Merge pull request #717 from cppalliance/real_cuda

.github/workflows/cuda.yml

@@ -0,0 +1,78 @@
+# Copyright 2024 Matt Borland
+# Distributed under the Boost Software License, Version 1.0.
+# (See accompanying file LICENSE_1_0.txt or copy at http://boost.org/LICENSE_1_0.txt)
+
+name: GPUs
+on:
+ push:
+ branches:
+ - master
+ - develop
+ - feature/**
+ pull_request:
+ release:
+ types: [published, created, edited]
+
+concurrency:
+ group: ${{ github.head_ref || github.run_id }}
+ cancel-in-progress: true
+
+jobs:
+ cuda-cmake-test:
+ strategy:
+ fail-fast: false
+
+ runs-on: gpu-runner-1
+
+ steps:
+ - uses: Jimver/cuda-toolkit@v0.2.16
+ id: cuda-toolkit
+ with:
+ cuda: '12.5.0'
+ method: 'network'
+ sub-packages: '["nvcc"]'
+
+ - name: Output CUDA information
+ run: |
+ echo "Installed cuda version is: ${{steps.cuda-toolkit.outputs.cuda}}"+
+ echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
+ nvcc -V
+ - uses: actions/checkout@v4
+
+ - name: Install Packages
+ run: |
+ sudo apt-get install -y cmake make
+ - name: Setup Boost
+ run: |
+ echo GITHUB_REPOSITORY: $GITHUB_REPOSITORY
+ LIBRARY=${GITHUB_REPOSITORY#*/}
+ echo LIBRARY: $LIBRARY
+ echo "LIBRARY=$LIBRARY" >> $GITHUB_ENV
+ echo GITHUB_BASE_REF: $GITHUB_BASE_REF
+ echo GITHUB_REF: $GITHUB_REF
+ REF=${GITHUB_BASE_REF:-$GITHUB_REF}
+ REF=${REF#refs/heads/}
+ echo REF: $REF
+ BOOST_BRANCH=develop && [ "$REF" == "master" ] && BOOST_BRANCH=master || true
+ echo BOOST_BRANCH: $BOOST_BRANCH
+ cd ..
+ git clone -b $BOOST_BRANCH --depth 1 https://github.com/boostorg/boost.git boost-root
+ cd boost-root
+ mkdir -p libs/$LIBRARY
+ cp -r $GITHUB_WORKSPACE/* libs/$LIBRARY
+ git submodule update --init tools/boostdep
+ python3 tools/boostdep/depinst/depinst.py --git_args "--jobs 3" $LIBRARY
+ - name: Configure
+ run: |
+ cd ../boost-root
+ mkdir __build__ && cd __build__
+ cmake -DBOOST_INCLUDE_LIBRARIES=$LIBRARY -DBUILD_TESTING=ON -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc -DBOOST_DECIMAL_ENABLE_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES=70 -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-12.5 ..
+ - name: Build tests
+ run: |
+ cd ../boost-root/__build__
+ cmake --build . --target tests -j $(nproc)
+ - name: Run tests
+ run: |
+ cd ../boost-root/__build__
+ ctest --output-on-failure --no-tests=error
+ 

include/boost/decimal/detail/add_impl.hpp

@@ -5,9 +5,12 @@
 #ifndef BOOST_DECIMAL_DETAIL_ADD_IMPL_HPP
 #define BOOST_DECIMAL_DETAIL_ADD_IMPL_HPP
 
+#include <boost/decimal/detail/config.hpp>
 #include <boost/decimal/detail/attributes.hpp>
 #include <boost/decimal/detail/apply_sign.hpp>
 #include <boost/decimal/detail/fenv_rounding.hpp>
+#include <boost/decimal/detail/type_traits.hpp>
+#include <boost/decimal/detail/numeric_limits.hpp>
 
