From 6c8a45c88c7dd32838cd039c8beef0c9043401d7 Mon Sep 17 00:00:00 2001
From: Allard Hendriksen <ahendriksen@nvidia.com>
Date: Tue, 17 Dec 2024 12:24:06 +0100
Subject: [PATCH 1/2] ptx: Add clusterlaunchcontrol

---
 .../ptx/instructions/clusterlaunchcontrol.rst | 11 ++++++
 .../__ptx/instructions/clusterlaunchcontrol.h | 37 +++++++++++++++++++
 libcudacxx/include/cuda/ptx                   |  1 +
 .../ptx.clusterlaunchcontrol.compile.pass.cpp | 22 +++++++++++
 4 files changed, 71 insertions(+)
 create mode 100644 docs/libcudacxx/ptx/instructions/clusterlaunchcontrol.rst
 create mode 100644 libcudacxx/include/cuda/__ptx/instructions/clusterlaunchcontrol.h
 create mode 100644 libcudacxx/test/libcudacxx/cuda/ptx/ptx.clusterlaunchcontrol.compile.pass.cpp

diff --git a/docs/libcudacxx/ptx/instructions/clusterlaunchcontrol.rst b/docs/libcudacxx/ptx/instructions/clusterlaunchcontrol.rst
new file mode 100644
index 00000000000..75fe44f6f22
--- /dev/null
+++ b/docs/libcudacxx/ptx/instructions/clusterlaunchcontrol.rst
@@ -0,0 +1,11 @@
+.. _libcudacxx-ptx-instructions-clusterlaunchcontrol:
+
+clusterlaunchcontrol
+====================
+
+-  PTX ISA:
+   `clusterlaunchcontrol.try_cancel <https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel>`__
+-  PTX ISA:
+   `clusterlaunchcontrol.query_cancel <https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-query-cancel>`__
+
+.. include:: generated/clusterlaunchcontrol.rst
diff --git a/libcudacxx/include/cuda/__ptx/instructions/clusterlaunchcontrol.h b/libcudacxx/include/cuda/__ptx/instructions/clusterlaunchcontrol.h
new file mode 100644
index 00000000000..b15cfddf4a0
--- /dev/null
+++ b/libcudacxx/include/cuda/__ptx/instructions/clusterlaunchcontrol.h
@@ -0,0 +1,37 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// Part of libcu++, the C++ Standard Library for your entire system,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _CUDA_PTX_CLUSTERLAUNCHCONTROL_H_
+#define _CUDA_PTX_CLUSTERLAUNCHCONTROL_H_
+
+#include <cuda/std/detail/__config>
+
+#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
+#  pragma GCC system_header
+#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
+#  pragma clang system_header
+#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
+#  pragma system_header
+#endif // no system header
+
+#include <cuda/__ptx/ptx_dot_variants.h>
+#include <cuda/__ptx/ptx_helper_functions.h>
+#include <cuda/std/cstdint>
+
+#include <nv/target> // __CUDA_MINIMUM_ARCH__ and friends
+
+_LIBCUDACXX_BEGIN_NAMESPACE_CUDA_PTX
+
+#include <cuda/__ptx/instructions/generated/clusterlaunchcontrol.h>
+
+_LIBCUDACXX_END_NAMESPACE_CUDA_PTX
+
+#endif // _CUDA_PTX_CLUSTERLAUNCHCONTROL_H_
diff --git a/libcudacxx/include/cuda/ptx b/libcudacxx/include/cuda/ptx
index 4798973df77..7087dd97d2a 100644
--- a/libcudacxx/include/cuda/ptx
+++ b/libcudacxx/include/cuda/ptx
@@ -70,6 +70,7 @@
  */
 
 #include <cuda/__ptx/instructions/barrier_cluster.h>
+#include <cuda/__ptx/instructions/clusterlaunchcontrol.h>
 #include <cuda/__ptx/instructions/cp_async_bulk.h>
 #include <cuda/__ptx/instructions/cp_async_bulk_commit_group.h>
 #include <cuda/__ptx/instructions/cp_async_bulk_tensor.h>
diff --git a/libcudacxx/test/libcudacxx/cuda/ptx/ptx.clusterlaunchcontrol.compile.pass.cpp b/libcudacxx/test/libcudacxx/cuda/ptx/ptx.clusterlaunchcontrol.compile.pass.cpp
new file mode 100644
index 00000000000..212414c4535
--- /dev/null
+++ b/libcudacxx/test/libcudacxx/cuda/ptx/ptx.clusterlaunchcontrol.compile.pass.cpp
@@ -0,0 +1,22 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of libcu++, the C++ Standard Library for your entire system,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+// UNSUPPORTED: libcpp-has-no-threads
+
+// <cuda/ptx>
+
+#include <cuda/ptx>
+#include <cuda/std/utility>
+
+#include "generated/clusterlaunchcontrol.h"
+
+int main(int, char**)
+{
+  return 0;
+}

From 7cdf7cff77f914826245873ee9aa24630221dbd3 Mon Sep 17 00:00:00 2001
From: Bernhard Manfred Gruber <bernhardmgruber@gmail.com>
Date: Thu, 30 Jan 2025 11:13:21 +0100
Subject: [PATCH 2/2] docs fix

---
 docs/libcudacxx/ptx/instructions.rst | 1 +
 1 file changed, 1 insertion(+)

diff --git a/docs/libcudacxx/ptx/instructions.rst b/docs/libcudacxx/ptx/instructions.rst
index f0776974eec..32db843c28d 100644
--- a/docs/libcudacxx/ptx/instructions.rst
+++ b/docs/libcudacxx/ptx/instructions.rst
@@ -7,6 +7,7 @@ PTX Instructions
    :maxdepth: 1
 
    instructions/barrier_cluster
+   instructions/clusterlaunchcontrol
    instructions/cp_async_bulk
    instructions/cp_async_bulk_commit_group
    instructions/cp_async_bulk_wait_group