#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <omp.h>
#include <cublas_v2.h>
#include <cuda_runtime.h>
#include <assert.h>
#include "gpu.h"

gpu_data_sp* gpu_init_sp(
   int nO, int nV, int cholesky_mo_num,
   double* cc_space_v_oo_chol_, double* cc_space_v_ov_chol_,
   double* cc_space_v_vo_chol_, double* cc_space_v_vv_chol_,
   double* cc_space_v_oooo_, double* cc_space_v_vooo_, double* cc_space_v_voov_,
   double* cc_space_v_oovv_, double* cc_space_v_vvoo_,
   double* cc_space_v_oovo_, double* cc_space_v_ovvo_,
   double* cc_space_v_ovov_, double* cc_space_v_ovoo_,
   double* cc_space_f_oo_, double* cc_space_f_ov_,
   double* cc_space_f_vo_, double* cc_space_f_vv_)
{
    int ngpus = 1;
    if (MULTIGPU == 1) cudaGetDeviceCount(&ngpus);

    gpu_data_sp* data = (gpu_data_sp*) malloc ((size_t) ngpus*sizeof(gpu_data_sp));
    assert (data != NULL);

    #pragma omp parallel num_threads(ngpus)
    {
      cudaError_t cudaStat = cudaSuccess;
      size_t lda;

      int igpu = omp_get_thread_num();
      cudaSetDevice(igpu);
      cublasHandle_t handle;
      cublasCreate(&handle);

/*
      size_t limit;
      size_t mtotal, mfree;
      cudaStat = cudaMemGetInfo(&mfree, &mtotal);
      fprintf(stderr, "%s\nTotal:%lu\nFree :%lu\n", cudaGetErrorString(cudaStat), mtotal, mfree);
*/

      float* d_cc_space_v_oo_chol = NULL;
      lda = cholesky_mo_num * nO;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_oo_chol, lda * nO * sizeof(float));

      assert (cudaStat == cudaSuccess);
      float* cc_space_v_oo_chol = malloc((size_t) lda * nO * sizeof(float));
      assert (cc_space_v_oo_chol != NULL);
      for (size_t i=0 ; i<(size_t)  lda * nO ; ++i) {
        cc_space_v_oo_chol[i] = cc_space_v_oo_chol_[i];
      }
      cublasSetMatrix(cholesky_mo_num*nO, nO, sizeof(float), cc_space_v_oo_chol, lda, d_cc_space_v_oo_chol, lda);
      free(cc_space_v_oo_chol);

      float* d_cc_space_v_ov_chol = NULL;
      lda = cholesky_mo_num * nO;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_ov_chol, lda * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_ov_chol = malloc((size_t) lda * nV * sizeof(float));
      assert (cc_space_v_ov_chol != NULL);
      for (size_t i=0 ; i<(size_t)  lda * nV ; ++i) {
        cc_space_v_ov_chol[i] = cc_space_v_ov_chol_[i];
      }
      cublasSetMatrix(cholesky_mo_num*nO, nV, sizeof(float), cc_space_v_ov_chol, lda, d_cc_space_v_ov_chol, lda);
      free(cc_space_v_ov_chol);

      float* d_cc_space_v_vo_chol = NULL;
      lda = cholesky_mo_num * nV;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_vo_chol, lda * nO * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_vo_chol = malloc((size_t) lda * nO * sizeof(float));
      assert (cc_space_v_vo_chol != NULL);
      for (size_t i=0 ; i<(size_t)  lda * nO ; ++i) {
        cc_space_v_vo_chol[i] = cc_space_v_vo_chol_[i];
      }
      cublasSetMatrix(cholesky_mo_num*nV, nO, sizeof(float), cc_space_v_vo_chol, lda, d_cc_space_v_vo_chol, lda);
      free(cc_space_v_vo_chol);

      float* d_cc_space_v_vv_chol = NULL;
      lda = cholesky_mo_num * nV;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_vv_chol, lda * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_vv_chol = malloc((size_t) lda * nV * sizeof(float));
      assert (cc_space_v_vv_chol != NULL);
      for (size_t i=0 ; i< (size_t) lda * nV ; ++i) {
        cc_space_v_vv_chol[i] = cc_space_v_vv_chol_[i];
      }
      cublasSetMatrix(cholesky_mo_num*nV, nV, sizeof(float), cc_space_v_vv_chol, lda, d_cc_space_v_vv_chol, lda);
      free(cc_space_v_vv_chol);