 #ifndef BOOST_DECIMAL_BUILD_MODULE
 #include <cstdint>
@@ -18,9 +21,10 @@ namespace decimal {
 namespace detail {
 
 template <typename ReturnType, typename T, typename U>
-BOOST_DECIMAL_FORCE_INLINE constexpr auto d32_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
- T rhs_sig, U rhs_exp, bool rhs_sign,
- bool abs_lhs_bigger) noexcept -> ReturnType
+BOOST_DECIMAL_GPU_ENABLED BOOST_DECIMAL_FORCE_INLINE 
+constexpr auto d32_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
+ T rhs_sig, U rhs_exp, bool rhs_sign,
+ bool abs_lhs_bigger) noexcept -> ReturnType
 {
  using add_type = std::int_fast32_t;
 
@@ -64,7 +68,7 @@ BOOST_DECIMAL_FORCE_INLINE constexpr auto d32_add_impl(T lhs_sig, U lhs_exp, boo
 
  if (delta_exp <= 2)
  {
- sig_bigger *= pow10(static_cast<std::remove_reference_t<decltype(sig_bigger)>>(delta_exp));
+ sig_bigger *= pow10(static_cast<BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::remove_reference_t<decltype(sig_bigger)>>(delta_exp));
  exp_bigger -= delta_exp;
  delta_exp = 0;
  }
@@ -76,7 +80,7 @@ BOOST_DECIMAL_FORCE_INLINE constexpr auto d32_add_impl(T lhs_sig, U lhs_exp, boo
 
  if (delta_exp > 1)
  {
- sig_smaller /= pow10(static_cast<std::remove_reference_t<decltype(sig_smaller)>>(delta_exp - 1));
+ sig_smaller /= pow10(static_cast<BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::remove_reference_t<decltype(sig_smaller)>>(delta_exp - 1));
  delta_exp = 1;
  }
  }
@@ -103,8 +107,9 @@ BOOST_DECIMAL_FORCE_INLINE constexpr auto d32_add_impl(T lhs_sig, U lhs_exp, boo
 }
 
 template <typename ReturnType, typename T, typename U>
-BOOST_DECIMAL_FORCE_INLINE constexpr auto add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
- T rhs_sig, U rhs_exp, bool rhs_sign) noexcept -> ReturnType
+BOOST_DECIMAL_FORCE_INLINE BOOST_DECIMAL_GPU_ENABLED
+constexpr auto add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
+ T rhs_sig, U rhs_exp, bool rhs_sign) noexcept -> ReturnType
 {
  const bool sign {lhs_sign};
 
@@ -176,9 +181,9 @@ BOOST_DECIMAL_FORCE_INLINE constexpr auto add_impl(T lhs_sig, U lhs_exp, bool lh
 }
 
 template <typename ReturnType, typename T, typename U>
-constexpr auto d64_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
- T rhs_sig, U rhs_exp, bool rhs_sign,
- bool abs_lhs_bigger) noexcept -> ReturnType
+BOOST_DECIMAL_GPU_ENABLED constexpr auto d64_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
+  T rhs_sig, U rhs_exp, bool rhs_sign,
+  bool abs_lhs_bigger) noexcept -> ReturnType
 {
  using add_type = std::int_fast64_t;
 
@@ -261,8 +266,8 @@ constexpr auto d64_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
 
 template <typename ReturnType, BOOST_DECIMAL_INTEGRAL T1, BOOST_DECIMAL_INTEGRAL U1,
  BOOST_DECIMAL_INTEGRAL T2, BOOST_DECIMAL_INTEGRAL U2>
-constexpr auto d128_add_impl(T1 lhs_sig, U1 lhs_exp, bool lhs_sign,
- T2 rhs_sig, U2 rhs_exp, bool rhs_sign) noexcept -> ReturnType
+BOOST_DECIMAL_GPU_ENABLED constexpr auto d128_add_impl(T1 lhs_sig, U1 lhs_exp, bool lhs_sign,
+  T2 rhs_sig, U2 rhs_exp, bool rhs_sign) noexcept -> ReturnType
 {
  const bool sign {lhs_sign};
 
@@ -284,7 +289,7 @@ constexpr auto d128_add_impl(T1 lhs_sig, U1 lhs_exp, bool lhs_sign,
  //
  // e.g. 1.234567e5 + 9.876543e-2 = 1.234568e5
 
