Merge pull request #363 from bluescarni/pr/single_prec

Support for single-precision computations

.github/workflows/gha_ci.yml

@@ -38,7 +38,7 @@ jobs:
           mkdir build
           cd build
           cmake ../ -G "Visual Studio 16 2019" -A x64 -DHEYOKA_BUILD_TESTS=yes -DHEYOKA_WITH_MPPP=yes -DHEYOKA_BUILD_TUTORIALS=ON -DHEYOKA_ENABLE_IPO=yes -DBoost_NO_BOOST_CMAKE=ON -DHEYOKA_WITH_SLEEF=yes -DMPPP_GMP_INCLUDE_DIR=C:\Miniconda\envs\test\Library\include -DMPPP_GMP_LIBRARY=C:\Miniconda\envs\test\Library\lib\mpir.lib
-          cmake --build . --config Release
+          cmake --build . --config Release -j2
           copy Release\heyoka.dll test\Release\
           ctest -j4 -V -C Release
   windows_2019_llvm13:
@@ -60,7 +60,7 @@ jobs:
           mkdir build
           cd build
           cmake ../ -G "Visual Studio 16 2019" -A x64 -DHEYOKA_BUILD_TESTS=yes -DHEYOKA_WITH_MPPP=yes -DHEYOKA_BUILD_TUTORIALS=ON -DHEYOKA_ENABLE_IPO=yes -DBoost_NO_BOOST_CMAKE=ON -DHEYOKA_WITH_SLEEF=yes -DMPPP_GMP_INCLUDE_DIR=C:\Miniconda\envs\test\Library\include -DMPPP_GMP_LIBRARY=C:\Miniconda\envs\test\Library\lib\mpir.lib
-          cmake --build . --config Release
+          cmake --build . --config Release -j2
           copy Release\heyoka.dll test\Release\
           ctest -j4 -V -C Release
   windows_2019_llvm14:
@@ -82,7 +82,7 @@ jobs:
           mkdir build
           cd build
           cmake ../ -G "Visual Studio 16 2019" -A x64 -DHEYOKA_BUILD_TESTS=yes -DHEYOKA_WITH_MPPP=yes -DHEYOKA_BUILD_TUTORIALS=ON -DHEYOKA_ENABLE_IPO=yes -DBoost_NO_BOOST_CMAKE=ON -DHEYOKA_WITH_SLEEF=yes -DMPPP_GMP_INCLUDE_DIR=C:\Miniconda\envs\test\Library\include -DMPPP_GMP_LIBRARY=C:\Miniconda\envs\test\Library\lib\mpir.lib
-          cmake --build . --config Release
+          cmake --build . --config Release -j2
           copy Release\heyoka.dll test\Release\
           ctest -j4 -V -C Release
   conda_release_static:

README.md

@@ -40,7 +40,7 @@ heyoka is a C++ library for the integration of ordinary differential equations
 (ODEs) via Taylor's method, based on automatic differentiation techniques and aggressive just-in-time
 compilation via [LLVM](https://llvm.org/). Notable features include:
 
-* support for double-precision, extended-precision (80-bit and 128-bit),
+* support for single-precision, double-precision, extended-precision (80-bit and 128-bit),
   and arbitrary-precision floating-point types,
 * the ability to maintain machine precision accuracy over
   tens of billions of timesteps,

doc/changelog.rst

@@ -7,6 +7,8 @@ Changelog
 New
 ~~~
 
+- Add support for single-precision computations
+  (`#363 <https://github.com/bluescarni/heyoka/pull/363>`__).
 - Add model implementing the ELP2000 analytical lunar theory
   (`#362 <https://github.com/bluescarni/heyoka/pull/362>`__).
 

doc/index.rst

@@ -17,7 +17,7 @@ heyoka is a C++ library for the integration of ordinary differential equations
 on automatic differentiation techniques and aggressive just-in-time
 compilation via `LLVM <https://llvm.org/>`__. Notable features include:
 
-* support for double-precision, extended-precision (80-bit and 128-bit),
+* support for single-precision, double-precision, extended-precision (80-bit and 128-bit),
   and arbitrary-precision floating-point types,
 * the ability to maintain machine precision accuracy over
   tens of billions of timesteps,
@@ -110,7 +110,7 @@ license. The authors are Francesco Biscani and
 Dario Izzo (European Space Agency).
 
 .. toctree::
-   :maxdepth: 2
+   :maxdepth: 1
 
    install.rst
    basic_tutorials.rst

doc/install.rst

@@ -64,7 +64,8 @@ installing from source (the minimum required version is 3.18).
 Support for extended precision
 ``````````````````````````````
 
-Whereas in heyoka double-precision computations are always supported, support for extended-precision
+Whereas in heyoka single-precision and double-precision computations are always supported via the
+``float`` and ``double`` types respectively, support for extended-precision
 computations varies depending on the software/hardware platform.
 
 80-bit precision
@@ -81,7 +82,7 @@ heyoka (and all its dependencies) have been compiled with a compiler supporting
 ^^^^^^^^^^^^^^^^^
 
 On platforms where ``long double`` is a quadruple-precision floating-point datatype (e.g., 64-bit ARM),
-quadruple-precision integrations are always supported. Otherwise,
+quadruple-precision integrations are always supported via ``long double``. Otherwise,
 on platforms such as x86-64, quadruple-precision computations are supported if:
 
 * the nonstandard ``__float128`` floating-point type is

doc/tut_extended_precision.rst

@@ -4,7 +4,7 @@ Computations in extended precision
 ==================================
 
 As hinted in the :ref:`installation instructions <ep_support>`, heyoka supports computations
-not only in double precision, but also in extended precision. Specifically, heyoka currently supports:
+not only in single and double precision, but also in extended precision. Specifically, heyoka currently supports:
 
 * the 80-bit IEEE `extended-precision format <https://en.wikipedia.org/wiki/Extended_precision>`__ (~21 decimal digits),
 * the 128-bit IEEE `quadruple-precision format <https://en.wikipedia.org/wiki/Quadruple-precision_floating-point_format>`__ (~36 decimal digits).

include/heyoka/detail/event_detection.hpp

@@ -41,6 +41,9 @@ inline T taylor_deduce_cooldown(T, T)
     static_assert(always_false_v<T>, "Unhandled type");
 }
 
+template <>
+float taylor_deduce_cooldown(float, float);
+
 template <>
 double taylor_deduce_cooldown(double, double);
 

include/heyoka/detail/type_traits.hpp

@@ -76,6 +76,10 @@ template <typename>
 struct is_supported_fp : std::false_type {
 };
 
+template <>
+struct is_supported_fp<float> : std::true_type {
+};
+
 template <>
 struct is_supported_fp<double> : std::true_type {
 };

include/heyoka/expression.hpp

@@ -81,6 +81,7 @@ class HEYOKA_DLL_PUBLIC expression
 public:
     expression();
 
+    explicit expression(float);
     explicit expression(double);
     explicit expression(long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -125,6 +126,9 @@ HEYOKA_DLL_PUBLIC bool is_fixed(const expression &);
 inline namespace literals
 {
 
+HEYOKA_DLL_PUBLIC expression operator""_flt(long double);
+HEYOKA_DLL_PUBLIC expression operator""_flt(unsigned long long);
+
 HEYOKA_DLL_PUBLIC expression operator""_dbl(long double);
 HEYOKA_DLL_PUBLIC expression operator""_dbl(unsigned long long);
 
@@ -257,6 +261,7 @@ HEYOKA_DLL_PUBLIC expression operator+(expression);
 HEYOKA_DLL_PUBLIC expression operator-(const expression &);
 
 HEYOKA_DLL_PUBLIC expression operator+(const expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression operator+(const expression &, float);
 HEYOKA_DLL_PUBLIC expression operator+(const expression &, double);
 HEYOKA_DLL_PUBLIC expression operator+(const expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -265,6 +270,7 @@ HEYOKA_DLL_PUBLIC expression operator+(const expression &, mppp::real128);
 #if defined(HEYOKA_HAVE_REAL)
 HEYOKA_DLL_PUBLIC expression operator+(const expression &, mppp::real);
 #endif
+HEYOKA_DLL_PUBLIC expression operator+(float, const expression &);
 HEYOKA_DLL_PUBLIC expression operator+(double, const expression &);
 HEYOKA_DLL_PUBLIC expression operator+(long double, const expression &);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -276,13 +282,15 @@ HEYOKA_DLL_PUBLIC expression operator+(mppp::real, const expression &);
 
 HEYOKA_DLL_PUBLIC expression operator-(const expression &, const expression &);
 HEYOKA_DLL_PUBLIC expression operator-(const expression &, double);
+HEYOKA_DLL_PUBLIC expression operator-(const expression &, float);
 HEYOKA_DLL_PUBLIC expression operator-(const expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
 HEYOKA_DLL_PUBLIC expression operator-(const expression &, mppp::real128);
 #endif
 #if defined(HEYOKA_HAVE_REAL)
 HEYOKA_DLL_PUBLIC expression operator-(const expression &, mppp::real);
 #endif
+HEYOKA_DLL_PUBLIC expression operator-(float, const expression &);
 HEYOKA_DLL_PUBLIC expression operator-(double, const expression &);
 HEYOKA_DLL_PUBLIC expression operator-(long double, const expression &);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -294,13 +302,15 @@ HEYOKA_DLL_PUBLIC expression operator-(mppp::real, const expression &);
 
 HEYOKA_DLL_PUBLIC expression operator*(const expression &, const expression &);
 HEYOKA_DLL_PUBLIC expression operator*(const expression &, double);
+HEYOKA_DLL_PUBLIC expression operator*(const expression &, float);
 HEYOKA_DLL_PUBLIC expression operator*(const expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
 HEYOKA_DLL_PUBLIC expression operator*(const expression &, mppp::real128);
 #endif
 #if defined(HEYOKA_HAVE_REAL)
 HEYOKA_DLL_PUBLIC expression operator*(const expression &, mppp::real);
 #endif
+HEYOKA_DLL_PUBLIC expression operator*(float, const expression &);
 HEYOKA_DLL_PUBLIC expression operator*(double, const expression &);
 HEYOKA_DLL_PUBLIC expression operator*(long double, const expression &);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -311,6 +321,7 @@ HEYOKA_DLL_PUBLIC expression operator*(mppp::real, const expression &);
 #endif
 
 HEYOKA_DLL_PUBLIC expression operator/(const expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression operator/(const expression &, float);
 HEYOKA_DLL_PUBLIC expression operator/(const expression &, double);
 HEYOKA_DLL_PUBLIC expression operator/(const expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -319,6 +330,7 @@ HEYOKA_DLL_PUBLIC expression operator/(const expression &, mppp::real128);
 #if defined(HEYOKA_HAVE_REAL)
 HEYOKA_DLL_PUBLIC expression operator/(const expression &, mppp::real);
 #endif
+HEYOKA_DLL_PUBLIC expression operator/(float, const expression &);
 HEYOKA_DLL_PUBLIC expression operator/(double, const expression &);
 HEYOKA_DLL_PUBLIC expression operator/(long double, const expression &);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -329,6 +341,7 @@ HEYOKA_DLL_PUBLIC expression operator/(mppp::real, const expression &);
 #endif
 
 HEYOKA_DLL_PUBLIC expression &operator+=(expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression &operator+=(expression &, float);
 HEYOKA_DLL_PUBLIC expression &operator+=(expression &, double);
 HEYOKA_DLL_PUBLIC expression &operator+=(expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -339,6 +352,7 @@ HEYOKA_DLL_PUBLIC expression &operator+=(expression &, mppp::real);
 #endif
 
 HEYOKA_DLL_PUBLIC expression &operator-=(expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression &operator-=(expression &, float);
 HEYOKA_DLL_PUBLIC expression &operator-=(expression &, double);
 HEYOKA_DLL_PUBLIC expression &operator-=(expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -349,6 +363,7 @@ HEYOKA_DLL_PUBLIC expression &operator-=(expression &, mppp::real);
 #endif
 
 HEYOKA_DLL_PUBLIC expression &operator*=(expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression &operator*=(expression &, float);
 HEYOKA_DLL_PUBLIC expression &operator*=(expression &, double);
 HEYOKA_DLL_PUBLIC expression &operator*=(expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)
@@ -359,6 +374,7 @@ HEYOKA_DLL_PUBLIC expression &operator*=(expression &, mppp::real);
 #endif
 
 HEYOKA_DLL_PUBLIC expression &operator/=(expression &, const expression &);
+HEYOKA_DLL_PUBLIC expression &operator/=(expression &, float);
 HEYOKA_DLL_PUBLIC expression &operator/=(expression &, double);
 HEYOKA_DLL_PUBLIC expression &operator/=(expression &, long double);
 #if defined(HEYOKA_HAVE_REAL128)

include/heyoka/llvm_state.hpp

@@ -61,6 +61,7 @@ struct target_features {
     bool vsx = false;
     bool vsx3 = false;
     // Recommended SIMD sizes.
+    std::uint32_t simd_size_flt = 1;
     std::uint32_t simd_size_dbl = 1;
     std::uint32_t simd_size_ldbl = 1;
 #if defined(HEYOKA_HAVE_REAL128)
@@ -99,6 +100,9 @@ inline std::uint32_t recommended_simd_size()
     return 0;
 }
 
+template <>
+HEYOKA_DLL_PUBLIC std::uint32_t recommended_simd_size<float>();
+
 template <>
 HEYOKA_DLL_PUBLIC std::uint32_t recommended_simd_size<double>();
 

include/heyoka/math/atan2.hpp

@@ -75,6 +75,7 @@ class HEYOKA_DLL_PUBLIC atan2_impl : public func_base
 
 HEYOKA_DLL_PUBLIC expression atan2(expression, expression);
 
