CUDAToolkit: No targets now depend on the CUDA runtime

It is not a requirement to have shared|static consistent across your
CUDA libraries (e.g curand, nppc ) and your CUDA runtime library.
It is entirely allowable to use a static nppc and a shared runtime.

Modules/FindCUDAToolkit.cmake

@@ -122,7 +122,6 @@ CUDA Runtime Library
 
 The CUDA Runtime library (cudart) are what most applications will typically
 need to link against to make any calls such as `cudaMalloc`, and `cudaFree`.
-They are an explicit dependency of almost every library.
 
 Targets Created:
 
@@ -729,10 +728,8 @@ if(CUDAToolkit_FOUND)
 
   foreach (cuda_lib cublas cufft cufftw curand cusolver cusparse nvgraph nvjpeg)
     find_and_add_cuda_import_lib(${cuda_lib})
-    add_cuda_link_dependency(${cuda_lib} cudart)
 
     find_and_add_cuda_import_lib(${cuda_lib}_static)
-    add_cuda_link_dependency(${cuda_lib}_static cudart_static)
   endforeach()
 
   # cuSOLVER depends on cuBLAS, and cuSPARSE
@@ -772,8 +769,6 @@ if(CUDAToolkit_FOUND)
   endif()
   find_and_add_cuda_import_lib(nvToolsExt nvToolsExt nvToolsExt64)
 
-  add_cuda_link_dependency(nvToolsExt cudart)
-
   find_and_add_cuda_import_lib(OpenCL)
 
   find_and_add_cuda_import_lib(culibos)

Tests/Cuda/CMakeLists.txt

@@ -13,4 +13,12 @@ ADD_TEST_MACRO(Cuda.Toolkit Toolkit)
 ADD_TEST_MACRO(Cuda.IncludePathNoToolkit IncludePathNoToolkit)
 ADD_TEST_MACRO(Cuda.ProperDeviceLibraries ProperDeviceLibraries)
 ADD_TEST_MACRO(Cuda.ProperLinkFlags ProperLinkFlags)
+ADD_TEST_MACRO(Cuda.SharedRuntimePlusToolkit SharedRuntimePlusToolkit)
+
+# The CUDA only ships the shared version of the toolkit libraries
+# on windows
+if(NOT WIN32)
+  ADD_TEST_MACRO(Cuda.StaticRuntimePlusToolkit StaticRuntimePlusToolkit)
+endif()
+
 ADD_TEST_MACRO(Cuda.WithC CudaWithC)

Tests/Cuda/SharedRuntimePlusToolkit/CMakeLists.txt

@@ -0,0 +1,35 @@
+cmake_minimum_required(VERSION 3.15)
+project(SharedRuntimePlusToolkit CXX)
+
+#Goal for this example:
+# Validate that with c++ we can use some components of the CUDA toolkit, and
+# specify the cuda runtime
+find_package(CUDAToolkit REQUIRED)
+
+add_library(Common OBJECT curand.cpp nppif.cpp)
+target_link_libraries(Common PRIVATE CUDA::toolkit)
+set_target_properties(Common PROPERTIES POSITION_INDEPENDENT_CODE ON)
+
+#shared runtime with shared toolkit libraries
+add_library(SharedToolkit SHARED shared.cpp)
+target_link_libraries(SharedToolkit PRIVATE Common PUBLIC CUDA::curand CUDA::nppif)
+target_link_libraries(SharedToolkit PUBLIC CUDA::cudart)
+
+# The CUDA only ships the shared version of the toolkit libraries
+# on windows
+if(NOT WIN32)
+  #shared runtime with static toolkit libraries
+  add_library(StaticToolkit SHARED static.cpp)
+  target_link_libraries(StaticToolkit PRIVATE Common CUDA::curand_static CUDA::nppif_static)
+  target_link_libraries(StaticToolkit PUBLIC CUDA::cudart)
+
+  #static runtime with mixed toolkit libraries
+  add_library(MixedToolkit SHARED mixed.cpp)
+  target_link_libraries(MixedToolkit PRIVATE Common CUDA::curand_static CUDA::nppif)
+  target_link_libraries(MixedToolkit PUBLIC CUDA::cudart)
+endif()
+
+add_executable(SharedRuntimePlusToolkit main.cpp)
+target_link_libraries(SharedRuntimePlusToolkit PRIVATE SharedToolkit
+                      $<TARGET_NAME_IF_EXISTS:StaticToolkit>
+                      $<TARGET_NAME_IF_EXISTS:MixedToolkit>)

