bpo-46841: Use inline caching for attribute accesses (GH-31640)

10 files changed, 292 insertions, 270 deletions

diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
index b9671d0..25c31a1 100644
--- a/Include/internal/pycore_code.h
+++ b/Include/internal/pycore_code.h
@@ -20,14 +20,8 @@ typedef struct {
     uint32_t version;
 } _PyAdaptiveEntry;
 
-
 typedef struct {
-    uint32_t tp_version;
-    uint32_t dk_version;
-} _PyAttrCache;
-
-typedef struct {
-    /* Borrowed ref in LOAD_METHOD */
+    /* Borrowed ref */
     PyObject *obj;
 } _PyObjectCache;
 
@@ -51,7 +45,6 @@ typedef struct {
 typedef union {
     _PyEntryZero zero;
     _PyAdaptiveEntry adaptive;
-    _PyAttrCache attr;
     _PyObjectCache obj;
     _PyCallCache call;
 } SpecializedCacheEntry;
@@ -65,8 +58,7 @@ typedef union {
 typedef struct {
     _Py_CODEUNIT counter;
     _Py_CODEUNIT index;
-    _Py_CODEUNIT module_keys_version;
-    _Py_CODEUNIT _m1;
+    _Py_CODEUNIT module_keys_version[2];
     _Py_CODEUNIT builtin_keys_version;
 } _PyLoadGlobalCache;
 
@@ -94,13 +86,32 @@ typedef struct {
 
 typedef struct {
     _Py_CODEUNIT counter;
-    _Py_CODEUNIT type_version;
-    _Py_CODEUNIT _t1;
+    _Py_CODEUNIT type_version[2];
     _Py_CODEUNIT func_version;
 } _PyBinarySubscrCache;
 
 #define INLINE_CACHE_ENTRIES_BINARY_SUBSCR CACHE_ENTRIES(_PyBinarySubscrCache)
 
+typedef struct {
+    _Py_CODEUNIT counter;
+    _Py_CODEUNIT version[2];
+    _Py_CODEUNIT index;
+} _PyAttrCache;
+
+#define INLINE_CACHE_ENTRIES_LOAD_ATTR CACHE_ENTRIES(_PyAttrCache)
+
+#define INLINE_CACHE_ENTRIES_STORE_ATTR CACHE_ENTRIES(_PyAttrCache)
+
+typedef struct {
+    _Py_CODEUNIT counter;
+    _Py_CODEUNIT type_version[2];
+    _Py_CODEUNIT dict_offset;
+    _Py_CODEUNIT keys_version[2];
+    _Py_CODEUNIT descr[4];
+} _PyLoadMethodCache;
+
+#define INLINE_CACHE_ENTRIES_LOAD_METHOD CACHE_ENTRIES(_PyLoadMethodCache)
+
 /* Maximum size of code to quicken, in code units. */
 #define MAX_SIZE_TO_QUICKEN 5000
 
@@ -328,10 +339,13 @@ cache_backoff(_PyAdaptiveEntry *entry) {
 
 /* Specialization functions */
 
-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_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr,
+                                   PyObject *name);
+extern int _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr,
+                                    PyObject *name);
 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_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr,
+                                     PyObject *name);
 extern int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_CODEUNIT *instr);
 extern int _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *instr);
 extern int _Py_Specialize_Call(PyObject *callable, _Py_CODEUNIT *instr, int nargs,
@@ -416,34 +430,107 @@ extern PyObject* _Py_GetSpecializationStats(void);
 #ifdef WORDS_BIGENDIAN
 
 static inline void
-write32(uint16_t *p, uint32_t val)
+write_u32(uint16_t *p, uint32_t val)
 {
-    p[0] = val >> 16;
-    p[1] = (uint16_t)val;
+    p[0] = (uint16_t)(val >> 16);
+    p[1] = (uint16_t)(val >>  0);
+}
+
+static inline void
+write_u64(uint16_t *p, uint64_t val)
+{
+    p[0] = (uint16_t)(val >> 48);
+    p[1] = (uint16_t)(val >> 32);
+    p[2] = (uint16_t)(val >> 16);
+    p[3] = (uint16_t)(val >>  0);
 }
 
 static inline uint32_t
-read32(uint16_t *p)
+read_u32(uint16_t *p)
+{
+    uint32_t val = 0;
+    val |= (uint32_t)p[0] << 16;
+    val |= (uint32_t)p[1] <<  0;
+    return val;
+}
+
+static inline uint64_t
+read_u64(uint16_t *p)
 {
-    return (p[0] << 16) | p[1];
+    uint64_t val = 0;
+    val |= (uint64_t)p[0] << 48;
+    val |= (uint64_t)p[1] << 32;
+    val |= (uint64_t)p[2] << 16;
+    val |= (uint64_t)p[3] <<  0;
+    return val;
 }
 
 #else
 
 static inline void
-write32(uint16_t *p, uint32_t val)
+write_u32(uint16_t *p, uint32_t val)
+{
+    p[0] = (uint16_t)(val >>  0);
+    p[1] = (uint16_t)(val >> 16);
+}
+
+static inline void
+write_u64(uint16_t *p, uint64_t val)
 {
-    p[0] = (uint16_t)val;
-    p[1] = val >> 16;
+    p[0] = (uint16_t)(val >>  0);
+    p[1] = (uint16_t)(val >> 16);
+    p[2] = (uint16_t)(val >> 32);
+    p[3] = (uint16_t)(val >> 48);
 }
 
 static inline uint32_t
-read32(uint16_t *p)
+read_u32(uint16_t *p)
+{
+    uint32_t val = 0;
+    val |= (uint32_t)p[0] <<  0;
+    val |= (uint32_t)p[1] << 16;
+    return val;
+}
+
+static inline uint64_t
+read_u64(uint16_t *p)
+{
+    uint64_t val = 0;
+    val |= (uint64_t)p[0] <<  0;
+    val |= (uint64_t)p[1] << 16;
+    val |= (uint64_t)p[2] << 32;
+    val |= (uint64_t)p[3] << 48;
+    return val;
+}
+
+#endif
+
+static inline void
+write_obj(uint16_t *p, PyObject *obj)
 {
-    return p[0] | (p[1] << 16);
+    uintptr_t val = (uintptr_t)obj;
+#if SIZEOF_VOID_P == 8
+    write_u64(p, val);
+#elif SIZEOF_VOID_P == 4
+    write_u32(p, val);
+#else
+    #error "SIZEOF_VOID_P must be 4 or 8"
+#endif
 }
 
+static inline PyObject *
+read_obj(uint16_t *p)
+{
+    uintptr_t val;
+#if SIZEOF_VOID_P == 8
+    val = read_u64(p);
+#elif SIZEOF_VOID_P == 4
+    val = read_u32(p);
+#else
+    #error "SIZEOF_VOID_P must be 4 or 8"
 #endif
+    return (PyObject *)val;
+}
 
