qp_plugins_scemama/devel/ccsd_gpu/gpu_init.c

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

gpu_data* gpu_init(
   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;
    cudaGetDeviceCount(&ngpus);

    gpu_data* data = (gpu_data*) malloc (ngpus*sizeof(gpu_data));

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

      cublasCreate(&handle);

      double* d_cc_space_v_oo_chol;
      lda = cholesky_mo_num * nO;
      cudaMalloc((void **)&d_cc_space_v_oo_chol, lda * nO * sizeof(double));
      cublasSetMatrix(cholesky_mo_num*nO, nO, sizeof(double), cc_space_v_oo_chol, lda, d_cc_space_v_oo_chol, lda);

      double* d_cc_space_v_ov_chol;
      lda = cholesky_mo_num * nO;
      cudaMalloc((void **)&d_cc_space_v_ov_chol, lda * nV * sizeof(double));
      cublasSetMatrix(cholesky_mo_num*nO, nV, sizeof(double), cc_space_v_ov_chol, lda, d_cc_space_v_ov_chol, lda);

      double* d_cc_space_v_vo_chol;
      lda = cholesky_mo_num * nV;
      cudaMalloc((void **)&d_cc_space_v_vo_chol, lda * nO * sizeof(double));
      cublasSetMatrix(cholesky_mo_num*nV, nO, sizeof(double), cc_space_v_vo_chol, lda, d_cc_space_v_vo_chol, lda);

      double* d_cc_space_v_vv_chol;
      lda = cholesky_mo_num * nV;
      cudaMalloc((void **)&d_cc_space_v_vv_chol, lda * nV * sizeof(double));
      cublasSetMatrix(cholesky_mo_num*nV, nV, sizeof(double), cc_space_v_vv_chol, lda, d_cc_space_v_vv_chol, lda);

      double* d_cc_space_v_oooo;
      cudaMalloc((void**)&d_cc_space_v_oooo, nO*nO*nO*nO*sizeof(double));
      cublasSetMatrix(nO*nO, nO*nO, sizeof(double), cc_space_v_oooo, nO*nO, d_cc_space_v_oooo, nO*nO);

      double* d_cc_space_v_vooo;
      cudaMalloc((void**)&d_cc_space_v_vooo, nV*nO*nO*nO*sizeof(double));
      cublasSetMatrix(nV*nO, nO*nO, sizeof(double), cc_space_v_vooo, nV*nO, d_cc_space_v_vooo, nV*nO);

      double* d_cc_space_v_voov;
      cudaMalloc((void**)&d_cc_space_v_voov, nV*nO*nO*nV*sizeof(double));
      cublasSetMatrix(nV*nO, nO*nV, sizeof(double), cc_space_v_voov, nV*nO, d_cc_space_v_voov, nV*nO);

      double* d_cc_space_v_oovv;
      cudaMalloc((void**)&d_cc_space_v_oovv, nO*nO*nV*nV*sizeof(double));
      cublasSetMatrix(nO*nO, nV*nV, sizeof(double), cc_space_v_oovv, nO*nO, d_cc_space_v_oovv, nO*nO);

      double* d_cc_space_v_vvoo;
      cudaMalloc((void**)&d_cc_space_v_vvoo, nV*nV*nO*nO*sizeof(double));
      cublasSetMatrix(nV*nV, nO*nO, sizeof(double), cc_space_v_vvoo, nV*nV, d_cc_space_v_vvoo, nV*nV);

      double* d_cc_space_v_oovo;
      lda = nO*nO;
      cudaMalloc((void **)&d_cc_space_v_oovo, nO*nO*nV*nO * sizeof(double));
      cublasSetMatrix(lda, nV*nO, sizeof(double), cc_space_v_oovo, lda, d_cc_space_v_oovo, lda);

      double* d_cc_space_v_ovvo;
      lda = nO*nV;
      cudaMalloc((void **)&d_cc_space_v_ovvo, nO*nV*nV*nO * sizeof(double));
      cublasSetMatrix(lda, nV*nO, sizeof(double), cc_space_v_ovvo, lda, d_cc_space_v_ovvo, lda);

      double* d_cc_space_v_ovov;
      lda = nO*nV;
      cudaMalloc((void **)&d_cc_space_v_ovov, nO*nV*nV*nO * sizeof(double));
      cublasSetMatrix(lda, nV*nO, sizeof(double), cc_space_v_ovov, lda, d_cc_space_v_ovov, lda);

      double* d_cc_space_v_ovoo;
      lda = nO*nV;
      cudaMalloc((void **)&d_cc_space_v_ovoo, nO*nV*nO*nO * sizeof(double));
      cublasSetMatrix(lda, nO*nO, sizeof(double), cc_space_v_ovoo, lda, d_cc_space_v_ovoo, lda);

      double* d_cc_space_f_oo;
      cudaMalloc((void**)&d_cc_space_f_oo, nO*nO*sizeof(double));
      cublasSetMatrix(nO, nO, sizeof(double), cc_space_f_oo, nO, d_cc_space_f_oo, nO);

      double* d_cc_space_f_vo;
      cudaMalloc((void**)&d_cc_space_f_vo, nV*nO*sizeof(double));
      cublasSetMatrix(nV, nO, sizeof(double), cc_space_f_vo, nV, d_cc_space_f_vo, nV);

      double* d_cc_space_f_ov;
      cudaMalloc((void**)&d_cc_space_f_ov, nV*nO*sizeof(double));
      cublasSetMatrix(nO, nV, sizeof(double), cc_space_f_ov, nO, d_cc_space_f_ov, nO);

      double* d_cc_space_f_vv;
      cudaMalloc((void**)&d_cc_space_f_vv, nV*nV*sizeof(double));
      cublasSetMatrix(nV, nV, sizeof(double), cc_space_f_vv, nV, d_cc_space_f_vv, nV);

      double* d_tau;
      lda = nO * nO;
      cudaMalloc((void **)&d_tau, lda * nV * nV * sizeof(double));

      double* d_tau_x;
      lda = nO * nO;
      cudaMalloc((void **)&d_tau_x, lda * nV * nV * sizeof(double));

      double* d_t1;
      cudaMalloc((void **)&d_t1, nO * nV * sizeof(double));

      double* d_t2;
      cudaMalloc((void **)&d_t2, nO*nO*nV*nV * sizeof(double));

      double* d_H_oo;
      cudaMalloc((void **)&d_H_oo, nO * nO * sizeof(double));

      double* d_H_vo;
      cudaMalloc((void **)&d_H_vo, nV * nO * sizeof(double));

      double* d_H_vv;
      cudaMalloc((void **)&d_H_vv, nV * nV * sizeof(double));

      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;
}