pythonGH-90997: Shrink the LOAD_GLOBAL caches (python#102569)

brandtbucher · web-flow · commit 08b67fb34f45 · 2023-03-10T17:01:16.000-08:00
diff --git a/Doc/library/dis.rst b/Doc/library/dis.rst
@@ -57,9 +57,9 @@ the following command can be used to display the disassembly of
      2           0 RESUME                   0
    <BLANKLINE>
      3           2 LOAD_GLOBAL              1 (NULL + len)
-                14 LOAD_FAST                0 (alist)
-                16 CALL                     1
-                26 RETURN_VALUE
+                12 LOAD_FAST                0 (alist)
+                14 CALL                     1
+                24 RETURN_VALUE
 
 (The "2" is a line number).
 
diff --git a/Include/internal/pycore_code.h b/Include/internal/pycore_code.h
@@ -20,7 +20,7 @@ extern "C" {
 typedef struct {
     uint16_t counter;
     uint16_t index;
-    uint16_t module_keys_version[2];
+    uint16_t module_keys_version;
     uint16_t builtin_keys_version;
 } _PyLoadGlobalCache;
 
diff --git a/Include/internal/pycore_opcode.h b/Include/internal/pycore_opcode.h
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
@@ -431,12 +431,17 @@ def _write_atomic(path, data, mode=0o666):
 #     Python 3.12a5 3515 (Embed jump mask in COMPARE_OP oparg)
 #     Python 3.12a5 3516 (Add COMPARE_AND_BRANCH instruction)
 #     Python 3.12a5 3517 (Change YIELD_VALUE oparg to exception block depth)
-#     Python 3.12a5 3518 (Add RETURN_CONST instruction)
-#     Python 3.12a5 3519 (Modify SEND instruction)
-#     Python 3.12a5 3520 (Remove PREP_RERAISE_STAR, add CALL_INTRINSIC_2)
+#     Python 3.12a6 3518 (Add RETURN_CONST instruction)
+#     Python 3.12a6 3519 (Modify SEND instruction)
+#     Python 3.12a6 3520 (Remove PREP_RERAISE_STAR, add CALL_INTRINSIC_2)
+#     Python 3.12a7 3521 (Shrink the LOAD_GLOBAL caches)
 
 #     Python 3.13 will start with 3550
 
+#     Please don't copy-paste the same pre-release tag for new entries above!!!
+#     You should always use the *upcoming* tag. For example, if 3.12a6 came out
+#     a week ago, I should put "Python 3.12a7" next to my new magic number.
+
 # MAGIC must change whenever the bytecode emitted by the compiler may no
 # longer be understood by older implementations of the eval loop (usually
 # due to the addition of new opcodes).
@@ -446,7 +451,7 @@ def _write_atomic(path, data, mode=0o666):
 # Whenever MAGIC_NUMBER is changed, the ranges in the magic_values array
 # in PC/launcher.c must also be updated.
 
-MAGIC_NUMBER = (3520).to_bytes(2, 'little') + b'\r\n'
+MAGIC_NUMBER = (3521).to_bytes(2, 'little') + b'\r\n'
 
 _RAW_MAGIC_NUMBER = int.from_bytes(MAGIC_NUMBER, 'little')  # For import.c
 
diff --git a/Lib/opcode.py b/Lib/opcode.py
@@ -390,7 +390,7 @@ def pseudo_op(name, op, real_ops):
     "LOAD_GLOBAL": {
         "counter": 1,
         "index": 1,
-        "module_keys_version": 2,
+        "module_keys_version": 1,
         "builtin_keys_version": 1,
     },
     "BINARY_OP": {
diff --git a/Lib/test/test_dis.py b/Lib/test/test_dis.py
diff --git a/Misc/NEWS.d/next/Core and Builtins/2023-03-09-13-57-35.gh-issue-90997.J-Yhn2.rst b/Misc/NEWS.d/next/Core and Builtins/2023-03-09-13-57-35.gh-issue-90997.J-Yhn2.rst
@@ -0,0 +1,2 @@
+Shrink the number of inline :opcode:`CACHE` entries used by
+:opcode:`LOAD_GLOBAL`.
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
@@ -1078,7 +1078,7 @@ dummy_func(
             LOAD_GLOBAL_BUILTIN,
         };
 
-        inst(LOAD_GLOBAL, (unused/1, unused/1, unused/2, unused/1 -- null if (oparg & 1), v)) {
+        inst(LOAD_GLOBAL, (unused/1, unused/1, unused/1, unused/1 -- null if (oparg & 1), v)) {
             #if ENABLE_SPECIALIZATION
             _PyLoadGlobalCache *cache = (_PyLoadGlobalCache *)next_instr;
             if (ADAPTIVE_COUNTER_IS_ZERO(cache->counter)) {
@@ -1133,7 +1133,7 @@ dummy_func(
             null = NULL;
         }
 
