Add a freestanding mode to allow building without any host standard library support

cub/cub/agent/agent_histogram.cuh

@@ -37,7 +37,7 @@ enum BlockHistogramMemoryPreference
   BLEND
 };
 
-#if !_CCCL_COMPILER(NVRTC)
+#if _CCCL_HOSTED()
 inline ::std::ostream& operator<<(::std::ostream& os, BlockHistogramMemoryPreference mempref)
 {
   switch (mempref)
@@ -52,7 +52,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, BlockHistogramMemoryPrefer
       return os << "<unknown BlockHistogramMemoryPreference: " << static_cast<int>(mempref) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC)
+#endif // _CCCL_HOSTED()
 
 //! Parameterizable tuning policy type for AgentHistogram
 //!

cub/cub/agent/agent_radix_sort_onesweep.cuh

@@ -55,7 +55,7 @@ enum RadixSortStoreAlgorithm
   RADIX_SORT_STORE_ALIGNED
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, RadixSortStoreAlgorithm algo)
 {
   switch (algo)
@@ -68,7 +68,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, RadixSortStoreAlgorithm al
       return os << "<unknown RadixSortStoreAlgorithm: " << static_cast<int>(algo) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 template <int NominalBlockThreads4B,
           int NominalItemsPerThread4B,

cub/cub/block/block_load.cuh

@@ -723,7 +723,7 @@ enum BlockLoadAlgorithm
   BLOCK_LOAD_WARP_TRANSPOSE_TIMESLICED,
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, BlockLoadAlgorithm algo)
 {
   switch (algo)
@@ -744,7 +744,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, BlockLoadAlgorithm algo)
       return os << "<unknown BlockLoadAlgorithm: " << static_cast<int>(algo) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 //! @rst
 //! The BlockLoad class provides :ref:`collective <collective-primitives>` data movement methods for loading a linear

cub/cub/block/block_radix_rank.cuh

@@ -71,7 +71,7 @@ enum RadixRankAlgorithm
   RADIX_RANK_MATCH_EARLY_COUNTS_ATOMIC_OR
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, RadixRankAlgorithm algo)
 {
   switch (algo)
@@ -90,7 +90,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, RadixRankAlgorithm algo)
       return os << "<unknown RadixRankAlgorithm: " << static_cast<int>(algo) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 /** Empty callback implementation */
 template <int BINS_PER_THREAD>

cub/cub/block/block_reduce.cuh

@@ -149,7 +149,7 @@ enum BlockReduceAlgorithm
   BLOCK_REDUCE_WARP_REDUCTIONS_NONDETERMINISTIC,
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, const BlockReduceAlgorithm& alg)
 {
   switch (alg)
@@ -166,7 +166,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, const BlockReduceAlgorithm
       return os << "<unknown BlockReduceAlgorithm: " << static_cast<int>(alg) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 //! @rst
 //! The BlockReduce class provides :ref:`collective <collective-primitives>` methods for computing a

cub/cub/block/block_scan.cuh

@@ -100,7 +100,7 @@ enum BlockScanAlgorithm
   BLOCK_SCAN_WARP_SCANS,
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, BlockScanAlgorithm algo)
 {
   switch (algo)
@@ -115,7 +115,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, BlockScanAlgorithm algo)
       return os << "<unknown BlockScanAlgorithm: " << static_cast<int>(algo) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 //! @rst
 //! The BlockScan class provides :ref:`collective <collective-primitives>` methods for computing a parallel prefix

cub/cub/block/block_store.cuh

@@ -542,7 +542,7 @@ enum BlockStoreAlgorithm
   BLOCK_STORE_WARP_TRANSPOSE_TIMESLICED,
 };
 
-#if !_CCCL_COMPILER(NVRTC) && !defined(_CCCL_DOXYGEN_INVOKED)
+#if _CCCL_HOSTED() && !defined(_CCCL_DOXYGEN_INVOKED)
 inline ::std::ostream& operator<<(::std::ostream& os, BlockStoreAlgorithm algo)
 {
   switch (algo)
@@ -563,7 +563,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, BlockStoreAlgorithm algo)
       return os << "<unknown BlockStoreAlgorithm: " << static_cast<int>(algo) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC) && !_CCCL_DOXYGEN_INVOKED
+#endif // _CCCL_HOSTED() && !_CCCL_DOXYGEN_INVOKED
 
 //! @rst
 //! The BlockStore class provides :ref:`collective <collective-primitives>` data movement

cub/cub/config.cuh

@@ -24,6 +24,6 @@
 #include <cub/util_macro.cuh> // IWYU pragma: export
 #include <cub/util_namespace.cuh> // IWYU pragma: export
 
