cuda-cccl 0.1.3.2.0.dev271__cp312-cp312-manylinux_2_26_x86_64.whl → 0.2.1__cp312-cp312-manylinux_2_26_x86_64.whl

cuda/cccl/_cuda_version_utils.py

@@ -6,9 +6,6 @@
 CUDA version detection utilities shared across the cccl package.
 """
 
-import os
-import shutil
-from pathlib import Path
 from typing import Optional
 
 import cuda.bindings
@@ -19,25 +16,6 @@ def detect_cuda_version() -> Optional[int]:
     return int(cuda_version.split(".")[0])
 
 
-def get_cuda_path() -> Optional[Path]:
-    """Get the CUDA installation path."""
-    cuda_path_str = os.environ.get("CUDA_PATH")
-    if cuda_path_str:
-        cuda_path = Path(cuda_path_str)
-        if cuda_path.exists():
-            return cuda_path
-
-    nvcc_path = shutil.which("nvcc")
-    if nvcc_path:
-        return Path(nvcc_path).parent.parent
-
-    default_path = Path("/usr/local/cuda")
-    if default_path.exists():
-        return default_path
-
-    return None
-
-
 def get_recommended_extra(cuda_version: Optional[int]) -> str:
     """Get the recommended pip extra for the detected CUDA version."""
     if cuda_version == 13:

cuda/cccl/headers/include/cub/agent/agent_batch_memcpy.cuh

@@ -52,9 +52,13 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/cmath>
+#include <cuda/__cmath/ceil_div.h>
+#include <cuda/__cmath/round_up.h>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/type_identity.h>
 #include <cuda/std/cstdint>
-#include <cuda/std/type_traits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_histogram.cuh

@@ -49,7 +49,9 @@
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_pointer.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_radix_sort_histogram.cuh

@@ -50,8 +50,11 @@
 #include <cub/util_math.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/ptx>
-#include <cuda/std/__algorithm_>
+#include <cuda/__cmath/ceil_div.h>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__algorithm/max.h>
+#include <cuda/std/__algorithm/min.h>
+#include <cuda/std/__functional/operations.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_radix_sort_onesweep.cuh

@@ -49,8 +49,10 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/ptx>
-#include <cuda/std/type_traits>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_radix_sort_upsweep.cuh

@@ -52,8 +52,9 @@
 #include <cub/util_type.cuh>
 #include <cub/warp/warp_reduce.cuh>
 
-#include <cuda/ptx>
-#include <cuda/std/__algorithm_>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__algorithm/max.h>
+#include <cuda/std/__algorithm/min.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_reduce.cuh

@@ -50,9 +50,12 @@
 #include <cub/util_device.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/memory>
-#include <cuda/std/functional>
-#include <cuda/std/type_traits>
+#include <cuda/__memory/is_aligned.h>
+#include <cuda/std/__algorithm/min.h>
+#include <cuda/std/__functional/identity.h>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/is_pointer.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_reduce_by_key.cuh

@@ -50,8 +50,10 @@
 #include <cub/block/block_store.cuh>
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 
-#include <cuda/std/type_traits>
-
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/is_pointer.h>
+#include <cuda/std/__type_traits/is_same.h>
 CUB_NAMESPACE_BEGIN
 
 /******************************************************************************

cuda/cccl/headers/include/cub/agent/agent_rle.cuh

@@ -50,11 +50,14 @@
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_scan.cuh>
 #include <cub/block/block_store.cuh>
-#include <cub/grid/grid_queue.cuh>
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 
-#include <cuda/ptx>
-#include <cuda/std/type_traits>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_pointer.h>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_scan.cuh

@@ -47,10 +47,11 @@
 #include <cub/block/block_load.cuh>
 #include <cub/block/block_scan.cuh>
 #include <cub/block/block_store.cuh>
-#include <cub/grid/grid_queue.cuh>
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/is_pointer.h>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_scan_by_key.cuh

@@ -50,7 +50,11 @@
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_pointer.h>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_select_if.cuh

