From 5a6b74c2019e47f6fe077fa9df8d95a46e4df6e5 Mon Sep 17 00:00:00 2001
From: Neil Schemenauer <nas@arctrix.com>
Date: Tue, 9 Jul 2019 23:19:03 -0700
Subject: [PATCH 1/3] Compute allocated blocks in _Py_GetAllocatedBlocks().

Keeping an account of allocated blocks slows down _PyObject_Malloc()
and _PyObject_Free() by a measureable amount.  Have
_Py_GetAllocatedBlocks() iterate over the arenas to sum up the
allocated blocks for pymalloc.
---
 Objects/obmalloc.c | 29 ++++++++++++++++++++++-------
 1 file changed, 22 insertions(+), 7 deletions(-)

diff --git a/Objects/obmalloc.c b/Objects/obmalloc.c
index 622da3ad08fcd4..9066fb0cc9b5f0 100644
--- a/Objects/obmalloc.c
+++ b/Objects/obmalloc.c
@@ -1206,12 +1206,29 @@ static size_t ntimes_arena_allocated = 0;
 /* High water mark (max value ever seen) for narenas_currently_allocated. */
 static size_t narenas_highwater = 0;
 
-static Py_ssize_t _Py_AllocatedBlocks = 0;
+static Py_ssize_t raw_allocated_blocks;
 
 Py_ssize_t
 _Py_GetAllocatedBlocks(void)
 {
-    return _Py_AllocatedBlocks;
+    Py_ssize_t n = raw_allocated_blocks;
+    /* add up allocated blocks for used pools */
+    for (uint i = 0; i < maxarenas; ++i) {
+        uintptr_t base = arenas[i].address;
+
+        /* Skip arenas which are not allocated. */
+        if (arenas[i].address == (uintptr_t)NULL)
+            continue;
+
+        /* visit every pool in the arena */
+        assert(base <= (uintptr_t) arenas[i].pool_address);
+        for (uint j = 0; base < (uintptr_t) arenas[i].pool_address;
+             ++j, base += POOL_SIZE) {
+            poolp p = (poolp)base;
+            n += p->ref.count;
+        }
+    }
+    return n;
 }
 
 
@@ -1622,13 +1639,12 @@ _PyObject_Malloc(void *ctx, size_t nbytes)
 {
     void* ptr;
     if (pymalloc_alloc(ctx, &ptr, nbytes)) {
-        _Py_AllocatedBlocks++;
         return ptr;
     }
 
     ptr = PyMem_RawMalloc(nbytes);
     if (ptr != NULL) {
-        _Py_AllocatedBlocks++;
+        raw_allocated_blocks++;
     }
     return ptr;
 }
@@ -1644,13 +1660,12 @@ _PyObject_Calloc(void *ctx, size_t nelem, size_t elsize)
 
     if (pymalloc_alloc(ctx, &ptr, nbytes)) {
         memset(ptr, 0, nbytes);
-        _Py_AllocatedBlocks++;
         return ptr;
     }
 
     ptr = PyMem_RawCalloc(nelem, elsize);
     if (ptr != NULL) {
-        _Py_AllocatedBlocks++;
+        raw_allocated_blocks++;
     }
     return ptr;
 }
@@ -1899,10 +1914,10 @@ _PyObject_Free(void *ctx, void *p)
         return;
     }
 
-    _Py_AllocatedBlocks--;
     if (!pymalloc_free(ctx, p)) {
         /* pymalloc didn't allocate this address */
         PyMem_RawFree(p);
+        raw_allocated_blocks--;
     }
 }
 

From 05d3912bba9da7858b760829d913f061d7c2756d Mon Sep 17 00:00:00 2001
From: Neil Schemenauer <nas@arctrix.com>
Date: Wed, 10 Jul 2019 09:49:18 -0700
Subject: [PATCH 2/3] Need to align arena base address.  Remove unused var 'j'.

---
 Objects/obmalloc.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/Objects/obmalloc.c b/Objects/obmalloc.c
index 9066fb0cc9b5f0..bb154c76ab1814 100644
--- a/Objects/obmalloc.c
+++ b/Objects/obmalloc.c
@@ -1214,16 +1214,16 @@ _Py_GetAllocatedBlocks(void)
     Py_ssize_t n = raw_allocated_blocks;
     /* add up allocated blocks for used pools */
     for (uint i = 0; i < maxarenas; ++i) {
-        uintptr_t base = arenas[i].address;
-
         /* Skip arenas which are not allocated. */
-        if (arenas[i].address == (uintptr_t)NULL)
+        if (arenas[i].address == NULL) {
             continue;
+        }
+
+        uintptr_t base = (uintptr_t)_Py_ALIGN_UP(arenas[i].address, POOL_SIZE);
 
         /* visit every pool in the arena */
         assert(base <= (uintptr_t) arenas[i].pool_address);
-        for (uint j = 0; base < (uintptr_t) arenas[i].pool_address;
-             ++j, base += POOL_SIZE) {
+        for (; base < (uintptr_t) arenas[i].pool_address; base += POOL_SIZE) {
             poolp p = (poolp)base;
             n += p->ref.count;
         }

From 5dc07709bd17d6e8eb8c5eb279f6b6aa2efee647 Mon Sep 17 00:00:00 2001
From: Neil Schemenauer <nas@arctrix.com>
Date: Wed, 10 Jul 2019 09:57:08 -0700
Subject: [PATCH 3/3] Add news.

---
 .../Core and Builtins/2019-07-10-09-56-47.bpo-37537.OkB0wd.rst | 3 +++
 1 file changed, 3 insertions(+)
 create mode 100644 Misc/NEWS.d/next/Core and Builtins/2019-07-10-09-56-47.bpo-37537.OkB0wd.rst

diff --git a/Misc/NEWS.d/next/Core and Builtins/2019-07-10-09-56-47.bpo-37537.OkB0wd.rst b/Misc/NEWS.d/next/Core and Builtins/2019-07-10-09-56-47.bpo-37537.OkB0wd.rst
new file mode 100644
index 00000000000000..abf874475d6d9c
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2019-07-10-09-56-47.bpo-37537.OkB0wd.rst	
@@ -0,0 +1,3 @@
+Compute allocated pymalloc blocks inside _Py_GetAllocatedBlocks().  This
+slows down _Py_GetAllocatedBlocks() but gives a small speedup to
+_PyObject_Malloc() and _PyObject_Free().