From 55413b7f7e7246bd399db352f3191d92cc08fe44 Mon Sep 17 00:00:00 2001
From: fbusato <fbusato@nvidia.com>
Date: Wed, 29 Jan 2025 12:29:36 -0800
Subject: [PATCH] add float128 again

---
 .../cuda/std/__cccl/extended_data_types.h     | 24 +++++++++-
 libcudacxx/include/cuda/std/__cccl/os.h       | 48 +++++++++++++++++++
 .../macros/extended_data_types.fp128.fail.cpp | 23 +++++++++
 .../macros/extended_data_types.pass.cpp       |  4 ++
 .../libcxx/macros/os.compile.pass.cpp         | 44 +++++++++++++++++
 5 files changed, 141 insertions(+), 2 deletions(-)
 create mode 100644 libcudacxx/include/cuda/std/__cccl/os.h
 create mode 100644 libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.fp128.fail.cpp
 create mode 100644 libcudacxx/test/libcudacxx/libcxx/macros/os.compile.pass.cpp

diff --git a/libcudacxx/include/cuda/std/__cccl/extended_data_types.h b/libcudacxx/include/cuda/std/__cccl/extended_data_types.h
index 573b22d6ea0..034ebc51b5f 100644
--- a/libcudacxx/include/cuda/std/__cccl/extended_data_types.h
+++ b/libcudacxx/include/cuda/std/__cccl/extended_data_types.h
@@ -23,12 +23,13 @@
 #endif // no system header
 
 #include <cuda/std/__cccl/diagnostic.h>
+#include <cuda/std/__cccl/os.h>
 #include <cuda/std/__cccl/preprocessor.h>
 
 #if !defined(_CCCL_DISABLE_INT128)
-#  if _CCCL_COMPILER(NVRTC) && defined(__CUDACC_RTC_INT128__) && (defined(__linux__) || defined(__LP64__))
+#  if _CCCL_COMPILER(NVRTC) && defined(__CUDACC_RTC_INT128__) && _CCCL_OS(LINUX)
 #    define _CCCL_HAS_INT128() 1
-#  elif defined(__SIZEOF_INT128__) && (defined(__linux__) || defined(__LP64__))
+#  elif defined(__SIZEOF_INT128__) && _CCCL_OS(LINUX)
 #    define _CCCL_HAS_INT128() 1
 #  else
 #    define _CCCL_HAS_INT128() 0
@@ -61,4 +62,23 @@
 #  define _CCCL_HAS_NVFP8() 0
 #endif // !defined(_CCCL_DISABLE_NVFP8_SUPPORT)
 
