Roll optimise() into compile()

CMakeLists.txt

@@ -11,7 +11,7 @@ if(NOT CMAKE_BUILD_TYPE)
 	FORCE)
 endif()
 
-project(heyoka VERSION 1.0.0 LANGUAGES CXX C)
+project(heyoka VERSION 1.1.0 LANGUAGES CXX C)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake" "${CMAKE_CURRENT_SOURCE_DIR}/cmake/yacma")
 
@@ -293,7 +293,7 @@ if(HEYOKA_WITH_SLEEF)
 endif()
 
 # Setup the heyoka ABI version number.
-set(HEYOKA_ABI_VERSION 22)
+set(HEYOKA_ABI_VERSION 23)
 
 if(HEYOKA_BUILD_STATIC_LIBRARY)
     # Setup of the heyoka static library.

benchmark/large_cfunc.cpp

@@ -56,7 +56,6 @@ int main()
 
     add_cfunc<double>(s, "en", {en}, kw::vars = {x, y, z, vx, vy, vz}, kw::compact_mode = true);
 
-    s.optimise();
     s.compile();
 
     [[maybe_unused]] auto fn = s.jit_lookup("en");

benchmark/penc_comparison.cpp

@@ -133,9 +133,6 @@ void run_benchmark(unsigned order)
     // Verify.
     s.verify_function(f);
 
-    // Run the optimisation pass.
-    s.optimise();
-
     // Compile.
     s.compile();
 

doc/changelog.rst

@@ -1,6 +1,38 @@
 Changelog
 =========
 
+1.1.0 (unreleased)
+------------------
+
+New
+~~~
+
+- It is now possible to get the LLVM bitcode of
+  an ``llvm_state``
+  (`#339 <https://github.com/bluescarni/heyoka/pull/339>`__).
+
+Changes
+~~~~~~~
+
+- The LLVM bitcode is now used internally (instead of the textual
+  representation of the IR) when copying and serialising
+  an ``llvm_state``
+  (`#339 <https://github.com/bluescarni/heyoka/pull/339>`__).
+- The optimisation pass in an ``llvm_state`` is now automatically
+  called during compilation
+  (`#339 <https://github.com/bluescarni/heyoka/pull/339>`__).
+
+Fix
+~~~
+
+- Fix the object file of an ``llvm_state`` not being
+  preserved during copy and deserialisation
+  (`#339 <https://github.com/bluescarni/heyoka/pull/339>`__).
+- Fix LLVM module name not being preserved during
+  copy and deserialisation of ``llvm_state``
+  (`#339 <https://github.com/bluescarni/heyoka/pull/339>`__).
+- Fix broken link in the docs.
+
 1.0.0 (2023-08-10)
 ------------------
 

include/heyoka/llvm_state.hpp

@@ -133,6 +133,7 @@ class HEYOKA_DLL_PUBLIC llvm_state
     std::unique_ptr<ir_builder> m_builder;
     unsigned m_opt_level;
     std::string m_ir_snapshot;
+    std::string m_bc_snapshot;
     bool m_fast_math;
     bool m_force_avx512;
     std::string m_module_name;
@@ -210,19 +211,20 @@ class HEYOKA_DLL_PUBLIC llvm_state
     // end of a constructor.
     HEYOKA_DLL_LOCAL void ctor_setup_math_flags();
 
