From fe3c1af26e82dc9015950ed325089fe564187756 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:14:31 +0100
Subject: [PATCH 01/30] gh-137838: Move _PyUOpInstruction
 buffer[UOP_MAX_TRACE_LENGTH] to _PyThreadStateImpl

---
 Include/internal/pycore_optimizer.h           | 30 +----------------
 Include/internal/pycore_tstate.h              | 32 ++++++++++++++++++-
 ...-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst |  2 ++
 Python/optimizer.c                            |  3 +-
 4 files changed, 36 insertions(+), 31 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst

diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
index 94d01999f68d9d..4c05c4c1df0ca9 100644
--- a/Include/internal/pycore_optimizer.h
+++ b/Include/internal/pycore_optimizer.h
@@ -11,6 +11,7 @@ extern "C" {
 #include "pycore_typedefs.h"      // _PyInterpreterFrame
 #include "pycore_uop_ids.h"
 #include "pycore_stackref.h"      // _PyStackRef
+#include "pycore_tstate.h"       // _PyUOpInstruction
 #include <stdbool.h>
 
 
@@ -41,32 +42,6 @@ typedef struct {
     PyCodeObject *code;  // Weak (NULL if no corresponding ENTER_EXECUTOR).
 } _PyVMData;
 
-/* Depending on the format,
- * the 32 bits between the oparg and operand are:
- * UOP_FORMAT_TARGET:
- *    uint32_t target;
- * UOP_FORMAT_JUMP
- *    uint16_t jump_target;
- *    uint16_t error_target;
- */
-typedef struct {
-    uint16_t opcode:15;
-    uint16_t format:1;
-    uint16_t oparg;
-    union {
-        uint32_t target;
-        struct {
-            uint16_t jump_target;
-            uint16_t error_target;
-        };
-    };
-    uint64_t operand0;  // A cache entry
-    uint64_t operand1;
-#ifdef Py_STATS
-    uint64_t execution_count;
-#endif
-} _PyUOpInstruction;
-
 typedef struct _PyExitData {
     uint32_t target;
     uint16_t index;
@@ -118,9 +93,6 @@ PyAPI_FUNC(void) _Py_Executors_InvalidateCold(PyInterpreterState *interp);
 // trace_run_counter is greater than this value.
 #define JIT_CLEANUP_THRESHOLD 100000
 
-// This is the length of the trace we project initially.
-#define UOP_MAX_TRACE_LENGTH 1200
-
 #define TRACE_STACK_SIZE 5
 
 int _Py_uop_analyze_and_optimize(_PyInterpreterFrame *frame,
diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index bad968428c73a1..5291b70b8662c6 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -13,7 +13,6 @@ extern "C" {
 #include "pycore_mimalloc.h"        // struct _mimalloc_thread_state
 #include "pycore_qsbr.h"            // struct qsbr
 
-
 #ifdef Py_GIL_DISABLED
 struct _gc_thread_state {
     /* Thread-local allocation count. */
@@ -21,6 +20,36 @@ struct _gc_thread_state {
 };
 #endif
 
+/* Depending on the format,
+ * the 32 bits between the oparg and operand are:
+ * UOP_FORMAT_TARGET:
+ *    uint32_t target;
+ * UOP_FORMAT_JUMP
+ *    uint16_t jump_target;
+ *    uint16_t error_target;
+ */
+typedef struct _PyUOpInstruction{
+    uint16_t opcode:15;
+    uint16_t format:1;
+    uint16_t oparg;
+    union {
+        uint32_t target;
+        struct {
+            uint16_t jump_target;
+            uint16_t error_target;
+        };
+    };
+    uint64_t operand0;  // A cache entry
+    uint64_t operand1;
+#ifdef Py_STATS
+    uint64_t execution_count;
+#endif
+} _PyUOpInstruction;
+
+// This is the length of the trace we project initially.
+#define UOP_MAX_TRACE_LENGTH 1200
+
+
 // Every PyThreadState is actually allocated as a _PyThreadStateImpl. The
 // PyThreadState fields are exposed as part of the C API, although most fields
 // are intended to be private. The _PyThreadStateImpl fields not exposed.
@@ -75,6 +104,7 @@ typedef struct _PyThreadStateImpl {
 #if defined(Py_REF_DEBUG) && defined(Py_GIL_DISABLED)
     Py_ssize_t reftotal;  // this thread's total refcount operations
 #endif
+    struct _PyUOpInstruction buffer[1200];
 
 } _PyThreadStateImpl;
 
diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst b/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst
new file mode 100644
index 00000000000000..fdede2efd1ae01
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst
@@ -0,0 +1,2 @@
+Move _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH] to _PyThreadStateImpl.
+Patch By Donghee Na.
diff --git a/Python/optimizer.c b/Python/optimizer.c
index b82c790ffa9e69..6a895a907cbd80 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1280,7 +1280,8 @@ uop_optimize(
 {
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
-    _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
+    _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
+    _PyUOpInstruction *buffer = tstate->buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
     bool is_noopt = true;

From cd868b9eec9d4e8bd8cbbd413129bbd7dee998ea Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:16:29 +0100
Subject: [PATCH 02/30] nit

---
 Include/internal/pycore_tstate.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index 5291b70b8662c6..c14cc212ce4aa6 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -104,7 +104,10 @@ typedef struct _PyThreadStateImpl {
 #if defined(Py_REF_DEBUG) && defined(Py_GIL_DISABLED)
     Py_ssize_t reftotal;  // this thread's total refcount operations
 #endif
-    struct _PyUOpInstruction buffer[1200];
+
+#ifdef _Py_TIER2
+    struct _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
+#endif
 
 } _PyThreadStateImpl;
 

From daf47305a0a5938ad6c217acb6f13e21db6590dc Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:18:41 +0100
Subject: [PATCH 03/30] nit

---
 Include/internal/pycore_tstate.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index c14cc212ce4aa6..9ee7c941890ba0 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -13,6 +13,7 @@ extern "C" {
 #include "pycore_mimalloc.h"        // struct _mimalloc_thread_state
 #include "pycore_qsbr.h"            // struct qsbr
 
+
 #ifdef Py_GIL_DISABLED
 struct _gc_thread_state {
     /* Thread-local allocation count. */

From e09af56ae2b059666af95a1765c4592989b0311b Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:43:10 +0100
Subject: [PATCH 04/30] Update

---
 .../2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst              | 2 --
 1 file changed, 2 deletions(-)
 delete mode 100644 Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst

diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst b/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst
deleted file mode 100644
index fdede2efd1ae01..00000000000000
--- a/Misc/NEWS.d/next/Core_and_Builtins/2025-09-15-13-14-20.gh-issue-137838.Ju-vRW.rst
+++ /dev/null
@@ -1,2 +0,0 @@
-Move _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH] to _PyThreadStateImpl.
-Patch By Donghee Na.