 #ifdef __cplusplus
 }
diff --git a/Include/opcode.h b/Include/opcode.h
index f6330d9..110f8c3 100644
--- a/Include/opcode.h
+++ b/Include/opcode.h
@@ -183,7 +183,6 @@ extern "C" {
 #define LOAD_FAST__LOAD_CONST           173
 #define LOAD_CONST__LOAD_FAST           174
 #define STORE_FAST__STORE_FAST          175
-#define LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE 176
 #define DO_TRACING                      255
 
 extern const uint8_t _PyOpcode_InlineCacheEntries[256];
@@ -213,9 +212,12 @@ static const uint32_t _PyOpcode_Jump[8] = {
 const uint8_t _PyOpcode_InlineCacheEntries[256] = {
     [BINARY_SUBSCR] = 4,
     [UNPACK_SEQUENCE] = 1,
+    [STORE_ATTR] = 4,
+    [LOAD_ATTR] = 4,
     [COMPARE_OP] = 2,
     [LOAD_GLOBAL] = 5,
     [BINARY_OP] = 1,
+    [LOAD_METHOD] = 10,
 };
 #endif /* OPCODE_TABLES */
 
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
index dd1f6ff..9d36bc2 100644
--- a/Lib/importlib/_bootstrap_external.py
+++ b/Lib/importlib/_bootstrap_external.py
@@ -390,6 +390,8 @@ _code_type = type(_write_atomic.__code__)
 #     Python 3.11a5 3481 (Use inline cache for BINARY_OP)
 #     Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE and LOAD_GLOBAL)
 #     Python 3.11a5 3483 (Use inline caching for COMPARE_OP and BINARY_SUBSCR)
+#     Python 3.11a5 3484 (Use inline caching for LOAD_ATTR, LOAD_METHOD, and 
+#                         STORE_ATTR)
 
 #     Python 3.12 will start with magic number 3500
 
@@ -404,7 +406,7 @@ _code_type = type(_write_atomic.__code__)
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3483).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3484).to_bytes(2, 'little') + b'\r\n'
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
 _PYCACHE = '__pycache__'
diff --git a/Lib/opcode.py b/Lib/opcode.py
index 9b08562..f6e2dec 100644
--- a/Lib/opcode.py
+++ b/Lib/opcode.py
@@ -112,7 +112,7 @@ name_op('DELETE_NAME', 91)      # ""
 def_op('UNPACK_SEQUENCE', 92, 1)   # Number of tuple items
 jrel_op('FOR_ITER', 93)
 def_op('UNPACK_EX', 94)
-name_op('STORE_ATTR', 95)       # Index in name list
+name_op('STORE_ATTR', 95, 4)       # Index in name list
 name_op('DELETE_ATTR', 96)      # ""
 name_op('STORE_GLOBAL', 97)     # ""
 name_op('DELETE_GLOBAL', 98)    # ""
@@ -124,7 +124,7 @@ def_op('BUILD_TUPLE', 102)      # Number of tuple items
 def_op('BUILD_LIST', 103)       # Number of list items
 def_op('BUILD_SET', 104)        # Number of set items
 def_op('BUILD_MAP', 105)        # Number of dict entries
-name_op('LOAD_ATTR', 106)       # Index in name list
+name_op('LOAD_ATTR', 106, 4)       # Index in name list
 def_op('COMPARE_OP', 107, 2)       # Comparison operator
 hascompare.append(107)
 name_op('IMPORT_NAME', 108)     # Index in name list
@@ -186,7 +186,7 @@ def_op('FORMAT_VALUE', 155)
 def_op('BUILD_CONST_KEY_MAP', 156)
 def_op('BUILD_STRING', 157)
 
-name_op('LOAD_METHOD', 160)
+name_op('LOAD_METHOD', 160, 10)
 
 def_op('LIST_EXTEND', 162)
 def_op('SET_UPDATE', 163)
@@ -301,7 +301,6 @@ _specialized_instructions = [
     "LOAD_FAST__LOAD_CONST",
     "LOAD_CONST__LOAD_FAST",
     "STORE_FAST__STORE_FAST",
-    "LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE",
 ]
 _specialization_stats = [
     "success",
diff --git a/Lib/test/test_dis.py b/Lib/test/test_dis.py
index 8de2ed0..7e0542a 100644
--- a/Lib/test/test_dis.py
+++ b/Lib/test/test_dis.py
@@ -375,7 +375,7 @@ dis_traceback = """\
         >> PUSH_EXC_INFO
 
 %3d        LOAD_GLOBAL              0 (Exception)
-           JUMP_IF_NOT_EXC_MATCH    31 (to 62)
+           JUMP_IF_NOT_EXC_MATCH    35 (to 70)
            STORE_FAST               0 (e)
 
 %3d        LOAD_FAST                0 (e)
diff --git a/Misc/NEWS.d/next/Core and Builtins/2022-03-01-17-47-58.bpo-46841.inYQlU.rst b/Misc/NEWS.d/next/Core and Builtins/2022-03-01-17-47-58.bpo-46841.inYQlU.rst
new file mode 100644
index 0000000..0e7beb0
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2022-03-01-17-47-58.bpo-46841.inYQlU.rst
@@ -0,0 +1,2 @@
+Use inline caching for :opcode:`LOAD_ATTR`, :opcode:`LOAD_METHOD`, and
+:opcode:`STORE_ATTR`.
diff --git a/Programs/test_frozenmain.h b/Programs/test_frozenmain.h
index 3fef981..4ebab4f 100644
--- a/Programs/test_frozenmain.h
+++ b/Programs/test_frozenmain.h
@@ -1,14 +1,15 @@
 // Auto-generated by Programs/freeze_test_frozenmain.py
 unsigned char M_test_frozenmain[] = {
     227,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,
-    0,0,0,0,0,115,120,0,0,0,151,0,100,0,100,1,
+    0,0,0,0,0,115,136,0,0,0,151,0,100,0,100,1,
     108,0,90,0,100,0,100,1,108,1,90,1,2,0,101,2,
     100,2,166,1,171,1,1,0,2,0,101,2,100,3,101,0,
-    106,3,166,2,171,2,1,0,2,0,101,1,106,4,166,0,
+    106,3,3,0,3,0,3,0,3,0,166,2,171,2,1,0,
+    2,0,101,1,106,4,3,0,3,0,3,0,3,0,166,0,
     171,0,100,4,25,0,3,0,3,0,3,0,3,0,90,5,
     100,5,68,0,93,20,90,6,2,0,101,2,100,6,101,6,
     155,0,100,7,101,5,101,6,25,0,3,0,3,0,3,0,
-    3,0,155,0,157,4,166,1,171,1,1,0,113,37,100,1,
+    3,0,155,0,157,4,166,1,171,1,1,0,113,45,100,1,
     83,0,41,8,233,0,0,0,0,78,122,18,70,114,111,122,
     101,110,32,72,101,108,108,111,32,87,111,114,108,100,122,8,
     115,121,115,46,97,114,103,118,218,6,99,111,110,102,105,103,
@@ -25,12 +26,13 @@ unsigned char M_test_frozenmain[] = {
     0,0,250,18,116,101,115,116,95,102,114,111,122,101,110,109,
     97,105,110,46,112,121,250,8,60,109,111,100,117,108,101,62,
     114,11,0,0,0,1,0,0,0,115,18,0,0,0,2,128,
-    8,3,8,1,12,2,16,1,24,1,8,1,38,7,4,249,
-    115,20,0,0,0,2,128,8,3,8,1,12,2,16,1,24,
-    1,2,7,4,1,2,249,42,7,115,120,0,0,0,0,0,
+    8,3,8,1,12,2,24,1,32,1,8,1,38,7,4,249,
+    115,20,0,0,0,2,128,8,3,8,1,12,2,24,1,32,
+    1,2,7,4,1,2,249,42,7,115,136,0,0,0,0,0,
     1,11,1,11,1,11,1,11,1,25,1,25,1,25,1,25,
     1,6,1,6,7,27,1,28,1,28,1,28,1,6,1,6,
-    7,17,19,22,19,27,1,28,1,28,1,28,10,39,10,27,
+    7,17,19,22,19,27,19,27,19,27,19,27,19,27,1,28,
+    1,28,1,28,10,39,10,27,10,39,10,39,10,39,10,39,
     10,39,10,41,10,41,42,50,10,51,10,51,10,51,10,51,
     10,51,1,7,12,2,1,42,1,42,5,8,5,10,5,10,
     11,41,21,24,11,41,11,41,28,34,35,38,28,39,28,39,
diff --git a/Python/ceval.c b/Python/ceval.c
index c86b744..915ab93 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -1448,16 +1448,15 @@ eval_frame_handle_pending(PyThreadState *tstate)
 
