Add `MKLMemory` class to expose MKL allocated memory via Python buffer protocol by ndgrigorian · Pull Request #182 · IntelPython/mkl-service

ndgrigorian · 2026-04-12T04:25:01Z

This PR proposes the introduction of _mkl_memory.pyx, which implements an MKLMemory class that exposes memory allocated via mkl_malloc and mkl_calloc to Python via the buffer protocol

The class uses an atomic counter incremented as __getbuffer__ and __releasebuffer__ are called to track the views on the buffer to permit use of mkl_realloc in the object (via realloc method). This concept was adapted from the PEP which revised the buffer protocol which proposed this kind of approach to tracking views on a buffer

Closes #18

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds a new MKLMemory Cython extension type backed by MKL’s allocator, exposes it from the top-level mkl package, and introduces tests/build changes to support C11 atomics and nogil MKL calls.

Changes:

Introduce mkl._mkl_memory with MKLMemory (allocation, buffer protocol, pickling, realloc).
Add pytest coverage for allocation, buffer protocol, and pickling behavior.
Update MKL C-API declarations/build to support nogil calls and C11 atomics (plus MSVC flag).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
mkl/tests/test_mkl_memory.py	Adds tests for MKLMemory creation, buffer protocol, and pickling behavior.
mkl/_py_mkl_service.pyx	Releases the GIL around MKL buffer-free calls.
mkl/_mkl_service.pxd	Marks MKL externs as `nogil` and adds malloc/calloc/realloc/free declarations.
mkl/_mkl_memory.pyx	Adds the new `MKLMemory` Cython extension implementing allocation + buffer protocol + pickling.
mkl/init.py	Exposes `MKLMemory` at the package top level.
meson.build	Enables C11, adds MSVC atomics flag, and builds the new `_mkl_memory` extension.

exposes Python buffer protocol

as atomics are a c11+ feature, specific flags are needed to enable on window

also address issues with undeclared variables and rename MKLMemory class members

ndgrigorian · 2026-09-03T21:38:15Z

@antonwolfy
Would be nice to get this reviewed and see if it can make it in the next release

antonwolfy · 2026-09-04T09:44:25Z

@@ -0,0 +1,271 @@
+# Copyright (c) 2018, Intel Corporation


Should be a year when file is created

Suggested change

# Copyright (c) 2018, Intel Corporation

# Copyright (c) 2026, Intel Corporation

antonwolfy · 2026-09-04T09:45:19Z

    subdir: 'mkl'
 )

+py.extension_module(


Please populate the changelog

antonwolfy · 2026-09-04T10:00:10Z

+            if new_nbytes <= 0:
+                raise ValueError("New number of bytes must be positive.")
+
+            with nogil:


There might be use-after-free issue:

Thread B (holding GIL): passes exported_buffers == 0 and the refcount check.

Thread B enters with nogil: → releases the GIL → calls mkl_realloc.

Thread A acquires the GIL, runs memoryview(mem) → __getbuffer__ reads the old _memory_ptr into buffer.buf, bumps the counter.

Thread B's mkl_realloc frees/moves the old block, reacquires GIL, sets _memory_ptr = p.

Thread A now holds a view into freed memory → UAF.

antonwolfy · 2026-09-04T10:08:19Z

+
+            if (p):
+                self._memory_ptr = p
+                self._nbytes = num * size


Needs to validate there is no num > PY_SSIZE_T_MAX / size

antonwolfy · 2026-09-04T10:13:30Z

+        with nogil:
+            memcpy(self._memory_ptr, other_mem._memory_ptr, self._nbytes)
+
+    def __cinit__(self, *args, **kwargs):


No check on name of alignment keyword.
Any typo in name, like MKLMemory(1024, alignmnet=128) (instead of alignment) will be silently ignored and fallback on 64.

Also no type-check before the Py_ssize_t conversion

antonwolfy · 2026-09-04T10:36:56Z

+        return self._nbytes
+
+    @property
+    def size(self):


That duplicates nbytes.
Also that collides with the constructor's .size parameter, which means something different.
And that collides with the NumPy convention where .size is the element count and .nbytes is the byte count.

antonwolfy · 2026-09-04T10:38:54Z

 cc = meson.get_compiler('c')
+if cc.get_id() == 'msvc'
+    add_project_arguments(
+        '/experimental:c11atomics',


/experimental:c11atomics requires VS 2022 17.5+ and fails hard on older toolchains.
Should we document the floor?