From be0b25b0a7415660b441d74bdd5b70897880662f Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:59:04 +0100
Subject: [PATCH 05/30] Address code review

---
 Include/internal/pycore_tstate.h   | 30 +-----------------------------
 Makefile.pre.in                    |  1 +
 PCbuild/pythoncore.vcxproj.filters |  9 +++++++++
 3 files changed, 11 insertions(+), 29 deletions(-)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index 9ee7c941890ba0..fd3a2928b8cab3 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -12,6 +12,7 @@ extern "C" {
 #include "pycore_freelist_state.h"  // struct _Py_freelists
 #include "pycore_mimalloc.h"        // struct _mimalloc_thread_state
 #include "pycore_qsbr.h"            // struct qsbr
+#include "pycore_uop.h"             // struct _PyUOpInstruction
 
 
 #ifdef Py_GIL_DISABLED
@@ -21,35 +22,6 @@ struct _gc_thread_state {
 };
 #endif
 
-/* Depending on the format,
- * the 32 bits between the oparg and operand are:
- * UOP_FORMAT_TARGET:
- *    uint32_t target;
- * UOP_FORMAT_JUMP
- *    uint16_t jump_target;
- *    uint16_t error_target;
- */
-typedef struct _PyUOpInstruction{
-    uint16_t opcode:15;
-    uint16_t format:1;
-    uint16_t oparg;
-    union {
-        uint32_t target;
-        struct {
-            uint16_t jump_target;
-            uint16_t error_target;
-        };
-    };
-    uint64_t operand0;  // A cache entry
-    uint64_t operand1;
-#ifdef Py_STATS
-    uint64_t execution_count;
-#endif
-} _PyUOpInstruction;
-
-// This is the length of the trace we project initially.
-#define UOP_MAX_TRACE_LENGTH 1200
-
 
 // Every PyThreadState is actually allocated as a _PyThreadStateImpl. The
 // PyThreadState fields are exposed as part of the C API, although most fields
diff --git a/Makefile.pre.in b/Makefile.pre.in
index 34bd4540efb0b8..f1fe50c88ebde6 100644
--- a/Makefile.pre.in
+++ b/Makefile.pre.in
@@ -1435,6 +1435,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_unicodeobject_generated.h \
 		$(srcdir)/Include/internal/pycore_unionobject.h \
 		$(srcdir)/Include/internal/pycore_uniqueid.h \
+		$(srcdir)/Include/internal/pycore_uop.h \
 		$(srcdir)/Include/internal/pycore_uop_ids.h \
 		$(srcdir)/Include/internal/pycore_uop_metadata.h \
 		$(srcdir)/Include/internal/pycore_warnings.h \
diff --git a/PCbuild/pythoncore.vcxproj.filters b/PCbuild/pythoncore.vcxproj.filters
index e9eedfd1312fae..1868b222f18534 100644
--- a/PCbuild/pythoncore.vcxproj.filters
+++ b/PCbuild/pythoncore.vcxproj.filters
@@ -882,6 +882,15 @@
     <ClInclude Include="..\Include\internal\pycore_uniqueid.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_uop.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_uop_ids.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_uop_metadata.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
     <ClInclude Include="..\Include\internal\mimalloc\mimalloc.h">
       <Filter>Include\internal\mimalloc</Filter>
     </ClInclude>

From 61b32eea977987c0ba051010160f2c7895fa8971 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 13:59:47 +0100
Subject: [PATCH 06/30] Add Include/internal/pycore_uop.h

---
 Include/internal/pycore_uop.h | 44 +++++++++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)
 create mode 100644 Include/internal/pycore_uop.h

diff --git a/Include/internal/pycore_uop.h b/Include/internal/pycore_uop.h
new file mode 100644
index 00000000000000..79c2bd355e3700
--- /dev/null
+++ b/Include/internal/pycore_uop.h
@@ -0,0 +1,44 @@
+#ifndef Py_CORE_UOP_H
+#define Py_CORE_UOP_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef Py_BUILD_CORE
+#  error "this header requires Py_BUILD_CORE define"
+#endif
+
+#include <stdint.h> 
+/* Depending on the format,
+ * the 32 bits between the oparg and operand are:
+ * UOP_FORMAT_TARGET:
+ *    uint32_t target;
+ * UOP_FORMAT_JUMP
+ *    uint16_t jump_target;
+ *    uint16_t error_target;
+ */
+typedef struct _PyUOpInstruction{
+    uint16_t opcode:15;
+    uint16_t format:1;
+    uint16_t oparg;
+    union {
+        uint32_t target;
+        struct {
+            uint16_t jump_target;
+            uint16_t error_target;
+        };
+    };
+    uint64_t operand0;  // A cache entry
+    uint64_t operand1;
+#ifdef Py_STATS
+    uint64_t execution_count;
+#endif
+} _PyUOpInstruction;
+
+// This is the length of the trace we project initially.
+#define UOP_MAX_TRACE_LENGTH 1200
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* !Py_INTERNAL_UOP_H */

From a39b4c1a5abe802c0234673bf51c79676ea5ed14 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 14:04:02 +0100
Subject: [PATCH 07/30] Update

---
 Include/internal/pycore_uop.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Include/internal/pycore_uop.h b/Include/internal/pycore_uop.h
index 79c2bd355e3700..b569c46ccb926f 100644
--- a/Include/internal/pycore_uop.h
+++ b/Include/internal/pycore_uop.h
@@ -8,7 +8,7 @@ extern "C" {
 #  error "this header requires Py_BUILD_CORE define"
 #endif
 
-#include <stdint.h> 
+#include <stdint.h>
 /* Depending on the format,
  * the 32 bits between the oparg and operand are:
  * UOP_FORMAT_TARGET:

From f32f7a70e7c3d063c4633f3c407dbd3053c3ef46 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na92@gmail.com>
Date: Mon, 15 Sep 2025 14:10:08 +0100
Subject: [PATCH 08/30] Update Include/internal/pycore_tstate.h

---
 Include/internal/pycore_tstate.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index fd3a2928b8cab3..b4177261d694ee 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -22,7 +22,6 @@ struct _gc_thread_state {
 };
 #endif
 
-
 // Every PyThreadState is actually allocated as a _PyThreadStateImpl. The
 // PyThreadState fields are exposed as part of the C API, although most fields
 // are intended to be private. The _PyThreadStateImpl fields not exposed.