-        inst(LOAD_GLOBAL_MODULE, (unused/1, index/1, version/2, unused/1 -- null if (oparg & 1), res)) {
+        inst(LOAD_GLOBAL_MODULE, (unused/1, index/1, version/1, unused/1 -- null if (oparg & 1), res)) {
             assert(cframe.use_tracing == 0);
             DEOPT_IF(!PyDict_CheckExact(GLOBALS()), LOAD_GLOBAL);
             PyDictObject *dict = (PyDictObject *)GLOBALS();
@@ -1147,7 +1147,7 @@ dummy_func(
             null = NULL;
         }
 
-        inst(LOAD_GLOBAL_BUILTIN, (unused/1, index/1, mod_version/2, bltn_version/1 -- null if (oparg & 1), res)) {
+        inst(LOAD_GLOBAL_BUILTIN, (unused/1, index/1, mod_version/1, bltn_version/1 -- null if (oparg & 1), res)) {
             assert(cframe.use_tracing == 0);
             DEOPT_IF(!PyDict_CheckExact(GLOBALS()), LOAD_GLOBAL);
             DEOPT_IF(!PyDict_CheckExact(BUILTINS()), LOAD_GLOBAL);
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
diff --git a/Python/opcode_metadata.h b/Python/opcode_metadata.h
@@ -708,7 +708,7 @@ _PyOpcode_num_pushed(int opcode, int oparg, bool jump) {
 #endif
 
 enum Direction { DIR_NONE, DIR_READ, DIR_WRITE };
-enum InstructionFormat { INSTR_FMT_IB, INSTR_FMT_IBC, INSTR_FMT_IBC0, INSTR_FMT_IBC000, INSTR_FMT_IBC0000, INSTR_FMT_IBC00000000, INSTR_FMT_IBIB, INSTR_FMT_IX, INSTR_FMT_IXC, INSTR_FMT_IXC000 };
+enum InstructionFormat { INSTR_FMT_IB, INSTR_FMT_IBC, INSTR_FMT_IBC0, INSTR_FMT_IBC000, INSTR_FMT_IBC00000000, INSTR_FMT_IBIB, INSTR_FMT_IX, INSTR_FMT_IXC, INSTR_FMT_IXC000 };
 struct opcode_metadata {
     enum Direction dir_op1;
     enum Direction dir_op2;
@@ -790,9 +790,9 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[256] = {
     [STORE_GLOBAL] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [DELETE_GLOBAL] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [LOAD_NAME] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
-    [LOAD_GLOBAL] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0000 },
-    [LOAD_GLOBAL_MODULE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0000 },
-    [LOAD_GLOBAL_BUILTIN] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0000 },
+    [LOAD_GLOBAL] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC000 },
+    [LOAD_GLOBAL_MODULE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC000 },
+    [LOAD_GLOBAL_BUILTIN] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC000 },
     [DELETE_FAST] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [MAKE_CELL] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [DELETE_DEREF] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
diff --git a/Python/specialize.c b/Python/specialize.c
@@ -1148,8 +1148,12 @@ _Py_Specialize_LoadGlobal(
             SPECIALIZATION_FAIL(LOAD_GLOBAL, SPEC_FAIL_OUT_OF_VERSIONS);
             goto fail;
         }
+        if (keys_version != (uint16_t)keys_version) {
+            SPECIALIZATION_FAIL(LOAD_GLOBAL, SPEC_FAIL_OUT_OF_RANGE);
+            goto fail;
+        }
         cache->index = (uint16_t)index;
-        write_u32(cache->module_keys_version, keys_version);
+        cache->module_keys_version = (uint16_t)keys_version;
         instr->op.code = LOAD_GLOBAL_MODULE;
         goto success;
     }
@@ -1177,6 +1181,10 @@ _Py_Specialize_LoadGlobal(
         SPECIALIZATION_FAIL(LOAD_GLOBAL, SPEC_FAIL_OUT_OF_VERSIONS);
         goto fail;
     }
+    if (globals_version != (uint16_t)globals_version) {
+        SPECIALIZATION_FAIL(LOAD_GLOBAL, SPEC_FAIL_OUT_OF_RANGE);
+        goto fail;
+    }
     uint32_t builtins_version = _PyDictKeys_GetVersionForCurrentState(
             interp, builtin_keys);
     if (builtins_version == 0) {
@@ -1188,7 +1196,7 @@ _Py_Specialize_LoadGlobal(
         goto fail;
     }
     cache->index = (uint16_t)index;
-    write_u32(cache->module_keys_version, globals_version);
+    cache->module_keys_version = (uint16_t)globals_version;
     cache->builtin_keys_version = (uint16_t)builtins_version;
     instr->op.code = LOAD_GLOBAL_BUILTIN;
     goto success;

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1,2 @@`
	`1`	+Shrink the number of inline :opcode:`CACHE` entries used by
	`2`	+:opcode:`LOAD_GLOBAL`.