      float* d_cc_space_v_oooo = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_v_oooo, nO*nO*nO*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_oooo = malloc((size_t) nO*nO*nO*nO*sizeof(float));
      assert (cc_space_v_oooo != NULL);
      for (size_t i=0 ; i<(size_t) nO*nO*nO*nO ; ++i) {
        cc_space_v_oooo[i] = cc_space_v_oooo_[i];
      }
      cublasSetMatrix(nO*nO, nO*nO, sizeof(float), cc_space_v_oooo, nO*nO, d_cc_space_v_oooo, nO*nO);
      free(cc_space_v_oooo);

      float* d_cc_space_v_vooo = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_v_vooo, nV*nO*nO*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_vooo = malloc((size_t) nV*nO*nO*nO*sizeof(float));
      assert (cc_space_v_vooo != NULL);
      for (size_t i=0 ; i<(size_t) nV*nO*nO*nO; ++i) {
        cc_space_v_vooo[i] = cc_space_v_vooo_[i];
      }
      cublasSetMatrix(nV*nO, nO*nO, sizeof(float), cc_space_v_vooo, nV*nO, d_cc_space_v_vooo, nV*nO);
      free(cc_space_v_vooo);

      float* d_cc_space_v_voov = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_v_voov, nV*nO*nO*nV*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_voov = malloc((size_t) nV*nO*nO*nV*sizeof(float));
      assert (cc_space_v_voov != NULL);
      for (size_t i=0 ; i<(size_t) nV*nO*nO*nV; ++i) {
        cc_space_v_voov[i] = cc_space_v_voov_[i];
      }
      cublasSetMatrix(nV*nO, nO*nV, sizeof(float), cc_space_v_voov, nV*nO, d_cc_space_v_voov, nV*nO);
      free(cc_space_v_voov);

      float* d_cc_space_v_oovv = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_v_oovv, nO*nO*nV*nV*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_oovv = malloc((size_t) nO*nO*nV*nV*sizeof(float));
      assert (cc_space_v_oovv != NULL);
      for (size_t i=0 ; i<(size_t) nO*nO*nV*nV; ++i) {
        cc_space_v_oovv[i] = cc_space_v_oovv_[i];
      }
      cublasSetMatrix(nO*nO, nV*nV, sizeof(float), cc_space_v_oovv, nO*nO, d_cc_space_v_oovv, nO*nO);
      free(cc_space_v_oovv);

      float* d_cc_space_v_vvoo = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_v_vvoo, nV*nV*nO*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_vvoo = malloc((size_t) nV*nV*nO*nO*sizeof(float));
      assert (cc_space_v_vvoo != NULL);
      for (size_t i=0 ; i<(size_t) nV*nV*nO*nO; ++i) {
        cc_space_v_vvoo[i] = cc_space_v_vvoo_[i];
      }
      cublasSetMatrix(nV*nV, nO*nO, sizeof(float), cc_space_v_vvoo, nV*nV, d_cc_space_v_vvoo, nV*nV);
      free(cc_space_v_vvoo);

      float* d_cc_space_v_oovo = NULL;
      lda = nO*nO;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_oovo, nO*nO*nV*nO * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_oovo = malloc((size_t) nO*nO*nV*nO * sizeof(float));
      assert (cc_space_v_oovo != NULL);
      for (size_t i=0 ; i<(size_t)  nO*nO*nV*nO ; ++i) {
        cc_space_v_oovo[i] = cc_space_v_oovo_[i];
      }
      cublasSetMatrix(lda, nV*nO, sizeof(float), cc_space_v_oovo, lda, d_cc_space_v_oovo, lda);
      free(cc_space_v_oovo);

      float* d_cc_space_v_ovvo = NULL;
      lda = nO*nV;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_ovvo, nO*nV*nV*nO * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_ovvo = malloc((size_t) nO*nV*nV*nO * sizeof(float));
      assert (cc_space_v_ovvo != NULL);
      for (size_t i=0 ; i<(size_t)  nO*nV*nV*nO ; ++i) {
        cc_space_v_ovvo[i] = cc_space_v_ovvo_[i];
      }
      cublasSetMatrix(lda, nV*nO, sizeof(float), cc_space_v_ovvo, lda, d_cc_space_v_ovvo, lda);
      free(cc_space_v_ovvo);