From fd15eb866ffa1ea5dd9236ebc264f14fcdf62812 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 15:37:07 +0100
Subject: [PATCH 09/30] nit

---
 Include/internal/pycore_optimizer.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
index 4c05c4c1df0ca9..188abab8c9625a 100644
--- a/Include/internal/pycore_optimizer.h
+++ b/Include/internal/pycore_optimizer.h
@@ -11,7 +11,7 @@ extern "C" {
 #include "pycore_typedefs.h"      // _PyInterpreterFrame
 #include "pycore_uop_ids.h"
 #include "pycore_stackref.h"      // _PyStackRef
-#include "pycore_tstate.h"       // _PyUOpInstruction
+#include "pycore_uop.h"       // _PyUOpInstruction
 #include <stdbool.h>
 
 

From 471832b1d3c6d9b8467d636f5b3adb505c9c303b Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Mon, 15 Sep 2025 15:37:36 +0100
Subject: [PATCH 10/30] nit

---
 Include/internal/pycore_optimizer.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
index 188abab8c9625a..c1a6b7abbf5de4 100644
--- a/Include/internal/pycore_optimizer.h
+++ b/Include/internal/pycore_optimizer.h
@@ -9,9 +9,9 @@ extern "C" {
 #endif
 
 #include "pycore_typedefs.h"      // _PyInterpreterFrame
+#include "pycore_uop.h"           // _PyUOpInstruction
 #include "pycore_uop_ids.h"
 #include "pycore_stackref.h"      // _PyStackRef
-#include "pycore_uop.h"       // _PyUOpInstruction
 #include <stdbool.h>
 
 

From d12b70596db96107082bd411242834f077aff57a Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 11:58:02 +0100
Subject: [PATCH 11/30] Use lazy allocation

---
 Include/internal/pycore_tstate.h | 2 +-
 Python/optimizer.c               | 7 +++++++
 Python/pylifecycle.c             | 5 +++++
 3 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index b4177261d694ee..25b6b16918041f 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -78,7 +78,7 @@ typedef struct _PyThreadStateImpl {
 #endif
 
 #ifdef _Py_TIER2
-    struct _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
+    struct _PyUOpInstruction *buffer;
 #endif
 
 } _PyThreadStateImpl;
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 6a895a907cbd80..16874373931371 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1281,6 +1281,13 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
+    if (tstate->buffer == NULL) {
+        tstate->buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
+        if (tstate->buffer == NULL) {
+            PyErr_NoMemory();
+            return -1;
+        }
+    }
     _PyUOpInstruction *buffer = tstate->buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index 51a777077d8255..eab92e8b80901d 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -1704,6 +1704,11 @@ finalize_modules(PyThreadState *tstate)
     interp->jit = false;
 #ifdef _Py_TIER2
     _Py_Executors_InvalidateAll(interp, 0);
+    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
+    if (_tstate->buffer != NULL) {
+        PyMem_RawFree(_tstate->buffer);
+        _tstate->buffer = NULL;
+    }
 #endif
 
     // Stop watching __builtin__ modifications

From c628ffc5efe2def222e097614dddd27fbe1fa5ad Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 13:51:34 +0100
Subject: [PATCH 12/30] Address code review

---
 Python/pylifecycle.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index eab92e8b80901d..a13bd532b91148 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -863,6 +863,10 @@ pycore_interp_init(PyThreadState *tstate)
     if (_tstate->c_stack_hard_limit == 0) {
         _Py_InitializeRecursionLimits(tstate);
     }
+#ifdef _Py_TIER2
+     // Ensure the buffer is to be set as NULL for MSVC
+    _tstate->buffer = NULL;
+#endif
     PyInterpreterState *interp = tstate->interp;
     PyStatus status;
     PyObject *sysmod = NULL;

From b17541646c404e52ac87e61c47f60ae3d1c65fa8 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 14:04:49 +0100
Subject: [PATCH 13/30] Address code review

---
 Include/internal/pycore_tstate.h |  2 +-
 Python/optimizer.c               |  9 ++++-----
 Python/pylifecycle.c             |  9 ---------
 Python/pystate.c                 | 13 +++++++++++++
 4 files changed, 18 insertions(+), 15 deletions(-)

diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index 25b6b16918041f..2ad02b5be6999f 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -78,7 +78,7 @@ typedef struct _PyThreadStateImpl {
 #endif
 
 #ifdef _Py_TIER2
-    struct _PyUOpInstruction *buffer;
+    struct _PyUOpInstruction *jit_uop_buffer;
 #endif
 
 } _PyThreadStateImpl;
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 16874373931371..f902f696545708 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1281,14 +1281,13 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
-    if (tstate->buffer == NULL) {
-        tstate->buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
-        if (tstate->buffer == NULL) {
-            PyErr_NoMemory();
+    if (tstate->jit_uop_buffer == NULL) {
+        tstate->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
+        if (tstate->jit_uop_buffer == NULL) {
             return -1;
         }
     }
-    _PyUOpInstruction *buffer = tstate->buffer;
+    _PyUOpInstruction *buffer = tstate->jit_uop_buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
     bool is_noopt = true;
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index a13bd532b91148..51a777077d8255 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -863,10 +863,6 @@ pycore_interp_init(PyThreadState *tstate)
     if (_tstate->c_stack_hard_limit == 0) {
         _Py_InitializeRecursionLimits(tstate);
     }
-#ifdef _Py_TIER2
-     // Ensure the buffer is to be set as NULL for MSVC
-    _tstate->buffer = NULL;
-#endif
     PyInterpreterState *interp = tstate->interp;
     PyStatus status;
     PyObject *sysmod = NULL;
@@ -1708,11 +1704,6 @@ finalize_modules(PyThreadState *tstate)
     interp->jit = false;
 #ifdef _Py_TIER2
     _Py_Executors_InvalidateAll(interp, 0);
-    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
-    if (_tstate->buffer != NULL) {
-        PyMem_RawFree(_tstate->buffer);
-        _tstate->buffer = NULL;
-    }
 #endif
 
     // Stop watching __builtin__ modifications
diff --git a/Python/pystate.c b/Python/pystate.c
index 2465d8667472dc..4308ff844b05b8 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -1536,6 +1536,11 @@ new_threadstate(PyInterpreterState *interp, int whence)
     }
 #endif
 
+#ifdef _Py_TIER2
+     // Ensure the buffer is to be set as NULL for MSVC
+    tstate->jit_uop_buffer = NULL;
+#endif
+
     /* We serialize concurrent creation to protect global state. */
     HEAD_LOCK(interp->runtime);
 
@@ -1726,6 +1731,14 @@ PyThreadState_Clear(PyThreadState *tstate)
     _Py_ClearTLBCIndex((_PyThreadStateImpl *)tstate);
 #endif
 
+#ifdef _Py_TIER2
+    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
+    if (_tstate->jit_uop_buffer != NULL) {
+        PyMem_RawFree(_tstate->jit_uop_buffer);
+        _tstate->jit_uop_buffer = NULL;
+    }
+#endif
+
     // Merge our queue of pointers to be freed into the interpreter queue.
     _PyMem_AbandonDelayed(tstate);
 