antonwolfy · 2026-09-04T10:40:12Z

+    def __repr__(self):
+        return (
+            f"<MKL memory allocation of {self._nbytes} bytes at "
+            f"{hex(<object>(<size_t>self._memory_ptr))}>"


It can be simplified:

Suggested change

f"{hex(<object>(<size_t>self._memory_ptr))}>"

f"{hex(self._pointer)}>"

Both forms cythonized to the same:

with_cast — hex(<object>(<size_t>x)): __pyx_t_1 = __Pyx_PyLong_FromSize_t(((size_t)__pyx_v_x)); ... __pyx_t_2 = __Pyx_PyNumber_Hex(__pyx_t_1); ... without_cast — hex(<size_t>x): __pyx_t_1 = __Pyx_PyLong_FromSize_t(((size_t)__pyx_v_x)); ... __pyx_t_2 = __Pyx_PyNumber_Hex(__pyx_t_1); ...

antonwolfy · 2026-09-04T10:50:42Z

@@ -0,0 +1,271 @@
+# Copyright (c) 2018, Intel Corporation


Should we update AGENTS.md, .github/copilot-instructions.md and mkl/tests/AGENTS.md with the new test file?

The the same for _py_mkl_service files

antonwolfy · 2026-09-04T11:31:57Z


 cc = meson.get_compiler('c')
+if cc.get_id() == 'msvc'
+    add_project_arguments(


That is only needed for _py_mkl_service. We probably should not add it everywhere, considering it's experimental:

mkl_memory_c_args = c_args if cc.get_id() == 'msvc' mkl_memory_c_args += '/experimental:c11atomics' endif

ndgrigorian force-pushed the feature/add-mkl-memory branch 5 times, most recently from aed1ff6 to a99e626 Compare April 12, 2026 07:19

ndgrigorian marked this pull request as ready for review April 12, 2026 09:03

ndgrigorian requested review from antonwolfy, jharlow-intel and xaleryb as code owners April 12, 2026 09:03

ndgrigorian force-pushed the use-meson-build branch from ef09add to ac6e9fc Compare May 10, 2026 01:12

ndgrigorian force-pushed the feature/add-mkl-memory branch 3 times, most recently from b2cc6fc to 00fea26 Compare May 10, 2026 02:02

ndgrigorian force-pushed the use-meson-build branch from ac6e9fc to b39b64b Compare May 15, 2026 20:08

Base automatically changed from use-meson-build to master July 21, 2026 14:49

Copilot AI lite review requested due to automatic review settings August 6, 2026 15:58

ndgrigorian force-pushed the feature/add-mkl-memory branch from 00fea26 to c764a81 Compare August 6, 2026 15:58

ndgrigorian requested a review from vlad-perevezentsev as a code owner August 6, 2026 15:58

Copilot AI reviewed Aug 6, 2026

View reviewed changes

ndgrigorian added 11 commits September 3, 2026 14:03

add MKLMemory object, backed with mkl_malloc memory

7bd5c95

exposes Python buffer protocol

add realloc to MKLMemory

d1d7211

as atomics are a c11+ feature, specific flags are needed to enable on window

overload MKLMemory constructor to use mkl_calloc

c5fdcd5

add info properties to MKLMemory

b5ea076

add pickling support for MKLMemory

20995e7

propagate alignment in MKLMemory

f85d0b5

add tests for MKLMemory class

ffa398f

add nogil to MKL functions for freeing buffers

42b59b0

fix meson.build rpath

910d982

mark _mkl_memory free-threading compatible

4b88d62

update meson.build

8cc7b43

ndgrigorian force-pushed the feature/add-mkl-memory branch from 0d4ccfb to 0c75d30 Compare September 3, 2026 21:04

align realloc behavior to NumPy

6aae4fb

also address issues with undeclared variables and rename MKLMemory class members

ndgrigorian force-pushed the feature/add-mkl-memory branch from 0c75d30 to 6aae4fb Compare September 3, 2026 21:34

antonwolfy added this to the 2.9.0 release milestone Sep 4, 2026

antonwolfy reviewed Sep 4, 2026

View reviewed changes

	# Copyright (c) 2018, Intel Corporation
	# Copyright (c) 2026, Intel Corporation

	f"{hex(<object>(<size_t>self._memory_ptr))}>"
	f"{hex(self._pointer)}>"

Uh oh!

Conversation

ndgrigorian commented Apr 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ndgrigorian commented Sep 3, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ndgrigorian commented Apr 12, 2026 •

edited

Loading