stdgpu 1.3.0

This release of stdgpu introduces a new experimental HIP backend adding support for AMD GPUs, significant improvements to the API documentation as well as many new code examples, the integration of clang-tidy and cppcheck in the CI, as well as a tremendous amount of warning fixes to enable clean builds at very high warning levels.

New Features & Enhancements

• General: Add experimental HIP backend #121 #143
• General: Add support for Compute Capability 3.0 in CUDA backend #153
• General: Add clang-tidy support #129 #138
• General: Add cppcheck support #149
• General: Add CI job for documentation creation #109
• General: Deprecate misleading/obsolete cmake options #103
• atomic: Make all operations follow sequentially consistent ordering #176
• atomic: Add backend documentation of template parameter #177
• atomic: Cleanup backend-specific internals of CUDA backend #152
• bit: Add ceil2 and floor2 functions #105
• bit: Rename functions to match most recent draft of C++20 #110
• bitset: Remove dependency to cstdlib #145
• cstddef: Hide initializers for clearer documentation #166
• cstdlib: Deprecate sizedivPow2 #161
• limits: Add implementation for non-specialized template and documentation for every type #167
• memory: Add construct_at function #95
• memory: Cleanup global variables and simplify allocate/deallocate logic #104
• memory: Improve construct* and destroy* unit tests #175
• platform: Add automatic dispatching of backend-specific definitions #119
• platform: Change detection of device code for OpenMP #174
• ranges: Add size() and empty() functions as well as additional constructors #122
• ranges: Add index64_t constructor and deprecate index_t version #102
• unordered_map,unordered_set: Improve robustness of Fibonacci Hashing #111
• README,doc: Significantly improve introduction, examples, and documentation #114 #116 #162 #165 #170 #171 #172 #181
• doc: Group all class and function definitions into modules #169
• doc: Cleanup unnecessary documentation #168
• examples: Add many new examples and improve existing ones #173
• test: Disable unused GMock #160
• cmake: Make installable package relocatable #180
• cmake: Add option to treat warnings as errors #108
• cmake: Generate compile flags more robustly #128
• cmake: Simplify architecture flag generation in CUDA backend #154
• cmake: Install backend-specific find modules in subdirectories #117
• cmake: Update support for CMake 3.17+ #123

Bug Fixes

• General: Increase warning level and fix conversion and float-equal warnings #98
• General: Increase MSVC warning level and fix related warnings #107 #156
• General: Fix Clang warnings #91 #147
• General: Fix format warnings #101
• General: Fix sign-conversion warnings #100
• General: Fix shadow warnings #90
• General: Fix numerous clang-tidy warnings #130 #131 #132 #133 #134 #135 #136 #137 #140 #141
• examples: Pass containers by reference for OpenMP backend #182
• src,test: Improve consistency and cleanup includes #118
• test: Fix missing namespace for uint8_t #142
• test: Pass containers by const reference to functors #158
• test: Fix double-promotion warnings in backend code #151
• test: Fix conversion warning and missing namespace #124
• test: Fix missing include in device_info cpp files #120
• bit: Fix potential negative bit shift in unit test #159
• bit,bitset: Fix missing post-conditions and remove unnecessary dependency #112
• bitset: Fix deprecated-copy warning #144
• compiler: Fix NVCC detection #155
• compiler,platform: Use unique numbers as internal macro definitions #139
• contract: Enforce user semicolon for all possible expansions #148 #150
• limits: Suppress long double device code warning with MSVC #178
• platform: Move STDGPU_HAS_CXX_17 to compiler #146
• ranges: Fix compilation with 64-bit index type #157
• ranges: Fix compilation error with select functor #125
• deque,vector: Fix overflow in test #99
• doc: Fix several minor documentation bugs #164
• scripts: Use released thrust version #126
• cmake: Fix error with unspecified build type #179
• cmake: Fix parsing of thrust version #163
• cmake: Workaround bug in imported rocthrust target name #127
• cmake: Properly handle CUDA toolkit dependency #96
• cmake: Add missing dependency checks in package config #94
• cmake: Fix selection of header files for installation #93
• cmake: Fix inconsistent thrust detection across the backends #92
• CI: Fix codecov task #113
• CI: Fix potentially missing OpenMP runtime package #106

Deprecated Features

• bit: ispow2(), log2pow2(), mod2()
• cstdlib: sizedivPow2(std::size_t, std::size_t), sizediv_t
• memory: safe_pinned_host_allocator, default_allocator_traits
• mutex: mutex_ref
• ranges: device_range(T*, index_t), host_range(T*, index_t), non-const begin() and end() member functions
• unordered_map,unordered_set: createDeviceObject(index_t, index_t), excess_count(), total_count()
• CMake Configuration Options: STDGPU_ENABLE_AUXILIARY_ARRAY_WARNING, STDGPU_ENABLE_MANAGED_ARRAY_WARNING, STDGPU_USE_FAST_DESTROY, STDGPU_USE_FIBONACCI_HASHING