From ad15aaf4c2cc3468705109b66202b85cd30f7984 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 14:08:41 +0100
Subject: [PATCH 14/30] Address code review

---
 Python/pystate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Python/pystate.c b/Python/pystate.c
index 4308ff844b05b8..8680156a21e59b 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -1537,7 +1537,7 @@ new_threadstate(PyInterpreterState *interp, int whence)
 #endif
 
 #ifdef _Py_TIER2
-     // Ensure the buffer is to be set as NULL for MSVC
+     // Ensure the buffer is to be set as NULL.
     tstate->jit_uop_buffer = NULL;
 #endif
 

From 07e6ba166565289538efc6fbef7f1adc46c48efa Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 14:09:54 +0100
Subject: [PATCH 15/30] Address code review

---
 Python/optimizer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Python/optimizer.c b/Python/optimizer.c
index f902f696545708..aec68bbca5398e 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1284,7 +1284,7 @@ uop_optimize(
     if (tstate->jit_uop_buffer == NULL) {
         tstate->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
         if (tstate->jit_uop_buffer == NULL) {
-            return -1;
+            return 0;
         }
     }
     _PyUOpInstruction *buffer = tstate->jit_uop_buffer;

From f13260e806b1771f6cf56e09519dd70a105498bc Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 22:40:11 +0100
Subject: [PATCH 16/30] Address code review

---
 Include/internal/pycore_interp_structs.h |  5 +++++
 Include/internal/pycore_tstate.h         |  5 -----
 Python/optimizer.c                       | 10 +++++-----
 Python/pystate.c                         | 22 +++++++++-------------
 4 files changed, 19 insertions(+), 23 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index fa9568ab4d0e85..e6b11c3224a2b8 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -14,6 +14,7 @@ extern "C" {
 #include "pycore_structs.h"       // PyHamtObject
 #include "pycore_tstate.h"        // _PyThreadStateImpl
 #include "pycore_typedefs.h"      // _PyRuntimeState
+#include "pycore_uop.h"           // struct _PyUOpInstruction
 
 
 #define CODE_MAX_WATCHERS 8
@@ -898,6 +899,10 @@ struct _is {
     struct _stoptheworld_state stoptheworld;
     struct _qsbr_shared qsbr;
 
+#ifdef _Py_TIER2
+    struct _PyUOpInstruction *jit_uop_buffer;
+#endif
+
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
     struct _brc_state brc;  // biased reference counting state
diff --git a/Include/internal/pycore_tstate.h b/Include/internal/pycore_tstate.h
index 2ad02b5be6999f..bad968428c73a1 100644
--- a/Include/internal/pycore_tstate.h
+++ b/Include/internal/pycore_tstate.h
@@ -12,7 +12,6 @@ extern "C" {
 #include "pycore_freelist_state.h"  // struct _Py_freelists
 #include "pycore_mimalloc.h"        // struct _mimalloc_thread_state
 #include "pycore_qsbr.h"            // struct qsbr
-#include "pycore_uop.h"             // struct _PyUOpInstruction
 
 
 #ifdef Py_GIL_DISABLED
@@ -77,10 +76,6 @@ typedef struct _PyThreadStateImpl {
     Py_ssize_t reftotal;  // this thread's total refcount operations
 #endif
 
-#ifdef _Py_TIER2
-    struct _PyUOpInstruction *jit_uop_buffer;
-#endif
-
 } _PyThreadStateImpl;
 
 #ifdef __cplusplus
diff --git a/Python/optimizer.c b/Python/optimizer.c
index aec68bbca5398e..259cdb0e698af4 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1280,14 +1280,14 @@ uop_optimize(
 {
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
-    _PyThreadStateImpl *tstate = (_PyThreadStateImpl *)_PyThreadState_GET();
-    if (tstate->jit_uop_buffer == NULL) {
-        tstate->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
-        if (tstate->jit_uop_buffer == NULL) {
+    PyInterpreterState *interp = _PyInterpreterState_GET();
+    if (interp->jit_uop_buffer == NULL) {
+        interp->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
+        if (interp->jit_uop_buffer == NULL) {
             return 0;
         }
     }
-    _PyUOpInstruction *buffer = tstate->jit_uop_buffer;
+    _PyUOpInstruction *buffer = interp->jit_uop_buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
     bool is_noopt = true;
diff --git a/Python/pystate.c b/Python/pystate.c
index 8680156a21e59b..f45087824c1b3d 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -556,6 +556,11 @@ init_interpreter(PyInterpreterState *interp,
 #ifdef Py_GIL_DISABLED
     _Py_brc_init_state(interp);
 #endif
+
+#ifdef _Py_TIER2
+     // Ensure the buffer is to be set as NULL.
+    interp->jit_uop_buffer = NULL;
+#endif
     llist_init(&interp->mem_free_queue.head);
     llist_init(&interp->asyncio_tasks_head);
     interp->asyncio_tasks_lock = (PyMutex){0};
@@ -803,6 +808,10 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
+    if (interp->jit_uop_buffer != NULL) {
+        PyMem_RawFree(interp->jit_uop_buffer);
+        interp->jit_uop_buffer = NULL;
+    }
 #endif
     _PyAST_Fini(interp);
     _PyAtExit_Fini(interp);
@@ -1536,11 +1545,6 @@ new_threadstate(PyInterpreterState *interp, int whence)
     }
 #endif
 
-#ifdef _Py_TIER2
-     // Ensure the buffer is to be set as NULL.
-    tstate->jit_uop_buffer = NULL;
-#endif
-
     /* We serialize concurrent creation to protect global state. */
     HEAD_LOCK(interp->runtime);
 
@@ -1731,14 +1735,6 @@ PyThreadState_Clear(PyThreadState *tstate)
     _Py_ClearTLBCIndex((_PyThreadStateImpl *)tstate);
 #endif
 
-#ifdef _Py_TIER2
-    _PyThreadStateImpl *_tstate = (_PyThreadStateImpl *)tstate;
-    if (_tstate->jit_uop_buffer != NULL) {
-        PyMem_RawFree(_tstate->jit_uop_buffer);
-        _tstate->jit_uop_buffer = NULL;
-    }
-#endif
-
     // Merge our queue of pointers to be freed into the interpreter queue.
     _PyMem_AbandonDelayed(tstate);
 

From ab39a55e57981dd1fe6ad2454257add987d1dcaa Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 23:08:36 +0100
Subject: [PATCH 17/30] test