@@ -50,11 +50,17 @@
 #include <cub/block/block_scan.cuh>
 #include <cub/block/block_store.cuh>
 #include <cub/device/dispatch/dispatch_common.cuh>
-#include <cub/grid/grid_queue.cuh>
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_callable.h>
+#include <cuda/std/__type_traits/is_pointer.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/cstdint>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/agent_three_way_partition.cuh

@@ -45,7 +45,10 @@
 #include <cub/block/block_store.cuh>
 #include <cub/iterator/cache_modified_input_iterator.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/is_pointer.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/agent/single_pass_scan_operators.cuh

@@ -51,7 +51,8 @@
 #include <cub/util_temporary_storage.cuh>
 #include <cub/warp/warp_reduce.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
 
 #include <nv/target>
 
@@ -1178,13 +1179,13 @@ struct ReduceByKeyScanTileState<ValueT, KeyT, true>
 #endif // _CCCL_DOXYGEN_INVOKED
 
 /******************************************************************************
- * Prefix call-back operator for coupling local block scan within a
+ * Prefix callback operator for coupling local block scan within a
  * block-cooperative scan
  ******************************************************************************/
 
 /**
- * Stateful block-scan prefix functor.  Provides the the running prefix for
- * the current tile by using the call-back warp to wait on on
+ * Stateful block-scan prefix functor.  Provides the running prefix for
+ * the current tile by using the callback warp to wait for
  * aggregates/prefixes from predecessor tiles to become available.
  *
  * @tparam DelayConstructorT

cuda/cccl/headers/include/cub/block/block_exchange.cuh

@@ -47,7 +47,9 @@
 #include <cub/util_type.cuh>
 #include <cub/warp/warp_exchange.cuh>
 
-#include <cuda/ptx>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__algorithm/min.h>
+#include <cuda/std/__type_traits/integral_constant.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_histogram.cuh

@@ -48,7 +48,7 @@
 #include <cub/block/specializations/block_histogram_sort.cuh>
 #include <cub/util_ptx.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/conditional.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_merge_sort.cuh

@@ -43,9 +43,8 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/__algorithm/max.h>
 #include <cuda/std/__algorithm/min.h>
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 
@@ -384,7 +383,7 @@ public:
        int valid_items,
        KeyT oob_default)
   {
-    if (IS_LAST_TILE)
+    if constexpr (IS_LAST_TILE)
     {
       // if last tile, find valid max_key
       // and fill the remaining keys with it
@@ -418,8 +417,8 @@ public:
     for (int target_merged_threads_number = 2; target_merged_threads_number <= NUM_THREADS;
          target_merged_threads_number *= 2)
     {
-      int merged_threads_number = target_merged_threads_number / 2;
-      int mask                  = target_merged_threads_number - 1;
+      const int merged_threads_number = target_merged_threads_number / 2;
+      const int mask                  = target_merged_threads_number - 1;
 
       Sync();
 
@@ -436,23 +435,23 @@ public:
 
       int indices[ITEMS_PER_THREAD];
 
-      int first_thread_idx_in_thread_group_being_merged = ~mask & linear_tid;
-      int start = ITEMS_PER_THREAD * first_thread_idx_in_thread_group_being_merged;
-      int size  = ITEMS_PER_THREAD * merged_threads_number;
+      const int first_thread_idx_in_thread_group_being_merged = ~mask & linear_tid;
+      const int start = ITEMS_PER_THREAD * first_thread_idx_in_thread_group_being_merged;
+      const int size  = ITEMS_PER_THREAD * merged_threads_number;
 
-      int thread_idx_in_thread_group_being_merged = mask & linear_tid;
+      const int thread_idx_in_thread_group_being_merged = mask & linear_tid;
 
-      int diag = (::cuda::std::min) (valid_items, ITEMS_PER_THREAD * thread_idx_in_thread_group_being_merged);
+      const int diag = (::cuda::std::min) (valid_items, ITEMS_PER_THREAD * thread_idx_in_thread_group_being_merged);
 
