v2.8.2

What's Changed

• [Version] Update branch/2.8.x to v2.8.2 by @github-actions in #4079
• Ignore Wmaybe-uninitialized in dispatch_reduce.cuh. by @bdice in #4054
• backport: fix numeric_limits digits for nvfp8/6/4 (#4070) by @miscco in #4130
• [BACKPORT]: Avoid compiler issue with MSVC and span constructor by @miscco in #4127

Full Changelog: v2.8.1...v2.8.2

v2.8.1

What's Changed

• Backport to 2.8: NVHPC fixes by @bernhardmgruber in #4021
• [Backport 2.8.x] [cuda::ptx] Fix .cta_group::2 definition (#4038) by @wmaxey in #4044
• [Version] Update branch/2.8.x to v2.8.1 by @github-actions in #4049

Full Changelog: v2.8.0...v2.8.1

CCCL 2.8.0

Release highlights

• With the increase of Python APIs, we changed our name from CUDA C++ Core Libraries to CUDA Core Compute Libraries
• This is the last release supporting C++11 and C++14. CCCL 3.0 will require C++17.
• We deprecated many entities and plan to remove them in CCCL 3.0, see the migration guide
• Internal assertions are now turned on by default for users in CMake Debug builds. Non-Debug and non-CMake builds are not affected.
• Many CUB algorithms were tuned for Blackwell GPUs

New C++ APIs

• thurst::transform_inclusive_scan with init value
• thrust::universal_host_pinned_vector, a Thrust vector with managed and pinned memory
• cub::DeviceTransform using bulk copy on Hopper+ or prefetch, used by thrust::transform
• cub::DeviceFor::ForEachInExtents to iterate over a multidimensional index space described by cuda::std::extents
• cuda::mr::cuda_async_memory_resource, a memory resource for stream-ordered allocations
• cuda::get_device_address, which returns the device address of a device object
• New function objects cuda::minimum and cuda::maximum
• New standard C++ functions in cuda::std::: :assume_aligned, source_location, dims, ignore, invoke_r, byteswap
• Exposure of several new PTX instructions in cuda::ptx: cp.async.mbarrier.arrive, mbarrier.expect_tx, clusterlaunchcontrol.*, st.bulk, multimem.ld_reduce, multimem.st, multimem.red, and many tcgen05 instructions
• cuda::experimental::basic_any: a utility for defining type-erasing wrappers in terms of an interface description
• Add general support for sm_101, sm_101a and sm_120 architectures
• The following CUB APIs support large number of items (> 2^32): DeviceSelect, DevicePartition, DeviceScan::*ByKey, DeviceReduce::{ArgMin,ArgMax}, DeviceTransform
• cuda::std::numeric_limits now supports FP16 (half and bfloat16) and FP8 types
• cuda::is_floating_point_v supporting any standard and extended (FP16, FP8) floating point types

Fixes

• cuda::atomic_ref now supports 8 and 16 bit operations - #2255
• Atomics placed in thread local memory will no longer exhibit undefined behavior - #2586

