Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 1 addition & 4 deletions cub/cub/agent/agent_histogram.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -23,14 +23,11 @@
#include <cub/iterator/cache_modified_input_iterator.cuh>
#include <cub/util_type.cuh>

#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/integral_constant.h>
#include <cuda/std/__type_traits/is_pointer.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

enum BlockHistogramMemoryPreference
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/agent/agent_radix_sort_onesweep.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -30,14 +30,11 @@
#endif

#include <cuda/__ptx/instructions/get_sreg.h>
#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/integral_constant.h>
#include <cuda/std/__type_traits/is_same.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

/** \brief cub::RadixSortStoreAlgorithm enumerates different algorithms to write
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/block/block_load.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -22,12 +22,9 @@
#include <cub/util_ptx.cuh>
#include <cub/util_type.cuh>

#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__new/device_new.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

//! @name Blocked arrangement I/O (direct)
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/block/block_radix_rank.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -30,16 +30,13 @@
#include <cuda/std/__bit/integral.h>
#include <cuda/std/__bit/popcount.h>
#include <cuda/std/__functional/operations.h>
#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/is_same.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>
#include <cuda/std/span>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

//! @brief Radix ranking algorithm, the algorithm used to implement stable ranking of the
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/block/block_reduce.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -26,12 +26,9 @@
#include <cub/util_type.cuh>

#include <cuda/std/__functional/operations.h>
#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__type_traits/conditional.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

/******************************************************************************
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/block/block_scan.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -24,12 +24,9 @@
#include <cub/util_type.cuh>

#include <cuda/std/__functional/operations.h>
#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/__type_traits/conditional.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

/******************************************************************************
Expand Down
4 changes: 1 addition & 3 deletions cub/cub/block/block_store.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,7 @@
#include <cub/util_ptx.cuh>
#include <cub/util_type.cuh>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)
#include <cuda/std/__host_stdlib/ostream>

CUB_NAMESPACE_BEGIN

Expand Down
4 changes: 1 addition & 3 deletions cub/cub/detail/delay_constructor.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -15,9 +15,7 @@

#include <cub/agent/single_pass_scan_operators.cuh>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif
#include <cuda/std/__host_stdlib/ostream>

CUB_NAMESPACE_BEGIN

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_adjacent_difference.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -28,12 +28,9 @@
#include <cuda/__cmath/ceil_div.h>
#include <cuda/__device/arch_id.h>
#include <cuda/std/__functional/invoke.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/is_empty.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

CUB_NAMESPACE_BEGIN

namespace detail::adjacent_difference
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_batch_memcpy.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -31,15 +31,12 @@

#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/type_identity.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

CUB_NAMESPACE_BEGIN

enum class CopyAlg
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_find.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -32,10 +32,7 @@
#include <thrust/type_traits/unwrap_contiguous_iterator.h>

#include <cuda/__iterator/transform_iterator.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
#include <cuda/std/__host_stdlib/sstream>

CUB_NAMESPACE_BEGIN

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_merge.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -23,10 +23,7 @@
#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>

#include <cuda/std/__algorithm/min.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif
#include <cuda/std/__host_stdlib/sstream>

CUB_NAMESPACE_BEGIN
namespace detail::merge
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_merge_sort.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -24,13 +24,10 @@
#include <cuda/__cmath/ceil_div.h>
#include <cuda/__cmath/ilog.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/is_same.h>
#include <cuda/std/cstdint>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

CUB_NAMESPACE_BEGIN

namespace detail::merge_sort
Expand Down
4 changes: 1 addition & 3 deletions cub/cub/device/dispatch/dispatch_radix_sort.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -37,9 +37,7 @@
// TODO(bgruber): included for backward compatibility, remove in CCCL 4.0
#include <cub/device/dispatch/dispatch_segmented_radix_sort.cuh>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif
#include <cuda/std/__host_stdlib/sstream>

// suppress warnings triggered by #pragma unroll:
// "warning: loop not unrolled: the optimizer was unable to perform the requested transformation; the transformation
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_reduce.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -33,12 +33,9 @@

#include <cuda/std/__functional/identity.h>
#include <cuda/std/__functional/invoke.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/cstdint>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

