uop_get_target(inst);
         if (_PyUop_Flags[opcode] & (HAS_EXIT_FLAG | HAS_DEOPT_FLAG)) {
             uint16_t exit_op = (_PyUop_Flags[opcode] & HAS_EXIT_FLAG) ?
-                _SIDE_EXIT : _DEOPT;
+                _EXIT_TRACE : _DEOPT;
             int32_t jump_target = target;
             if (is_for_iter_test[opcode]) {
                 /* Target the POP_TOP immediately after the END_FOR,
@@ -1112,7 +1112,7 @@ sanity_check(_PyExecutorObject *executor)
                 CHECK(target_unused(opcode));
                 break;
             case UOP_FORMAT_EXIT:
-                CHECK(opcode == _SIDE_EXIT);
+                CHECK(opcode == _EXIT_TRACE);
                 CHECK(inst->exit_index < executor->exit_count);
                 break;
             case UOP_FORMAT_JUMP:
@@ -1138,9 +1138,9 @@ sanity_check(_PyExecutorObject *executor)
         uint16_t opcode = inst->opcode;
         CHECK(
             opcode == _DEOPT ||
-            opcode == _SIDE_EXIT ||
+            opcode == _EXIT_TRACE ||
             opcode == _ERROR_POP_N);
-        if (opcode == _SIDE_EXIT) {
+        if (opcode == _EXIT_TRACE) {
             CHECK(inst->format == UOP_FORMAT_EXIT);
         }
     }
@@ -1178,7 +1178,7 @@ make_executor_from_uops(_PyUOpInstruction *buffer, int length, const _PyBloomFil
         dest--;
         *dest = buffer[i];
         assert(opcode != _POP_JUMP_IF_FALSE && opcode != _POP_JUMP_IF_TRUE);
-        if (opcode == _SIDE_EXIT) {
+        if (opcode == _EXIT_TRACE) {
             executor->exits[next_exit].target = buffer[i].target;
             dest->exit_index = next_exit;
             dest->format = UOP_FORMAT_EXIT;
@@ -1398,14 +1398,13 @@ counter_optimize(
         return 0;
     }
     _Py_CODEUNIT *target = instr + 1 + _PyOpcode_Caches[JUMP_BACKWARD] - oparg;
-    _PyUOpInstruction buffer[5] = {
-        { .opcode = _START_EXECUTOR, .jump_target = 4, .format=UOP_FORMAT_JUMP },
+    _PyUOpInstruction buffer[4] = {
+        { .opcode = _START_EXECUTOR, .jump_target = 3, .format=UOP_FORMAT_JUMP },
         { .opcode = _LOAD_CONST_INLINE_BORROW, .operand = (uintptr_t)self },
         { .opcode = _INTERNAL_INCREMENT_OPT_COUNTER },
-        { .opcode = _EXIT_TRACE, .jump_target = 4, .format=UOP_FORMAT_JUMP },
-        { .opcode = _SIDE_EXIT, .target = (uint32_t)(target - _PyCode_CODE(code)), .format=UOP_FORMAT_TARGET }
+        { .opcode = _EXIT_TRACE, .target = (uint32_t)(target - _PyCode_CODE(code)), .format=UOP_FORMAT_TARGET }
     };
-    _PyExecutorObject *executor = make_executor_from_uops(buffer, 5, &EMPTY_FILTER);
+    _PyExecutorObject *executor = make_executor_from_uops(buffer, 4, &EMPTY_FILTER);
     if (executor == NULL) {
         return -1;
     }
diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h
index e680d76..b602d66 100644
--- a/Python/optimizer_cases.c.h
+++ b/Python/optimizer_cases.c.h
@@ -2140,10 +2140,6 @@
             break;
         }
 
-        case _SIDE_EXIT: {
-            break;
-        }
-
         case _ERROR_POP_N: {
             stack_pointer += -oparg;
             break;