What's Changed in Detail

• Adds benchmarks for DeviceSelect::Unique by @elstehle in #2359
• CUB - Enable DPX Reduction by @fbusato in #2286
• [CUDAX] add a small c++17 implementation of std::execution (aka P2300) by @ericniebler in #2301
• Add thurst::transform_inclusive_scan with init value by @gonidelis in #2326
• Widen histogram agent constructor to more types by @bernhardmgruber in #2380
• Use a constant for the amount of static SMEM by @bernhardmgruber in #2374
• Add cub::DeviceTransform by @bernhardmgruber in #2086
• Update toolkit to CTK 12.6 by @miscco in #2348
• implement make_integer_sequence in terms of intrinsics whenever possible by @ericniebler in #2384
• Implement cuda::mr::cuda_async_memory_resource by @miscco in #1637
• Drop implementation of thrust::pair and thrust::tuple by @miscco in #2395
• Pull out _LIBCUDACXX_UNREACHABLE into its own file by @miscco in #2399
• Share common compiler flags in new CCCL-level targets. by @alliepiper in #2386
• conditionally include <crt/host_defines.h> from __cccl/execution_space.h header by @ericniebler in #2406
• add some simple utilities for manipulating lists of types by @ericniebler in #2370
• Drop thrusts diagnostic suppression warnings by @miscco in #2392
• [PoC]: Implement cuda::experimental::uninitialized_async_buffer by @miscco in #1854
• Fix thrust package to work with newer FindOpenMP.cmake. by @alliepiper in #2421
• Introduce cccl_configure_target cmake function. by @alliepiper in #2388
• Fix sccache errors in RAPIDS builds by @trxcllnt in #2417
• Replace CUDA C++ Core Libraries with CUDA Core Compute Libraries (only in README.md). by @rwgk in #2424
• Minor cleanup with cuda/atomic by @miscco in #2418
• uninitialized_buffer::get_resource returns a ref to an any_resource that can be copied by @ericniebler in #2431
• Refactor cuda::ceil_div to take two different types by @miscco in #2376
• Reduce PR testing matrix. by @alliepiper in #2436
• Implement cudax::shared_resource by @miscco in #2398
• Increase the libcu++ timeout by @miscco in #2435
• Move c/include/cccl/.h files to c/include/cccl/c/.h by @rwgk in #2428
• Make any_resource emplacable by @miscco in #2425
• Fix issues with __host__ and __device__ definitions by @miscco in #2413
• Make bit_cast play nice with extended floating point types by @miscco in #2434
• Do not include our own string.h file by @miscco in #2444
• Move nightly time by @bdice in #2437
• Remove a ton of lines in thrust tests by @gonidelis in #2356
• [CUDAX] Add placeholder green context type and logical device that can hold both a green ctx and a device by @pciolkosz in #2446
• Fix typo in CCCLBuildCompilerTargets.cmake by @alliepiper in #2453
• Drop superflous compile definition from thrust tests by @miscco in #2450
• Consolidate packages and install rules by @alliepiper in #2456
• Prune CUB's ChainedPolicy by CUDA_ARCH_LIST by @bernhardmgruber in #2154
• fixes merge conflict for policy pruning by @elstehle in #2466
• Add CCCL_ENABLE_WERROR flag. by @alliepiper in #2463
• Add CUB tests for segmented sort/radix sort with 64-bit num. items and segments by @fbusato in #2254
• Propagate compiler flags down to libcu++ LIT tests by @Artem-B in #2420
• Drop remaining uses of _LIBCUDACXX_COMPILER_* by @miscco in #2467
• Avoid C++17 extension in c++11 tests by @miscco in #2469
• Add span to example and templated block size by @Kh4ster in #2470
• Drop Objective C++ support by @miscco in #2468
• removes superfluous template keyword in call to Dereference by @andrewcorrigan in #2482
• Improve build times in several heavyweight libcudacxx tests. by @wmaxey in #2478
• Drop __availability header by @miscco in #2484
• Replace a few more instances of CUDA C++ Core Libraries with CUDA Core Compute Libraries`. by @rwgk in #2447
• Fix common_type specialization for extended floating point types by @miscco in #2483
• Implement some CUDA API calls for async_memory_pool by @miscco in #2455
• Move cudax example project to CCCL project examples. by @alliepiper in #2462
• Disable system header for narrowing conversion check by @miscco in #2465
• Require resources to always provide at least one execution space property by @miscco in #2489
• Rework builtin handling by @miscco in #2461
• Disable execution checks for std::equal by @miscco in #2491
• replace _CCCL_ALWAYS_INLINE with _CCCL_FORCEINLINE by @ericniebler in #2439
• Drop 2 relative includes that snuck in by @miscco in #2492
• re-express the cudax::__tupl::__apply member to make nvc++ happy by @ericniebler in #2493
• Drop badly named _One_of concept by @miscco in #2490
• Unify assert handling in cccl by @miscco in #2382
• Reduce scope of Thrust linkage in cudax. by @alliepiper in #2496
• Centralize CPM logic. by @alliepiper in #2495
• Fix typo in presets. by @alliepiper in #2497
• Refactor away per-project TOPLEVEL flags. by @alliepiper in #2498
• [FEA]: Validate cuda.parallel type matching in build and execution by @rwgk in #2429
• avoid gcc optimizer bug by not force inlining part of thrust::transform by @ericniebler in #2509
• Cleanup and modularize <cuda/std/barrier> by @miscco in https://github.com/N...

CCCL 2.7.0

What’s New

C++

Thrust / CUB

• Inclusive scan now supports initial value #1940
• Inclusive and exclusive scan now support problem sizes exceeding 2^31 elements #2171
• New cub::DeviceMerge::MergeKeys and cub::DeviceMerge::MergePairs algorithms #1817
• New thrust::tabulate_output_iterator fancy iterator #2282

Libcudacxx

• Enable Assertions on host and device depending on users choice
• C++26 inplace_vector has been implemented and backported to C++14
• Improved support for extended floating point types __half and __nv_bfloat16 both for cmath functions and complex
• cuda::std::tuple is now trivially copyable if the stored types are trivially copyable
• Reworked our atomics implementation
• Improved <cuda/std/bit> conformance
• Implemented <cuda/std/bitset> and backported to C++14
• Implemented and backported C++20 bit_cast. It is available in all standard modes and constexpr with compiler support
• Various backports and constexpr improvements (bool_constant, cuda::std::max)
• Moved the experimental memory resources from <cuda/memory_resource> into <cuda/experimental/memory_resource.cuh>

Python

cuda.cooperative

Best practice of using CCCL to make your CUDA kernels easier to write and faster to execute is now available in Python through the cuda.cooperative module. This module currently supports block- and warp-level algorithms within numba.cuda kernels, offering speed-of-light reductions, prefix sums, radix, and merge sort. You can customize cuda.cooperative algorithms with user-defined data types and operators, implemented directly in Python.