Tests/Cuda/SharedRuntimePlusToolkit/curand.cpp

@@ -0,0 +1,65 @@
+// Comes from:
+// https://docs.nvidia.com/cuda/curand/host-api-overview.html#host-api-example
+
+#ifdef _WIN32
+#  define EXPORT __declspec(dllexport)
+#else
+#  define EXPORT
+#endif
+
+/*
+ * This program uses the host CURAND API to generate 100
+ * pseudorandom floats.
+ */
+#include <cuda.h>
+#include <curand.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#define CUDA_CALL(x)                                                          \
+  do {                                                                        \
+    if ((x) != cudaSuccess) {                                                 \
+      printf("Error at %s:%d\n", __FILE__, __LINE__);                         \
+      return EXIT_FAILURE;                                                    \
+    }                                                                         \
+  } while (0)
+#define CURAND_CALL(x)                                                        \
+  do {                                                                        \
+    if ((x) != CURAND_STATUS_SUCCESS) {                                       \
+      printf("Error at %s:%d\n", __FILE__, __LINE__);                         \
+      return EXIT_FAILURE;                                                    \
+    }                                                                         \
+  } while (0)
+
+EXPORT int curand_main()
+{
+  size_t n = 100;
+  size_t i;
+  curandGenerator_t gen;
+  float *devData, *hostData;
+
+  /* Allocate n floats on host */
+  hostData = (float*)calloc(n, sizeof(float));
+
+  /* Allocate n floats on device */
+  CUDA_CALL(cudaMalloc((void**)&devData, n * sizeof(float)));
+
+  /* Create pseudo-random number generator */
+  CURAND_CALL(curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT));
+
+  /* Set seed */
+  CURAND_CALL(curandSetPseudoRandomGeneratorSeed(gen, 1234ULL));
+
+  /* Generate n floats on device */
+  CURAND_CALL(curandGenerateUniform(gen, devData, n));
+
+  /* Copy device memory to host */
+  CUDA_CALL(
+    cudaMemcpy(hostData, devData, n * sizeof(float), cudaMemcpyDeviceToHost));
+
+  /* Cleanup */
+  CURAND_CALL(curandDestroyGenerator(gen));
+  CUDA_CALL(cudaFree(devData));
+  free(hostData);
+  return EXIT_SUCCESS;
+}

Tests/Cuda/SharedRuntimePlusToolkit/main.cpp

@@ -0,0 +1,23 @@
+
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+IMPORT int shared_version();
+int static_version()
+{
+  return 0;
+}
+int mixed_version()
+{
+  return 0;
+}
+#else
+int shared_version();
+int static_version();
+int mixed_version();
+#endif
+
+int main()
+{
+  return mixed_version() == 0 && shared_version() == 0 &&
+    static_version() == 0;
+}

Tests/Cuda/SharedRuntimePlusToolkit/mixed.cpp

@@ -0,0 +1,16 @@
+
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
+
+IMPORT int curand_main();
+IMPORT int nppif_main();
+
+EXPORT int mixed_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}

Tests/Cuda/SharedRuntimePlusToolkit/nppif.cpp

@@ -0,0 +1,92 @@
+// Comes from
+// https://devtalk.nvidia.com/default/topic/1037482/gpu-accelerated-libraries/help-me-help-you-with-modern-cmake-and-cuda-mwe-for-npp/post/5271066/#5271066
+
+#ifdef _WIN32
+#  define EXPORT __declspec(dllexport)
+#else
+#  define EXPORT
+#endif
+
+#include <cstdio>
+#include <iostream>
+
+#include <assert.h>
+#include <cuda_runtime_api.h>
+#include <nppi_filtering_functions.h>
+
+EXPORT int nppif_main()
+{
+  /**
+   * 8-bit unsigned single-channel 1D row convolution.
+   */
+  const int simgrows = 32;
+  const int simgcols = 32;
+  Npp8u *d_pSrc, *d_pDst;
+  const int nMaskSize = 3;
+  NppiSize oROI;
+  oROI.width = simgcols - nMaskSize;
+  oROI.height = simgrows;
+  const int simgsize = simgrows * simgcols * sizeof(d_pSrc[0]);
+  const int dimgsize = oROI.width * oROI.height * sizeof(d_pSrc[0]);
+  const int simgpix = simgrows * simgcols;
+  const int dimgpix = oROI.width * oROI.height;
+  const int nSrcStep = simgcols * sizeof(d_pSrc[0]);
+  const int nDstStep = oROI.width * sizeof(d_pDst[0]);
+  const int pixval = 1;
+  const int nDivisor = 1;
+  const Npp32s h_pKernel[nMaskSize] = { pixval, pixval, pixval };
+  Npp32s* d_pKernel;
+  const Npp32s nAnchor = 2;
+  cudaError_t err = cudaMalloc((void**)&d_pSrc, simgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMalloc((void**)&d_pDst, dimgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMalloc((void**)&d_pKernel, nMaskSize * sizeof(d_pKernel[0]));
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // set image to pixval initially
+  err = cudaMemset(d_pSrc, pixval, simgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMemset(d_pDst, 0, dimgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMemcpy(d_pKernel, h_pKernel, nMaskSize * sizeof(d_pKernel[0]),
+                   cudaMemcpyHostToDevice);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // copy src to dst
+  NppStatus ret =
+    nppiFilterRow_8u_C1R(d_pSrc, nSrcStep, d_pDst, nDstStep, oROI, d_pKernel,
+                         nMaskSize, nAnchor, nDivisor);
+  assert(ret == NPP_NO_ERROR);
+  Npp8u* h_imgres = new Npp8u[dimgpix];
+  err = cudaMemcpy(h_imgres, d_pDst, dimgsize, cudaMemcpyDeviceToHost);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // test for filtering
+  for (int i = 0; i < dimgpix; i++) {
+    if (h_imgres[i] != (pixval * pixval * nMaskSize)) {
+      fprintf(stderr, "h_imgres at index %d failed to match\n", i);
+      return 1;
+    }
+  }
+
+  return 0;
+}