 // shared by LOAD_ATTR_MODULE and LOAD_METHOD_MODULE
 #define LOAD_MODULE_ATTR_OR_METHOD(attr_or_method) \
-    SpecializedCacheEntry *caches = GET_CACHE(); \
-    _PyAdaptiveEntry *cache0 = &caches[0].adaptive; \
+    _PyAttrCache *cache = (_PyAttrCache *)next_instr; \
     DEOPT_IF(!PyModule_CheckExact(owner), LOAD_##attr_or_method); \
     PyDictObject *dict = (PyDictObject *)((PyModuleObject *)owner)->md_dict; \
     assert(dict != NULL); \
-    DEOPT_IF(dict->ma_keys->dk_version != cache0->version, \
-        LOAD_##attr_or_method); \
+    DEOPT_IF(dict->ma_keys->dk_version != read_u32(cache->version), \
+             LOAD_##attr_or_method); \
     assert(dict->ma_keys->dk_kind == DICT_KEYS_UNICODE); \
-    assert(cache0->index < dict->ma_keys->dk_nentries); \
-    PyDictUnicodeEntry *ep = DK_UNICODE_ENTRIES(dict->ma_keys) + cache0->index; \
+    assert(cache->index < dict->ma_keys->dk_nentries); \
+    PyDictUnicodeEntry *ep = DK_UNICODE_ENTRIES(dict->ma_keys) + cache->index; \
     res = ep->me_value; \
     DEOPT_IF(res == NULL, LOAD_##attr_or_method); \
     STAT_INC(LOAD_##attr_or_method, hit); \
@@ -2197,7 +2196,7 @@ handle_eval_breaker:
             PyObject *sub = TOP();
             PyObject *container = SECOND();
             _PyBinarySubscrCache *cache = (_PyBinarySubscrCache *)next_instr;
-            uint32_t type_version = read32(&cache->type_version);
+            uint32_t type_version = read_u32(cache->type_version);
             PyTypeObject *tp = Py_TYPE(container);
             DEOPT_IF(tp->tp_version_tag != type_version, BINARY_SUBSCR);
             assert(tp->tp_flags & Py_TPFLAGS_HEAPTYPE);
@@ -2849,8 +2848,10 @@ handle_eval_breaker:
             err = PyObject_SetAttr(owner, name, v);
             Py_DECREF(v);
             Py_DECREF(owner);
-            if (err != 0)
+            if (err != 0) {
                 goto error;
+            }
+            JUMPBY(INLINE_CACHE_ENTRIES_STORE_ATTR);
             DISPATCH();
         }
 
@@ -3028,7 +3029,7 @@ handle_eval_breaker:
             DEOPT_IF(!PyDict_CheckExact(GLOBALS()), LOAD_GLOBAL);
             PyDictObject *dict = (PyDictObject *)GLOBALS();
             _PyLoadGlobalCache *cache = (_PyLoadGlobalCache *)next_instr;
-            uint32_t version = read32(&cache->module_keys_version);
+            uint32_t version = read_u32(cache->module_keys_version);
             DEOPT_IF(dict->ma_keys->dk_version != version, LOAD_GLOBAL);
             assert(DK_IS_UNICODE(dict->ma_keys));
             PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dict->ma_keys);
@@ -3048,7 +3049,7 @@ handle_eval_breaker:
             PyDictObject *mdict = (PyDictObject *)GLOBALS();
             PyDictObject *bdict = (PyDictObject *)BUILTINS();
             _PyLoadGlobalCache *cache = (_PyLoadGlobalCache *)next_instr;
-            uint32_t mod_version = read32(&cache->module_keys_version);
+            uint32_t mod_version = read_u32(cache->module_keys_version);
             uint16_t bltn_version = cache->builtin_keys_version;
             DEOPT_IF(mdict->ma_keys->dk_version != mod_version, LOAD_GLOBAL);
             DEOPT_IF(bdict->ma_keys->dk_version != bltn_version, LOAD_GLOBAL);
@@ -3423,76 +3424,49 @@ handle_eval_breaker:
             }
             Py_DECREF(owner);
             SET_TOP(res);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
             DISPATCH();
         }
 
         TARGET(LOAD_ATTR_ADAPTIVE) {
             assert(cframe.use_tracing == 0);
-            SpecializedCacheEntry *cache = GET_CACHE();
-            if (cache->adaptive.counter == 0) {
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            if (cache->counter == 0) {
                 PyObject *owner = TOP();
-                PyObject *name = GETITEM(names, cache->adaptive.original_oparg);
+                PyObject *name = GETITEM(names, oparg);
                 next_instr--;
-                if (_Py_Specialize_LoadAttr(owner, next_instr, name, cache) < 0) {
+                if (_Py_Specialize_LoadAttr(owner, next_instr, name) < 0) {
                     goto error;
                 }
                 DISPATCH();
             }
             else {
                 STAT_INC(LOAD_ATTR, deferred);
-                cache->adaptive.counter--;
-                oparg = cache->adaptive.original_oparg;
+                cache->counter--;
                 JUMP_TO_INSTRUCTION(LOAD_ATTR);
             }
         }
 