// TODO(bgruber): included to not break users when moving DeviceSegmentedReduce to its own file. Remove in CCCL 4.0.
#include <cub/device/dispatch/dispatch_fixed_size_segmented_reduce.cuh>
#include <cub/device/dispatch/dispatch_segmented_reduce.cuh>
Expand Down
4 changes: 1 addition & 3 deletions cub/cub/device/dispatch/dispatch_reduce_by_key.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -32,9 +32,7 @@

#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
#include <cuda/std/__host_stdlib/sstream>

CUB_NAMESPACE_BEGIN

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_rle.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -30,13 +30,10 @@

#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_scan.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@
#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__functional/invoke.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__iterator/readable_traits.h>
#include <cuda/std/__memory/construct_at.h>
#include <cuda/std/__type_traits/conditional.h>
Expand All @@ -53,10 +54,6 @@
#include <cuda_runtime_api.h>
#include <cudaTypedefs.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

CUB_NAMESPACE_BEGIN

namespace detail::scan
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_segmented_reduce.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -25,16 +25,13 @@

#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/is_empty.h>
#include <cuda/std/__type_traits/type_identity.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

CUB_NAMESPACE_BEGIN

namespace detail::segmented_reduce
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_segmented_sort.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -29,17 +29,14 @@

#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__iterator/reverse_iterator.h>
#include <cuda/std/__type_traits/is_same.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

#include <nv/target>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

CUB_NAMESPACE_BEGIN

namespace detail::segmented_sort
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_select_if.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -34,17 +34,14 @@

#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__utility/swap.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

#include <nv/target>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

_CCCL_DIAG_PUSH
_CCCL_DIAG_SUPPRESS_GCC("-Wattributes") // __visibility__ attribute ignored
_CCCL_DIAG_SUPPRESS_NVHPC(attribute_requires_external_linkage)
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -24,13 +24,10 @@
#include <thrust/system/cuda/detail/core/triple_chevron_launch.h>

#include <cuda/std/__functional/invoke.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/conditional.h>
#include <cuda/std/__type_traits/is_same.h>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

CUB_NAMESPACE_BEGIN

namespace detail::reduce_by_key
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_three_way_partition.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -25,16 +25,13 @@
#include <cuda/__cmath/ceil_div.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__utility/swap.h>
#include <cuda/std/cstdint>
#include <cuda/std/limits>

#include <nv/target>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)

CUB_NAMESPACE_BEGIN

namespace detail::three_way_partition
Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/dispatch_transform.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@
#include <cuda/std/__algorithm/clamp.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__host_stdlib/sstream>
#include <cuda/std/__type_traits/integral_constant.h>
#include <cuda/std/__type_traits/is_same.h>
#include <cuda/std/__type_traits/void_t.h>
Expand All @@ -45,10 +46,6 @@
#include <cuda/std/optional>
#include <cuda/std/tuple>

#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
# include <sstream>
#endif

// On Windows, the `if CUB_DETAIL_CONSTEXPR_ISH` results in `warning C4702: unreachable code`.
_CCCL_DIAG_PUSH
_CCCL_DIAG_SUPPRESS_MSVC(4702)
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -18,10 +18,7 @@
#include <cub/util_math.cuh>

#include <cuda/__device/arch_id.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif
#include <cuda/std/__host_stdlib/ostream>

CUB_NAMESPACE_BEGIN

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/tuning/tuning_batch_memcpy.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -18,10 +18,7 @@
#include <cub/device/dispatch/tuning/common.cuh>

#include <cuda/__device/arch_id.h>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)
#include <cuda/std/__host_stdlib/ostream>

CUB_NAMESPACE_BEGIN

Expand Down
5 changes: 1 addition & 4 deletions cub/cub/device/dispatch/tuning/tuning_find.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -17,12 +17,9 @@
#include <cub/util_device.cuh>

#include <cuda/__device/arch_id.h>
#include <cuda/std/__host_stdlib/ostream>
#include <cuda/std/concepts>

#if !_CCCL_COMPILER(NVRTC)
# include <ostream>
#endif // !_CCCL_COMPILER(NVRTC)

CUB_NAMESPACE_BEGIN

namespace detail::find
Expand Down
Loading
Loading