Block and warp-level cooperative algorithms are now available in Python #1973.
Experimental versions of reduce, scan, merge and radix sort are available in numba.cuda kernels.

cuda.parallel

Apart from device-side cooperative algorithms, CCCL 2.7 provides an experimental version of host-side parallel algorithms as part of the cuda.parallel module. This release includes parallel reduction.

What's Changed

• Fix documentation generation for thrust::pair by @bernhardmgruber in #1976
• Correct typo in a launch configuration header name by @pciolkosz in #1972
• Fix thrust::sort for large problem sizes by @gevtushenko in #1952
• Avoid SIGPIPE when truncating verbose output in CI scripts. by @alliepiper in #1971
• Clarify compiler support by @bernhardmgruber in #1970
• Experimental Python cooperative algorithms by @gevtushenko in #1973
• [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci in #1928
• Guard against an overflow in sort tests by @bernhardmgruber in #1980
• Remove obsolete Thrust function traits by @bernhardmgruber in #1962
• Python: Add version string & wheel build command by @leofang in #1985
• Add device inclusive scan with init_value by @gonidelis in #1845
• Fix BWUtil report on early exit by @gonidelis in #1994
• Use libcu++ void_t everywhere by @bernhardmgruber in #1977
• Drop zipped_binary_op by @bernhardmgruber in #1988
• Clarify PtxVersion and SmVersion by @bernhardmgruber in #2004
• More simplifications for CUB util_device by @bernhardmgruber in #1948
• fix some typos in <cuda/stream_ref> by @ericniebler in #2003
• Add CI slack notifications. by @alliepiper in #1961
• Allow nightly workflow to be manually invoked. by @alliepiper in #2007
• Need to use a different approach to reuse secrets in reusable workflows vs. actions. by @alliepiper in #2008
• Enable RAPIDS builds for manually dispatched workflows. by @alliepiper in #2009
• clean up complex.inl by @ZelboK in #1655
• Add github token to nightly workflow-results action. by @alliepiper in #2012
• Remove obsolete build system glue from the Thrust/CUB submodule structure. by @alliepiper in #2016
• Benchmark thrust::copy with non-trivially relocatable type by @bernhardmgruber in #1989
• Make bool_constant available in C++11 by @bernhardmgruber in #1997
• Spell value initialization where used in thrust vectors by @bernhardmgruber in #1990
• Do no redefine __ELF__ macro by @miscco in #2018
• Port thrust::merge[_by_key] to CUB by @bernhardmgruber in #1817
• Simplify some pointer traits by @bernhardmgruber in #2020
• Simplify test data setup by @bernhardmgruber in #2023
• Add tests to ensure that we properly propagate common_type for complex types by @miscco in #2025
• Update Thrust CMake README to use CCCL repo. by @alliepiper in #2026
• Include container toolkit in manual prereqs by @bryevdv in #2064
• Avoid ADL issues with thrust::distance by @miscco in #2053
• Simplify thrust::detail::wrapped_function by @bernhardmgruber in #2019
• Add a test for Thrust scan with non-commutative op by @bernhardmgruber in #2024
• Update memory_resource docs by @miscco in #1883
• Temporarily switch nightly H100 CI to build-only. by @alliepiper in #2060
• Do not rely on conversions between float and extended floating point types by @miscco in #2046
• experimental wrapper types for cudaEvent_t that provide a modern C++ interface. by @ericniebler in #2017
• [CUDAX] Add a dummy device struct for now by @pciolkosz in #2066
• Allow (somewhat) different input value types for merge by @bernhardmgruber in #2075
• Avoid ::result_type for partial sums in TBB reduce_by_key by @bernhardmgruber in #1998
• Fix formatting by @bernhardmgruber in #2090
• Rename and refactor transform_iterator_base by @bernhardmgruber in #1987
• Benchmark analysis: Print all top rows when asked for by @bernhardmgruber in #2089
• Makes user-provided functors in our examples use __device__ instead of CUB_RUNTIME_FUNCTION by @elstehle in #2088
• Separate cuda/experimental when sorting includes by @bernhardmgruber in #2094
• add support to cudax::device for querying a device's attributes by @ericniebler in #2084
• [CUDAX] Add experimental owning abstraction for cudaStream_t by @pciolkosz in #2093
• Do not query NVRTC for cuda runtime header by @miscco in #2102
• Cleanup CUB block/thread load and exchange by @bernhardmgruber in #1946
• Improve binary function objects and replace thrust implementation by @srinivasyadav18 in #1872
• Replace _LIBCUDACXX_CPO_ACCESSIBILITY with _CCCL_GLOBAL_VARIABLE by @miscco in #1881
• Add script to update RAPIDS version. by @bdice in #2082
• Update bad links by @bryevdv in #2080
• Fix line break issues that break doxygen code examples by @miscco in #2103
• Add internal wrapper for cuda driver APIs by @pciolkosz in #2070
• Use common_type for complex pow by @miscco in #1800
• [CUDAX] rename device to device_ref, add immovable device as a place to cache properties by @ericniebler in #2110
• Use the float flavors of the cmath functions in the extended floating point fallbacks by @miscco in #2106
• [PoC]: Implement cuda::experimental::uninitialized_buffer by @miscco in #1831
• Ensure that we avoid ABI Version conflics by @miscco in #2137
• Ensure that cuda_memory_resource allocates memory on the proper device by @miscco in #2073
• Clarify compatibility wrt. template specializations by @bernhardmgruber in #2138
• Implement a cudax::get_stream CPO by @miscco in #2135
• Make cuda::std::tuple trivially copyable by @miscco in #2127
• Fix missing copy of docs artifacts by @miscco in #2162
  ...

