Merge pull request #4 from blackwer/optimize-encoder

Fix `unique` bug and add IPO

CMakeLists.txt

@@ -3,6 +3,14 @@ project(simple_ans VERSION 0.1.0 LANGUAGES CXX)
 
 set(CMAKE_CXX_RELEASE_FLAGS "-O3 -march=native")
 
+include(CheckIPOSupported)
+check_ipo_supported(RESULT ipo_supported OUTPUT ipo_error)
+if (ipo_supported)
+  message(STATUS "IPO/LTO is supported")
+else()
+  message(WARNING "IPO/LTO is not supported: ${ipo_error}")
+endif()
+
 # Find pybind11
 find_package(pybind11 REQUIRED)
 
@@ -47,6 +55,11 @@ else()
     target_compile_options(_simple_ans PRIVATE -Wall -Wextra -Wpedantic)
 endif()
 
+if(ipo_supported AND (CMAKE_BUILD_TYPE STREQUAL "Release"))
+  set_target_properties(_simple_ans PROPERTIES INTERPROCEDURAL_OPTIMIZATION TRUE)
+endif()
+
+
 # Install rules
 include(GNUInstallDirs)
 install(TARGETS _simple_ans

simple_ans/cpp/simple_ans.hpp

@@ -51,7 +51,8 @@ void ans_decode_t(T* output,
 
 namespace simple_ans
 {
-constexpr int lookup_array_threshold = std::numeric_limits<uint16_t>::max() + 1;
+constexpr int unique_array_threshold = static_cast<int>(std::numeric_limits<uint16_t>::max()) + 1;
+constexpr int lookup_array_threshold = unique_array_threshold;
 
 template <typename T>
 std::tuple<std::vector<T>, std::vector<uint64_t>> unique_with_counts(const T* values, size_t n)
@@ -75,15 +76,15 @@ std::tuple<std::vector<T>, std::vector<uint64_t>> unique_with_counts(const T* va
         max_value = std::max(max_value, static_cast<int64_t>(values[i]));
     }
 
-    if ((max_value - min_value + 1) <= lookup_array_threshold)
+    if ((max_value - min_value + 1) <= unique_array_threshold)
     {
         std::vector<uint64_t> raw_counts(max_value - min_value + 1);
         for (size_t i = 0; i < n; ++i)
         {
             raw_counts[values[i] - min_value]++;
         }
 
-        for (size_t i = 0; i < counts.size(); ++i)
+        for (size_t i = 0; i < raw_counts.size(); ++i)
         {
             if (raw_counts[i])
             {