init_cuda.cu

Go to the documentation of this file.

/*
 * Copyright (C) 2010-2022 The ESPResSo project
 *
 * This file is part of ESPResSo.
 *
 * ESPResSo is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * ESPResSo is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
#include "init.hpp"
#include "utils.cuh"
 
#include <cuda.h>
#include <cuda_runtime.h>
 
#include <cstring>
#include <memory>
#include <string>
 
#if defined(OMPI_MPI_H) || defined(_MPI_H)
#error CU-file includes mpi.h! This should not happen!
#endif
 
#ifdef ESPRESSO_CUDA
 
/** \name minimally required compute capability. */
/**@{*/
static int constexpr computeCapabilityMinMajor = 3;
static int constexpr computeCapabilityMinMinor = 0;
/**@}*/
 
void cuda_init() { CUDA_CHECK(cudaStreamCreate(&stream[0])) }
 
int cuda_get_n_gpus() {
  int deviceCount;
  CUDA_CHECK(cudaGetDeviceCount(&deviceCount))
  return deviceCount;
}
 
bool cuda_check_gpu_compute_capability(int dev) {
  cudaDeviceProp deviceProp;
  CUDA_CHECK(cudaGetDeviceProperties(&deviceProp, dev))
  return (deviceProp.major < computeCapabilityMinMajor or
          (deviceProp.major == computeCapabilityMinMajor and
           deviceProp.minor < computeCapabilityMinMinor));
}
 
std::string cuda_get_gpu_name(int dev) {
  cudaDeviceProp deviceProp;
  CUDA_CHECK(cudaGetDeviceProperties(&deviceProp, dev))
  return {deviceProp.name};
}
 
EspressoGpuDevice cuda_get_device_props(const int dev) {
  cudaDeviceProp deviceProp;
  CUDA_CHECK(cudaGetDeviceProperties(&deviceProp, dev))
  auto const [node, hostname] = detail::get_node_info();
  return {dev,
          deviceProp.name,
          hostname,
          node,
          deviceProp.major,
          deviceProp.minor,
          deviceProp.totalGlobalMem,
          deviceProp.multiProcessorCount};
}
 
void cuda_set_device(int dev) {
  CUDA_CHECK(cudaSetDevice(dev))
  CUDA_CHECK(cudaStreamDestroy(stream[0]))
  CUDA_CHECK(cudaStreamCreate(&stream[0]))
}
 
int cuda_get_device() {
  int dev;
  CUDA_CHECK(cudaGetDevice(&dev))
  return dev;
}
 
bool cuda_test_device_access() {
  auto const deleter = [](int *p) { cudaFree(reinterpret_cast<void *>(p)); };
  int *ptr = nullptr;
  int h = 42;
  CUDA_CHECK(cudaMalloc(reinterpret_cast<void **>(&ptr), sizeof(int)));
  std::unique_ptr<int, decltype(deleter)> d(ptr, deleter);
  CUDA_CHECK(cudaMemcpy(d.get(), &h, sizeof(int), cudaMemcpyHostToDevice));
  h = 0;
  CUDA_CHECK(cudaMemcpy(&h, d.get(), sizeof(int), cudaMemcpyDeviceToHost));
  return h != 42;
}
 
void cuda_check_device() {
  if (cuda_get_n_gpus() == 0) {
    throw cuda_runtime_error("No GPU was found.");
  }
  auto const devID = cuda_get_device();
  auto const incompatible = cuda_check_gpu_compute_capability(devID);
  auto const communication_failure = cuda_test_device_access();
  if (incompatible or communication_failure) {
    throw cuda_runtime_error("CUDA device " + std::to_string(devID) +
                             " is not capable of running ESPResSo.");
  }
}
 
#endif /* defined(ESPRESSO_CUDA) */