-        TARGET(LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE) {
-            assert(cframe.use_tracing == 0);
-            PyObject *owner = GETLOCAL(oparg); // borrowed
-            if (owner == NULL) {
-                goto unbound_local_error;
-            }
-            // GET_CACHE(), but for the following opcode
-            assert(_Py_OPCODE(*next_instr) == LOAD_ATTR_INSTANCE_VALUE);
-            SpecializedCacheEntry *caches = _GetSpecializedCacheEntryForInstruction(
-                first_instr, INSTR_OFFSET() + 1, _Py_OPARG(*next_instr));
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            PyTypeObject *tp = Py_TYPE(owner);
-            // These DEOPT_IF miss branches do PUSH(Py_NewRef(owner)).
-            DEOPT_IF(tp->tp_version_tag != cache0->version,
-                     LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE);
-            assert(tp->tp_dictoffset < 0);
-            assert(tp->tp_flags & Py_TPFLAGS_MANAGED_DICT);
-            PyDictValues *values = *_PyObject_ValuesPointer(owner);
-            DEOPT_IF(values == NULL, LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE);
-            PyObject *res = values->values[cache0->index];
-            DEOPT_IF(res == NULL, LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE);
-            STAT_INC(LOAD_ATTR, hit);
-            PUSH(Py_NewRef(res));
-            next_instr++;
-            NOTRACE_DISPATCH();
-        }
-
         TARGET(LOAD_ATTR_INSTANCE_VALUE) {
             assert(cframe.use_tracing == 0);
             PyObject *owner = TOP();
             PyObject *res;
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, LOAD_ATTR_INSTANCE_VALUE);
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, LOAD_ATTR);
             assert(tp->tp_dictoffset < 0);
             assert(tp->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictValues *values = *_PyObject_ValuesPointer(owner);
-            DEOPT_IF(values == NULL, LOAD_ATTR_INSTANCE_VALUE);
-            res = values->values[cache0->index];
-            DEOPT_IF(res == NULL, LOAD_ATTR_INSTANCE_VALUE);
+            DEOPT_IF(values == NULL, LOAD_ATTR);
+            res = values->values[cache->index];
+            DEOPT_IF(res == NULL, LOAD_ATTR);
             STAT_INC(LOAD_ATTR, hit);
             Py_INCREF(res);
             SET_TOP(res);
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -3504,6 +3478,7 @@ handle_eval_breaker:
             LOAD_MODULE_ATTR_OR_METHOD(ATTR);
             SET_TOP(res);
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -3512,16 +3487,16 @@ handle_eval_breaker:
             PyObject *owner = TOP();
             PyObject *res;
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, LOAD_ATTR);
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, LOAD_ATTR);
             assert(tp->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictObject *dict = *(PyDictObject **)_PyObject_ManagedDictPointer(owner);
             DEOPT_IF(dict == NULL, LOAD_ATTR);
             assert(PyDict_CheckExact((PyObject *)dict));
-            PyObject *name = GETITEM(names, cache0->original_oparg);
-            uint16_t hint = cache0->index;
+            PyObject *name = GETITEM(names, oparg);
+            uint16_t hint = cache->index;
             DEOPT_IF(hint >= (size_t)dict->ma_keys->dk_nentries, LOAD_ATTR);
             if (DK_IS_UNICODE(dict->ma_keys)) {
                 PyDictUnicodeEntry *ep = DK_UNICODE_ENTRIES(dict->ma_keys) + hint;
@@ -3538,6 +3513,7 @@ handle_eval_breaker:
             Py_INCREF(res);
             SET_TOP(res);
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -3546,36 +3522,36 @@ handle_eval_breaker:
             PyObject *owner = TOP();
             PyObject *res;
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, LOAD_ATTR);
-            char *addr = (char *)owner + cache0->index;
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, LOAD_ATTR);
+            char *addr = (char *)owner + cache->index;
             res = *(PyObject **)addr;
             DEOPT_IF(res == NULL, LOAD_ATTR);
             STAT_INC(LOAD_ATTR, hit);
             Py_INCREF(res);
             SET_TOP(res);
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_ATTR);
             NOTRACE_DISPATCH();
         }
 
         TARGET(STORE_ATTR_ADAPTIVE) {
             assert(cframe.use_tracing == 0);
-            SpecializedCacheEntry *cache = GET_CACHE();
-            if (cache->adaptive.counter == 0) {
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            if (cache->counter == 0) {
                 PyObject *owner = TOP();
-                PyObject *name = GETITEM(names, cache->adaptive.original_oparg);
+                PyObject *name = GETITEM(names, oparg);
                 next_instr--;
-                if (_Py_Specialize_StoreAttr(owner, next_instr, name, cache) < 0) {
+                if (_Py_Specialize_StoreAttr(owner, next_instr, name) < 0) {
                     goto error;
                 }
                 DISPATCH();
             }
             else {
                 STAT_INC(STORE_ATTR, deferred);
-                cache->adaptive.counter--;
-                oparg = cache->adaptive.original_oparg;
+                cache->counter--;
                 JUMP_TO_INSTRUCTION(STORE_ATTR);
             }
         }
@@ -3584,15 +3560,15 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *owner = TOP();
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, STORE_ATTR);
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, STORE_ATTR);
             assert(tp->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictValues *values = *_PyObject_ValuesPointer(owner);
             DEOPT_IF(values == NULL, STORE_ATTR);
             STAT_INC(STORE_ATTR, hit);
-            Py_ssize_t index = cache0->index;
+            Py_ssize_t index = cache->index;
             STACK_SHRINK(1);
             PyObject *value = POP();
             PyObject *old_value = values->values[index];
@@ -3604,6 +3580,7 @@ handle_eval_breaker:
                 Py_DECREF(old_value);
             }
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_STORE_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -3611,16 +3588,16 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *owner = TOP();
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, STORE_ATTR);
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, STORE_ATTR);
             assert(tp->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictObject *dict = *(PyDictObject **)_PyObject_ManagedDictPointer(owner);
             DEOPT_IF(dict == NULL, STORE_ATTR);
             assert(PyDict_CheckExact((PyObject *)dict));
-            PyObject *name = GETITEM(names, cache0->original_oparg);
-            uint16_t hint = cache0->index;
+            PyObject *name = GETITEM(names, oparg);
+            uint16_t hint = cache->index;
             DEOPT_IF(hint >= (size_t)dict->ma_keys->dk_nentries, STORE_ATTR);
             PyObject *value, *old_value;
             if (DK_IS_UNICODE(dict->ma_keys)) {
@@ -3650,6 +3627,7 @@ handle_eval_breaker:
             /* PEP 509 */
             dict->ma_version_tag = DICT_NEXT_VERSION();
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_STORE_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -3657,11 +3635,11 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *owner = TOP();
             PyTypeObject *tp = Py_TYPE(owner);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            assert(cache0->version != 0);
-            DEOPT_IF(tp->tp_version_tag != cache0->version, STORE_ATTR);
-            char *addr = (char *)owner + cache0->index;
+            _PyAttrCache *cache = (_PyAttrCache *)next_instr;
+            uint32_t type_version = read_u32(cache->version);
+            assert(type_version != 0);
+            DEOPT_IF(tp->tp_version_tag != type_version, STORE_ATTR);
+            char *addr = (char *)owner + cache->index;
             STAT_INC(STORE_ATTR, hit);
             STACK_SHRINK(1);
             PyObject *value = POP();
@@ -3669,6 +3647,7 @@ handle_eval_breaker:
             *(PyObject **)addr = value;
             Py_XDECREF(old_value);
             Py_DECREF(owner);
+            JUMPBY(INLINE_CACHE_ENTRIES_STORE_ATTR);
             NOTRACE_DISPATCH();
         }
 