CCCL 2.6.1

This release includes backports for PRs #2332 and #2341. Please see release 2.6.0 for the full list of changes included in the release.

What's Changed

• Backport PR #2332 and #2341 by @wmaxey in #2368

Full Changelog: v2.6.0...v2.6.1

CCCL 2.6.0

What's Changed

• Restrict active histogram channels to channel count by @bernhardmgruber in #1796
• Cleanup internal thrust CUDA utils by @bernhardmgruber in #1802
• Use variadic interfaces in agent launcher by @bernhardmgruber in #1804
• Use nullptr over NULL by @bernhardmgruber in #1805
• Rework the documentation to be build with sphinx by @miscco in #1753
• Let Catch2 report cudaError descriptions by @bernhardmgruber in #1808
• Check size-querying CUB API invocation in tests by @bernhardmgruber in #1809
• Update docs link by @gevtushenko in #1812
• Add missing inline specifiers by @bernhardmgruber in #1813
• Upgrade actions that use node16 to versions that use node20 by @trxcllnt in #1779
• Document NVTX range behavior during graph capture by @bernhardmgruber in #1814
• Clean up AliasTemporaries by @bernhardmgruber in #1815
• Drop removed clang-tidy option by @bernhardmgruber in #1810
• Exclude docs from cccl infra changes. by @alliepiper in #1821
• Clean up thrust merge unit tests by @bernhardmgruber in #1819
• Fix atomic performance regressions by avoiding use of memcpy with natively supported atomic types. by @wmaxey in #1801
• Clean up merge_by_key and merge_key_value tests by @bernhardmgruber in #1824
• Restore the old thrust api documentation in rst by @miscco in #1818
• Drop all internal implementations of exceptions by @miscco in #1806
• Fix span for non-ranges by @miscco in #1836
• Cleanup thrust test special types by @bernhardmgruber in #1837
• Add inclusive_scan with initial value support (warp/block) by @gonidelis in #1749
• Fix loading from incorrect URI on 404 page. by @wmaxey in #1843
• Port CUB temporary storage layout test to Catch2 by @bernhardmgruber in #1835
• Port CUB thread operators test to Catch2 by @bernhardmgruber in #1834
• Adds ceil_div by @gonzalobg in #1825
• Split workflow into multiple dispatch groups to avoid skipped jobs. by @alliepiper in #1797
• Fix broken CUB doc build and add 404 page to Sphinx. by @wmaxey in #1846
• Port CUB thread sort test to Catch2 by @bernhardmgruber in #1838
• Cleanup CUB temporary storage layout test by @bernhardmgruber in #1848
• Propogate error when docsbuild fails, add docs build to CI. by @alliepiper in #1852
• Cleanup CUB util_macro.cuh by @bernhardmgruber in #1849
• Provide libcu++ transparent functors in C++11 by @bernhardmgruber in #1851
• Roll back upload-pages-artifact to v2. by @alliepiper in #1861
• Port CUB iterator test to Catch2 by @bernhardmgruber in #1822
• Symbol visibility is now invariant in regards to __cuda_std__ definition by @robertmaynard in #1832
• Add dimensions description functionality to CUDA Experimental library by @pciolkosz in #1743
• Document Asynchronous Operations by @gonzalobg in #1781
• Remove cpp11_required.h by @bernhardmgruber in #1860
• Add workflow to build RAPIDS from source with local CCCL by @trxcllnt in #1667
• Refactor CI matrix. by @alliepiper in #1844
• Adds tests for large number of items in cub::DeviceScan by @elstehle in #1830
• Make CUB test launch wrappers functor instances by @bernhardmgruber in #1850
• Improve CUB test overview docs by @bernhardmgruber in #1867
• Skip devcontainer validation jobs if not needed. by @alliepiper in #1853
• Improve CUB device-scope documentation by @bernhardmgruber in #1862