- BOOST_DECIMAL_IF_CONSTEXPR (std::numeric_limits<T2>::digits10 > std::numeric_limits<std::uint64_t>::digits10)
+ BOOST_DECIMAL_IF_CONSTEXPR (boost::decimal::detail::numeric_limits<T2>::digits10 > boost::decimal::detail::numeric_limits<std::uint64_t>::digits10)
  {
  if (rhs_sig >= detail::uint128 {UINT64_C(0xF684DF56C3E0), UINT64_C(0x1BC6C73200000000)})
  {
@@ -341,9 +346,9 @@ constexpr auto d128_add_impl(T1 lhs_sig, U1 lhs_exp, bool lhs_sign,
 }
 
 template <typename ReturnType, typename T, typename U>
-constexpr auto d128_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
- T rhs_sig, U rhs_exp, bool rhs_sign,
- bool abs_lhs_bigger) noexcept -> ReturnType
+BOOST_DECIMAL_GPU_ENABLED constexpr auto d128_add_impl(T lhs_sig, U lhs_exp, bool lhs_sign,
+  T rhs_sig, U rhs_exp, bool rhs_sign,
+  bool abs_lhs_bigger) noexcept -> ReturnType
 {
  auto delta_exp {lhs_exp > rhs_exp ? lhs_exp - rhs_exp : rhs_exp - lhs_exp};
 

include/boost/decimal/detail/apply_sign.hpp

@@ -22,39 +22,39 @@ namespace boost { namespace decimal { namespace detail {
 
 template <typename Integer, typename Unsigned_Integer = detail::make_unsigned_t<Integer>,
  std::enable_if_t<detail::is_signed_v<Integer>, bool> = true>
-constexpr auto apply_sign(Integer val) noexcept -> Unsigned_Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto apply_sign(Integer val) noexcept -> Unsigned_Integer
 {
  return static_cast<Unsigned_Integer>(-(static_cast<Unsigned_Integer>(val)));
 }
 
 template <typename Unsigned_Integer, std::enable_if_t<!detail::is_signed_v<Unsigned_Integer>, bool> = true>
-constexpr auto apply_sign(Unsigned_Integer val) noexcept -> Unsigned_Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto apply_sign(Unsigned_Integer val) noexcept -> Unsigned_Integer
 {
  return val;
 }
 
 template <typename Integer, typename Unsigned_Integer = detail::make_unsigned_t<Integer>,
  std::enable_if_t<detail::is_signed_v<Integer>, bool> = true>
-constexpr auto make_positive_unsigned(Integer val) noexcept -> Unsigned_Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto make_positive_unsigned(Integer val) noexcept -> Unsigned_Integer
 {
  return static_cast<Unsigned_Integer>(val < static_cast<Integer>(static_cast<std::int8_t>(0)) ? apply_sign(val) : static_cast<Unsigned_Integer>(val));
 }
 
 template <typename Unsigned_Integer, std::enable_if_t<!detail::is_signed_v<Unsigned_Integer>, bool> = true>
-constexpr auto make_positive_unsigned(Unsigned_Integer val) noexcept -> Unsigned_Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto make_positive_unsigned(Unsigned_Integer val) noexcept -> Unsigned_Integer
 {
  return val;
 }
 
 template <typename Integer, std::enable_if_t<detail::is_signed_v<Integer>, bool> = true>
-constexpr auto make_signed_value(Integer val, bool sign) noexcept -> Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto make_signed_value(Integer val, bool sign) noexcept -> Integer
 {
  return sign ? -val : val;
 }
 
 template <typename Unsigned_Integer, typename Integer = detail::make_signed_t<Unsigned_Integer>,
  std::enable_if_t<!detail::is_signed_v<Unsigned_Integer>, bool> = true>
-constexpr auto make_signed_value(Unsigned_Integer val, bool sign) noexcept -> Integer
+BOOST_DECIMAL_GPU_ENABLED constexpr auto make_signed_value(Unsigned_Integer val, bool sign) noexcept -> Integer
 {
  const auto signed_val {static_cast<Integer>(val)};
  return sign ? -signed_val : signed_val;

