From 3ea90bc4a5042b1e2f43bfc1af857cbe3efa00e8 Mon Sep 17 00:00:00 2001
From: Anthony Scemama <scemama@irsamc.ups-tlse.fr>
Date: Wed, 6 Apr 2022 17:11:21 +0200
Subject: [PATCH] OpenMP

---
 org/qmckl_jastrow.org | 32 +++++++++++---------------------
 1 file changed, 11 insertions(+), 21 deletions(-)

diff --git a/org/qmckl_jastrow.org b/org/qmckl_jastrow.org
index 9173a43..1fd0d47 100644
--- a/org/qmckl_jastrow.org
+++ b/org/qmckl_jastrow.org
@@ -5969,13 +5969,13 @@ qmckl_compute_tmp_c_omp_offload (const qmckl_context context,
   const int64_t size_e = walk_num*(cord_num+1)*elec_num*elec_num;
   const int64_t size_n = walk_num*(cord_num+1)*nucl_num*elec_num; 
 
-#pragma omp parallel copyout(tmp_c [0:size_tmp_c]) copyin(een_rescaled_e[0:size_e], een_rescaled_n[0:size_n])
-  {
-#pragma omp loop independent gang worker vector
-    for (int64_t i=0 ; i<size_tmp_c ; ++i)
+  for (int64_t i=0 ; i<size_tmp_c ; ++i)
       tmp_c[i] = 0.;
 
-#pragma omp loop independent gang worker vector collapse(5)
+#pragma omp target data map(to:een_rescaled_e[0:size_e], \
+                               een_rescaled_n[0:size_n]) \
+                        map(tofrom:tmp_c[0:size_tmp_c])
+#pragma omp target teams distribute parallel for collapse(5)
     for (int nw=0; nw < walk_num; ++nw) {
       for (int i=0; i<cord_num; ++i){
 
@@ -5998,7 +5998,6 @@ qmckl_compute_tmp_c_omp_offload (const qmckl_context context,
         }
       }
     }
-  }
 
   return QMCKL_SUCCESS;
 }
@@ -6314,8 +6313,6 @@ qmckl_compute_dtmp_c_hpc (const qmckl_context context,
   const int64_t bf = elec_num*nucl_num*(cord_num+1);
   const int64_t cf = elec_num*4*nucl_num*(cord_num+1);
 
-  printf("COUCOU\n");
-
 #ifdef HAVE_OPENMP
 #pragma omp parallel for collapse(2)
 #endif
@@ -6398,10 +6395,6 @@ qmckl_exit_code qmckl_compute_dtmp_c_acc_offload (
 
   #pragma acc parallel copyout(dtmp_c [0:size_dtmp_c]) copyin(een_rescaled_e_deriv_e[0:size_e], een_rescaled_n[0:size_n])
   {
-      #pragma acc loop independent gang worker vector
-      for (int64_t i=0 ; i<size_dtmp_c ; ++i)
-          dtmp_c[i] = 0.;
-
       #pragma loop independent gang worker vector collapse(6)
       for (int nw=0; nw < walk_num; nw++) {
         for (int i=0; i < cord_num; i++) {
@@ -6413,7 +6406,7 @@ qmckl_exit_code qmckl_compute_dtmp_c_acc_offload (
                 for(int m=0; m<elec_num; m++) {
           
                   // Single reduction
-                        dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] = 0;
+                        dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] = 0.;
                   for(int n=0; n<elec_num; n++){
                 dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] =
                 dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] +
@@ -6499,13 +6492,11 @@ qmckl_exit_code qmckl_compute_dtmp_c_omp_offload (
   const int64_t size_n = walk_num*(cord_num+1)*nucl_num*elec_num; 
   const int64_t size_e = walk_num*(cord_num+1)*elec_num*4*elec_num;
 
-  #pragma omp parallel copyout(dtmp_c [0:size_dtmp_c]) copyin(een_rescaled_e_deriv_e[0:size_e], een_rescaled_n[0:size_n])
-  {
-      #pragma omp target
-      for (int64_t i=0 ; i<size_dtmp_c ; ++i)
-          dtmp_c[i] = 0.;
 
-      #pragma loop independent gang worker vector collapse(6)
+#pragma omp target data map(to:een_rescaled_e_deriv_e[0:size_e], \
+                               een_rescaled_n[0:size_n]), \
+                        map(tofrom:dtmp_c[0:size_dtmp_c])
+#pragma omp target teams distribute parallel for collapse(6)
       for (int nw=0; nw < walk_num; nw++) {
         for (int i=0; i < cord_num; i++) {
 
@@ -6516,7 +6507,7 @@ qmckl_exit_code qmckl_compute_dtmp_c_omp_offload (
                 for(int m=0; m<elec_num; m++) {
                   
                   // Single reduction
-                  dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] = 0;
+                  dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] = 0.;
                   for(int n=0; n<elec_num; n++){
                     dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] =
                       dtmp_c[m + l * stride_l_d + k * stride_k_d + j * stride_j_d + i * stride_i_d + nw * stride_nw_d] +
@@ -6529,7 +6520,6 @@ qmckl_exit_code qmckl_compute_dtmp_c_omp_offload (
           }
         }
       }
-  }
 
   return QMCKL_SUCCESS;
 }