Remove ability to build kvikio without CUDA (#829)

This change supersedes #822 and makes CUDA a requirement for building kvikio. We have no compelling use case for this support outside of [legate](https://docs.nvidia.com/legate/latest/api/cpp/generated/group/group__io-kvikio.html), and after discussion with the team they are happy to sunset this support as well.

Resolves #736 
Resolves #485

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #829

Author: vyasr

cpp/CMakeLists.txt

@@ -48,7 +48,6 @@ option(KvikIO_BUILD_BENCHMARKS "Configure CMake to build benchmarks" ON)
 option(KvikIO_BUILD_EXAMPLES "Configure CMake to build examples" ON)
 option(KvikIO_BUILD_TESTS "Configure CMake to build tests" ON)
 option(KvikIO_REMOTE_SUPPORT "Configure CMake to build with remote IO support" ON)
-option(KvikIO_CUDA_SUPPORT "Configure CMake to build with CUDA support" ON)
 
 # ##################################################################################################
 # * conda environment ------------------------------------------------------------------------------
@@ -72,66 +71,65 @@ if(KvikIO_REMOTE_SUPPORT)
   endif()
 endif()
 
-set(cuFile_FOUND 0)
-if(KvikIO_CUDA_SUPPORT)
-  rapids_find_package(
-    CUDAToolkit REQUIRED
-    BUILD_EXPORT_SET kvikio-exports
-    INSTALL_EXPORT_SET kvikio-exports
+# CUDA is now required
+rapids_find_package(
+  CUDAToolkit REQUIRED
+  BUILD_EXPORT_SET kvikio-exports
+  INSTALL_EXPORT_SET kvikio-exports
+)
+include(cmake/thirdparty/get_nvtx.cmake)
+
+if(NOT TARGET CUDA::cuFile)
+  set(cuFile_FOUND 0)
+  message(
+    WARNING "Cannot find cuFile - KvikIO will still work but won't use GPUDirect Storage (GDS)"
   )
-  include(cmake/thirdparty/get_nvtx.cmake)
-
-  if(NOT TARGET CUDA::cuFile)
-    message(
-      WARNING "Cannot find cuFile - KvikIO will still work but won't use GPUDirect Storage (GDS)"
-    )
-  else()
-    set(cuFile_FOUND 1)
-
-    # Check API support
-    try_compile(
-      cuFile_BATCH_API_FOUND SOURCE_FROM_CONTENT
-      batch.cpp
-      [[#include <cufile.h>
-      int main() {
-        cuFileBatchIOSetUp(nullptr, 0);
-        return 0;
-      }
-      ]]
-      LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
-      OUTPUT_VARIABLE batch_output
-    )
-    message(STATUS "Found cuFile Batch API: ${cuFile_BATCH_API_FOUND}")
-    try_compile(
-      cuFile_STREAM_API_FOUND SOURCE_FROM_CONTENT
-      stream.cpp
-      [[#include <cufile.h>
-      int main() {
-        CUfileHandle_t fh;
-        CUstream stream;
-        cuFileReadAsync(fh, nullptr, nullptr, nullptr, nullptr, nullptr, stream);
-        return 0;
-      }
-      ]]
-      LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
-      OUTPUT_VARIABLE stream_output
-    )
-    message(STATUS "Found cuFile Stream API: ${cuFile_STREAM_API_FOUND}")
-    try_compile(
-      cuFile_VERSION_API_FOUND SOURCE_FROM_CONTENT
-      version.cpp
-      [[#include <cufile.h>
-      int main() {
-        int version;
-        cuFileGetVersion(&version);
-        return 0;
-      }
-      ]]
-      LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
-      OUTPUT_VARIABLE version_output
-    )
-    message(STATUS "Found cuFile Version API: ${cuFile_VERSION_API_FOUND}")
-  endif()
+else()
+  set(cuFile_FOUND 1)
+
+  # Check API support
+  try_compile(
+    cuFile_BATCH_API_FOUND SOURCE_FROM_CONTENT
+    batch.cpp
+    [[#include <cufile.h>
+    int main() {
+      cuFileBatchIOSetUp(nullptr, 0);
+      return 0;
+    }
+    ]]
+    LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
+    OUTPUT_VARIABLE batch_output
+  )
+  message(STATUS "Found cuFile Batch API: ${cuFile_BATCH_API_FOUND}")
+  try_compile(
+    cuFile_STREAM_API_FOUND SOURCE_FROM_CONTENT
+    stream.cpp
+    [[#include <cufile.h>
+    int main() {
+      CUfileHandle_t fh;
+      CUstream stream;
+      cuFileReadAsync(fh, nullptr, nullptr, nullptr, nullptr, nullptr, stream);
+      return 0;
+    }
+    ]]
+    LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
+    OUTPUT_VARIABLE stream_output
+  )
+  message(STATUS "Found cuFile Stream API: ${cuFile_STREAM_API_FOUND}")
+  try_compile(
+    cuFile_VERSION_API_FOUND SOURCE_FROM_CONTENT
+    version.cpp
+    [[#include <cufile.h>
+    int main() {
+      int version;
+      cuFileGetVersion(&version);
+      return 0;
+    }
+    ]]
+    LINK_LIBRARIES CUDA::cuFile rt ${CMAKE_DL_LIBS}
+    OUTPUT_VARIABLE version_output
+  )
+  message(STATUS "Found cuFile Version API: ${cuFile_VERSION_API_FOUND}")
 endif()
 