Tests/Cuda/SharedRuntimePlusToolkit/shared.cpp

@@ -0,0 +1,16 @@
+
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
+
+int curand_main();
+int nppif_main();
+
+EXPORT int shared_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}

Tests/Cuda/SharedRuntimePlusToolkit/static.cpp

@@ -0,0 +1,16 @@
+
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
+
+IMPORT int curand_main();
+IMPORT int nppif_main();
+
+EXPORT int static_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}

Tests/Cuda/StaticRuntimePlusToolkit/CMakeLists.txt

@@ -0,0 +1,29 @@
+cmake_minimum_required(VERSION 3.15)
+project(StaticRuntimePlusToolkit CXX)
+
+#Goal for this example:
+# Validate that with c++ we can use some components of the CUDA toolkit, and
+# specify the cuda runtime
+find_package(CUDAToolkit REQUIRED)
+
+add_library(Common OBJECT curand.cpp nppif.cpp)
+target_link_libraries(Common PRIVATE CUDA::toolkit)
+set_target_properties(Common PROPERTIES POSITION_INDEPENDENT_CODE ON)
+
+#static runtime with shared toolkit libraries
+add_library(SharedToolkit SHARED shared.cpp)
+target_link_libraries(SharedToolkit PRIVATE Common PUBLIC CUDA::curand CUDA::nppif)
+target_link_libraries(SharedToolkit PUBLIC CUDA::cudart_static)
+
+#static runtime with static toolkit libraries
+add_library(StaticToolkit SHARED static.cpp)
+target_link_libraries(StaticToolkit PRIVATE Common CUDA::curand_static CUDA::nppif_static)
+target_link_libraries(StaticToolkit PUBLIC CUDA::cudart_static)
+
+#static runtime with mixed toolkit libraries
+add_library(MixedToolkit SHARED mixed.cpp)
+target_link_libraries(MixedToolkit PRIVATE Common CUDA::curand CUDA::nppif_static)
+target_link_libraries(MixedToolkit PUBLIC CUDA::cudart_static)
+
+add_executable(StaticRuntimePlusToolkit main.cpp)
+target_link_libraries(StaticRuntimePlusToolkit PRIVATE SharedToolkit StaticToolkit MixedToolkit)

Tests/Cuda/StaticRuntimePlusToolkit/curand.cpp

@@ -0,0 +1,59 @@
+// Comes from:
+// https://docs.nvidia.com/cuda/curand/host-api-overview.html#host-api-example
+
+/*
+ * This program uses the host CURAND API to generate 100
+ * pseudorandom floats.
+ */
+#include <cuda.h>
+#include <curand.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#define CUDA_CALL(x)                                                          \
+  do {                                                                        \
+    if ((x) != cudaSuccess) {                                                 \
+      printf("Error at %s:%d\n", __FILE__, __LINE__);                         \
+      return EXIT_FAILURE;                                                    \
+    }                                                                         \
+  } while (0)
+#define CURAND_CALL(x)                                                        \
+  do {                                                                        \
+    if ((x) != CURAND_STATUS_SUCCESS) {                                       \
+      printf("Error at %s:%d\n", __FILE__, __LINE__);                         \
+      return EXIT_FAILURE;                                                    \
+    }                                                                         \
+  } while (0)
+
+int curand_main()
+{
+  size_t n = 100;
+  size_t i;
+  curandGenerator_t gen;
+  float *devData, *hostData;
+
+  /* Allocate n floats on host */
+  hostData = (float*)calloc(n, sizeof(float));
+
+  /* Allocate n floats on device */
+  CUDA_CALL(cudaMalloc((void**)&devData, n * sizeof(float)));
+
+  /* Create pseudo-random number generator */
+  CURAND_CALL(curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT));
+
+  /* Set seed */
+  CURAND_CALL(curandSetPseudoRandomGeneratorSeed(gen, 1234ULL));
+
+  /* Generate n floats on device */
+  CURAND_CALL(curandGenerateUniform(gen, devData, n));
+
+  /* Copy device memory to host */
+  CUDA_CALL(
+    cudaMemcpy(hostData, devData, n * sizeof(float), cudaMemcpyDeviceToHost));
+
+  /* Cleanup */
+  CURAND_CALL(curandDestroyGenerator(gen));
+  CUDA_CALL(cudaFree(devData));
+  free(hostData);
+  return EXIT_SUCCESS;
+}