-      int keys1_beg = (::cuda::std::min) (valid_items, start);
-      int keys1_end = (::cuda::std::min) (valid_items, keys1_beg + size);
-      int keys2_beg = keys1_end;
-      int keys2_end = (::cuda::std::min) (valid_items, keys2_beg + size);
+      const int keys1_beg = (::cuda::std::min) (valid_items, start);
+      const int keys1_end = (::cuda::std::min) (valid_items, keys1_beg + size);
+      const int keys2_beg = keys1_end;
+      const int keys2_end = (::cuda::std::min) (valid_items, keys2_beg + size);
 
-      int keys1_count = keys1_end - keys1_beg;
-      int keys2_count = keys2_end - keys2_beg;
+      const int keys1_count = keys1_end - keys1_beg;
+      const int keys2_count = keys2_end - keys2_beg;
 
-      int partition_diag = MergePath(
+      const int partition_diag = MergePath(
         &temp_storage.keys_shared[keys1_beg],
         &temp_storage.keys_shared[keys2_beg],
         keys1_count,
@@ -460,12 +459,12 @@ public:
         diag,
         compare_op);
 
-      int keys1_beg_loc   = keys1_beg + partition_diag;
-      int keys1_end_loc   = keys1_end;
-      int keys2_beg_loc   = keys2_beg + diag - partition_diag;
-      int keys2_end_loc   = keys2_end;
-      int keys1_count_loc = keys1_end_loc - keys1_beg_loc;
-      int keys2_count_loc = keys2_end_loc - keys2_beg_loc;
+      const int keys1_beg_loc   = keys1_beg + partition_diag;
+      const int keys1_end_loc   = keys1_end;
+      const int keys2_beg_loc   = keys2_beg + diag - partition_diag;
+      const int keys2_end_loc   = keys2_end;
+      const int keys1_count_loc = keys1_end_loc - keys1_beg_loc;
+      const int keys2_count_loc = keys2_end_loc - keys2_beg_loc;
       SerialMerge(
         &temp_storage.keys_shared[0],
         keys1_beg_loc,
@@ -477,7 +476,7 @@ public:
         compare_op,
         oob_default);
 
-      if (!KEYS_ONLY)
+      if constexpr (!KEYS_ONLY)
       {
         Sync();
 

cuda/cccl/headers/include/cub/block/block_radix_rank.cuh

@@ -48,12 +48,14 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/ptx>
-#include <cuda/std/__algorithm_>
+#include <cuda/__ptx/instructions/get_sreg.h>
+#include <cuda/std/__algorithm/max.h>
+#include <cuda/std/__functional/operations.h>
+#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>
-#include <cuda/std/type_traits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_radix_sort.cuh

@@ -50,8 +50,11 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/__algorithm_>
-#include <cuda/std/type_traits>
+#include <cuda/std/__algorithm/min.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_convertible.h>
+#include <cuda/std/__type_traits/is_same.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_reduce.cuh

@@ -49,7 +49,8 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_run_length_decode.cuh

@@ -44,10 +44,7 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/__algorithm/max.h>
 #include <cuda/std/__algorithm/min.h>
-#include <cuda/std/limits>
-#include <cuda/std/type_traits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/block_scan.cuh

@@ -47,7 +47,8 @@
 #include <cub/util_ptx.cuh>
 #include <cub/util_type.cuh>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__functional/operations.h>
+#include <cuda/std/__type_traits/conditional.h>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/radix_rank_sort_operations.cuh

@@ -49,14 +49,18 @@
 
 #include <thrust/type_traits/integer_sequence.h>
 
-#include <cuda/bit>
-#include <cuda/functional>
-#include <cuda/std/__algorithm/max.h>
+#include <cuda/__bit/bitfield.h>
+#include <cuda/__type_traits/is_floating_point.h>
+#include <cuda/__utility/static_for.h>
 #include <cuda/std/__algorithm/min.h>
+#include <cuda/std/__functional/invoke.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/__type_traits/remove_cv.h>
+#include <cuda/std/__type_traits/void_t.h>
 #include <cuda/std/cstdint>
 #include <cuda/std/tuple>
-#include <cuda/std/type_traits>
-#include <cuda/type_traits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/block/specializations/block_reduce_warp_reductions.cuh