+HEYOKA_DLL_PUBLIC expression atan2(expression, float);
 HEYOKA_DLL_PUBLIC expression atan2(expression, double);
 HEYOKA_DLL_PUBLIC expression atan2(expression, long double);
 
@@ -90,6 +91,7 @@ HEYOKA_DLL_PUBLIC expression atan2(expression, mppp::real);
 
 #endif
 
+HEYOKA_DLL_PUBLIC expression atan2(float, expression);
 HEYOKA_DLL_PUBLIC expression atan2(double, expression);
 HEYOKA_DLL_PUBLIC expression atan2(long double, expression);
 

include/heyoka/math/kepDE.hpp

@@ -72,6 +72,7 @@ HEYOKA_DLL_PUBLIC expression kepDE(expression, expression, expression);
     HEYOKA_DLL_PUBLIC expression kepDE(expression, type, expression);                                                  \
     HEYOKA_DLL_PUBLIC expression kepDE(type, expression, expression)
 
+HEYOKA_DECLARE_KEPDE_OVERLOADS(float);
 HEYOKA_DECLARE_KEPDE_OVERLOADS(double);
 HEYOKA_DECLARE_KEPDE_OVERLOADS(long double);
 

include/heyoka/math/kepE.hpp

@@ -77,6 +77,7 @@ HEYOKA_DLL_PUBLIC expression kepE(expression, expression);
     HEYOKA_DLL_PUBLIC expression kepE(expression, type);                                                               \
     HEYOKA_DLL_PUBLIC expression kepE(type, expression);
 
+HEYOKA_DECLARE_KEPE_OVERLOADS(float);
 HEYOKA_DECLARE_KEPE_OVERLOADS(double);
 HEYOKA_DECLARE_KEPE_OVERLOADS(long double);
 

include/heyoka/math/kepF.hpp

@@ -81,6 +81,7 @@ HEYOKA_DLL_PUBLIC expression kepF(expression, expression, expression);
     HEYOKA_DLL_PUBLIC expression kepF(expression, type, expression);                                                   \
     HEYOKA_DLL_PUBLIC expression kepF(type, expression, expression)
 
+HEYOKA_DECLARE_KEPF_OVERLOADS(float);
 HEYOKA_DECLARE_KEPF_OVERLOADS(double);
 HEYOKA_DECLARE_KEPF_OVERLOADS(long double);
 

include/heyoka/math/pow.hpp

@@ -114,6 +114,7 @@ pow_eval_algo get_pow_eval_algo(const pow_impl &);
 } // namespace detail
 
 HEYOKA_DLL_PUBLIC expression pow(expression, expression);
+HEYOKA_DLL_PUBLIC expression pow(expression, float);
 HEYOKA_DLL_PUBLIC expression pow(expression, double);
 HEYOKA_DLL_PUBLIC expression pow(expression, long double);
 

include/heyoka/step_callback.hpp

@@ -121,7 +121,7 @@ struct HEYOKA_DLL_PUBLIC_INLINE_CLASS step_callback_inner final : step_callback_
     void serialize(Archive &ar, unsigned)
     {
         ar &boost::serialization::base_object<step_callback_inner_base<TA>>(*this);
-        ar &m_value;
+        ar & m_value;
     }
 };
 
@@ -135,7 +135,7 @@ class HEYOKA_DLL_PUBLIC step_callback_impl
     template <typename Archive>
     void serialize(Archive &ar, unsigned)
     {
-        ar &m_ptr;
+        ar & m_ptr;
     }
 
     // Meta-programming for the generic ctor.
@@ -226,6 +226,9 @@ using step_callback_batch = detail::step_callback_impl<taylor_adaptive_batch<T>>
 HEYOKA_END_NAMESPACE
 
 // Disable Boost.Serialization tracking for the implementation details of step_callback.
+BOOST_CLASS_TRACKING(heyoka::detail::step_callback_inner_base<heyoka::taylor_adaptive<float>>,
+                     boost::serialization::track_never)
+
 BOOST_CLASS_TRACKING(heyoka::detail::step_callback_inner_base<heyoka::taylor_adaptive<double>>,
                      boost::serialization::track_never)
 
@@ -246,6 +249,9 @@ BOOST_CLASS_TRACKING(heyoka::detail::step_callback_inner_base<heyoka::taylor_ada
 
 #endif
 
+BOOST_CLASS_TRACKING(heyoka::detail::step_callback_inner_base<heyoka::taylor_adaptive_batch<float>>,
+                     boost::serialization::track_never)
+
 BOOST_CLASS_TRACKING(heyoka::detail::step_callback_inner_base<heyoka::taylor_adaptive_batch<double>>,
                      boost::serialization::track_never)