-#if !_CCCL_COMPILER(NVRTC)
+#if _CCCL_HOSTED()
 #  include <cuda/__nvtx/nvtx.h>
-#endif // !_CCCL_COMPILER(NVRTC)
+#endif // _CCCL_HOSTED()

cub/cub/detail/delay_constructor.cuh

@@ -34,7 +34,7 @@ enum class delay_constructor_kind
   reduce_by_key
 };
 
-#if !_CCCL_COMPILER(NVRTC)
+#if _CCCL_HOSTED()
 inline ::std::ostream& operator<<(::std::ostream& os, delay_constructor_kind kind)
 {
   switch (kind)
@@ -61,7 +61,7 @@ inline ::std::ostream& operator<<(::std::ostream& os, delay_constructor_kind kin
       return os << "<unknown delay_constructor_kind: " << static_cast<int>(kind) << ">";
   }
 }
-#endif // !_CCCL_COMPILER(NVRTC)
+#endif // _CCCL_HOSTED()
 
 struct delay_constructor_policy
 {
@@ -79,13 +79,13 @@ struct delay_constructor_policy
     return !(lhs == rhs);
   }
 
-#if !_CCCL_COMPILER(NVRTC)
+#if _CCCL_HOSTED()
   friend ::std::ostream& operator<<(::std::ostream& os, const delay_constructor_policy& p)
   {
     return os << "delay_constructor_policy { .kind = " << p.kind << ", .delay = " << p.delay
               << ", .l2_write_latency = " << p.l2_write_latency << " }";
   }
-#endif // !_CCCL_COMPILER(NVRTC)
+#endif // _CCCL_HOSTED()
 };
 
 template <typename DelayConstructor>

cub/cub/device/dispatch/dispatch_adjacent_difference.cuh

@@ -344,12 +344,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
   }
 
   const adjacent_difference_policy active_policy = policy_selector(arch_id);
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (::std::stringstream ss; ss << active_policy;
      _CubLog("Dispatching DeviceAdjacentDifference to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   const int tile_size = active_policy.block_threads * active_policy.items_per_thread;
   const int num_tiles = static_cast<int>(::cuda::ceil_div(num_items, tile_size));

cub/cub/device/dispatch/dispatch_batch_memcpy.cuh

@@ -316,12 +316,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
   }
   const batch_memcpy_policy active_policy = policy_selector(arch_id);
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (::std::stringstream ss; ss << active_policy; _CubLog(
        "Dispatching DeviceBatchMemcpy to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   enum : uint32_t
   {

cub/cub/device/dispatch/dispatch_find.cuh

@@ -114,11 +114,11 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
 
   const find_policy active_policy = policy_selector(arch_id);
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(NV_IS_HOST,
                (std::stringstream ss; ss << active_policy;
                 _CubLog("Dispatching DeviceFind to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   const int tile_size = active_policy.block_threads * active_policy.items_per_thread;
   const int num_tiles = static_cast<int>(::cuda::ceil_div(num_items, tile_size));

cub/cub/device/dispatch/dispatch_histogram.cuh

@@ -200,11 +200,11 @@ CUB_RUNTIME_FUNCTION _CCCL_VISIBILITY_HIDDEN _CCCL_FORCEINLINE auto dispatch(
 
   const histogram_policy active_policy = policy_selector(arch_id);
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(NV_IS_HOST,
                (std::stringstream ss; ss << active_policy;
                 _CubLog("Dispatching DeviceHistogram to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   const auto init_kernel = kernel_source.template HistogramInitKernel<PolicySelector>();
   auto sweep_kernel      = [&] {

cub/cub/device/dispatch/dispatch_merge.cuh

@@ -216,12 +216,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
   }
 
   return dispatch_arch(policy_selector, arch_id, [&](auto policy_getter) {
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
     NV_IF_TARGET(
       NV_IS_HOST,
       (std::stringstream ss; ss << policy_getter();
        _CubLog("Dispatching DeviceMerge to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
     static_assert(::cuda::std::is_empty_v<decltype(policy_getter)>);
     using AgentT = typename choose_merge_agent<

cub/cub/device/dispatch/dispatch_merge_sort.cuh

@@ -507,12 +507,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
   constexpr merge_sort_policy active_policy = vsmem_adapted_agents::policy;
 #endif // CUB_DEFINE_RUNTIME_POLICIES
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
     NV_IF_TARGET(
       NV_IS_HOST,
       (std::stringstream ss; ss << active_policy;
        _CubLog("Dispatching DeviceMergeSort to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
     const auto tile_size = active_policy.items_per_tile();
     const auto num_tiles = ::cuda::ceil_div(num_items, tile_size);