• Make integer sequence et al. available in C++11 by @bernhardmgruber in #1859
• Minimize template instantiations in CUB thread_load by @bernhardmgruber in #1857
• Create major version 2.6.0 by @wmaxey in #1880
• Drop facilities deprecated in CUB 1.x by @bernhardmgruber in #1868
• Make thrust::sort use radix sort with more comparators by @bernhardmgruber in #1884
• Make cuda::ptx::*_multicast pass on all architectures by @ahendriksen in #1874
• Replace typedef by alias declarations in CUB by @bernhardmgruber in #1885
• Remove legacy benchmarks and other dvs/p4 remnants by @alliepiper in #1901
• Qualify call to distance in thrust::async_reduce by @bernhardmgruber in #1904
• Rename CUB uninitialized_copy by @bernhardmgruber in #1913
• Sanitizer fixes by @alliepiper in #1916
• Use c2h::vectors in all non-example CUB tests by @bernhardmgruber in #1914
• Renamed overlooked uninitialized_copy by @bernhardmgruber in #1920
• Add assert implementation for device side testing by @pciolkosz in #1918
• Thrust and CUB: README: Fix copy-paste from libcu++ and links by @pauleonix in #1878
• Follow-up fixes to CUB iterator test by @bernhardmgruber in #1875
• Replace typedef by alias declarations in Thrust by @bernhardmgruber in #1915
• Cleanup CUB util_type.cuh by @bernhardmgruber in #1863
• Fix include for in cub/util_type.cuh by @bernhardmgruber in #1929
• Fix issues with comments in the concept emulation by @miscco in #1931
• Deprecate and reduce use of old functional stuff by @bernhardmgruber in #1925
• Deprecate more nested aliases in thrust functors by @bernhardmgruber in #1932
• Fix various typos in CUB documentation and comments. by @brycelelbach in #1933
• Add BabelStream flavors as thrust::transform benchmarks by @bernhardmgruber in #1921
• Some cleanup in Thrust config headers by @bernhardmgruber in #1934
• Update to CUDA 12.5 containers by @jrhemstad in #1935
• Check that the current version of CMake supports policy 141 before se… by @alliepiper in #1924
• Fix memmove optimization by @miscco in #1937
• Fixes thrust::unique_by_key examples by @elstehle in #1943
• Use only explicit NVTX3 V1 API in CUB by @bernhardmgruber in #1751
• Suppress a clang warning on array size computation by @bernhardmgruber in #1942
• Add a benchmark for thrust::equal by @bernhardmgruber in #1944
• Strip prefix paths to improve doc rendering by @bdice in #1954
• Modernize Thrust's alignment.h and triple_chevron_launch by @bernhardmgruber in #1905
• Restore RAPIDS devcontainer by @bdice in #1955
• Fix for in-place DeviceSelect & thrust::remove_if by @elstehle in #1782
• Drop Thrust's cstdint.h by @bernhardmgruber in #1959
• Use make_devcontainers.sh --clean when validating. by @alliepiper in #1963
• Fix missing binary_pred in thrust::unique_by_key by @bernhardmgruber in #1957
• cuda::launch and launch configuration object with minimal functionality by @pciolkosz in #1950
• Backport PR #2046 - Fixing FP16 conversions. by @wmaxey in #2222

Full Changelog: v2.5.0...v2.6.0

CCCL 2.5.0

What's New

This release includes several notable improvements and new features:

• CUB device-level algorithms now support NVTX ranges in Nsight Systems. This integration makes it easier to identify and analyze the time spent in CUB algorithms. Please note that profiling with this feature requires at least C++14.
• We have added new cub::DeviceSelect::FlaggedIf API, which allows you to select items based on applying a predicate to flags. This addition provides more flexibility and control over item selection.

