Refactor limits and climits (#3221)

* implement builtins for huge val, nan and nans * change `INFINITY` and `NAN` implementation for NVRTC
NVIDIA · Jan 9, 2025 · 1503b25 · 1503b25
1 parent ad5e7b2
commit 1503b25
Show file tree

Hide file tree

Showing 18 changed files with 678 additions and 1,314 deletions.
diff --git a/libcudacxx/include/cuda/std/__cccl/builtin.h b/libcudacxx/include/cuda/std/__cccl/builtin.h
@@ -288,6 +288,48 @@
 #  define _CCCL_BUILTIN_LINE() __LINE__
 #endif // _CCCL_CUDACC_BELOW(11, 3)
 
+#if _CCCL_CHECK_BUILTIN(builtin_huge_valf) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_HUGE_VALF() __builtin_huge_valf()
+#endif // _CCCL_CHECK_BUILTIN(builtin_huge_valf)
+
+#if _CCCL_CHECK_BUILTIN(builtin_huge_val) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_HUGE_VAL() __builtin_huge_val()
+#endif // _CCCL_CHECK_BUILTIN(builtin_huge_val)
+
+#if _CCCL_CHECK_BUILTIN(builtin_huge_vall) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_HUGE_VALL() __builtin_huge_vall()
+#elif _CCCL_COMPILER(MSVC)
+#  define _CCCL_BUILTIN_HUGE_VALL() static_cast<long double>(__builtin_huge_val())
+#endif // _CCCL_CHECK_BUILTIN(builtin_huge_vall)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nanf) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NANF(...) __builtin_nanf(__VA_ARGS__)
+#endif // _CCCL_CHECK_BUILTIN(builtin_nanf)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nan) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NAN(...) __builtin_nan(__VA_ARGS__)
+#endif // _CCCL_CHECK_BUILTIN(builtin_nan)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nanl) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NANL(...) __builtin_nanl(__VA_ARGS__)
+#elif _CCCL_COMPILER(MSVC)
+#  define _CCCL_BUILTIN_NANL(...) static_cast<long double>(__builtin_nan(__VA_ARGS__))
+#endif // _CCCL_CHECK_BUILTIN(builtin_nanl)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nansf) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NANSF(...) __builtin_nansf(__VA_ARGS__)
+#endif // _CCCL_CHECK_BUILTIN(builtin_nansf)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nans) || _CCCL_COMPILER(MSVC) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NANS(...) __builtin_nans(__VA_ARGS__)
+#endif // _CCCL_CHECK_BUILTIN(builtin_nans)
+
+#if _CCCL_CHECK_BUILTIN(builtin_nansl) || _CCCL_COMPILER(GCC, <, 10)
+#  define _CCCL_BUILTIN_NANSL(...) __builtin_nansl(__VA_ARGS__)
+#elif _CCCL_COMPILER(MSVC)
+#  define _CCCL_BUILTIN_NANSL(...) static_cast<long double>(__builtin_nans(__VA_ARGS__))
+#endif // _CCCL_CHECK_BUILTIN(builtin_nansl)
+
 #if _CCCL_CHECK_BUILTIN(builtin_log) || _CCCL_COMPILER(GCC)
 #  define _CCCL_BUILTIN_LOGF(...) __builtin_logf(__VA_ARGS__)
 #  define _CCCL_BUILTIN_LOG(...)  __builtin_log(__VA_ARGS__)

diff --git a/libcudacxx/include/cuda/std/__cuda/climits_prelude.h b/libcudacxx/include/cuda/std/__cuda/climits_prelude.h
diff --git a/...include/support/win32/limits_msvc_win32.h → ...xx/include/cuda/std/__limits/msvc_win32.h b/...include/support/win32/limits_msvc_win32.h → ...xx/include/cuda/std/__limits/msvc_win32.h
@@ -1,14 +1,16 @@
 // -*- C++ -*-
-//===------------------ support/win32/limits_msvc_win32.h -----------------===//
+//===----------------------------------------------------------------------===//
 //
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// 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 _LIBCUDACXX_SUPPORT_WIN32_LIMITS_MSVC_WIN32_H
-#define _LIBCUDACXX_SUPPORT_WIN32_LIMITS_MSVC_WIN32_H
+#ifndef _LIBCUDACXX___LIMITS_MSVC_WIN32_H
+#define _LIBCUDACXX___LIMITS_MSVC_WIN32_H
 