---
 Python/pystate.c | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/Python/pystate.c b/Python/pystate.c
index f45087824c1b3d..23307cbc23e04d 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -556,11 +556,6 @@ init_interpreter(PyInterpreterState *interp,
 #ifdef Py_GIL_DISABLED
     _Py_brc_init_state(interp);
 #endif
-
-#ifdef _Py_TIER2
-     // Ensure the buffer is to be set as NULL.
-    interp->jit_uop_buffer = NULL;
-#endif
     llist_init(&interp->mem_free_queue.head);
     llist_init(&interp->asyncio_tasks_head);
     interp->asyncio_tasks_lock = (PyMutex){0};

From 60541dc982ee99830ed4aa16fe857143e419bd35 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 23:27:55 +0100
Subject: [PATCH 18/30] Revert "test"

This reverts commit ab39a55e57981dd1fe6ad2454257add987d1dcaa.
---
 Python/pystate.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/Python/pystate.c b/Python/pystate.c
index 23307cbc23e04d..f45087824c1b3d 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -556,6 +556,11 @@ init_interpreter(PyInterpreterState *interp,
 #ifdef Py_GIL_DISABLED
     _Py_brc_init_state(interp);
 #endif
+
+#ifdef _Py_TIER2
+     // Ensure the buffer is to be set as NULL.
+    interp->jit_uop_buffer = NULL;
+#endif
     llist_init(&interp->mem_free_queue.head);
     llist_init(&interp->asyncio_tasks_head);
     interp->asyncio_tasks_lock = (PyMutex){0};

From 877f3a90efad8641a253f3b81ef597f8486efe14 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 23:28:20 +0100
Subject: [PATCH 19/30] Just for test

---
 Python/pystate.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/Python/pystate.c b/Python/pystate.c
index f45087824c1b3d..85c0a421bb1595 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -808,10 +808,6 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
-    if (interp->jit_uop_buffer != NULL) {
-        PyMem_RawFree(interp->jit_uop_buffer);
-        interp->jit_uop_buffer = NULL;
-    }
 #endif
     _PyAST_Fini(interp);
     _PyAtExit_Fini(interp);

From 9af48c2d66324394275072f541d85b2446e39d4c Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Tue, 16 Sep 2025 23:48:03 +0100
Subject: [PATCH 20/30] Revert "Just for test"

This reverts commit 877f3a90efad8641a253f3b81ef597f8486efe14.
---
 Python/pystate.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/Python/pystate.c b/Python/pystate.c
index 85c0a421bb1595..f45087824c1b3d 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -808,6 +808,10 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
+    if (interp->jit_uop_buffer != NULL) {
+        PyMem_RawFree(interp->jit_uop_buffer);
+        interp->jit_uop_buffer = NULL;
+    }
 #endif
     _PyAST_Fini(interp);
     _PyAtExit_Fini(interp);

From 989fa8e2cada9417fa7f41041f3e234ecfdbb0eb Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 06:40:01 +0100
Subject: [PATCH 21/30] Test statically allocated

---
 Include/internal/pycore_interp_structs.h | 2 +-
 Python/optimizer.c                       | 2 ++
 Python/pystate.c                         | 4 +++-
 3 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index e6b11c3224a2b8..86c906bd21b9e4 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -900,7 +900,7 @@ struct _is {
     struct _qsbr_shared qsbr;
 
 #ifdef _Py_TIER2
-    struct _PyUOpInstruction *jit_uop_buffer;
+    struct _PyUOpInstruction jit_uop_buffer[UOP_MAX_TRACE_LENGTH];
 #endif
 
 #if defined(Py_GIL_DISABLED)
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 259cdb0e698af4..3c04b0c8fe422e 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1281,12 +1281,14 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     PyInterpreterState *interp = _PyInterpreterState_GET();
+    /**
     if (interp->jit_uop_buffer == NULL) {
         interp->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
         if (interp->jit_uop_buffer == NULL) {
             return 0;
         }
     }
+    **/
     _PyUOpInstruction *buffer = interp->jit_uop_buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
diff --git a/Python/pystate.c b/Python/pystate.c
index f45087824c1b3d..b1f729a339a752 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -559,7 +559,7 @@ init_interpreter(PyInterpreterState *interp,
 
 #ifdef _Py_TIER2
      // Ensure the buffer is to be set as NULL.
-    interp->jit_uop_buffer = NULL;
+    // interp->jit_uop_buffer = NULL;
 #endif
     llist_init(&interp->mem_free_queue.head);
     llist_init(&interp->asyncio_tasks_head);
@@ -808,10 +808,12 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
+    /**
     if (interp->jit_uop_buffer != NULL) {
         PyMem_RawFree(interp->jit_uop_buffer);
         interp->jit_uop_buffer = NULL;
     }
+    **/
 #endif
     _PyAST_Fini(interp);
     _PyAtExit_Fini(interp);

From 4dff8168a23f4ae7355147fd540ffd5b2ea81f7f Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 10:16:11 +0100
Subject: [PATCH 22/30] Revert "Test statically allocated"

This reverts commit 989fa8e2cada9417fa7f41041f3e234ecfdbb0eb.
---
 Include/internal/pycore_interp_structs.h | 2 +-
 Python/optimizer.c                       | 2 --
 Python/pystate.c                         | 4 +---
 3 files changed, 2 insertions(+), 6 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index 86c906bd21b9e4..e6b11c3224a2b8 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -900,7 +900,7 @@ struct _is {
     struct _qsbr_shared qsbr;
 
 #ifdef _Py_TIER2
-    struct _PyUOpInstruction jit_uop_buffer[UOP_MAX_TRACE_LENGTH];
+    struct _PyUOpInstruction *jit_uop_buffer;
 #endif
 
 #if defined(Py_GIL_DISABLED)
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 3c04b0c8fe422e..259cdb0e698af4 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1281,14 +1281,12 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     PyInterpreterState *interp = _PyInterpreterState_GET();
-    /**
     if (interp->jit_uop_buffer == NULL) {
         interp->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
         if (interp->jit_uop_buffer == NULL) {
             return 0;
         }
     }
-    **/
     _PyUOpInstruction *buffer = interp->jit_uop_buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
diff --git a/Python/pystate.c b/Python/pystate.c
index b1f729a339a752..f45087824c1b3d 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -559,7 +559,7 @@ init_interpreter(PyInterpreterState *interp,
 
 #ifdef _Py_TIER2
      // Ensure the buffer is to be set as NULL.
-    // interp->jit_uop_buffer = NULL;
+    interp->jit_uop_buffer = NULL;
 #endif
     llist_init(&interp->mem_free_queue.head);
     llist_init(&interp->asyncio_tasks_head);
@@ -808,12 +808,10 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
-    /**
     if (interp->jit_uop_buffer != NULL) {
         PyMem_RawFree(interp->jit_uop_buffer);
         interp->jit_uop_buffer = NULL;
     }
-    **/
 #endif
     _PyAST_Fini(interp);
     _PyAtExit_Fini(interp);

From 7e5261ce8970c4dc855c57b2f13e1f93016fb698 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 10:17:26 +0100
Subject: [PATCH 23/30] Make situation to isolate with the change

---
 Python/optimizer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Python/optimizer.c b/Python/optimizer.c
index 259cdb0e698af4..e4849a09883763 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1287,7 +1287,7 @@ uop_optimize(
             return 0;
         }
     }
-    _PyUOpInstruction *buffer = interp->jit_uop_buffer;
+    _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
     bool is_noopt = true;

From 08adf263a5f3c011cd022b7b5227ec4071d15d7d Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 11:23:25 +0100
Subject: [PATCH 24/30] fix

---
 Python/optimizer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Python/optimizer.c b/Python/optimizer.c
index e4849a09883763..259cdb0e698af4 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1287,7 +1287,7 @@ uop_optimize(
             return 0;
         }
     }
-    _PyUOpInstruction buffer[UOP_MAX_TRACE_LENGTH];
+    _PyUOpInstruction *buffer = interp->jit_uop_buffer;
     OPT_STAT_INC(attempts);
     char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE");
     bool is_noopt = true;

From 9051791f16e93357d4b4a4b6e9b74819591f89d5 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 11:23:34 +0100
Subject: [PATCH 25/30] fix

---
 Include/internal/pycore_interp_structs.h | 2 --
 1 file changed, 2 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index e6b11c3224a2b8..03e9c5807ea6f8 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -899,9 +899,7 @@ struct _is {
     struct _stoptheworld_state stoptheworld;
     struct _qsbr_shared qsbr;
 
-#ifdef _Py_TIER2
     struct _PyUOpInstruction *jit_uop_buffer;
-#endif
 
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;

From 928485ac0f40696ffd36aacf5a967f3f746a59de Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 13:36:50 +0100
Subject: [PATCH 26/30] Address code review

---
 Include/internal/pycore_uop.h | 1 +
 Python/optimizer.c            | 2 +-
 Python/pystate.c              | 3 ++-
 3 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/Include/internal/pycore_uop.h b/Include/internal/pycore_uop.h
index b569c46ccb926f..4abefd3b95d21a 100644
--- a/Include/internal/pycore_uop.h
+++ b/Include/internal/pycore_uop.h
@@ -37,6 +37,7 @@ typedef struct _PyUOpInstruction{
 
 // This is the length of the trace we project initially.
 #define UOP_MAX_TRACE_LENGTH 1200
+#define UOP_BUFFER_SIZE (UOP_MAX_TRACE_LENGTH * sizeof(_PyUOpInstruction))
 
 #ifdef __cplusplus
 }
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 259cdb0e698af4..e1b7a78a240ea8 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -1282,7 +1282,7 @@ uop_optimize(
     _Py_BloomFilter_Init(&dependencies);
     PyInterpreterState *interp = _PyInterpreterState_GET();
     if (interp->jit_uop_buffer == NULL) {
-        interp->jit_uop_buffer = (_PyUOpInstruction *)PyMem_RawMalloc(UOP_MAX_TRACE_LENGTH*sizeof(_PyUOpInstruction));
+        interp->jit_uop_buffer = (_PyUOpInstruction *)_PyObject_VirtualAlloc(UOP_BUFFER_SIZE);
         if (interp->jit_uop_buffer == NULL) {
             return 0;
         }
diff --git a/Python/pystate.c b/Python/pystate.c
index f45087824c1b3d..d0094aba04ca4a 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -23,6 +23,7 @@
 #include "pycore_runtime_init.h"  // _PyRuntimeState_INIT
 #include "pycore_stackref.h"      // Py_STACKREF_DEBUG
 #include "pycore_time.h"          // _PyTime_Init()
+#include "pycore_uop.h".          // UOP_BUFFER_SIZE
 #include "pycore_uniqueid.h"      // _PyObject_FinalizePerThreadRefcounts()
 
 
@@ -809,7 +810,7 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate)
 #ifdef _Py_TIER2
     _Py_ClearExecutorDeletionList(interp);
     if (interp->jit_uop_buffer != NULL) {
-        PyMem_RawFree(interp->jit_uop_buffer);
+        _PyObject_VirtualFree(interp->jit_uop_buffer, UOP_BUFFER_SIZE);
         interp->jit_uop_buffer = NULL;
     }
 #endif

From 0c2a74f920315ef8ca457cc96a97ff9d23ade353 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 13:43:32 +0100
Subject: [PATCH 27/30] nit

---
 Python/pystate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Python/pystate.c b/Python/pystate.c
index d0094aba04ca4a..9d771d2687bb0e 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -23,7 +23,7 @@
 #include "pycore_runtime_init.h"  // _PyRuntimeState_INIT
 #include "pycore_stackref.h"      // Py_STACKREF_DEBUG
 #include "pycore_time.h"          // _PyTime_Init()
-#include "pycore_uop.h".          // UOP_BUFFER_SIZE
+#include "pycore_uop.h"           // UOP_BUFFER_SIZE
 #include "pycore_uniqueid.h"      // _PyObject_FinalizePerThreadRefcounts()
 
 

From a49bcab0ad92be69fee5244cdbdaf22600b875c8 Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 13:58:46 +0100
Subject: [PATCH 28/30] Update

---
 Include/internal/pycore_interp_structs.h |  4 ++--
 Python/bytecodes.c                       |  2 ++
 Python/generated_cases.c.h               |  2 ++
 Python/optimizer.c                       | 18 +++++++++++++-----
 Python/pystate.c                         |  1 +
 5 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index 03e9c5807ea6f8..4c55770e01da4f 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -899,8 +899,6 @@ struct _is {
     struct _stoptheworld_state stoptheworld;
     struct _qsbr_shared qsbr;
 