 include(cmake/thirdparty/get_thread_pool.cmake)
@@ -181,7 +179,7 @@ add_library(kvikio::kvikio ALIAS kvikio)
 target_include_directories(
   kvikio
   PUBLIC "$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>"
-         "$<BUILD_INTERFACE:$<$<BOOL:${KvikIO_CUDA_SUPPORT}>:${CUDAToolkit_INCLUDE_DIRS}>>"
+         "$<BUILD_INTERFACE:${CUDAToolkit_INCLUDE_DIRS}>"
   INTERFACE "$<INSTALL_INTERFACE:include>"
 )
 
@@ -195,7 +193,6 @@ target_link_libraries(
 target_compile_definitions(
   kvikio
   PUBLIC $<$<BOOL:${KvikIO_REMOTE_SUPPORT}>:KVIKIO_LIBCURL_FOUND>
-         $<$<BOOL:${KvikIO_CUDA_SUPPORT}>:KVIKIO_CUDA_FOUND>
          $<$<BOOL:${cuFile_FOUND}>:KVIKIO_CUFILE_FOUND>
          $<$<BOOL:${cuFile_BATCH_API_FOUND}>:KVIKIO_CUFILE_BATCH_API_FOUND>
          $<$<BOOL:${cuFile_STREAM_API_FOUND}>:KVIKIO_CUFILE_STREAM_API_FOUND>
@@ -231,10 +228,7 @@ if(KvikIO_BUILD_EXAMPLES)
   add_subdirectory(examples)
 endif()
 
-if(CUDAToolkit_FOUND
-   AND KvikIO_BUILD_TESTS
-   AND CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME
-)
+if(KvikIO_BUILD_TESTS AND CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
   include(cmake/thirdparty/get_gtest.cmake)
 
   # include CTest module -- automatically calls enable_testing()
@@ -271,7 +265,6 @@ Provide targets for KvikIO.
 
 set(final_code_string
     "
-set(KvikIO_CUDA_SUPPORT [=[${KvikIO_CUDA_SUPPORT}]=])
 set(KvikIO_CUFILE_SUPPORT [=[${cuFile_FOUND}]=])
 set(KvikIO_REMOTE_SUPPORT [=[${KvikIO_REMOTE_SUPPORT}]=])
 "
@@ -280,13 +273,11 @@ string(
   APPEND
   final_code_string
   [=[
-if(KvikIO_CUDA_SUPPORT)
-  find_package(CUDAToolkit REQUIRED QUIET)
-  target_include_directories(kvikio::kvikio INTERFACE ${CUDAToolkit_INCLUDE_DIRS})
+find_package(CUDAToolkit REQUIRED QUIET)
+target_include_directories(kvikio::kvikio INTERFACE ${CUDAToolkit_INCLUDE_DIRS})
 