@@ -4425,25 +4404,25 @@ handle_eval_breaker:
                 Py_DECREF(obj);
                 PUSH(meth);
             }
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             DISPATCH();
         }
 
         TARGET(LOAD_METHOD_ADAPTIVE) {
             assert(cframe.use_tracing == 0);
-            SpecializedCacheEntry *cache = GET_CACHE();
-            if (cache->adaptive.counter == 0) {
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
+            if (cache->counter == 0) {
                 PyObject *owner = TOP();
-                PyObject *name = GETITEM(names, cache->adaptive.original_oparg);
+                PyObject *name = GETITEM(names, oparg);
                 next_instr--;
-                if (_Py_Specialize_LoadMethod(owner, next_instr, name, cache) < 0) {
+                if (_Py_Specialize_LoadMethod(owner, next_instr, name) < 0) {
                     goto error;
                 }
                 DISPATCH();
             }
             else {
                 STAT_INC(LOAD_METHOD, deferred);
-                cache->adaptive.counter--;
-                oparg = cache->adaptive.original_oparg;
+                cache->counter--;
                 JUMP_TO_INSTRUCTION(LOAD_METHOD);
             }
         }
@@ -4453,22 +4432,24 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *self = TOP();
             PyTypeObject *self_cls = Py_TYPE(self);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAttrCache *cache1 = &caches[-1].attr;
-            _PyObjectCache *cache2 = &caches[-2].obj;
-            assert(cache1->tp_version != 0);
-            DEOPT_IF(self_cls->tp_version_tag != cache1->tp_version, LOAD_METHOD);
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
+            uint32_t type_version = read_u32(cache->type_version);
+            assert(type_version != 0);
+            DEOPT_IF(self_cls->tp_version_tag != type_version, LOAD_METHOD);
             assert(self_cls->tp_flags & Py_TPFLAGS_MANAGED_DICT);
             PyDictObject *dict = *(PyDictObject**)_PyObject_ManagedDictPointer(self);
             DEOPT_IF(dict != NULL, LOAD_METHOD);
-            DEOPT_IF(((PyHeapTypeObject *)self_cls)->ht_cached_keys->dk_version != cache1->dk_version, LOAD_METHOD);
+            PyHeapTypeObject *self_heap_type = (PyHeapTypeObject *)self_cls;
+            DEOPT_IF(self_heap_type->ht_cached_keys->dk_version !=
+                     read_u32(cache->keys_version), LOAD_METHOD);
             STAT_INC(LOAD_METHOD, hit);
-            PyObject *res = cache2->obj;
+            PyObject *res = read_obj(cache->descr);
             assert(res != NULL);
             assert(_PyType_HasFeature(Py_TYPE(res), Py_TPFLAGS_METHOD_DESCRIPTOR));
             Py_INCREF(res);
             SET_TOP(res);
             PUSH(self);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             NOTRACE_DISPATCH();
         }
 
@@ -4478,14 +4459,12 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *self = TOP();
             PyTypeObject *self_cls = Py_TYPE(self);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAdaptiveEntry *cache0 = &caches[0].adaptive;
-            _PyAttrCache *cache1 = &caches[-1].attr;
-            _PyObjectCache *cache2 = &caches[-2].obj;
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
 
-            DEOPT_IF(self_cls->tp_version_tag != cache1->tp_version, LOAD_METHOD);
+            DEOPT_IF(self_cls->tp_version_tag != read_u32(cache->type_version),
+                     LOAD_METHOD);
             /* Treat index as a signed 16 bit value */
-            int dictoffset = *(int16_t *)&cache0->index;
+            int dictoffset = *(int16_t *)&cache->dict_offset;
             PyDictObject **dictptr = (PyDictObject**)(((char *)self)+dictoffset);
             assert(
                 dictoffset == MANAGED_DICT_OFFSET ||
@@ -4493,14 +4472,16 @@ handle_eval_breaker:
             );
             PyDictObject *dict = *dictptr;
             DEOPT_IF(dict == NULL, LOAD_METHOD);
-            DEOPT_IF(dict->ma_keys->dk_version != cache1->dk_version, LOAD_METHOD);
+            DEOPT_IF(dict->ma_keys->dk_version != read_u32(cache->keys_version),
+                     LOAD_METHOD);
             STAT_INC(LOAD_METHOD, hit);
-            PyObject *res = cache2->obj;
+            PyObject *res = read_obj(cache->descr);
             assert(res != NULL);
             assert(_PyType_HasFeature(Py_TYPE(res), Py_TPFLAGS_METHOD_DESCRIPTOR));
             Py_INCREF(res);
             SET_TOP(res);
             PUSH(self);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             NOTRACE_DISPATCH();
         }
 
@@ -4508,18 +4489,18 @@ handle_eval_breaker:
             assert(cframe.use_tracing == 0);
             PyObject *self = TOP();
             PyTypeObject *self_cls = Py_TYPE(self);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAttrCache *cache1 = &caches[-1].attr;
-            _PyObjectCache *cache2 = &caches[-2].obj;
-            DEOPT_IF(self_cls->tp_version_tag != cache1->tp_version, LOAD_METHOD);
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
+            uint32_t type_version = read_u32(cache->type_version);
+            DEOPT_IF(self_cls->tp_version_tag != type_version, LOAD_METHOD);
             assert(self_cls->tp_dictoffset == 0);
             STAT_INC(LOAD_METHOD, hit);
-            PyObject *res = cache2->obj;
+            PyObject *res = read_obj(cache->descr);
             assert(res != NULL);
             assert(_PyType_HasFeature(Py_TYPE(res), Py_TPFLAGS_METHOD_DESCRIPTOR));
             Py_INCREF(res);
             SET_TOP(res);
             PUSH(self);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             NOTRACE_DISPATCH();
         }
 
@@ -4532,29 +4513,30 @@ handle_eval_breaker:
             SET_TOP(NULL);
             Py_DECREF(owner);
             PUSH(res);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             NOTRACE_DISPATCH();
         }
 
