Add _CCCL_BUILTIN_PREFETCH (#3433)

NVIDIA · Jan 21, 2025 · d2857b1 · d2857b1
1 parent da9f6e3
commit d2857b1
Show file tree

Hide file tree

Showing 3 changed files with 37 additions and 7 deletions.
diff --git a/docs/cccl_development/macro.rst b/docs/cccl_development/macro.rst
@@ -264,13 +264,15 @@ Usage example:
 
 **Portable Builtin Macros**:
 
-+-----------------------------+--------------------------------------------+
-| ``_CCCL_UNREACHABLE()``     | Portable ``__builtin_unreachable()``       |
-+-----------------------------+--------------------------------------------+
-| ``_CCCL_BUILTIN_ASSUME(X)`` | Portable ``__builtin_assume(X)``           |
-+-----------------------------+--------------------------------------------+
-| ``_CCCL_BUILTIN_EXPECT(X)`` | Portable ``__builtin_expected(X)``         |
-+-----------------------------+--------------------------------------------+
++---------------------------------------+--------------------------------------------+
+| ``_CCCL_UNREACHABLE()``               | Portable ``__builtin_unreachable()``       |
++---------------------------------------+--------------------------------------------+
+| ``_CCCL_BUILTIN_ASSUME(X)``           | Portable ``__builtin_assume(X)``           |
++---------------------------------------+--------------------------------------------+
+| ``_CCCL_BUILTIN_EXPECT(X)``           | Portable ``__builtin_expected(X)``         |
++---------------------------------------+--------------------------------------------+
+| ``_CCCL_BUILTIN_PREFETCH(X[, Y, Z])`` | Portable ``__builtin_prefetch(X, Y, Z)``   |
++---------------------------------------+--------------------------------------------+
 
 **Portable Keyword Macros**
 

diff --git a/libcudacxx/include/cuda/std/__cccl/builtin.h b/libcudacxx/include/cuda/std/__cccl/builtin.h
@@ -111,6 +111,12 @@
     NV_IF_ELSE_TARGET(NV_IS_DEVICE, (__builtin_assume(__VA_ARGS__);), (__assume(__VA_ARGS__);))
 #endif // _CCCL_CHECK_BUILTIN(builtin_assume)
 
+#if _CCCL_CHECK_BUILTIN(builtin_prefetch) || _CCCL_COMPILER(GCC)
+#  define _CCCL_BUILTIN_PREFETCH(...) NV_IF_TARGET(NV_IS_HOST, __builtin_prefetch(__VA_ARGS__);)
+#else
+#  define _CCCL_BUILTIN_PREFETCH(...)
+#endif // _CCCL_CHECK_BUILTIN(builtin_prefetch)
+
 // NVCC prior to 11.2 cannot handle __builtin_assume
 #if _CCCL_CUDACC_BELOW(11, 2)
 #  undef _CCCL_BUILTIN_ASSUME

diff --git a/libcudacxx/test/libcudacxx/libcxx/macros/prefetch.compile.pass.cpp b/libcudacxx/test/libcudacxx/libcxx/macros/prefetch.compile.pass.cpp
@@ -0,0 +1,22 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, 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) 2023 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#include <cuda/std/detail/__config>
+
+#include <test_macros.h>
+
+int main(int, char**)
+{
+  int memory[8];
+  _CCCL_BUILTIN_PREFETCH(memory);
+  _CCCL_BUILTIN_PREFETCH(memory, /*read-only=*/0);
+  _CCCL_BUILTIN_PREFETCH(memory, /*read-only=*/0, /*medium cache utilization=*/1);
+  unused(memory);
+  return 0;
+}