-    struct _PyUOpInstruction *jit_uop_buffer;
-
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
     struct _brc_state brc;  // biased reference counting state
@@ -952,6 +950,8 @@ struct _is {
     struct callable_cache callable_cache;
     PyObject *common_consts[NUM_COMMON_CONSTANTS];
     bool jit;
+    bool compiling;
+    struct _PyUOpInstruction *jit_uop_buffer;
     struct _PyExecutorObject *executor_list_head;
     struct _PyExecutorObject *executor_deletion_list_head;
     struct _PyExecutorObject *cold_executor;
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 6c3609d293890f..2f1d240f11d6de 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2961,7 +2961,9 @@ dummy_func(
                     start--;
                 }
                 _PyExecutorObject *executor;
+                assert(!_PyInterpreterState_GET()->compiling);
                 int optimized = _PyOptimizer_Optimize(frame, start, &executor, 0);
+                assert(!_PyInterpreterState_GET()->compiling);
                 if (optimized <= 0) {
                     this_instr[1].counter = restart_backoff_counter(counter);
                     ERROR_IF(optimized < 0);
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index c1f6f5c85cdd88..048bd8273d7559 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -7651,9 +7651,11 @@
                         start--;
                     }
                     _PyExecutorObject *executor;
+                    assert(!_PyInterpreterState_GET()->compiling);
                     _PyFrame_SetStackPointer(frame, stack_pointer);
                     int optimized = _PyOptimizer_Optimize(frame, start, &executor, 0);
                     stack_pointer = _PyFrame_GetStackPointer(frame);
+                    assert(!_PyInterpreterState_GET()->compiling);
                     if (optimized <= 0) {
                         this_instr[1].counter = restart_backoff_counter(counter);
                         if (optimized < 0) {
diff --git a/Python/optimizer.c b/Python/optimizer.c
index e1b7a78a240ea8..021e257dcbd84f 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -116,7 +116,10 @@ _PyOptimizer_Optimize(
     _PyExecutorObject **executor_ptr, int chain_depth)
 {
     _PyStackRef *stack_pointer = frame->stackpointer;
-    assert(_PyInterpreterState_GET()->jit);
+    PyInterpreterState *interp = _PyInterpreterState_GET();
+    assert(interp->jit);
+    assert(!interp->compiling);
+    interp->compiling = true;
     // The first executor in a chain and the MAX_CHAIN_DEPTH'th executor *must*
     // make progress in order to avoid infinite loops or excessively-long
     // side-exit chains. We can only insert the executor into the bytecode if
@@ -125,12 +128,14 @@ _PyOptimizer_Optimize(
     bool progress_needed = chain_depth == 0;
     PyCodeObject *code = _PyFrame_GetCode(frame);
     assert(PyCode_Check(code));
+    int ret = 0;
     if (progress_needed && !has_space_for_executor(code, start)) {
-        return 0;
+        goto end;
     }
     int err = uop_optimize(frame, start, executor_ptr, (int)(stack_pointer - _PyFrame_Stackbase(frame)), progress_needed);
     if (err <= 0) {
-        return err;
+        ret = err;
+        goto end;
     }
     assert(*executor_ptr != NULL);
     if (progress_needed) {
@@ -143,7 +148,7 @@ _PyOptimizer_Optimize(
              * it might get confused by the executor disappearing,
              * but there is not much we can do about that here. */
             Py_DECREF(*executor_ptr);
-            return 0;
+            goto end;
         }
         insert_executor(code, start, index, *executor_ptr);
     }
@@ -152,7 +157,9 @@ _PyOptimizer_Optimize(
     }
     (*executor_ptr)->vm_data.chain_depth = chain_depth;
     assert((*executor_ptr)->vm_data.valid);
-    return 1;
+end:
+    interp->compiling = false;
+    return ret;
 }
 
 static _PyExecutorObject *
@@ -1281,6 +1288,7 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     PyInterpreterState *interp = _PyInterpreterState_GET();
+    assert(interp->compiling);
     if (interp->jit_uop_buffer == NULL) {
         interp->jit_uop_buffer = (_PyUOpInstruction *)_PyObject_VirtualAlloc(UOP_BUFFER_SIZE);
         if (interp->jit_uop_buffer == NULL) {
diff --git a/Python/pystate.c b/Python/pystate.c
index 9d771d2687bb0e..5305fd47791ec4 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -577,6 +577,7 @@ init_interpreter(PyInterpreterState *interp,
     }
     interp->_code_object_generation = 0;
     interp->jit = false;
+    interp->compiling = false;
     interp->executor_list_head = NULL;
     interp->executor_deletion_list_head = NULL;
     interp->executor_deletion_list_remaining_capacity = 0;

From f88ee3be9b110195e7364c09f591fdfdc0f1136e Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 14:32:50 +0100
Subject: [PATCH 29/30] Revert "Update"

This reverts commit a49bcab0ad92be69fee5244cdbdaf22600b875c8.
---
 Include/internal/pycore_interp_structs.h |  4 ++--
 Python/bytecodes.c                       |  2 --
 Python/generated_cases.c.h               |  2 --
 Python/optimizer.c                       | 18 +++++-------------
 Python/pystate.c                         |  1 -
 5 files changed, 7 insertions(+), 20 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index 4c55770e01da4f..03e9c5807ea6f8 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -899,6 +899,8 @@ struct _is {
     struct _stoptheworld_state stoptheworld;
     struct _qsbr_shared qsbr;
 
+    struct _PyUOpInstruction *jit_uop_buffer;
+
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
     struct _brc_state brc;  // biased reference counting state
@@ -950,8 +952,6 @@ struct _is {
     struct callable_cache callable_cache;
     PyObject *common_consts[NUM_COMMON_CONSTANTS];
     bool jit;
-    bool compiling;
-    struct _PyUOpInstruction *jit_uop_buffer;
     struct _PyExecutorObject *executor_list_head;
     struct _PyExecutorObject *executor_deletion_list_head;
     struct _PyExecutorObject *cold_executor;
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 2f1d240f11d6de..6c3609d293890f 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2961,9 +2961,7 @@ dummy_func(
                     start--;
                 }
                 _PyExecutorObject *executor;
-                assert(!_PyInterpreterState_GET()->compiling);
                 int optimized = _PyOptimizer_Optimize(frame, start, &executor, 0);
-                assert(!_PyInterpreterState_GET()->compiling);
                 if (optimized <= 0) {
                     this_instr[1].counter = restart_backoff_counter(counter);
                     ERROR_IF(optimized < 0);
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 048bd8273d7559..c1f6f5c85cdd88 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -7651,11 +7651,9 @@
                         start--;
                     }
                     _PyExecutorObject *executor;
-                    assert(!_PyInterpreterState_GET()->compiling);
                     _PyFrame_SetStackPointer(frame, stack_pointer);
                     int optimized = _PyOptimizer_Optimize(frame, start, &executor, 0);
                     stack_pointer = _PyFrame_GetStackPointer(frame);
-                    assert(!_PyInterpreterState_GET()->compiling);
                     if (optimized <= 0) {
                         this_instr[1].counter = restart_backoff_counter(counter);
                         if (optimized < 0) {
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 021e257dcbd84f..e1b7a78a240ea8 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -116,10 +116,7 @@ _PyOptimizer_Optimize(
     _PyExecutorObject **executor_ptr, int chain_depth)
 {
     _PyStackRef *stack_pointer = frame->stackpointer;
-    PyInterpreterState *interp = _PyInterpreterState_GET();
-    assert(interp->jit);
-    assert(!interp->compiling);
-    interp->compiling = true;
+    assert(_PyInterpreterState_GET()->jit);
     // The first executor in a chain and the MAX_CHAIN_DEPTH'th executor *must*
     // make progress in order to avoid infinite loops or excessively-long
     // side-exit chains. We can only insert the executor into the bytecode if
@@ -128,14 +125,12 @@ _PyOptimizer_Optimize(
     bool progress_needed = chain_depth == 0;
     PyCodeObject *code = _PyFrame_GetCode(frame);
     assert(PyCode_Check(code));
-    int ret = 0;
     if (progress_needed && !has_space_for_executor(code, start)) {
-        goto end;
+        return 0;
     }
     int err = uop_optimize(frame, start, executor_ptr, (int)(stack_pointer - _PyFrame_Stackbase(frame)), progress_needed);
     if (err <= 0) {
-        ret = err;
-        goto end;
+        return err;
     }
     assert(*executor_ptr != NULL);
     if (progress_needed) {
@@ -148,7 +143,7 @@ _PyOptimizer_Optimize(
              * it might get confused by the executor disappearing,
              * but there is not much we can do about that here. */
             Py_DECREF(*executor_ptr);
-            goto end;
+            return 0;
         }
         insert_executor(code, start, index, *executor_ptr);
     }
@@ -157,9 +152,7 @@ _PyOptimizer_Optimize(
     }
     (*executor_ptr)->vm_data.chain_depth = chain_depth;
     assert((*executor_ptr)->vm_data.valid);
-end:
-    interp->compiling = false;
-    return ret;
+    return 1;
 }
 
 static _PyExecutorObject *
@@ -1288,7 +1281,6 @@ uop_optimize(
     _PyBloomFilter dependencies;
     _Py_BloomFilter_Init(&dependencies);
     PyInterpreterState *interp = _PyInterpreterState_GET();
-    assert(interp->compiling);
     if (interp->jit_uop_buffer == NULL) {
         interp->jit_uop_buffer = (_PyUOpInstruction *)_PyObject_VirtualAlloc(UOP_BUFFER_SIZE);
         if (interp->jit_uop_buffer == NULL) {
diff --git a/Python/pystate.c b/Python/pystate.c
index 5305fd47791ec4..9d771d2687bb0e 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -577,7 +577,6 @@ init_interpreter(PyInterpreterState *interp,
     }
     interp->_code_object_generation = 0;
     interp->jit = false;
-    interp->compiling = false;
     interp->executor_list_head = NULL;
     interp->executor_deletion_list_head = NULL;
     interp->executor_deletion_list_remaining_capacity = 0;

From 05825f57e68f3d6f4c677dc2ca0b473397f15e7c Mon Sep 17 00:00:00 2001
From: Donghee Na <donghee.na@python.org>
Date: Wed, 17 Sep 2025 15:02:02 +0100
Subject: [PATCH 30/30] fix

---
 Include/internal/pycore_interp_structs.h | 4 ++--
 Python/optimizer.c                       | 9 ++++++++-
 Python/pylifecycle.c                     | 1 +
 Python/pystate.c                         | 1 +
 4 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index 03e9c5807ea6f8..4c55770e01da4f 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -899,8 +899,6 @@ struct _is {
     struct _stoptheworld_state stoptheworld;
     struct _qsbr_shared qsbr;
 
-    struct _PyUOpInstruction *jit_uop_buffer;
-
 #if defined(Py_GIL_DISABLED)
     struct _mimalloc_interp_state mimalloc;
     struct _brc_state brc;  // biased reference counting state
@@ -952,6 +950,8 @@ struct _is {
     struct callable_cache callable_cache;
     PyObject *common_consts[NUM_COMMON_CONSTANTS];
     bool jit;
+    bool compiling;
+    struct _PyUOpInstruction *jit_uop_buffer;
     struct _PyExecutorObject *executor_list_head;
     struct _PyExecutorObject *executor_deletion_list_head;
     struct _PyExecutorObject *cold_executor;
diff --git a/Python/optimizer.c b/Python/optimizer.c
index e1b7a78a240ea8..6f27b521018c8c 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -116,7 +116,10 @@ _PyOptimizer_Optimize(
     _PyExecutorObject **executor_ptr, int chain_depth)
 {
     _PyStackRef *stack_pointer = frame->stackpointer;
-    assert(_PyInterpreterState_GET()->jit);
+    PyInterpreterState *interp = _PyInterpreterState_GET();
+    assert(interp->jit);
+    assert(!interp->compiling);
+    interp->compiling = true;
     // The first executor in a chain and the MAX_CHAIN_DEPTH'th executor *must*
     // make progress in order to avoid infinite loops or excessively-long
     // side-exit chains. We can only insert the executor into the bytecode if
@@ -126,10 +129,12 @@ _PyOptimizer_Optimize(
     PyCodeObject *code = _PyFrame_GetCode(frame);
     assert(PyCode_Check(code));
     if (progress_needed && !has_space_for_executor(code, start)) {
+        interp->compiling = false;
         return 0;
     }
     int err = uop_optimize(frame, start, executor_ptr, (int)(stack_pointer - _PyFrame_Stackbase(frame)), progress_needed);
     if (err <= 0) {
+        interp->compiling = false;
         return err;
     }
     assert(*executor_ptr != NULL);
@@ -143,6 +148,7 @@ _PyOptimizer_Optimize(
              * it might get confused by the executor disappearing,
              * but there is not much we can do about that here. */
             Py_DECREF(*executor_ptr);
+            interp->compiling = false;
             return 0;
         }
         insert_executor(code, start, index, *executor_ptr);
@@ -152,6 +158,7 @@ _PyOptimizer_Optimize(
     }
     (*executor_ptr)->vm_data.chain_depth = chain_depth;
     assert((*executor_ptr)->vm_data.valid);
+    interp->compiling = false;
     return 1;
 }
 
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index 51a777077d8255..20d985f59e72eb 100644
--- a/Python/pylifecycle.c
+++ b/Python/pylifecycle.c
@@ -1702,6 +1702,7 @@ finalize_modules(PyThreadState *tstate)
 
     // Invalidate all executors and turn off JIT:
     interp->jit = false;
+    interp->compiling = false;
 #ifdef _Py_TIER2
     _Py_Executors_InvalidateAll(interp, 0);
 #endif
diff --git a/Python/pystate.c b/Python/pystate.c
index 9d771d2687bb0e..5305fd47791ec4 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -577,6 +577,7 @@ init_interpreter(PyInterpreterState *interp,
     }
     interp->_code_object_generation = 0;
     interp->jit = false;
+    interp->compiling = false;
     interp->executor_list_head = NULL;
     interp->executor_deletion_list_head = NULL;
     interp->executor_deletion_list_remaining_capacity = 0;