#include "sortlib.cuh"
#include <algorithm>
#include <cassert>
#include <cstdio>
#include <vector>

constexpr size_t length = 2097152;
constexpr long TEST_LENGTH = 32'768;
std::vector<unsigned long long> population_vector, sample_vector,
    sample_vector_into_cuda;

#define CHECK_CUDA_ERR(x)                                                      \
  {                                                                            \
    auto ret_ = (x);                                                           \
    if (ret_ != 0) {                                                           \
      printf("%d: CUDA ERR: %d\n", __LINE__, ret_);                            \
    }                                                                          \
  }

unsigned long long max_value = 0, min_value = 0xfffffffff;

inline void store_into_vector(unsigned long long value) {
  if (max_value < value) {
    max_value = value;
  } else if (min_value > value) {
    min_value = value;
  }
  population_vector.push_back(value);
}

typedef unsigned long long key_type;

typedef unsigned long long *key_type_ptr;

constexpr long SAMPLE_LENGTH = 1024;

__device__ key_type *cudaSampleItem, *cudaPopulationItem;
__device__ bool *cdf_result;
__device__ unsigned insert_value;

__global__ void initStorage(unsigned scale, unsigned test_size) {
  CHECK_CUDA_ERR(cudaFree(cdf_result));
  CHECK_CUDA_ERR(cudaMalloc(&cdf_result, scale * test_size * sizeof(bool)));
  memset(cdf_result, 0, scale * test_size * sizeof(bool));
  insert_value = 0;
}

__global__ void initCuda(key_type *sample_item, key_type *population_item) {
  cudaSampleItem = sample_item;
  cudaPopulationItem = population_item;
  cdf_result = nullptr;
}

__global__ void kernel(unsigned long step, const double slice_size) {
  for (int i = 0; i < step; i++) {
    auto tid = step * (blockIdx.x * blockDim.x + threadIdx.x) + i;

    // printf("%d\n", tid);
    auto index = (int)(FactorySort::cdf(cudaSampleItem, length,
                                        cudaPopulationItem[tid]) /
                       slice_size);

    cdf_result[index] = true;
  }
}

__global__ void print_function() {

  // printf("%u\n", insert_value);
#ifdef TEST_BOUNDS
  printf("%u %u\n", index_min, index_max);
#endif
}

__global__ void freeStorage() {
  cudaFree(cudaSampleItem);
  cudaFree(cudaPopulationItem);
  cudaFree(cdf_result);
}

typedef std::pair<int, int> dim_pair_type;
const dim_pair_type DIM_PAIR[] = {dim_pair_type(32, 32)};

void run_kernel(size_t test_size) {
  for (auto &pair : DIM_PAIR) {
    dim3 grid_dim = pair.first, block_dim = pair.second;
    unsigned long step = test_size / (grid_dim.x * block_dim.x);
    printf("current dim: %d %d %lu\n", pair.first, pair.second, step);
    for (int scale = 2; scale <= 16; scale += 2) {
      const unsigned long split_size = test_size * scale;
      const auto slice_size = 1.0 / (double)split_size;
      printf("scale: %i ", scale);
      initStorage<<<1, 1>>>(scale, test_size);
      cudaDeviceSynchronize();
      cudaEvent_t start, stop;
      cudaEventCreate(&start);
      cudaEventCreate(&stop);
      cudaEventRecord(start, nullptr);
      kernel<<<grid_dim, block_dim>>>(step, slice_size);

      cudaDeviceSynchronize();
      cudaEventRecord(stop, nullptr);
      cudaEventSynchronize(stop);
      float time;
      cudaEventElapsedTime(&time, start, stop);
      cudaEventDestroy(start);
      cudaEventDestroy(stop);
      if (time == 0) {
        // printf("last error: %u\n", cudaGetLastError());
      }
      printf("time: %lf\n", time);

      print_function<<<1, 1>>>();
      cudaDeviceSynchronize();
    }
  }
}

int main() {

  auto test_size = TEST_LENGTH;
  FILE *file = fopen("normal_distribution.txt", "r");
  assert(file);
  for (long long i; fscanf(file, "%lld ", &i) != EOF; store_into_vector(i))
    ;
  fclose(file);

  printf("test size: %d\n", test_size);
  sample_vector = std::vector<key_type>(
      population_vector.begin(), population_vector.begin() + SAMPLE_LENGTH - 2);
  sample_vector.push_back(min_value);
  sample_vector.push_back(max_value);
  std::sort(sample_vector.begin(), sample_vector.end());

  key_type *cudaSample = nullptr, *cudaPopulation;
  cudaMalloc(&cudaSample, sizeof(key_type) * SAMPLE_LENGTH);
  assert(SAMPLE_LENGTH == sample_vector.size());
  cudaMemcpy(cudaSample, sample_vector.data(), sizeof(key_type) * SAMPLE_LENGTH,
             cudaMemcpyHostToDevice);
  cudaMalloc(&cudaPopulation, sizeof(key_type) * test_size);
  cudaMemcpy(cudaPopulation, population_vector.data() + SAMPLE_LENGTH,
             sizeof(key_type) * test_size, cudaMemcpyHostToDevice);

  // auto custom_sort = CustomSort(SAMPLE_LENGTH, sizeof(long) * 8);
  // custom_sort.testCalculation();
  // testCustomCalculation<<<1, 1>>>(SAMPLE_LENGTH);
  // cudaDeviceSynchronize();
  run_kernel(test_size);

  freeStorage<<<1, 1>>>();
  cudaDeviceSynchronize();
}