-public:
-    llvm_state();
-    // NOTE: enable the kwargs ctor only if:
+    // Meta-programming for the kwargs ctor. Enabled if:
     // - there is at least 1 argument (i.e., cannot act as a def ctor),
     // - if there is only 1 argument, it cannot be of type llvm_state
     //   (so that it does not interfere with copy/move ctors).
-    template <typename... KwArgs,
-              std::enable_if_t<
-                  (sizeof...(KwArgs) > 0u)
-                      && (sizeof...(KwArgs) > 1u
-                          || std::conjunction_v<std::negation<std::is_same<detail::uncvref_t<KwArgs>, llvm_state>>...>),
-                  int>
-              = 0>
+    template <typename... KwArgs>
+    using kwargs_ctor_enabler = std::enable_if_t<
+        (sizeof...(KwArgs) > 0u)
+            && (sizeof...(KwArgs) > 1u
+                || std::conjunction_v<std::negation<std::is_same<detail::uncvref_t<KwArgs>, llvm_state>>...>),
+        int>;
+
+public:
+    llvm_state();
+    template <typename... KwArgs, kwargs_ctor_enabler<KwArgs...> = 0>
     explicit llvm_state(KwArgs &&...kw_args) : llvm_state(kw_args_ctor_impl(std::forward<KwArgs>(kw_args)...))
     {
     }
@@ -246,6 +248,7 @@ class HEYOKA_DLL_PUBLIC llvm_state
     [[nodiscard]] bool force_avx512() const;
 
     [[nodiscard]] std::string get_ir() const;
+    [[nodiscard]] std::string get_bc() const;
     void dump_object_code(const std::string &) const;
     [[nodiscard]] const std::string &get_object_code() const;
 
@@ -255,6 +258,7 @@ class HEYOKA_DLL_PUBLIC llvm_state
     void optimise();
 
     [[nodiscard]] bool is_compiled() const;
+    [[nodiscard]] bool has_object_code() const;
 
     void compile();
 
@@ -265,9 +269,10 @@ class HEYOKA_DLL_PUBLIC llvm_state
 
 HEYOKA_END_NAMESPACE
 
-// Current archive version is 2. Changelog:
+// Archive version changelog:
 // - version 1: got rid of the inline_functions setting;
-// - version 2: added the force_avx512 setting.
-BOOST_CLASS_VERSION(heyoka::llvm_state, 2)
+// - version 2: added the force_avx512 setting;
+// - version 3: added the bitcode snapshot.
+BOOST_CLASS_VERSION(heyoka::llvm_state, 3)
 
 #endif

include/heyoka/taylor.hpp

@@ -134,7 +134,7 @@ taylor_c_diff_func_name_args(llvm::LLVMContext &, llvm::Type *, const std::strin
 // Add a function for computing the dense output
 // via polynomial evaluation.
 void taylor_add_d_out_function(llvm_state &, llvm::Type *, std::uint32_t, std::uint32_t, std::uint32_t, bool,
-                               bool = true, bool = true);
+                               bool = true);
 
 } // namespace detail
 

src/detail/event_detection.cpp

@@ -911,9 +911,6 @@ taylor_adaptive<T>::ed_data::ed_data(llvm_state s, std::vector<t_event_t> tes, s
     // Add the function for the fast exclusion check.
     detail::llvm_add_fex_check(m_state, fp_t, order, 1);
 
-    // Run the optimisation pass.
-    m_state.optimise();
-
     // Compile.
     m_state.compile();
 
@@ -1548,9 +1545,6 @@ taylor_adaptive_batch<T>::ed_data::ed_data(llvm_state s, std::vector<t_event_t>
     // NOTE: the fast exclusion check is vectorised.
     detail::llvm_add_fex_check(m_state, fp_t, order, batch_size);
 
-    // Run the optimisation pass.
-    m_state.optimise();
-
     // Compile.
     m_state.compile();
 

src/detail/llvm_helpers.cpp

@@ -939,7 +939,7 @@ std::string llvm_type_name(llvm::Type *t)
 
     t->print(ostr, false, true);
 
-    return ostr.str();
+    return std::move(ostr.str());
 }
 
 // This function will return true if:

src/expression_cfunc.cpp

@@ -1722,9 +1722,6 @@ auto add_cfunc_impl(llvm_state &s, const std::string &name, const F &fn, std::ui
     // Restore the original insertion block.
     builder.SetInsertPoint(orig_bb);
 
-    // Run the optimisation pass.
-    s.optimise();
-
     return dc;
 }