+#if !defined(_CCCL_DISABLE_FLOAT128)
+#  if _CCCL_COMPILER(NVRTC) && defined(__CUDACC_RTC_FLOAT128__) && _CCCL_OS(LINUX)
+#    if !defined(__CUDA_ARCH__) || (defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 1000)
+#      define _CCCL_HAS_FLOAT128() 1
+#    else
+#      define _CCCL_HAS_FLOAT128() 0
+#    endif
+// NVC++ support float128 only in host code
+#  elif (defined(__SIZEOF_FLOAT128__) || defined(__FLOAT128__)) && _CCCL_OS(LINUX) && !_CCCL_CUDA_COMPILER(NVHPC)
+#    if !defined(__CUDA_ARCH__) || (defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 1000)
+#      define _CCCL_HAS_FLOAT128() 1
+#    else
+#      define _CCCL_HAS_FLOAT128() 0
+#    endif
+#  else
+#    define _CCCL_HAS_FLOAT128() 0
+#  endif
+#endif // !defined(_CCCL_DISABLE_FLOAT128)
+
 #endif // __CCCL_EXTENDED_DATA_TYPES_H
diff --git a/libcudacxx/include/cuda/std/__cccl/os.h b/libcudacxx/include/cuda/std/__cccl/os.h
new file mode 100644
index 00000000000..a3909f5d755
--- /dev/null
+++ b/libcudacxx/include/cuda/std/__cccl/os.h
@@ -0,0 +1,48 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2025 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CCCL_OS_H
+#define __CCCL_OS_H
+
+// The header provides the following macros to determine the host architecture:
+//
+// _CCCL_OS(WINDOWS)
+// _CCCL_OS(LINUX)
+// _CCCL_OS(ANDROID)
+// _CCCL_OS(QNX)
+
+// Determine the host compiler and its version
+#if defined(_WIN32) || defined(_WIN64) /* _WIN64 for NVRTC */
+#  define _CCCL_OS_WINDOWS_() 1
+#else
+#  define _CCCL_OS_WINDOWS_() 0
+#endif
+
+#if defined(__linux__) || defined(__LP64__) /* __LP64__ for NVRTC */
+#  define _CCCL_OS_LINUX_() 1
+#else
+#  define _CCCL_OS_LINUX_() 0
+#endif
+
+#if defined(__ANDROID__)
+#  define _CCCL_OS_ANDROID_() 1
+#else
+#  define _CCCL_OS_ANDROID_() 0
+#endif
+
+#if defined(__QNX__) || defined(__QNXNTO__)
+#  define _CCCL_OS_QNX_() 1
+#else
+#  define _CCCL_OS_QNX_() 0
+#endif
+
+#define _CCCL_OS(...) _CCCL_OS_##__VA_ARGS__##_()
+
+#endif // __CCCL_OS_H
diff --git a/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.fp128.fail.cpp b/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.fp128.fail.cpp
new file mode 100644
index 00000000000..5cebaa96168
--- /dev/null
+++ b/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.fp128.fail.cpp
@@ -0,0 +1,23 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2025 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+#include <cuda/std/__cccl/extended_data_types.h>
+
+#include "test_macros.h"
+
+int main(int, char**)
+{
+#if !_CCCL_HAS_FLOAT128()
+  __float128 x4 = __float128(3.14) + __float128(3.14);
+  unused(x4);
+#else
+  static_assert(false);
+#endif
+  return 0;
+}
diff --git a/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.pass.cpp b/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.pass.cpp
index e8c770e0074..3b230d8710b 100644
--- a/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.pass.cpp
+++ b/libcudacxx/test/libcudacxx/libcxx/macros/extended_data_types.pass.cpp
@@ -39,6 +39,10 @@ int main(int, char**)
 #if defined(_CCCL_HAS_NVBF16)
   auto x4 = __nv_bfloat16(1.0f);
   unused(x4);
+#endif
+#if _CCCL_HAS_FLOAT128()
+  __float128 x5 = __float128(3.14) + __float128(3.14);
+  unused(x5);
 #endif
   return 0;
 }
diff --git a/libcudacxx/test/libcudacxx/libcxx/macros/os.compile.pass.cpp b/libcudacxx/test/libcudacxx/libcxx/macros/os.compile.pass.cpp
new file mode 100644
index 00000000000..9bc788526c9
--- /dev/null
+++ b/libcudacxx/test/libcudacxx/libcxx/macros/os.compile.pass.cpp
@@ -0,0 +1,44 @@
+//===----------------------------------------------------------------------===//
+//
+// 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/__cccl/os.h>
+
+#if !defined(__CUDACC_RTC__)
+#  if _CCCL_OS(WINDOWS)
+#    include <windows.h>
+#  endif
+
+#  if _CCCL_OS(LINUX)
+#    include <unistd.h>
+#  endif
+
+#  if _CCCL_OS(ANDROID)
+#    include <android/api-level.h>
+#  endif
+
+#  if _CCCL_OS(QNX)
+#    include <qnx.h>
+#  endif
+#endif
+
+int main(int, char**)
+{
+  static_assert(_CCCL_OS(WINDOWS) + _CCCL_OS(LINUX) == 1, "");
+#if _CCCL_OS(ANDROID) || _CCCL_OS(QNX)
+  static_assert(_CCCL_OS(LINUX) == 1, "");
+  static_assert(_CCCL_OS(ANDROID) + _CCCL_OS(QNX) == 1, "");
+#endif
+#if _CCCL_OS(LINUX)
+  static_assert(_CCCL_OS(WINDOWS) == 0, "");
+#endif
+#if _CCCL_OS(WINDOWS)
+  static_assert(_CCCL_OS(ANDROID) + _CCCL_OS(QNX) + _CCCL_OS(LINUX) == 0, "");
+#endif
+  return 0;
+}