bpo-46841: Move the cache for LOAD_GLOBAL inline. (GH-31575)

Author: markshannon

Include/internal/pycore_code.h

@@ -26,11 +26,6 @@ typedef struct {
     uint32_t dk_version;
 } _PyAttrCache;
 
-typedef struct {
-    uint32_t module_keys_version;
-    uint32_t builtin_keys_version;
-} _PyLoadGlobalCache;
-
 typedef struct {
     /* Borrowed ref in LOAD_METHOD */
     PyObject *obj;
@@ -57,23 +52,35 @@ typedef union {
     _PyEntryZero zero;
     _PyAdaptiveEntry adaptive;
     _PyAttrCache attr;
-    _PyLoadGlobalCache load_global;
     _PyObjectCache obj;
     _PyCallCache call;
 } SpecializedCacheEntry;
 
 #define INSTRUCTIONS_PER_ENTRY (sizeof(SpecializedCacheEntry)/sizeof(_Py_CODEUNIT))
 
+/* Inline caches */
+
+#define CACHE_ENTRIES(cache) (sizeof(cache)/sizeof(_Py_CODEUNIT))
+
+typedef struct {
+    _Py_CODEUNIT counter;
+    _Py_CODEUNIT index;
+    _Py_CODEUNIT module_keys_version;
+    _Py_CODEUNIT _m1;
+    _Py_CODEUNIT builtin_keys_version;
+} _PyLoadGlobalCache;
+
+#define INLINE_CACHE_ENTRIES_LOAD_GLOBAL CACHE_ENTRIES(_PyLoadGlobalCache)
+
 typedef struct {
     _Py_CODEUNIT counter;
 } _PyBinaryOpCache;
 
+#define INLINE_CACHE_ENTRIES_BINARY_OP CACHE_ENTRIES(_PyBinaryOpCache)
 typedef struct {
     _Py_CODEUNIT counter;
 } _PyUnpackSequenceCache;
 
-#define INLINE_CACHE_ENTRIES_BINARY_OP \
-    (sizeof(_PyBinaryOpCache) / sizeof(_Py_CODEUNIT))
 
 #define INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE \
     (sizeof(_PyUnpackSequenceCache) / sizeof(_Py_CODEUNIT))
@@ -307,7 +314,7 @@ cache_backoff(_PyAdaptiveEntry *entry) {
 
 extern int _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
 extern int _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
-extern int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
+extern int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNIT *instr, PyObject *name);
 extern int _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
 extern int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_CODEUNIT *instr, SpecializedCacheEntry *cache);
 extern int _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *instr);
@@ -388,6 +395,38 @@ extern PyObject* _Py_GetSpecializationStats(void);
 #define OBJECT_STAT_INC(name) ((void)0)
 #endif
 
+// Cache values are only valid in memory, so use native endianness.
+#ifdef WORDS_BIGENDIAN
+
+static inline void
+write32(uint16_t *p, uint32_t val)
+{
+    p[0] = val >> 16;
+    p[1] = (uint16_t)val;
+}
+
+static inline uint32_t
+read32(uint16_t *p)
+{
+    return (p[0] << 16) | p[1];
+}
+
+#else
+
+static inline void
+write32(uint16_t *p, uint32_t val)
+{
+    p[0] = (uint16_t)val;
+    p[1] = val >> 16;
+}
+
+static inline uint32_t
+read32(uint16_t *p)
+{
+    return p[0] | (p[1] << 16);
+}
+
+#endif
 
 #ifdef __cplusplus
 }

Include/opcode.h

@@ -387,8 +387,8 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.11a5 3478 (New CALL opcodes)
 #     Python 3.11a5 3479 (Add PUSH_NULL opcode)
 #     Python 3.11a5 3480 (New CALL opcodes, second iteration)
-#     Python 3.11a5 3481 (Use inline CACHE instructions)
-#     Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE)
+#     Python 3.11a5 3481 (Use inline cache for BINARY_OP)
+#     Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE and LOAD_GLOBAL)
 
 #     Python 3.12 will start with magic number 3500
 

Lib/importlib/_bootstrap_external.py

@@ -135,7 +135,7 @@ def jabs_op(name, op, entries=0):
 jabs_op('JUMP_ABSOLUTE', 113)        # ""
 jabs_op('POP_JUMP_IF_FALSE', 114)    # ""
 jabs_op('POP_JUMP_IF_TRUE', 115)     # ""
-name_op('LOAD_GLOBAL', 116)     # Index in name list
+name_op('LOAD_GLOBAL', 116, 5)     # Index in name list
 def_op('IS_OP', 117)
 def_op('CONTAINS_OP', 118)
 def_op('RERAISE', 119)
@@ -198,6 +198,7 @@ def jabs_op(name, op, entries=0):
 def_op('KW_NAMES', 172)
 hasconst.append(172)
 
+
 del def_op, name_op, jrel_op, jabs_op
 
 _nb_ops = [