         TARGET(LOAD_METHOD_CLASS) {
             /* LOAD_METHOD, for class methods */
             assert(cframe.use_tracing == 0);
-            SpecializedCacheEntry *caches = GET_CACHE();
-            _PyAttrCache *cache1 = &caches[-1].attr;
-            _PyObjectCache *cache2 = &caches[-2].obj;
+            _PyLoadMethodCache *cache = (_PyLoadMethodCache *)next_instr;
 
             PyObject *cls = TOP();
             DEOPT_IF(!PyType_Check(cls), LOAD_METHOD);
-            DEOPT_IF(((PyTypeObject *)cls)->tp_version_tag != cache1->tp_version,
-                LOAD_METHOD);
-            assert(cache1->tp_version != 0);
+            uint32_t type_version = read_u32(cache->type_version);
+            DEOPT_IF(((PyTypeObject *)cls)->tp_version_tag != type_version,
+                     LOAD_METHOD);
+            assert(type_version != 0);
 
             STAT_INC(LOAD_METHOD, hit);
-            PyObject *res = cache2->obj;
+            PyObject *res = read_obj(cache->descr);
             assert(res != NULL);
             Py_INCREF(res);
             SET_TOP(NULL);
             Py_DECREF(cls);
             PUSH(res);
+            JUMPBY(INLINE_CACHE_ENTRIES_LOAD_METHOD);
             NOTRACE_DISPATCH();
         }
 
@@ -5607,10 +5589,10 @@ opname ## _miss: \
         JUMP_TO_INSTRUCTION(opname); \
     }
 
-MISS_WITH_CACHE(LOAD_ATTR)
-MISS_WITH_CACHE(STORE_ATTR)
+MISS_WITH_INLINE_CACHE(LOAD_ATTR)
+MISS_WITH_INLINE_CACHE(STORE_ATTR)
 MISS_WITH_INLINE_CACHE(LOAD_GLOBAL)
-MISS_WITH_CACHE(LOAD_METHOD)
+MISS_WITH_INLINE_CACHE(LOAD_METHOD)
 MISS_WITH_CACHE(PRECALL)
 MISS_WITH_CACHE(CALL)
 MISS_WITH_INLINE_CACHE(BINARY_OP)
@@ -5619,52 +5601,6 @@ MISS_WITH_INLINE_CACHE(BINARY_SUBSCR)
 MISS_WITH_INLINE_CACHE(UNPACK_SEQUENCE)
 MISS_WITH_OPARG_COUNTER(STORE_SUBSCR)
 
-LOAD_ATTR_INSTANCE_VALUE_miss:
-        {
-            // Special-cased so that if LOAD_ATTR_INSTANCE_VALUE
-            // gets replaced, then any preceeding
-            // LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE gets replaced as well
-            STAT_INC(LOAD_ATTR_INSTANCE_VALUE, miss);
-            STAT_INC(LOAD_ATTR, miss);
-            _PyAdaptiveEntry *cache = &GET_CACHE()->adaptive;
-            cache->counter--;
-            if (cache->counter == 0) {
-                next_instr[-1] = _Py_MAKECODEUNIT(LOAD_ATTR_ADAPTIVE, _Py_OPARG(next_instr[-1]));
-                if (_Py_OPCODE(next_instr[-2]) == LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE) {
-                    next_instr[-2] = _Py_MAKECODEUNIT(LOAD_FAST, _Py_OPARG(next_instr[-2]));
-                    if (_Py_OPCODE(next_instr[-3]) == LOAD_FAST) {
-                        next_instr[-3] =  _Py_MAKECODEUNIT(LOAD_FAST__LOAD_FAST, _Py_OPARG(next_instr[-3]));
-                    }
-                }
-                STAT_INC(LOAD_ATTR, deopt);
-                cache_backoff(cache);
-            }
-            oparg = cache->original_oparg;
-            JUMP_TO_INSTRUCTION(LOAD_ATTR);
-        }
-
-LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE_miss:
-        {
-            // This is special-cased because we have a superinstruction
-            // that includes a specialized instruction.
-            // If the specialized portion misses, carry out
-            // the first instruction, then perform a miss
-            // for the second instruction as usual.
-
-            // Do LOAD_FAST
-            {
-                PyObject *value = GETLOCAL(oparg);
-                assert(value != NULL); // Already checked if unbound
-                Py_INCREF(value);
-                PUSH(value);
-                NEXTOPARG();
-                next_instr++;
-            }
-
-            // Now we are in the correct state for LOAD_ATTR
-            goto LOAD_ATTR_INSTANCE_VALUE_miss;
-        }
-
 binary_subscr_dict_error:
         {
             PyObject *sub = POP();
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index d463e30..7be7b16 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -175,7 +175,7 @@ static void *opcode_targets[256] = {
     &&TARGET_LOAD_FAST__LOAD_CONST,
     &&TARGET_LOAD_CONST__LOAD_FAST,
     &&TARGET_STORE_FAST__STORE_FAST,
-    &&TARGET_LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE,
+    &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
diff --git a/Python/specialize.c b/Python/specialize.c
index 4a94aaf..66dce8c 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -58,12 +58,9 @@ static uint8_t adaptive_opcodes[256] = {
 
 /* The number of cache entries required for a "family" of instructions. */
 static uint8_t cache_requirements[256] = {
-    [LOAD_ATTR] = 1,  // _PyAdaptiveEntry
-    [LOAD_METHOD] = 3, /* _PyAdaptiveEntry, _PyAttrCache and _PyObjectCache */
     [STORE_SUBSCR] = 0,
     [CALL] = 2, /* _PyAdaptiveEntry and _PyObjectCache/_PyCallCache */
     [PRECALL] = 2, /* _PyAdaptiveEntry and _PyObjectCache/_PyCallCache */
-    [STORE_ATTR] = 1,  // _PyAdaptiveEntry
 };
 