What's Changed

• Clean up libcu++ docs landing page by @jrhemstad in #1492
• PTX: Add cuda::ptx::elect_sync by @ahendriksen in #1537
• Print a summary of all tests sorted by execution time. by @alliepiper in #1539
• Fix unused variable warning for __can_use_complete_tx by @wmaxey in #1547
• Fix usage of naked array with 0 elements in sm90 barrier tests. by @wmaxey in #1546
• Add support for stream operators for complex by @miscco in #1538
• Fix __half for older architectures by @miscco in #1543
• Feat 565 remove redundant thrust dialect conditional by @ZelboK in #566
• fix missing device hint in WarpMergeSort Documentation by @MARD1NO in #1553
• Minor fixes and additions on cub developer guides by @gonidelis in #1559
• Consolidate handling of constexpr and if constexpr by @miscco in #1562
• Ensure that cuda::aligned_size_t is usable in a constexpr context by @miscco in #1564
• Group CUB docs by @gevtushenko in #1565
• Update toolkit to 12.4 by @miscco in #1554
• Work around change in cuTensorMapEncode by @miscco in #1567
• Remove stdlib arg from .clangd. by @alliepiper in #1569
• Add the DeviceSelect::FlaggedIf algorithm by @gonidelis in #1533
• Catch2 segmented sort by @alliepiper in #1484
• Do not emit diagnostic with extended device lambdas with preserved re… by @Revaj in #1495
• Use absolute includes for libcu++ by @miscco in #1560
• [NFC] Modularize <exception> by @miscco in #199
• Add test support for launching kernels with cluster size > 1 by @ahendriksen in #416
• Fix typo in README.md by @bprb in #1574
• [FEA]: Modularize <cuda/memory_resource> by @miscco in #1532
• Cleanup_complex by @miscco in #1555
• Add missing comma in barrier __try_wait by @miscco in #1593
• Segmented sort test fix by @alliepiper in #1591
• Add pre-commit configuration by @bdice in #1596
• Preserve .devcontainer/img/ when cleaning. by @alliepiper in #1604
• Add some documentation for recent additions to libcu++ by @miscco in #1594
• Ensure cuda::std::nullopt is visible in device code by @trxcllnt in #1598
• Fix ordering of alignas and __shared__ by @miscco in #1601
• Update Thrust CI tests. by @alliepiper in #1605
• Implement tuple interface for cuda vector types by @miscco in #1410
• Inspect PR changes to determine if subproject builds are needed. by @alliepiper in #1572
• Apply clang-format to cub by @bdice in #1602
• Add missing non-volatile atomic overloads. by @wmaxey in #1582
• Drop unused libcxx files by @miscco in #1606
• Apply formatting to libcudacxx by @miscco in #1610
• Add conda documentation to the README. by @bdice in #1581
• Allow jobs to be skipped. by @alliepiper in #1611
• Make libcu++ work with exceptions by @miscco in #1607
• Implement cuda::mr::cuda_memory_resource by @miscco in #1578
• Implement cuda::mr::managed_memory_resource by @miscco in #1579
• Apply formatting to thrust by @miscco in #1616
• Update example_device_radix_sort.cu by @eriktedhamre in #1608
• Implement cuda::mr::pinned_memory_resource by @miscco in #1580
• Set the devcontainers to format on save. by @miscco in #1624
• Enable internal use of std::allocator related functionality by @miscco in #1583
• Adds tests for large number of items for cub::DeviceSelect by @elstehle in #1612
• Add pre-commit docs to CONTRIBUTING.md. by @bdice in #1627
• Move visibility attributes to cccl by @miscco in #1595
• Work around thrust/memory.h circular include by @dkolsen-pgi in #1634
• Fix mbarrier.init addressing by @ahendriksen in #1636
• Trim trailing whitespace and normalize newlines. by @bdice in #1633
• Add a git-blame-ignore-revs file by @miscco in #1629
• Revert "PTX: Add cuda::ptx::elect_sync (#1537)" by @ahendriksen in #1638
• Address potential oob in cub when passing in an invalid device counter by @miscco in #1641
• Allow ninja_summary to fail by @jrhemstad in #1644
• Mostly flatten the folder structure of libcu++ by @miscco in #1630
• Make --cmake-options="" always override others. by @alliepiper in #1648
• Fix invalid _CCCL_CUDACC definition for clang cuda by @miscco in #1656
• Add missing #pragma once in some headers by @bernhardmgruber in #1668
• Add NVTX ranges for all CUB algorithms by @bernhardmgruber in #1657
• Implement LWG-3843 and LWG-3940 by @miscco in #1621
• Modularize <memory> by @miscco in #1639
• Expose <cuda/std/numeric> to be publicly available by @miscco in #1671
• Add nsight support for automated debugging by @gonidelis in #1660
• Format core headers by @miscco in #1670
• Guard resource_ref and friends behind feature flag by @miscco in #1675
• Create major version 2.5.0 by @wmaxey in #1677
• Install CUB headers with .hpp extension by @bernhardmgruber in #1687
• Update CMakePresets.json by @alliepiper in #1686
• Fix deprecated status by @gevtushenko in #1692
• Test combined internal/user-side use of NVTX by @bernhardmgruber in #1690
• CI Overhaul, new nightly workflow by @alliepiper in #1654
• Fix CMake option handling. by @alliepiper in #1698
• Fix issues that came up with building cuDF with main by @miscco in #1643
• Drop new properties until we are certain about the design by @miscco in #1681
• Remove more uses of __cuda_std__ by @miscco in #1669
• Fix usage of result_of in thrust by @miscco in #1705
• Fix thrust::optional<T&>::emplace() by @Snektron in #1707
• Remove old f(void) function signatures by @bernhardmgruber in #1708
• Fix code sample in README and docs by @pauleonix in #1652
• Format libcudacxx/include files without extensions by @bdice in #1676
• Several improvements to zip_iterator/zip_function by @bernhardmgruber in #1710
• Expose thrust's contiguous iterator unwrap helpers by @bernhardmgruber in #1717
• Fix flakey heterogeneous tests by @wmaxey in #1712
• Ensure that we can use cuda::std::optional with types that are not __host__ __device__ by @miscco in #1663
• Fix a typo in barrier docs and update the godbolt link by @PointKernel in #1718
• Massively improve test times in heterogeneous atomics tests by @wmaxey in #1719
• Consolidate more common functi...

v2.4.0

What’s New

We are still hard at work in CCCL on paying down lots of technical debt, improving infrastructure, and various other simplifications as part of the unification of Thrust/CUB/libcu++. In addition to various fixes and documentation improvements, the following notable improvements have been made to Thrust, CUB, and libcudacxx.

Thrust

As part of our kernel consolidation effort, kernels of thrust::unique_by_key, thrust::copy_if, and thrust::partition algorithms are now consolidated in CUB. Kernel consolidation achieves two goals. First, it delivers the latest optimizations of CUB algorithms to Thrust users. Apart from the performance improvements, it introduces support of large problem sizes (64-bit offsets) into Thrust algorithms.

CUB

• cub::DeviceSelect::UniqueByKey now supports equality operator and large problem sizes.
• New cub::DeviceFor family of algorithms goes beyond conventional cub::DeviceFor::ForEach. cub::DeviceFor::ForEachCopy can provide you with additional performance benefits from vectorized memory accesses.
• Many CUB algorithms now support CUDA graph capture mode.

libcudacxx

• Added new cuda::ptx namespace with wrappers for inline-PTX instructions
• cuda::std::complex specializations for CUDA types bfloat and half.

What's Changed

• Implement remaining ranges iterator concepts and modernize array by @miscco in #627
• Fix C++11 support of recently added tests by @ahendriksen in #651
• Update CUDA newest to CTK 12.3 by @jrhemstad in #629
• Add cuda::ptx::* namespace by @ahendriksen in #574
• The test seems to pass just fine by @miscco in #654
• Fixes discard_memory compilation failure for pre-Volta by @elstehle in #637
• Reduce benchmarking time by @gevtushenko in #657
• Add CCCL_VERSION and script for updating version by @jrhemstad in #652
• Fixes compiler error for extended fp type data gen by @elstehle in #666
• fixup ___CUDA_VPTX -> _CUDA_VPTX by @wmaxey in #664
• Attempt to WAR CUB / RDC / MSVC issue by @gevtushenko in #669
• Rework our system header approach to be more error proof by @miscco in #661
• Project automation - fix sync action and draft setting step by @jarmak-nv in #625
• Fix fallback when checking git repo by @wmaxey in #1085
• Currently the verbose option does not work beacuse of a typo in the argument handling by @miscco in #1088
• Adds virtual shared memory helper and tests by @elstehle in #619
• Add cuda::ptx::st_async by @ahendriksen in #1078
• Add cuda::ptx::red_async by @ahendriksen in #1080
• Remove libcudacxx symlinks by @wmaxey in #1075
• Move PTX tests that missed the symlink PR by @wmaxey in #1098
• Fix truncation of constant value by @gevtushenko in #1097
• Add cuda::ptx:mbarrier_{try/test}_wait{_parity} by @ahendriksen in #674
• Initial CUB/NVRTC support by @gevtushenko in #1081
• Fix cuda::ptx::red.async for int32_t types by @ahendriksen in #1102
• Fix local test runs with lit by @miscco in #1108
• Fix config when only non-CDPv1 arches are enabled. by @alliepiper in #1109
• Do not replace the sccache binary for windows by @miscco in #1115
• Test cuda graph capture by @gevtushenko in #1112
• Fix overflow bug for >2^32 elements in thrust::shuffle by @djns99 in #1074
• Introduce CUB transform reduce by @gevtushenko in #1091
• Add infrastructure for compile-time CUB tests by @gevtushenko in #1124
• Fix GCC6 / FP8 warning by @gevtushenko in #1130
• Fix thrust transform reduce bench by @gevtushenko in #1133
• Fix ptx.st.async.compile.pass.cpp failing in C++11. by @wmaxey in #1132
• Fix _LIBCUDACXX_UNREACHABLE for old MSVC by @miscco in #1114
• Allow filtering P0 benchmarks by @gevtushenko in #1135
• Update barrier_arrive_tx.md docs by @gonzalobg in #1147
• Update std iterators by @miscco in #672
• Fix argument name in windows CI by @miscco in #1145