 #if defined(__MINGW32__)
 #  error "This header complements the Microsoft C Runtime library, and should not be included otherwise."
@@ -19,8 +21,6 @@
 
 #include <float.h> // limit constants
 #include <limits.h> // CHAR_BIT
-#include <math.h> // HUGE_VAL
-#include <ymath.h> // internal MSVC header providing the needed functionality
 
 #define __CHAR_BIT__ CHAR_BIT
 
@@ -63,15 +63,4 @@
 // predefined by MinGW GCC
 #define __LDBL_DENORM_MIN__ 3.64519953188247460253e-4951L
 
-// __builtin replacements/workarounds
-#if _MSC_VER < 1934
-#  define __builtin_huge_vall()    _LInf._Long_double
-#  define __builtin_nanl(__dummmy) _LNan._Long_double
-#  define __builtin_nansl(__dummy) _LSnan._Long_double
-#else
-#  define __builtin_huge_vall() __builtin_huge_val()
-#  define __builtin_nanl(__v)   __builtin_nan(__v)
-#  define __builtin_nansl(__v)  __builtin_nans(__v)
-#endif
-
-#endif // _LIBCUDACXX_SUPPORT_WIN32_LIMITS_MSVC_WIN32_H
+#endif // _LIBCUDACXX___LIMITS_MSVC_WIN32_H
diff --git a/libcudacxx/include/cuda/std/__memory/allocator_traits.h b/libcudacxx/include/cuda/std/__memory/allocator_traits.h
@@ -28,6 +28,8 @@
 #include <cuda/std/__type_traits/is_copy_constructible.h>
 #include <cuda/std/__type_traits/is_empty.h>
 #include <cuda/std/__type_traits/is_move_constructible.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/__type_traits/is_trivially_move_constructible.h>
 #include <cuda/std/__type_traits/make_unsigned.h>
 #include <cuda/std/__type_traits/remove_reference.h>
 #include <cuda/std/__type_traits/void_t.h>

diff --git a/libcudacxx/include/cuda/std/__memory/temporary_buffer.h b/libcudacxx/include/cuda/std/__memory/temporary_buffer.h
@@ -26,10 +26,11 @@
 #include <cuda/std/__iterator/iterator_traits.h>
 #include <cuda/std/__memory/addressof.h>
 #include <cuda/std/__new_>
+#include <cuda/std/__type_traits/alignment_of.h>
 #include <cuda/std/__utility/move.h>
 #include <cuda/std/__utility/pair.h>
 #include <cuda/std/cstddef>
-#include <cuda/std/detail/libcxx/include/limits>
+#include <cuda/std/limits>
 
 _LIBCUDACXX_BEGIN_NAMESPACE_STD
 

diff --git a/libcudacxx/include/cuda/std/climits b/libcudacxx/include/cuda/std/climits
@@ -23,7 +23,70 @@
 