 Py_ssize_t _Py_QuickenedCount = 0;
@@ -641,11 +638,10 @@ initial_counter_value(void) {
 
 
 static int
-specialize_module_load_attr(
-    PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
-    _PyAdaptiveEntry *cache0, int opcode,
-    int opcode_module)
+specialize_module_load_attr(PyObject *owner, _Py_CODEUNIT *instr,
+                            PyObject *name, int opcode, int opcode_module)
 {
+    _PyAttrCache *cache = (_PyAttrCache *)(instr + 1);
     PyModuleObject *m = (PyModuleObject *)owner;
     PyObject *value = NULL;
     assert((owner->ob_type->tp_flags & Py_TPFLAGS_MANAGED_DICT) == 0);
@@ -676,8 +672,8 @@ specialize_module_load_attr(
         SPECIALIZATION_FAIL(opcode, SPEC_FAIL_OUT_OF_VERSIONS);
         return -1;
     }
-    cache0->version = keys_version;
-    cache0->index = (uint16_t)index;
+    write_u32(cache->version, keys_version);
+    cache->index = (uint16_t)index;
     *instr = _Py_MAKECODEUNIT(opcode_module, _Py_OPARG(*instr));
     return 0;
 }
@@ -765,7 +761,6 @@ static int
 specialize_dict_access(
     PyObject *owner, _Py_CODEUNIT *instr, PyTypeObject *type,
     DescriptorClassification kind, PyObject *name,
-    _PyAdaptiveEntry *cache0,
     int base_op, int values_op, int hint_op)
 {
     assert(kind == NON_OVERRIDING || kind == NON_DESCRIPTOR || kind == ABSENT ||
@@ -775,6 +770,7 @@ specialize_dict_access(
         SPECIALIZATION_FAIL(base_op, SPEC_FAIL_ATTR_NOT_MANAGED_DICT);
         return 0;
     }
+    _PyAttrCache *cache = (_PyAttrCache *)(instr + 1);
     PyObject **dictptr = _PyObject_ManagedDictPointer(owner);
     PyDictObject *dict = (PyDictObject *)*dictptr;
     if (dict == NULL) {
@@ -787,8 +783,8 @@ specialize_dict_access(
             SPECIALIZATION_FAIL(base_op, SPEC_FAIL_OUT_OF_RANGE);
             return 0;
         }
-        cache0->version = type->tp_version_tag;
-        cache0->index = (uint16_t)index;
+        write_u32(cache->version, type->tp_version_tag);
+        cache->index = (uint16_t)index;
         *instr = _Py_MAKECODEUNIT(values_op, _Py_OPARG(*instr));
     }
     else {
@@ -804,20 +800,22 @@ specialize_dict_access(
             SPECIALIZATION_FAIL(base_op, SPEC_FAIL_OUT_OF_RANGE);
             return 0;
         }
-        cache0->index = (uint16_t)hint;
-        cache0->version = type->tp_version_tag;
+        cache->index = (uint16_t)hint;
+        write_u32(cache->version, type->tp_version_tag);
         *instr = _Py_MAKECODEUNIT(hint_op, _Py_OPARG(*instr));
     }
     return 1;
 }
 
 int
-_Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache)
+_Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name)
 {
-    _PyAdaptiveEntry *cache0 = &cache->adaptive;
+    assert(_PyOpcode_InlineCacheEntries[LOAD_ATTR] ==
+           INLINE_CACHE_ENTRIES_LOAD_ATTR);
+    _PyAttrCache *cache = (_PyAttrCache *)(instr + 1);
     if (PyModule_CheckExact(owner)) {
-        int err = specialize_module_load_attr(owner, instr, name, cache0,
-            LOAD_ATTR, LOAD_ATTR_MODULE);
+        int err = specialize_module_load_attr(owner, instr, name, LOAD_ATTR,
+                                              LOAD_ATTR_MODULE);
         if (err) {
             goto fail;
         }
@@ -856,8 +854,8 @@ _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, Sp
             }
             assert(dmem->type == T_OBJECT_EX);
             assert(offset > 0);
-            cache0->index = (uint16_t)offset;
-            cache0->version = type->tp_version_tag;
+            cache->index = (uint16_t)offset;
+            write_u32(cache->version, type->tp_version_tag);
             *instr = _Py_MAKECODEUNIT(LOAD_ATTR_SLOT, _Py_OPARG(*instr));
             goto success;
         }
@@ -865,8 +863,8 @@ _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, Sp
         {
             Py_ssize_t offset = offsetof(PyObject, ob_type);
             assert(offset == (uint16_t)offset);
-            cache0->index = (uint16_t)offset;
-            cache0->version = type->tp_version_tag;
+            cache->index = (uint16_t)offset;
+            write_u32(cache->version, type->tp_version_tag);
             *instr = _Py_MAKECODEUNIT(LOAD_ATTR_SLOT, _Py_OPARG(*instr));
             goto success;
         }
@@ -887,41 +885,33 @@ _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, Sp
             break;
     }
     int err = specialize_dict_access(
-        owner, instr, type, kind, name, cache0,
+        owner, instr, type, kind, name,
         LOAD_ATTR, LOAD_ATTR_INSTANCE_VALUE, LOAD_ATTR_WITH_HINT
     );
     if (err < 0) {
         return -1;
     }
     if (err) {
-        if (_Py_OPCODE(instr[0]) == LOAD_ATTR_INSTANCE_VALUE) {
-            // Note: instr[-1] exists because there's something on the stack,
-            // and instr[-2] exists because there's at least a RESUME as well.
-            if (_Py_OPCODE(instr[-1]) == LOAD_FAST) {
-                instr[-1] = _Py_MAKECODEUNIT(LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE, _Py_OPARG(instr[-1]));
-                if (_Py_OPCODE(instr[-2]) == LOAD_FAST__LOAD_FAST) {
-                    instr[-2] = _Py_MAKECODEUNIT(LOAD_FAST, _Py_OPARG(instr[-2]));
-                }
-            }
-        }
         goto success;
     }
 fail:
     STAT_INC(LOAD_ATTR, failure);
     assert(!PyErr_Occurred());
-    cache_backoff(cache0);
+    cache->counter = ADAPTIVE_CACHE_BACKOFF;
     return 0;
 success:
     STAT_INC(LOAD_ATTR, success);
     assert(!PyErr_Occurred());
-    cache0->counter = initial_counter_value();
+    cache->counter = initial_counter_value();
     return 0;
 }
 
 int
-_Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache)
+_Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name)
 {
-    _PyAdaptiveEntry *cache0 = &cache->adaptive;
+    assert(_PyOpcode_InlineCacheEntries[STORE_ATTR] ==
+           INLINE_CACHE_ENTRIES_STORE_ATTR);
+    _PyAttrCache *cache = (_PyAttrCache *)(instr + 1);
     PyTypeObject *type = Py_TYPE(owner);
     if (PyModule_CheckExact(owner)) {
         SPECIALIZATION_FAIL(STORE_ATTR, SPEC_FAIL_OVERRIDDEN);
@@ -954,8 +944,8 @@ _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, S
             }
             assert(dmem->type == T_OBJECT_EX);
             assert(offset > 0);
-            cache0->index = (uint16_t)offset;
-            cache0->version = type->tp_version_tag;
+            cache->index = (uint16_t)offset;
+            write_u32(cache->version, type->tp_version_tag);
             *instr = _Py_MAKECODEUNIT(STORE_ATTR_SLOT, _Py_OPARG(*instr));
             goto success;
         }
@@ -978,7 +968,7 @@ _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, S
     }
 
     int err = specialize_dict_access(
-        owner, instr, type, kind, name, cache0,
+        owner, instr, type, kind, name,
         STORE_ATTR, STORE_ATTR_INSTANCE_VALUE, STORE_ATTR_WITH_HINT
     );
     if (err < 0) {
@@ -990,12 +980,12 @@ _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, S
 fail:
     STAT_INC(STORE_ATTR, failure);
     assert(!PyErr_Occurred());
-    cache_backoff(cache0);
+    cache->counter = ADAPTIVE_CACHE_BACKOFF;
     return 0;
 success:
     STAT_INC(STORE_ATTR, success);
     assert(!PyErr_Occurred());
-    cache0->counter = initial_counter_value();
+    cache->counter = initial_counter_value();
     return 0;
 }
 
