gh-106529: Make FOR_ITER a viable uop (#112134)

This uses the new mechanism whereby certain uops
are replaced by others during translation,
using the `_PyUop_Replacements` table.
We further special-case the `_FOR_ITER_TIER_TWO` uop
to update the deoptimization target to point
just past the corresponding `END_FOR` opcode.

Two tiny code cleanups are also part of this PR.

5 files changed, 60 insertions, 4 deletions

diff --git a/Python/abstract_interp_cases.c.h b/Python/abstract_interp_cases.c.h
index a2f6aa8..0d7fbe8 100644
--- a/Python/abstract_interp_cases.c.h
+++ b/Python/abstract_interp_cases.c.h
@@ -624,6 +624,12 @@
             break;
         }
 
+        case _FOR_ITER_TIER_TWO: {
+            STACK_GROW(1);
+            PARTITIONNODE_OVERWRITE((_Py_PARTITIONNODE_t *)PARTITIONNODE_NULLROOT, PEEK(-(-1)), true);
+            break;
+        }
+
         case _ITER_CHECK_LIST: {
             break;
         }
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index da98630..a1ca66e 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2369,7 +2369,7 @@ dummy_func(
             goto enter_tier_one;
         }
 
-       replaced op(_POP_JUMP_IF_FALSE, (unused/1, cond -- )) {
+        replaced op(_POP_JUMP_IF_FALSE, (unused/1, cond -- )) {
             assert(PyBool_Check(cond));
             int flag = Py_IsFalse(cond);
             #if ENABLE_SPECIALIZATION
@@ -2513,7 +2513,7 @@ dummy_func(
             #endif  /* ENABLE_SPECIALIZATION */
         }
 
-        op(_FOR_ITER, (iter -- iter, next)) {
+        replaced op(_FOR_ITER, (iter -- iter, next)) {
             /* before: [iter]; after: [iter, iter()] *or* [] (and jump over END_FOR.) */
             next = (*Py_TYPE(iter)->tp_iternext)(iter);
             if (next == NULL) {
@@ -2536,6 +2536,25 @@ dummy_func(
             // Common case: no jump, leave it to the code generator
         }
 
+        op(_FOR_ITER_TIER_TWO, (iter -- iter, next)) {
+            /* before: [iter]; after: [iter, iter()] *or* [] (and jump over END_FOR.) */
+            next = (*Py_TYPE(iter)->tp_iternext)(iter);
+            if (next == NULL) {
+                if (_PyErr_Occurred(tstate)) {
+                    if (!_PyErr_ExceptionMatches(tstate, PyExc_StopIteration)) {
+                        GOTO_ERROR(error);
+                    }
+                    _PyErr_Clear(tstate);
+                }
+                /* iterator ended normally */
+                Py_DECREF(iter);
+                STACK_SHRINK(1);
+                /* The translator sets the deopt target just past END_FOR */
+                DEOPT_IF(true);
+            }
+            // Common case: no jump, leave it to the code generator
+        }
+
         macro(FOR_ITER) = _SPECIALIZE_FOR_ITER + _FOR_ITER;
 
         inst(INSTRUMENTED_FOR_ITER, (unused/1 -- )) {
diff --git a/Python/ceval.c b/Python/ceval.c
index ba234dc..b5e8593 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -1074,7 +1074,7 @@ deoptimize:
     UOP_STAT_INC(opcode, miss);
     frame->return_offset = 0;  // Dispatch to frame->instr_ptr
     _PyFrame_SetStackPointer(frame, stack_pointer);
-    frame->instr_ptr = next_uop[-1].target + _PyCode_CODE((PyCodeObject *)frame->f_executable);
+    frame->instr_ptr = next_uop[-1].target + _PyCode_CODE(_PyFrame_GetCode(frame));
     Py_DECREF(current_executor);
     // Fall through
 // Jump here from ENTER_EXECUTOR
@@ -1085,7 +1085,7 @@ enter_tier_one:
 // Jump here from _EXIT_TRACE
 exit_trace:
     _PyFrame_SetStackPointer(frame, stack_pointer);
-    frame->instr_ptr = next_uop[-1].target + _PyCode_CODE((PyCodeObject *)frame->f_executable);
+    frame->instr_ptr = next_uop[-1].target + _PyCode_CODE(_PyFrame_GetCode(frame));
     Py_DECREF(current_executor);
     OPT_HIST(trace_uop_execution_counter, trace_run_length_hist);
     goto enter_tier_one;
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 4e29fb9..ae662b2 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -2101,6 +2101,31 @@
             break;
         }
 
+        case _FOR_ITER_TIER_TWO: {
+            PyObject *iter;
+            PyObject *next;
+            iter = stack_pointer[-1];
+            /* before: [iter]; after: [iter, iter()] *or* [] (and jump over END_FOR.) */
+            next = (*Py_TYPE(iter)->tp_iternext)(iter);
+            if (next == NULL) {
+                if (_PyErr_Occurred(tstate)) {
+                    if (!_PyErr_ExceptionMatches(tstate, PyExc_StopIteration)) {
+                        GOTO_ERROR(error);
+                    }
+                    _PyErr_Clear(tstate);
+                }
+                /* iterator ended normally */
+                Py_DECREF(iter);
+                STACK_SHRINK(1);
+                /* The translator sets the deopt target just past END_FOR */
+                DEOPT_IF(true, _FOR_ITER_TIER_TWO);
+            }
+            // Common case: no jump, leave it to the code generator
+            STACK_GROW(1);
+            stack_pointer[-1] = next;
+            break;
+        }
+
         case _ITER_CHECK_LIST: {
             PyObject *iter;
             iter = stack_pointer[-1];
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 64a15e0..261a5ff 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -392,6 +392,7 @@ _PyUop_Replacements[OPCODE_METADATA_SIZE] = {
     [_ITER_JUMP_RANGE] = _GUARD_NOT_EXHAUSTED_RANGE,
     [_ITER_JUMP_LIST] = _GUARD_NOT_EXHAUSTED_LIST,
     [_ITER_JUMP_TUPLE] = _GUARD_NOT_EXHAUSTED_TUPLE,
+    [_FOR_ITER] = _FOR_ITER_TIER_TWO,
 };
 
 static const uint16_t
@@ -620,6 +621,11 @@ top:  // Jump here after _PUSH_FRAME or likely branches
                                 }
                                 if (_PyUop_Replacements[uop]) {
                                     uop = _PyUop_Replacements[uop];
+                                    if (uop == _FOR_ITER_TIER_TWO) {
+                                        target += 1 + INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1;
+                                        assert(_PyCode_CODE(code)[target-1].op.code == END_FOR ||
+                                               _PyCode_CODE(code)[target-1].op.code == INSTRUMENTED_END_FOR);
+                                    }
                                 }
                                 break;
                             case OPARG_CACHE_1: