bpo-40521: Make async gen free lists per-interpreter (GH-20643)

Each interpreter now has its own asynchronous generator free lists:

* Move async gen free lists into PyInterpreterState.
* Move _PyAsyncGen_MAXFREELIST define to pycore_interp.h
* Add _Py_async_gen_state structure.
* Add tstate parameter to _PyAsyncGen_ClearFreeLists
  and _PyAsyncGen_Fini().

3 files changed, 20 insertions, 2 deletions

diff --git a/Include/internal/pycore_gc.h b/Include/internal/pycore_gc.h
index 3388b4d..ad2e552 100644
--- a/Include/internal/pycore_gc.h
+++ b/Include/internal/pycore_gc.h
@@ -170,7 +170,7 @@ extern void _PyTuple_ClearFreeList(PyThreadState *tstate);
 extern void _PyFloat_ClearFreeList(PyThreadState *tstate);
 extern void _PyList_ClearFreeList(PyThreadState *tstate);
 extern void _PyDict_ClearFreeList(void);
-extern void _PyAsyncGen_ClearFreeLists(void);
+extern void _PyAsyncGen_ClearFreeLists(PyThreadState *tstate);
 extern void _PyContext_ClearFreeList(void);
 
 #ifdef __cplusplus
diff --git a/Include/internal/pycore_interp.h b/Include/internal/pycore_interp.h
index 0eab246..d624218 100644
--- a/Include/internal/pycore_interp.h
+++ b/Include/internal/pycore_interp.h
@@ -108,6 +108,23 @@ struct _Py_frame_state {
     int numfree;
 };
 
+#ifndef _PyAsyncGen_MAXFREELIST
+#  define _PyAsyncGen_MAXFREELIST 80
+#endif
+
+struct _Py_async_gen_state {
+    /* Freelists boost performance 6-10%; they also reduce memory
+       fragmentation, as _PyAsyncGenWrappedValue and PyAsyncGenASend
+       are short-living objects that are instantiated for every
+       __anext__() call. */
+    struct _PyAsyncGenWrappedValue* value_freelist[_PyAsyncGen_MAXFREELIST];
+    int value_numfree;
+
+    struct PyAsyncGenASend* asend_freelist[_PyAsyncGen_MAXFREELIST];
+    int asend_numfree;
+};
+
+
 
 /* interpreter state */
 
@@ -205,6 +222,7 @@ struct _is {
     struct _Py_list_state list;
     struct _Py_float_state float_state;
     struct _Py_frame_state frame;
+    struct _Py_async_gen_state async_gen;
 
     /* Using a cache is very effective since typically only a single slice is
        created and then deleted again. */
diff --git a/Include/internal/pycore_pylifecycle.h b/Include/internal/pycore_pylifecycle.h
index 3c35ca2..3e36573 100644
--- a/Include/internal/pycore_pylifecycle.h
+++ b/Include/internal/pycore_pylifecycle.h
@@ -66,7 +66,7 @@ extern void _PySet_Fini(void);
 extern void _PyBytes_Fini(void);
 extern void _PyFloat_Fini(PyThreadState *tstate);
 extern void _PySlice_Fini(PyThreadState *tstate);
-extern void _PyAsyncGen_Fini(void);
+extern void _PyAsyncGen_Fini(PyThreadState *tstate);
 
 extern void PyOS_FiniInterrupts(void);