@@ -49,8 +49,8 @@
 #include <cub/warp/warp_reduce.cuh>
 
 #include <cuda/__cmath/ceil_div.h>
+#include <cuda/__ptx/instructions/get_sreg.h>
 #include <cuda/atomic>
-#include <cuda/ptx>
 #include <cuda/std/__algorithm/min.h>
 
 CUB_NAMESPACE_BEGIN

cuda/cccl/headers/include/cub/block/specializations/block_scan_warp_scans.cuh

@@ -47,8 +47,8 @@
 #include <cub/util_ptx.cuh>
 #include <cub/warp/warp_scan.cuh>
 
-#include <cuda/cmath>
-#include <cuda/ptx>
+#include <cuda/__cmath/ceil_div.h>
+#include <cuda/__ptx/instructions/get_sreg.h>
 
 CUB_NAMESPACE_BEGIN
 namespace detail

cuda/cccl/headers/include/cub/detail/array_utils.cuh

@@ -38,11 +38,12 @@
 #include <cub/detail/type_traits.cuh> // static_size_v
 #include <cub/util_namespace.cuh>
 
-#include <cuda/std/array> // array
-#include <cuda/std/cstddef> // size_t
-#include <cuda/std/iterator> // ::cuda::std::iter_value_t
-#include <cuda/std/type_traits> // _If
-#include <cuda/std/utility> // index_sequence
+#include <cuda/std/__iterator/iterator_traits.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/__utility/integer_sequence.h>
+#include <cuda/std/array>
+#include <cuda/std/cstddef>
 
 CUB_NAMESPACE_BEGIN
 namespace detail

cuda/cccl/headers/include/cub/detail/choose_offset.cuh

@@ -37,9 +37,15 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/std/__iterator/iterator_traits.h>
+#include <cuda/std/__type_traits/common_type.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/is_integral.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/__type_traits/is_unsigned.h>
+#include <cuda/std/__type_traits/remove_cv.h>
 #include <cuda/std/cstdint>
-#include <cuda/std/iterator>
-#include <cuda/std/type_traits>
+#include <cuda/std/limits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/detail/fast_modulo_division.cuh

@@ -40,12 +40,19 @@
 #include <cub/detail/type_traits.cuh> // implicit_prom_t
 #include <cub/util_type.cuh> // _CCCL_HAS_INT128()
 
-#include <cuda/cmath> // cuda::std::ceil_div
-#include <cuda/std/bit> // cuda::std::has_single_bit
+#include <cuda/__cmath/ceil_div.h>
+#include <cuda/std/__bit/has_single_bit.h>
+#include <cuda/std/__bit/integral.h>
+#include <cuda/std/__type_traits/conditional.h>
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__type_traits/integral_constant.h>
+#include <cuda/std/__type_traits/is_integral.h>
+#include <cuda/std/__type_traits/is_same.h>
+#include <cuda/std/__type_traits/is_signed.h>
+#include <cuda/std/__type_traits/make_unsigned.h>
 #include <cuda/std/climits> // CHAR_BIT
 #include <cuda/std/cstdint> // uint64_t
-#include <cuda/std/limits> // numeric_limits
-#include <cuda/std/type_traits> // ::cuda::std::is_integral
+#include <cuda/std/limits>
 
 #if defined(CCCL_ENABLE_DEVICE_ASSERTIONS)
 _CCCL_BEGIN_NV_DIAG_SUPPRESS(186) // pointless comparison of unsigned integer with zero

cuda/cccl/headers/include/cub/detail/integer_utils.cuh

@@ -15,9 +15,13 @@
 #include <cub/detail/unsafe_bitcast.cuh>
 #include <cub/thread/thread_operators.cuh> // is_cuda_minimum_maximum_v
 