Tests/Cuda/StaticRuntimePlusToolkit/main.cpp

@@ -0,0 +1,11 @@
+
+
+int shared_version();
+int static_version();
+int mixed_version();
+
+int main()
+{
+  return mixed_version() == 0 && shared_version() == 0 &&
+    static_version() == 0;
+}

Tests/Cuda/StaticRuntimePlusToolkit/mixed.cpp

@@ -0,0 +1,8 @@
+
+int curand_main();
+int nppif_main();
+
+int mixed_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}

Tests/Cuda/StaticRuntimePlusToolkit/nppif.cpp

@@ -0,0 +1,86 @@
+// Comes from
+// https://devtalk.nvidia.com/default/topic/1037482/gpu-accelerated-libraries/help-me-help-you-with-modern-cmake-and-cuda-mwe-for-npp/post/5271066/#5271066
+
+#include <cstdio>
+#include <iostream>
+
+#include <assert.h>
+#include <cuda_runtime_api.h>
+#include <nppi_filtering_functions.h>
+
+int nppif_main()
+{
+  /**
+   * 8-bit unsigned single-channel 1D row convolution.
+   */
+  const int simgrows = 32;
+  const int simgcols = 32;
+  Npp8u *d_pSrc, *d_pDst;
+  const int nMaskSize = 3;
+  NppiSize oROI;
+  oROI.width = simgcols - nMaskSize;
+  oROI.height = simgrows;
+  const int simgsize = simgrows * simgcols * sizeof(d_pSrc[0]);
+  const int dimgsize = oROI.width * oROI.height * sizeof(d_pSrc[0]);
+  const int simgpix = simgrows * simgcols;
+  const int dimgpix = oROI.width * oROI.height;
+  const int nSrcStep = simgcols * sizeof(d_pSrc[0]);
+  const int nDstStep = oROI.width * sizeof(d_pDst[0]);
+  const int pixval = 1;
+  const int nDivisor = 1;
+  const Npp32s h_pKernel[nMaskSize] = { pixval, pixval, pixval };
+  Npp32s* d_pKernel;
+  const Npp32s nAnchor = 2;
+  cudaError_t err = cudaMalloc((void**)&d_pSrc, simgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMalloc((void**)&d_pDst, dimgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMalloc((void**)&d_pKernel, nMaskSize * sizeof(d_pKernel[0]));
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // set image to pixval initially
+  err = cudaMemset(d_pSrc, pixval, simgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMemset(d_pDst, 0, dimgsize);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  err = cudaMemcpy(d_pKernel, h_pKernel, nMaskSize * sizeof(d_pKernel[0]),
+                   cudaMemcpyHostToDevice);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // copy src to dst
+  NppStatus ret =
+    nppiFilterRow_8u_C1R(d_pSrc, nSrcStep, d_pDst, nDstStep, oROI, d_pKernel,
+                         nMaskSize, nAnchor, nDivisor);
+  assert(ret == NPP_NO_ERROR);
+  Npp8u* h_imgres = new Npp8u[dimgpix];
+  err = cudaMemcpy(h_imgres, d_pDst, dimgsize, cudaMemcpyDeviceToHost);
+  if (err != cudaSuccess) {
+    fprintf(stderr, "Cuda error %d\n", __LINE__);
+    return 1;
+  }
+  // test for filtering
+  for (int i = 0; i < dimgpix; i++) {
+    if (h_imgres[i] != (pixval * pixval * nMaskSize)) {
+      fprintf(stderr, "h_imgres at index %d failed to match\n", i);
+      return 1;
+    }
+  }
+
+  return 0;
+}

Tests/Cuda/StaticRuntimePlusToolkit/shared.cpp

@@ -0,0 +1,8 @@
+
+int curand_main();
+int nppif_main();
+
+int shared_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}

Tests/Cuda/StaticRuntimePlusToolkit/static.cpp

@@ -0,0 +1,8 @@
+
+int curand_main();
+int nppif_main();
+
+int static_version()
+{
+  return curand_main() == 0 && nppif_main() == 0;
+}