      float* d_cc_space_v_ovov = NULL;
      lda = nO*nV;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_ovov, nO*nV*nV*nO * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_ovov = malloc((size_t) nO*nV*nV*nO * sizeof(float));
      assert (cc_space_v_ovov != NULL);
      for (size_t i=0 ; i<(size_t)  nO*nV*nV*nO ; ++i) {
        cc_space_v_ovov[i] = cc_space_v_ovov_[i];
      }
      cublasSetMatrix(lda, nV*nO, sizeof(float), cc_space_v_ovov, lda, d_cc_space_v_ovov, lda);
      free(cc_space_v_ovov);

      float* d_cc_space_v_ovoo = NULL;
      lda = nO*nV;
      cudaStat = cudaMalloc((void **)&d_cc_space_v_ovoo, nO*nV*nO*nO * sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_v_ovoo = malloc((size_t) nO*nV*nO*nO * sizeof(float));
      assert (cc_space_v_ovoo != NULL);
      for (size_t i=0 ; i<(size_t)  nO*nV*nO*nO ; ++i) {
        cc_space_v_ovoo[i] = cc_space_v_ovoo_[i];
      }
      cublasSetMatrix(lda, nO*nO, sizeof(float), cc_space_v_ovoo, lda, d_cc_space_v_ovoo, lda);
      free(cc_space_v_ovoo);

      float* d_cc_space_f_oo = NULL;
      cudaStat = cudaMalloc((void**)&d_cc_space_f_oo, nO*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_f_oo = malloc((size_t) nO*nO*sizeof(float));
      assert (cc_space_f_oo != NULL);
      for (size_t i=0 ; i<(size_t) nO*nO; ++i) {
        cc_space_f_oo[i] = cc_space_f_oo_[i];
      }
      cublasSetMatrix(nO, nO, sizeof(float), cc_space_f_oo, nO, d_cc_space_f_oo, nO);
      free(cc_space_f_oo);

      float* d_cc_space_f_vo = NULL;

      cudaStat = cudaMalloc((void**)&d_cc_space_f_vo, nV*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_f_vo = malloc((size_t) nV*nO*sizeof(float));
      assert (cc_space_f_vo != NULL);
      for (size_t i=0 ; i<(size_t) nV*nO; ++i) {
        cc_space_f_vo[i] = cc_space_f_vo_[i];
      }
      cublasSetMatrix(nV, nO, sizeof(float), cc_space_f_vo, nV, d_cc_space_f_vo, nV);
      free(cc_space_f_vo);

      float* d_cc_space_f_ov = NULL;

      cudaStat = cudaMalloc((void**)&d_cc_space_f_ov, nV*nO*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_f_ov = malloc((size_t) nV*nO*sizeof(float));
      assert (cc_space_f_ov != NULL);
      for (size_t i=0 ; i<(size_t) nV*nO; ++i) {
        cc_space_f_ov[i] = cc_space_f_ov_[i];
      }
      cublasSetMatrix(nO, nV, sizeof(float), cc_space_f_ov, nO, d_cc_space_f_ov, nO);
      free(cc_space_f_ov);

      float* d_cc_space_f_vv = NULL;

      cudaStat = cudaMalloc((void**)&d_cc_space_f_vv, nV*nV*sizeof(float));
      assert (cudaStat == cudaSuccess);
      float* cc_space_f_vv = malloc((size_t) nV*nV*sizeof(float));
      assert (cc_space_f_vv != NULL);
      for (size_t i=0 ; i<(size_t) nV*nV; ++i) {
        cc_space_f_vv[i] = cc_space_f_vv_[i];
      }
      cublasSetMatrix(nV, nV, sizeof(float), cc_space_f_vv, nV, d_cc_space_f_vv, nV);
      free(cc_space_f_vv);