include/boost/decimal/detail/cmath/abs.hpp

@@ -19,7 +19,7 @@ namespace boost {
 namespace decimal {
 
 BOOST_DECIMAL_EXPORT template <typename T>
-constexpr auto abs BOOST_DECIMAL_PREVENT_MACRO_SUBSTITUTION (T rhs) noexcept
+BOOST_DECIMAL_GPU_ENABLED constexpr auto abs BOOST_DECIMAL_PREVENT_MACRO_SUBSTITUTION (T rhs) noexcept
  BOOST_DECIMAL_REQUIRES(detail::is_decimal_floating_point_v, T)
 {
  return signbit(rhs) ? -rhs : rhs;

include/boost/decimal/detail/cmath/fpclassify.hpp

@@ -20,7 +20,7 @@ namespace boost {
 namespace decimal {
 
 BOOST_DECIMAL_EXPORT template <typename T>
-constexpr auto fpclassify BOOST_DECIMAL_PREVENT_MACRO_SUBSTITUTION (T rhs) noexcept
+BOOST_DECIMAL_GPU_ENABLED constexpr auto fpclassify BOOST_DECIMAL_PREVENT_MACRO_SUBSTITUTION (T rhs) noexcept
  BOOST_DECIMAL_REQUIRES_RETURN(detail::is_decimal_floating_point_v, T, int)
 {
  constexpr T zero {0, 0};

include/boost/decimal/detail/concepts.hpp

@@ -5,6 +5,7 @@
 #ifndef BOOST_DECIMAL_DETAIL_CONCEPTS
 #define BOOST_DECIMAL_DETAIL_CONCEPTS
 
+#include <boost/decimal/detail/config.hpp>
 #include <boost/decimal/detail/promotion.hpp>
 #include <boost/decimal/detail/type_traits.hpp>
 
@@ -413,23 +414,23 @@ concept execution_policy = std::is_execution_policy_v<std::remove_cvref_t<T>>;
 #endif
 
 #ifndef BOOST_DECIMAL_REQUIRES
-# define BOOST_DECIMAL_REQUIRES(X, T) -> std::enable_if_t<X<T>, T>
+# define BOOST_DECIMAL_REQUIRES(X, T) -> BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::enable_if_t<X<T>, T>
 #endif
 
 #ifndef BOOST_DECIMAL_REQUIRES_TWO
-# define BOOST_DECIMAL_REQUIRES_TWO(X1, T1, X2, T2) -> std::enable_if_t<X1<T1> && X2<T2>, detail::promote_args_t<T1, T2>>
+# define BOOST_DECIMAL_REQUIRES_TWO(X1, T1, X2, T2) -> BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::enable_if_t<X1<T1> && X2<T2>, detail::promote_args_t<T1, T2>>
 #endif
 
 #ifndef BOOST_DECIMAL_REQUIRES_TWO_RETURN
-# define BOOST_DECIMAL_REQUIRES_TWO_RETURN(X1, T1, X2, T2, ReturnType) -> std::enable_if_t<X1<T1> && X2<T2>, ReturnType>
+# define BOOST_DECIMAL_REQUIRES_TWO_RETURN(X1, T1, X2, T2, ReturnType) -> BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::enable_if_t<X1<T1> && X2<T2>, ReturnType>
 #endif
 
 #ifndef BOOST_DECIMAL_REQUIRES_THREE
-# define BOOST_DECIMAL_REQUIRES_THREE(X1, T1, X2, T2, X3, T3) -> std::enable_if_t<X1<T1> && X2<T2> && X3<T3>, detail::promote_args_t<T1, T2, T3>>
+# define BOOST_DECIMAL_REQUIRES_THREE(X1, T1, X2, T2, X3, T3) -> BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::enable_if_t<X1<T1> && X2<T2> && X3<T3>, detail::promote_args_t<T1, T2, T3>>
 #endif
 
 #ifndef BOOST_DECIMAL_REQUIRES_RETURN
-# define BOOST_DECIMAL_REQUIRES_RETURN(X, T, ReturnType) -> std::enable_if_t<X<T>, ReturnType>
+# define BOOST_DECIMAL_REQUIRES_RETURN(X, T, ReturnType) -> BOOST_DECIMAL_TYPE_TRAITS_NAMESPACE::enable_if_t<X<T>, ReturnType>
 #endif
 
 #endif //BOOST_DECIMAL_DETAIL_CONCEPTS