-  if(KvikIO_CUFILE_SUPPORT AND NOT TARGET CUDA::cuFile)
-    message(FATAL_ERROR "Compiled with cuFile support but cuFile not found")
-  endif()
+if(KvikIO_CUFILE_SUPPORT AND NOT TARGET CUDA::cuFile)
+  message(FATAL_ERROR "Compiled with cuFile support but cuFile not found")
 endif()
 ]=]
 )

cpp/include/kvikio/nvtx.hpp

@@ -17,16 +17,13 @@
 
 #include <cstdint>
 
-#ifdef KVIKIO_CUDA_FOUND
 #include <nvtx3/nvtx3.hpp>
-#endif
 
 #include <kvikio/shim/cuda.hpp>
 #include <kvikio/utils.hpp>
 
 namespace kvikio {
 
-#ifdef KVIKIO_CUDA_FOUND
 /**
  * @brief Tag type for libkvikio's NVTX domain.
  */
@@ -100,13 +97,7 @@ using nvtx_registered_string_type = nvtx3::registered_string_in<libkvikio_domain
   nvtx3::mark_in<kvikio::libkvikio_domain>(nvtx3::event_attributes{ \
     KVIKIO_REGISTER_STRING(message), nvtx3::payload{kvikio::convert_to_64bit(payload_v)}})
 
-#endif
-
-#ifdef KVIKIO_CUDA_FOUND
 using nvtx_color_type = nvtx3::color;
-#else
-using nvtx_color_type = int;
-#endif
 
 /**
  * @brief Utility singleton class for NVTX annotation.
@@ -181,13 +172,7 @@ class NvtxManager {
  * }
  * ```
  */
-#ifdef KVIKIO_CUDA_FOUND
 #define KVIKIO_NVTX_FUNC_RANGE(...) KVIKIO_NVTX_FUNC_RANGE_IMPL(__VA_ARGS__)
-#else
-#define KVIKIO_NVTX_FUNC_RANGE(...) \
-  do {                              \
-  } while (0)
-#endif
 
 /**
  * @brief Convenience macro for generating an NVTX scoped range in the `libkvikio` domain to
@@ -206,13 +191,7 @@ class NvtxManager {
  * }
  * ```
  */
-#ifdef KVIKIO_CUDA_FOUND
 #define KVIKIO_NVTX_SCOPED_RANGE(...) KVIKIO_NVTX_SCOPED_RANGE_IMPL(__VA_ARGS__)
-#else
-#define KVIKIO_NVTX_SCOPED_RANGE(message, payload, ...) \
-  do {                                                  \
-  } while (0)
-#endif
 
 /**
  * @brief Convenience macro for generating an NVTX marker in the `libkvikio` domain to annotate a
@@ -232,12 +211,6 @@ class NvtxManager {
  * }
  * ```
  */
-#ifdef KVIKIO_CUDA_FOUND
 #define KVIKIO_NVTX_MARKER(message, payload) KVIKIO_NVTX_MARKER_IMPL(message, payload)
-#else
-#define KVIKIO_NVTX_MARKER(message, payload) \
-  do {                                       \
-  } while (0)
-#endif
 
 }  // namespace kvikio

cpp/include/kvikio/shim/cuda.hpp

@@ -18,7 +18,7 @@
 #include <any>
 #include <functional>
 
-#include <kvikio/shim/cuda_h_wrapper.hpp>
+#include <cuda.h>
 #include <kvikio/shim/utils.hpp>
 #include <stdexcept>
 
@@ -134,10 +134,6 @@ class cudaAPI {
  *
  * @return The boolean answer
  */
-#ifdef KVIKIO_CUDA_FOUND
 bool is_cuda_available();
-#else
-constexpr bool is_cuda_available() { return false; }
-#endif
 
 }  // namespace kvikio

cpp/include/kvikio/shim/cuda_h_wrapper.hpp

@@ -17,7 +17,7 @@
 
 #include <sys/types.h>
 
-#include <kvikio/shim/cuda_h_wrapper.hpp>
+#include <cuda.h>
 
 /**
  * In order to support compilation when `cufile.h` isn't available, we

cpp/include/kvikio/shim/cufile_h_wrapper.hpp

@@ -73,11 +73,7 @@ template <typename T, std::enable_if_t<std::is_floating_point_v<T>>* = nullptr>
  * @param ptr Memory pointer to query
  * @return The boolean answer
  */
-#ifdef KVIKIO_CUDA_FOUND
 bool is_host_memory(void const* ptr);
-#else
-constexpr bool is_host_memory(void const* ptr) { return true; }
-#endif
 
 /**
  * @brief Return the device owning the pointer