@@ -1037,18 +1027,18 @@ load_method_fail_kind(DescriptorClassification kind)
 #endif
 
 static int
-specialize_class_load_method(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
-                           _PyAttrCache *cache1, _PyObjectCache *cache2)
+specialize_class_load_method(PyObject *owner, _Py_CODEUNIT *instr,
+                             PyObject *name)
 {
-
+    _PyLoadMethodCache *cache = (_PyLoadMethodCache *)(instr + 1);
     PyObject *descr = NULL;
     DescriptorClassification kind = 0;
     kind = analyze_descriptor((PyTypeObject *)owner, name, &descr, 0);
     switch (kind) {
         case METHOD:
         case NON_DESCRIPTOR:
-            cache1->tp_version = ((PyTypeObject *)owner)->tp_version_tag;
-            cache2->obj = descr;
+            write_u32(cache->type_version, ((PyTypeObject *)owner)->tp_version_tag);
+            write_obj(cache->descr, descr);
             *instr = _Py_MAKECODEUNIT(LOAD_METHOD_CLASS, _Py_OPARG(*instr));
             return 0;
 #ifdef Py_STATS
@@ -1078,16 +1068,18 @@ typedef enum {
 // can cause a significant drop in cache hits. A possible test is
 // python.exe -m test_typing test_re test_dis test_zlib.
 int
-_Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache)
+_Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name)
 {
-    _PyAdaptiveEntry *cache0 = &cache->adaptive;
-    _PyAttrCache *cache1 = &cache[-1].attr;
-    _PyObjectCache *cache2 = &cache[-2].obj;
+    assert(_PyOpcode_InlineCacheEntries[LOAD_METHOD] ==
+           INLINE_CACHE_ENTRIES_LOAD_METHOD);
+    _PyLoadMethodCache *cache = (_PyLoadMethodCache *)(instr + 1);
     PyTypeObject *owner_cls = Py_TYPE(owner);
 
     if (PyModule_CheckExact(owner)) {
-        int err = specialize_module_load_attr(owner, instr, name, cache0,
-            LOAD_METHOD, LOAD_METHOD_MODULE);
+        assert(INLINE_CACHE_ENTRIES_LOAD_ATTR <= 
+               INLINE_CACHE_ENTRIES_LOAD_METHOD);
+        int err = specialize_module_load_attr(owner, instr, name, LOAD_METHOD,
+                                              LOAD_METHOD_MODULE);
         if (err) {
             goto fail;
         }
@@ -1099,7 +1091,7 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
         }
     }
     if (PyType_Check(owner)) {
-        int err = specialize_class_load_method(owner, instr, name, cache1, cache2);
+        int err = specialize_class_load_method(owner, instr, name);
         if (err) {
             goto fail;
         }
@@ -1157,7 +1149,7 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
             SPECIALIZATION_FAIL(LOAD_METHOD, SPEC_FAIL_OUT_OF_VERSIONS);
             goto fail;
         }
-        cache1->dk_version = keys_version;
+        write_u32(cache->keys_version, keys_version);
     }
     switch(dictkind) {
         case NO_DICT:
@@ -1167,12 +1159,12 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
             *instr = _Py_MAKECODEUNIT(LOAD_METHOD_WITH_VALUES, _Py_OPARG(*instr));
             break;
         case MANAGED_DICT:
-            *(int16_t *)&cache0->index = (int16_t)MANAGED_DICT_OFFSET;
+            *(int16_t *)&cache->dict_offset = (int16_t)MANAGED_DICT_OFFSET;
             *instr = _Py_MAKECODEUNIT(LOAD_METHOD_WITH_DICT, _Py_OPARG(*instr));
             break;
         case OFFSET_DICT:
             assert(owner_cls->tp_dictoffset > 0 && owner_cls->tp_dictoffset <= INT16_MAX);
-            cache0->index = (uint16_t)owner_cls->tp_dictoffset;
+            cache->dict_offset = (uint16_t)owner_cls->tp_dictoffset;
             *instr = _Py_MAKECODEUNIT(LOAD_METHOD_WITH_DICT, _Py_OPARG(*instr));
             break;
     }
@@ -1190,18 +1182,18 @@ _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name,
     *  PyType_Modified usages in typeobject.c). The MCACHE has been
     *  working since Python 2.6 and it's battle-tested.
     */
-    cache1->tp_version = owner_cls->tp_version_tag;
-    cache2->obj = descr;
+    write_u32(cache->type_version, owner_cls->tp_version_tag);
+    write_obj(cache->descr, descr);
     // Fall through.
 success:
     STAT_INC(LOAD_METHOD, success);
     assert(!PyErr_Occurred());
-    cache0->counter = initial_counter_value();
+    cache->counter = initial_counter_value();
     return 0;
 fail:
     STAT_INC(LOAD_METHOD, failure);
     assert(!PyErr_Occurred());
-    cache_backoff(cache0);
+    cache->counter = ADAPTIVE_CACHE_BACKOFF;
     return 0;
 
 }
@@ -1238,7 +1230,7 @@ _Py_Specialize_LoadGlobal(
             goto fail;
         }
         cache->index = (uint16_t)index;
-        write32(&cache->module_keys_version, keys_version);
+        write_u32(cache->module_keys_version, keys_version);
         *instr = _Py_MAKECODEUNIT(LOAD_GLOBAL_MODULE, _Py_OPARG(*instr));
         goto success;
     }
@@ -1273,7 +1265,7 @@ _Py_Specialize_LoadGlobal(
         goto fail;
     }
     cache->index = (uint16_t)index;
-    write32(&cache->module_keys_version, globals_version);
+    write_u32(cache->module_keys_version, globals_version);
     cache->builtin_keys_version = (uint16_t)builtins_version;
     *instr = _Py_MAKECODEUNIT(LOAD_GLOBAL_BUILTIN, _Py_OPARG(*instr));
     goto success;
@@ -1393,7 +1385,7 @@ _Py_Specialize_BinarySubscr(
             goto fail;
         }
         assert(cls->tp_version_tag != 0);
-        write32(&cache->type_version, cls->tp_version_tag);
+        write_u32(cache->type_version, cls->tp_version_tag);
         int version = _PyFunction_GetVersionForCurrentState(func);
         if (version == 0 || version != (uint16_t)version) {
             SPECIALIZATION_FAIL(BINARY_SUBSCR, SPEC_FAIL_OUT_OF_VERSIONS);