cub/cub/device/dispatch/dispatch_radix_sort.cuh

@@ -1208,11 +1208,11 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE cudaError_t dispatch(
     return error;
   }
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(NV_IS_HOST,
                (std::stringstream ss; ss << policy_selector(arch_id);
                 _CubLog("Dispatching DeviceReduce to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   return dispatch_arch(policy_selector, arch_id, [&](auto policy_getter) {
     return DispatchRadixSort<Order, KeyT, ValueT, OffsetT, DecomposerT, fake_policy, KernelSource, KernelLauncherFactory>{

cub/cub/device/dispatch/dispatch_reduce.cuh

@@ -780,11 +780,11 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
   }
 
   const reduce_policy active_policy = policy_selector(arch_id);
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(NV_IS_HOST,
                (std::stringstream ss; ss << active_policy;
                 _CubLog("Dispatching DeviceReduce to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   // Check for small, single tile size
   if (num_items

cub/cub/device/dispatch/dispatch_reduce_by_key.cuh

@@ -709,12 +709,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE static cudaError_t dispatch(
   }
 
   return detail::dispatch_arch(policy_selector, arch_id, [&](auto policy_getter) {
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
     NV_IF_TARGET(
       NV_IS_HOST,
       (::std::stringstream ss; ss << policy_getter(); _CubLog(
          "Dispatching DeviceReduceByKey to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
     const auto [block_threads, items_per_thread, vsmem_per_block] = determine_threads_items_vsmem<
       decltype(policy_getter),

cub/cub/device/dispatch/dispatch_reduce_nondeterministic.cuh

@@ -184,12 +184,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch_nondeterministic(
   }
 
   const reduce_policy active_policy = policy_selector(arch_id);
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (std::stringstream ss; ss << active_policy; _CubLog(
        "Dispatching DeviceReduceNondeterministic to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   // No temp storage needed but keep API consistent
   if (d_temp_storage == nullptr)

cub/cub/device/dispatch/dispatch_rle.cuh

@@ -679,12 +679,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE static cudaError_t dispatch(
   }
 
   const non_trivial_runs::rle_non_trivial_runs_policy active_policy = policy_selector(arch_id);
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (::std::stringstream ss; ss << active_policy;
      _CubLog("Dispatching DeviceRle to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   const int block_threads    = active_policy.block_threads;
   const int items_per_thread = active_policy.items_per_thread;

cub/cub/device/dispatch/dispatch_scan.cuh

@@ -961,11 +961,11 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
     return error;
   }
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(NV_IS_HOST,
                (std::stringstream ss; ss << policy_selector(arch_id);
                 _CubLog("Dispatching DeviceScan to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   struct fake_policy
   {

cub/cub/device/dispatch/dispatch_scan_by_key.cuh

@@ -641,12 +641,12 @@ struct DispatchScanByKey
       return error;
     }
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
     NV_IF_TARGET(
       NV_IS_HOST,
       (::std::stringstream ss; ss << policy_selector(arch_id); _CubLog(
          "Dispatching DeviceScanByKey to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
     const detail::scan_by_key::scan_by_key_policy active_policy = policy_selector(arch_id);
 
@@ -733,12 +733,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
     return error;
   }
 
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (::std::stringstream ss; ss << policy_selector(arch_id);
      _CubLog("Dispatching DeviceScanByKey to arch %d with tuning: %s\n", static_cast<int>(arch_id), ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   struct fake_policy
   {

cub/cub/device/dispatch/dispatch_segmented_reduce.cuh

@@ -511,12 +511,12 @@ CUB_RUNTIME_FUNCTION _CCCL_FORCEINLINE auto dispatch(
   }
 
   const segmented_reduce_policy active_policy = policy_selector(arch_id);
-#if !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#if _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
   NV_IF_TARGET(
     NV_IS_HOST,
     (::std::stringstream ss; ss << active_policy;
      _CubLog("Dispatching DeviceSegmentedReduce to arch %d with tuning: %s\n", (int) arch_id, ss.str().c_str());))
-#endif // !_CCCL_COMPILER(NVRTC) && defined(CUB_DEBUG_LOG)
+#endif // _CCCL_HOSTED() && defined(CUB_DEBUG_LOG)
 
   // Compute segments_per_block based on max_segment_size hint
   int segments_per_block = 1;