-#include <cuda/std/cmath> // isnan
-#include <cuda/std/limits> // numeric_limits
-#include <cuda/type_traits> // is_floating_point_v
+#include <cuda/__type_traits/is_floating_point.h>
+#include <cuda/std/__cmath/isnan.h>
+#include <cuda/std/__type_traits/is_integer.h>
+#include <cuda/std/__type_traits/is_signed.h>
+#include <cuda/std/__type_traits/make_nbit_int.h>
+#include <cuda/std/__type_traits/make_unsigned.h>
+#include <cuda/std/limits>
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/detail/mdspan_utils.cuh

@@ -39,11 +39,11 @@
 
 #include <cub/detail/fast_modulo_division.cuh> // fast_div_mod
 
-#include <cuda/std/array> // cuda::std::array
-#include <cuda/std/cstddef> // size_t
+#include <cuda/std/__type_traits/make_unsigned.h>
+#include <cuda/std/__utility/integer_sequence.h>
+#include <cuda/std/array>
+#include <cuda/std/cstddef>
 #include <cuda/std/mdspan>
-#include <cuda/std/type_traits> // make_unsigned_t
-#include <cuda/std/utility> // index_sequence
 
 CUB_NAMESPACE_BEGIN
 

cuda/cccl/headers/include/cub/detail/ptx-json/json.h

@@ -32,9 +32,9 @@
 #include <cub/detail/ptx-json/string.h>
 #include <cub/detail/ptx-json/value.h>
 
+#include <cuda/std/__type_traits/enable_if.h>
+#include <cuda/std/__utility/integer_sequence.h>
 #include <cuda/std/cstddef>
-#include <cuda/std/type_traits>
-#include <cuda/std/utility>
 
 namespace ptx_json
 {
@@ -47,9 +47,10 @@ struct tagged_json<T, cuda::std::index_sequence<Is...>>
   template <typename V, typename = cuda::std::enable_if_t<is_object<V>::value || is_array<V>::value>>
   __noinline__ __device__ void operator=(V)
   {
-    asm volatile("cccl.ptx_json.begin(%0)\n\n" ::"C"(storage_helper<T.str[Is]...>::value) : "memory");
+    static constexpr char str[]{T.str[Is]...};
+    asm volatile("cccl.ptx_json.begin(%0)\n\n" ::"C"(str) : "memory");
     V::emit();
-    asm volatile("\ncccl.ptx_json.end(%0)" ::"C"(storage_helper<T.str[Is]...>::value) : "memory");
+    asm volatile("\ncccl.ptx_json.end(%0)" ::"C"(str) : "memory");
   }
 };
 

cuda/cccl/headers/include/cub/detail/ptx-json/object.h

@@ -30,7 +30,7 @@
 #include <cub/detail/ptx-json/string.h>
 #include <cub/detail/ptx-json/value.h>
 
-#include <cuda/std/type_traits>
+#include <cuda/std/__type_traits/integral_constant.h>
 
 namespace ptx_json
 {

cuda/cccl/headers/include/cub/detail/ptx-json/string.h

@@ -50,22 +50,4 @@ __forceinline__ __device__ void comma()
 {
   asm volatile("," ::: "memory");
 }
-
-#pragma nv_diag_suppress 177
-template <char... Cs>
-struct storage_helper
-{
-  // This, and the dance to invoke this through value_traits elsewhere, is necessary because the "C" inline assembly
-  // constraint supported by NVCC requires that its argument is a pointer to a constant array of type char; NVCC also
-  // doesn't allow passing raw character literals as pointer template arguments; and *also* it seems to look at the type
-  // of a containing object, not a subobject it is given, when passed in a pointer to an array inside a literal type.
-  // All of this means that we can't just pass strings, and *also* we can't just use the string<N>::array member above
-  // as the string literal; therefore, using the fact that the length of the string is a core constant expression in the
-  // definition of value_traits, we can generate a variadic pack that allows us to expand the contents of
-  // string<N>::array into a comma separated list of N chars. We can then plug that in as template arguments to
-  // storage_helper, which then can, as below, turn that into its own char array that NVCC accepts as an argument for a
-  // "C" inline assembly constraint.
-  static const constexpr char value[] = {Cs...};
-};
-#pragma nv_diag_default 177
 } // namespace ptx_json