Move launch API from cudax to libcu++ (#6667)

* Move launch API from cudax to libcu++

* Review feedback and test fixes

* More fixes

* GCC7 fix

* Update libcudacxx/include/cuda/__launch/configuration.h

* Fix old GCC

---------

Co-authored-by: David Bayer <[email protected]>
(cherry picked from commit 4ab39a7)

Author: 2 people

cudax/examples/simple_p2p.cu

@@ -130,7 +130,7 @@ void test_cross_device_access_from_kernel(
   dev1_stream.wait(dev0_stream);
 
   // Kernel launch configuration
-  auto config = cudax::distribute<512>(dev0_buffer.size());
+  auto config = cuda::distribute<512>(dev0_buffer.size());
 
   // Run kernel on GPU 1, reading input from the GPU 0 buffer, writing output to the GPU 1 buffer
   printf("Run kernel on GPU%d, taking source data from GPU%d and writing to "

cudax/examples/vector_add.cu

@@ -92,7 +92,7 @@ try
 
   // Define the kernel launch parameters
   constexpr int threadsPerBlock = 256;
-  auto config                   = cudax::distribute<threadsPerBlock>(numElements);
+  auto config                   = cuda::distribute<threadsPerBlock>(numElements);
 
   // Launch the vectorAdd kernel
   printf(

cudax/include/cuda/experimental/__execution/bulk.cuh

@@ -22,6 +22,7 @@
 #endif // no system header
 
 #include <cuda/__cmath/ceil_div.h>
+#include <cuda/__launch/configuration.h>
 #include <cuda/__utility/immovable.h>
 #include <cuda/std/__concepts/arithmetic.h>
 #include <cuda/std/__concepts/same_as.h>
@@ -45,7 +46,6 @@
 #include <cuda/experimental/__execution/transform_completion_signatures.cuh>
 #include <cuda/experimental/__execution/transform_sender.cuh>
 #include <cuda/experimental/__execution/type_traits.cuh>
-#include <cuda/experimental/__launch/configuration.cuh>
 
 #include <cuda/experimental/__execution/prologue.cuh>
 
@@ -73,7 +73,7 @@ struct _CCCL_TYPE_VISIBILITY_DEFAULT __attrs_t
   {
     constexpr int __block_threads = 256;
     const int __grid_blocks       = ::cuda::ceil_div(static_cast<int>(__shape), __block_threads);
-    return experimental::make_config(block_dims<__block_threads>(), grid_dims(__grid_blocks));
+    return make_config(block_dims<__block_threads>(), grid_dims(__grid_blocks));
   }
 
   using __launch_config_t = decltype(__get_launch_config(_Shape()));

cudax/include/cuda/experimental/__execution/queries.cuh

@@ -25,6 +25,7 @@ _CCCL_SUPPRESS_DEPRECATED_PUSH
 #include <cuda/std/__memory/allocator.h>
 _CCCL_SUPPRESS_DEPRECATED_POP
 
+#include <cuda/__launch/configuration.h>
 #include <cuda/std/__execution/env.h>
 #include <cuda/std/__type_traits/enable_if.h>
 #include <cuda/std/__type_traits/is_callable.h>
@@ -39,7 +40,6 @@ _CCCL_SUPPRESS_DEPRECATED_POP
 #include <cuda/experimental/__execution/stop_token.cuh>
 #include <cuda/experimental/__execution/type_traits.cuh>
 #include <cuda/experimental/__execution/utility.cuh>
-#include <cuda/experimental/__launch/configuration.cuh>
 
 #include <cuda/experimental/__execution/prologue.cuh>
 
@@ -314,14 +314,14 @@ _CCCL_GLOBAL_CONSTANT struct get_forward_progress_guarantee_t
 } get_forward_progress_guarantee{};
 
 // By default, CUDA kernels are launched with a single thread and a single block.
-using __single_threaded_config_base_t = decltype(experimental::make_config(grid_dims<1>(), block_dims<1>()));
+using __single_threaded_config_base_t = decltype(make_config(grid_dims<1>(), block_dims<1>()));
 
 // We hide the complicated type of the default launch configuration so diagnositics are
 // easier to read.
 struct __single_threaded_config_t : __single_threaded_config_base_t
 {
   _CCCL_HOST_API constexpr __single_threaded_config_t() noexcept
-      : __single_threaded_config_base_t{experimental::make_config(grid_dims<1>(), block_dims<1>())}
+      : __single_threaded_config_base_t{make_config(grid_dims<1>(), block_dims<1>())}
   {}
 };
 

cudax/include/cuda/experimental/__execution/stream/adaptor.cuh

@@ -21,6 +21,7 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/__launch/configuration.h>
 #include <cuda/__utility/immovable.h>
 #include <cuda/std/__concepts/concept_macros.h>
 #include <cuda/std/__memory/unique_ptr.h>
@@ -36,7 +37,6 @@
 #include <cuda/experimental/__execution/utility.cuh>
 #include <cuda/experimental/__execution/variant.cuh>
 #include <cuda/experimental/__execution/visit.cuh>
-#include <cuda/experimental/__launch/configuration.cuh>
 #include <cuda/experimental/__launch/launch.cuh>
 #include <cuda/experimental/__stream/stream_ref.cuh>