bpo-45510: Specialize BINARY_SUBTRACT (GH-29523)

3 files changed, 60 insertions, 12 deletions

diff --git a/Python/ceval.c b/Python/ceval.c
index c02e3517d4..e579edefdb 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -2009,6 +2009,41 @@ check_eval_breaker:
             DISPATCH();
         }
 
+        TARGET(BINARY_OP_SUBTRACT_INT) {
+            PyObject *left = SECOND();
+            PyObject *right = TOP();
+            DEOPT_IF(!PyLong_CheckExact(left), BINARY_OP);
+            DEOPT_IF(!PyLong_CheckExact(right), BINARY_OP);
+            STAT_INC(BINARY_OP, hit);
+            PyObject *sub = _PyLong_Subtract((PyLongObject *)left, (PyLongObject *)right);
+            SET_SECOND(sub);
+            Py_DECREF(right);
+            Py_DECREF(left);
+            STACK_SHRINK(1);
+            if (sub == NULL) {
+                goto error;
+            }
+            DISPATCH();
+        }
+
+        TARGET(BINARY_OP_SUBTRACT_FLOAT) {
+            PyObject *left = SECOND();
+            PyObject *right = TOP();
+            DEOPT_IF(!PyFloat_CheckExact(left), BINARY_OP);
+            DEOPT_IF(!PyFloat_CheckExact(right), BINARY_OP);
+            STAT_INC(BINARY_OP, hit);
+            double dsub = ((PyFloatObject *)left)->ob_fval - ((PyFloatObject *)right)->ob_fval;
+            PyObject *sub = PyFloat_FromDouble(dsub);
+            SET_SECOND(sub);
+            Py_DECREF(right);
+            Py_DECREF(left);
+            STACK_SHRINK(1);
+            if (sub == NULL) {
+                goto error;
+            }
+            DISPATCH();
+        }
+
         TARGET(BINARY_OP_ADD_UNICODE) {
             PyObject *left = SECOND();
             PyObject *right = TOP();
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index a57617e022..07852d167c 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -18,25 +18,27 @@ static void *opcode_targets[256] = {
     &&TARGET_BINARY_OP_INPLACE_ADD_UNICODE,
     &&TARGET_BINARY_OP_MULTIPLY_INT,
     &&TARGET_BINARY_OP_MULTIPLY_FLOAT,
+    &&TARGET_BINARY_OP_SUBTRACT_INT,
+    &&TARGET_BINARY_OP_SUBTRACT_FLOAT,
     &&TARGET_BINARY_SUBSCR_ADAPTIVE,
     &&TARGET_BINARY_SUBSCR_LIST_INT,
     &&TARGET_BINARY_SUBSCR_TUPLE_INT,
     &&TARGET_BINARY_SUBSCR_DICT,
+    &&TARGET_BINARY_SUBSCR,
     &&TARGET_CALL_FUNCTION_ADAPTIVE,
     &&TARGET_CALL_FUNCTION_BUILTIN_O,
-    &&TARGET_BINARY_SUBSCR,
     &&TARGET_CALL_FUNCTION_BUILTIN_FAST,
     &&TARGET_CALL_FUNCTION_LEN,
-    &&TARGET_CALL_FUNCTION_ISINSTANCE,
-    &&TARGET_CALL_FUNCTION_PY_SIMPLE,
     &&TARGET_GET_LEN,
     &&TARGET_MATCH_MAPPING,
     &&TARGET_MATCH_SEQUENCE,
     &&TARGET_MATCH_KEYS,
-    &&TARGET_JUMP_ABSOLUTE_QUICK,
+    &&TARGET_CALL_FUNCTION_ISINSTANCE,
     &&TARGET_PUSH_EXC_INFO,
-    &&TARGET_LOAD_ATTR_ADAPTIVE,
+    &&TARGET_CALL_FUNCTION_PY_SIMPLE,
     &&TARGET_POP_EXCEPT_AND_RERAISE,
+    &&TARGET_JUMP_ABSOLUTE_QUICK,
+    &&TARGET_LOAD_ATTR_ADAPTIVE,
     &&TARGET_LOAD_ATTR_INSTANCE_VALUE,
     &&TARGET_LOAD_ATTR_WITH_HINT,
     &&TARGET_LOAD_ATTR_SLOT,
@@ -46,27 +48,25 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_GLOBAL_BUILTIN,
     &&TARGET_LOAD_METHOD_ADAPTIVE,
     &&TARGET_LOAD_METHOD_CACHED,
-    &&TARGET_LOAD_METHOD_CLASS,
-    &&TARGET_LOAD_METHOD_MODULE,
     &&TARGET_WITH_EXCEPT_START,
     &&TARGET_GET_AITER,
     &&TARGET_GET_ANEXT,
     &&TARGET_BEFORE_ASYNC_WITH,
     &&TARGET_BEFORE_WITH,
     &&TARGET_END_ASYNC_FOR,
+    &&TARGET_LOAD_METHOD_CLASS,
+    &&TARGET_LOAD_METHOD_MODULE,
     &&TARGET_LOAD_METHOD_NO_DICT,
     &&TARGET_STORE_ATTR_ADAPTIVE,
     &&TARGET_STORE_ATTR_INSTANCE_VALUE,
-    &&TARGET_STORE_ATTR_SLOT,
-    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_STORE_SUBSCR,
     &&TARGET_DELETE_SUBSCR,
+    &&TARGET_STORE_ATTR_SLOT,
+    &&TARGET_STORE_ATTR_WITH_HINT,
     &&TARGET_LOAD_FAST__LOAD_FAST,
     &&TARGET_STORE_FAST__LOAD_FAST,
     &&TARGET_LOAD_FAST__LOAD_CONST,
     &&TARGET_LOAD_CONST__LOAD_FAST,
-    &&TARGET_STORE_FAST__STORE_FAST,
-    &&_unknown_opcode,
     &&TARGET_GET_ITER,
     &&TARGET_GET_YIELD_FROM_ITER,
     &&TARGET_PRINT_EXPR,
@@ -74,7 +74,7 @@ static void *opcode_targets[256] = {
     &&TARGET_YIELD_FROM,
     &&TARGET_GET_AWAITABLE,
     &&TARGET_LOAD_ASSERTION_ERROR,
-    &&_unknown_opcode,
+    &&TARGET_STORE_FAST__STORE_FAST,
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
diff --git a/Python/specialize.c b/Python/specialize.c
index cfc21bf70a..dd15de72d2 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -1424,6 +1424,19 @@ _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                 goto success;
             }
             break;
+        case NB_SUBTRACT:
+        case NB_INPLACE_SUBTRACT:
+            if (PyLong_CheckExact(lhs)) {
+                *instr = _Py_MAKECODEUNIT(BINARY_OP_SUBTRACT_INT,
+                                          _Py_OPARG(*instr));
+                goto success;
+            }
+            if (PyFloat_CheckExact(lhs)) {
+                *instr = _Py_MAKECODEUNIT(BINARY_OP_SUBTRACT_FLOAT,
+                                          _Py_OPARG(*instr));
+                goto success;
+            }
+            break;
         default:
             // These operators don't have any available specializations. Rather
             // than repeatedly attempting to specialize them, just convert them