gh-100239: specialize long tail of binary operations

Author: iritkatriel

Include/internal/pycore_code.h

@@ -100,6 +100,7 @@ typedef struct {
 
 typedef struct {
     _Py_BackoffCounter counter;
+    uint16_t external_cache[4];
 } _PyBinaryOpCache;
 
 #define INLINE_CACHE_ENTRIES_BINARY_OP CACHE_ENTRIES(_PyBinaryOpCache)
@@ -345,8 +346,8 @@ extern void _Py_Specialize_Call(_PyStackRef callable, _Py_CODEUNIT *instr,
                                 int nargs);
 extern void _Py_Specialize_CallKw(_PyStackRef callable, _Py_CODEUNIT *instr,
                                   int nargs);
-extern void _Py_Specialize_BinaryOp(_PyStackRef lhs, _PyStackRef rhs, _Py_CODEUNIT *instr,
-                                    int oparg, _PyStackRef *locals);
+extern int _Py_Specialize_BinaryOp(_PyStackRef lhs, _PyStackRef rhs, _Py_CODEUNIT *instr,
+                                   int oparg, _PyStackRef *locals);
 extern void _Py_Specialize_CompareOp(_PyStackRef lhs, _PyStackRef rhs,
                                      _Py_CODEUNIT *instr, int oparg);
 extern void _Py_Specialize_UnpackSequence(_PyStackRef seq, _Py_CODEUNIT *instr,
@@ -443,6 +444,12 @@ write_obj(uint16_t *p, PyObject *val)
     memcpy(p, &val, sizeof(val));
 }
 
+static inline void
+write_void(uint16_t *p, void *val)
+{
+    memcpy(p, &val, sizeof(val));
+}
+
 static inline uint16_t
 read_u16(uint16_t *p)
 {
@@ -473,6 +480,14 @@ read_obj(uint16_t *p)
     return val;
 }
 
+static inline void *
+read_void(uint16_t *p)
+{
+    void *val;
+    memcpy(&val, p, sizeof(val));
+    return val;
+}
+
 /* See Objects/exception_handling_notes.txt for details.
  */
 static inline unsigned char *
@@ -576,6 +591,17 @@ adaptive_counter_backoff(_Py_BackoffCounter counter) {
     return restart_backoff_counter(counter);
 }
 
+/* Specialization Extensions */
+
+/* callbacks for an external specialization */
+typedef int (*binaryopguardfunc)(PyObject *lhs, PyObject *rhs);
+typedef PyObject *(*binaryopactionfunc)(PyObject *lhs, PyObject *rhs);
+
+typedef struct _PyBinaryOpSpecializationDescr {
+    binaryopguardfunc guard;
+    binaryopactionfunc action;
+    struct _PyBinaryOpSpecializationDescr *prev, *next;  /* For the tstate linked list */
+} PyBinaryOpSpecializationDescr;
 
 /* Comparison bit masks. */
 

Include/internal/pycore_interp.h

@@ -287,6 +287,8 @@ struct _is {
     // _initial_thread should be the last field of PyInterpreterState.
     // See https://github.com/python/cpython/issues/127117.
 
+    PyBinaryOpSpecializationDescr* binary_op_specialization_list;
+
 #if !defined(Py_GIL_DISABLED) && defined(Py_STACKREF_DEBUG)
     uint64_t next_stackref;
     _Py_hashtable_t *stackref_debug_table;

Include/internal/pycore_magic_number.h

@@ -265,6 +265,7 @@ Known values:
     Python 3.14a4 3610 (Add VALUE_WITH_FAKE_GLOBALS format to annotationlib)
     Python 3.14a4 3611 (Add NOT_TAKEN instruction)
     Python 3.14a4 3612 (Add POP_ITER and INSTRUMENTED_POP_ITER)
+    Python 3.14a4 3613 (Add BINARY_OP_EXTEND)
 
     Python 3.15 will start with 3650
 
@@ -277,7 +278,7 @@ PC/launcher.c must also be updated.
 
 */
 
-#define PYC_MAGIC_NUMBER 3612
+#define PYC_MAGIC_NUMBER 3613
 /* This is equivalent to converting PYC_MAGIC_NUMBER to 2 bytes
    (little-endian) and then appending b'\r\n'. */
 #define PYC_MAGIC_NUMBER_TOKEN \