 _CCCL_PUSH_MACROS
 
-#include <cuda/std/detail/libcxx/include/climits>
+#if !_CCCL_COMPILER(NVRTC)
+#  include <climits>
+#else // ^^^ !_CCCL_COMPILER(NVRTC) ^^^ / vvv _CCCL_COMPILER(NVRTC) vvv
+#  define CHAR_BIT 8
+
+#  define SCHAR_MIN         (-128)
+#  define SCHAR_MAX         127
+#  define UCHAR_MAX         255
+#  define __CHAR_UNSIGNED__ ('\xff' > 0) // CURSED
+#  if __CHAR_UNSIGNED__
+#    define CHAR_MIN 0
+#    define CHAR_MAX UCHAR_MAX
+#  else
+#    define CHAR_MIN SCHAR_MIN
+#    define CHAR_MAX SCHAR_MAX
+#  endif
+#  define SHRT_MIN  (-SHRT_MAX - 1)
+#  define SHRT_MAX  0x7fff
+#  define USHRT_MAX 0xffff
+#  define INT_MIN   (-INT_MAX - 1)
+#  define INT_MAX   0x7fffffff
+#  define UINT_MAX  0xffffffff
+#  define LONG_MIN  (-LONG_MAX - 1)
+#  ifdef __LP64__
+#    define LONG_MAX  LLONG_MAX
+#    define ULONG_MAX ULLONG_MAX
+#  else
+#    define LONG_MAX  INT_MAX
+#    define ULONG_MAX UINT_MAX
+#  endif
+#  define LLONG_MIN  (-LLONG_MAX - 1)
+#  define LLONG_MAX  0x7fffffffffffffffLL
+#  define ULLONG_MAX 0xffffffffffffffffUL
+
+#  define __FLT_RADIX__      2
+#  define __FLT_MANT_DIG__   24
+#  define __FLT_DIG__        6
+#  define __FLT_MIN__        1.17549435082228750796873653722224568e-38F
+#  define __FLT_MAX__        3.40282346638528859811704183484516925e+38F
+#  define __FLT_EPSILON__    1.19209289550781250000000000000000000e-7F
+#  define __FLT_MIN_EXP__    (-125)
+#  define __FLT_MIN_10_EXP__ (-37)
+#  define __FLT_MAX_EXP__    128
+#  define __FLT_MAX_10_EXP__ 38
+#  define __FLT_DENORM_MIN__ 1.40129846432481707092372958328991613e-45F
+#  define __DBL_MANT_DIG__   53
+#  define __DBL_DIG__        15
+#  define __DBL_MIN__        2.22507385850720138309023271733240406e-308
+#  define __DBL_MAX__        1.79769313486231570814527423731704357e+308
+#  define __DBL_EPSILON__    2.22044604925031308084726333618164062e-16
+#  define __DBL_MIN_EXP__    (-1021)
+#  define __DBL_MIN_10_EXP__ (-307)
+#  define __DBL_MAX_EXP__    1024
+#  define __DBL_MAX_10_EXP__ 308
+#  define __DBL_DENORM_MIN__ 4.94065645841246544176568792868221372e-324
+#endif // _CCCL_COMPILER(NVRTC)
+
+#if _CCCL_COMPILER(MSVC)
+#  include <cuda/std/__limits/msvc_win32.h>
+#endif // _CCCL_COMPILER(MSVC)
+
+#ifndef __CHAR_BIT__
+#  define __CHAR_BIT__ 8
+#endif // !__CHAR_BIT__
 
 _CCCL_POP_MACROS
 

diff --git a/libcudacxx/include/cuda/std/detail/libcxx/include/climits b/libcudacxx/include/cuda/std/detail/libcxx/include/climits
diff --git a/libcudacxx/include/cuda/std/detail/libcxx/include/cmath b/libcudacxx/include/cuda/std/detail/libcxx/include/cmath
@@ -338,9 +338,8 @@ long double    truncl(long double x);
 #endif // _LIBCUDACXX_HAS_NVBF16
 
 #if _CCCL_COMPILER(NVRTC)
-#  include <cuda/std/climits>
-#  define INFINITY __builtin_huge_val()
-#  define NAN      __builtin_nan()
+#  define INFINITY _CUDA_VSTD::numeric_limits<float>::infinity()
+#  define NAN      _CUDA_VSTD::numeric_limits<float>::quiet_NaN()
 #endif // _CCCL_COMPILER(NVRTC)
 
 _CCCL_PUSH_MACROS