      float* d_tau = NULL;
      lda = nO * nO;
      cudaStat = cudaMalloc((void **)&d_tau, lda * nV * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_tau_x = NULL;
      lda = nO * nO;
      cudaStat = cudaMalloc((void **)&d_tau_x, lda * nV * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_t1 = NULL;
      cudaStat = cudaMalloc((void **)&d_t1, nO * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_t2 = NULL;
      cudaStat = cudaMalloc((void **)&d_t2, nO*nO*nV*nV * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_H_oo = NULL;
      cudaStat = cudaMalloc((void **)&d_H_oo, nO * nO * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_H_vo = NULL;
      cudaStat = cudaMalloc((void **)&d_H_vo, nV * nO * sizeof(float));
      assert (cudaStat == cudaSuccess);

      float* d_H_vv = NULL;
      cudaStat = cudaMalloc((void **)&d_H_vv, nV * nV * sizeof(float));
      assert (cudaStat == cudaSuccess);

      data[igpu].cc_space_v_oo_chol = d_cc_space_v_oo_chol;
      data[igpu].cc_space_v_ov_chol = d_cc_space_v_ov_chol;
      data[igpu].cc_space_v_vo_chol = d_cc_space_v_vo_chol;
      data[igpu].cc_space_v_vv_chol = d_cc_space_v_vv_chol;
      data[igpu].cc_space_v_oooo = d_cc_space_v_oooo;
      data[igpu].cc_space_v_vooo = d_cc_space_v_vooo;
      data[igpu].cc_space_v_voov = d_cc_space_v_voov;
      data[igpu].cc_space_v_oovv = d_cc_space_v_oovv;
      data[igpu].cc_space_v_vvoo = d_cc_space_v_vvoo;
      data[igpu].cc_space_v_oovo = d_cc_space_v_oovo;
      data[igpu].cc_space_v_ovvo = d_cc_space_v_ovvo;
      data[igpu].cc_space_v_ovov = d_cc_space_v_ovov;
      data[igpu].cc_space_v_ovoo = d_cc_space_v_ovoo;
      data[igpu].cc_space_f_oo   = d_cc_space_f_oo;
      data[igpu].cc_space_f_ov   = d_cc_space_f_ov;
      data[igpu].cc_space_f_vo   = d_cc_space_f_vo;
      data[igpu].cc_space_f_vv   = d_cc_space_f_vv;
      data[igpu].tau   = d_tau;
      data[igpu].tau_x = d_tau_x;
      data[igpu].t1    = d_t1;
      data[igpu].t2    = d_t2;
      data[igpu].H_oo  = d_H_oo;
      data[igpu].H_vo  = d_H_vo;
      data[igpu].H_vv  = d_H_vv;

      data[igpu].nO = nO;
      data[igpu].nV = nV;
      data[igpu].cholesky_mo_num = cholesky_mo_num;

    }
    return data;
}


void gpu_deinit_sp(gpu_data_sp* data)
{
    int ngpus = 1;
    if (MULTIGPU == 1) cudaGetDeviceCount(&ngpus);

    #pragma omp parallel num_threads(ngpus)
    {
      size_t lda;
      int igpu = omp_get_thread_num();
      cudaSetDevice(igpu);

      free(data[igpu].cc_space_v_oo_chol);
      free(data[igpu].cc_space_v_ov_chol);
      free(data[igpu].cc_space_v_vo_chol);
      free(data[igpu].cc_space_v_vv_chol);
      free(data[igpu].cc_space_v_oooo);
      free(data[igpu].cc_space_v_vooo);
      free(data[igpu].cc_space_v_voov);
      free(data[igpu].cc_space_v_oovv);
      free(data[igpu].cc_space_v_vvoo);
      free(data[igpu].cc_space_v_oovo);
      free(data[igpu].cc_space_v_ovvo);
      free(data[igpu].cc_space_v_ovov);
      free(data[igpu].cc_space_v_ovoo);
      free(data[igpu].cc_space_f_oo);
      free(data[igpu].cc_space_f_ov);
      free(data[igpu].cc_space_f_vo);
      free(data[igpu].cc_space_f_vv);
      free(data[igpu].tau);
      free(data[igpu].tau_x);
      free(data[igpu].t1);
      free(data[igpu].t2);
      free(data[igpu].H_oo);
      free(data[igpu].H_vo);
      free(data[igpu].H_vv);
    }
}