• Fix XFAIL condition for subsumption tests by @miscco in #1144
• Project Automation - remove draft automation + reduce permissions by @jarmak-nv in #1154
• Use rst in block-scope docs by @gevtushenko in #1150
• Fix errors when find_package(CCCL) is called twice. by @alliepiper in #1157
• Fix icc / cub by @gevtushenko in #1152
• Abort testing on unsupported dialect flags by @wmaxey in #1158
• Run with latest nvbench by @robertmaynard in #583
• Set finer-grain workflow permissions by @jrhemstad in #1163
• Port device docs to rst by @gevtushenko in #1160
• CI log improvements by @jrhemstad in #621
• Setup documentation and corresponding github action by @wmaxey in #1118
• Update Docs links in README.md by @wmaxey in #1169
• Fix GCC 13 by @gevtushenko in #1175
• Add missing exit from run-as-coder by @jrhemstad in #1176
• Adds new virtual shared memory facility to DeviceMergeSort by @elstehle in #1117
• Add first batch of Catch2 tests for DeviceRadixSort by @alliepiper in #1164
• Implement math functions for thrust::complex by @miscco in #1178
• Use anchors in matrix.yaml by @jrhemstad in #1193
• Ensure the targets that Thrust creates are global. by @robertmaynard in #1182
• Fix availability of is_constant_evaluated on old MSVC by @miscco in #1180
• Enable std::variant for libcu++ by @miscco in #1076
• Implement enable_borrowed_range by @miscco in #1196
• Reduce thrust benchmarks noise by @gevtushenko in #1203
• Prepare more algorithms by @miscco in #1161
• Add icc compiler to CI matrix by @jrhemstad in #1159
• Unify handling of dialects by @miscco in #1200
• Add argument to build/test scripts for additional cmake options by @jrhemstad in #620
• Move definitions of execution space macros into cccl by @miscco in #1199
• Adds new virtual shared memory facility to DeviceSelect::UniqueByKey by @elstehle in #1197
• Add Catch2 tests for cub::DeviceSegmentedRadixSort by @alliepiper in #1214
• Fix the example on README.md by @so298 in #1220
• Add missing overloads for thrust::pow by @miscco in #1222
• Fix 'nvc++ -stdpar' by @dkolsen-pgi in #1224
• Fix examples in reduce docs by @gevtushenko in #1230
• Do not benchmark small problem sizes by @gevtushenko in #1243
• Implement enable_view by @miscco in #1208
• Refactors thrust::unique_by_key to use cub::DeviceSelect::UniqueByKey by @elstehle in #1245
• Fix merge conflict from incoming PR by @miscco in #1250
• Disable fast-math for ICC by @miscco in #1252
• Fix a typo in thrust-config.cmake by @valgur in #1259
• Implement ranges::{c}begin and ranges::{c}end by @miscco in #1256
• Switch to entropy-based stopping criterion by @gevtushenko in #1280
• Fix a sync bug in stream_ref::wait by @PointKernel in #1238
• Silence some static asserts in ptx helpers by @miscco in #1257
• Restore docs images...

v2.3.2

What's Changed

• [BACKPORT]: Silence some static asserts in ptx helpers (#1257) by @miscco in #1284
• [BACKPORT]: Ensure that pair is trivially copyable (#1249) by @miscco in #1292
• [BACKPORT]: Properly test internal headers (#1258) by @miscco in #1299
• [Backport]: Fix errors when find_package(CCCL) is called twice. (#1157) by @miscco in #1298
• [BACKPORT] Fix MSVC issues (#1261) by @miscco in #1297
• [backport] thrust/mr: fix the case of reuising a block for a smaller alloc. (#1232) by @griwes in #1317
• [BACKPORT]: Fix ptx usage to account for PTX ISA availability (#1359) by @miscco in #1421
• Create patch 2.3.2 by @wmaxey in #1530

Full Changelog: v2.3.1...v2.3.2

v2.3.1

What's Changed

• [BACKPORT]: Fix bug in stream_ref::wait by @miscco in #1283
• Revert "Refactor thrust::complex as a struct derived from cuda::std::complex (#454)" by @miscco in #1286
• Create patch 2.3.1 by @wmaxey in #1287

Full Changelog: v2.3.0...v2.3.1