From d1dc35eaa4b37bee0e03f9dfb6c304f4f027e970 Mon Sep 17 00:00:00 2001
From: Anthony Scemama <scemama@irsamc.ups-tlse.fr>
Date: Wed, 6 Apr 2022 17:58:05 +0200
Subject: [PATCH] First working OpenMP version

---
 org/qmckl_jastrow.org | 26 ++++++++++----------------
 1 file changed, 10 insertions(+), 16 deletions(-)

diff --git a/org/qmckl_jastrow.org b/org/qmckl_jastrow.org
index dd2722a..79183a6 100644
--- a/org/qmckl_jastrow.org
+++ b/org/qmckl_jastrow.org
@@ -5869,10 +5869,6 @@ qmckl_exit_code qmckl_compute_tmp_c_acc_offload (const qmckl_context context,
 
 #pragma acc parallel copyout(tmp_c [0:size_tmp_c]) copyin(een_rescaled_e[0:size_e], een_rescaled_n[0:size_n])
   {
-#pragma acc loop independent gang worker vector
-    for (int64_t i=0 ; i<size_tmp_c ; ++i)
-      tmp_c[i] = 0.;
-
 #pragma acc loop independent gang worker vector collapse(5)
     for (int nw=0; nw < walk_num; ++nw) {
       for (int i=0; i<cord_num; ++i){
@@ -5883,7 +5879,7 @@ qmckl_exit_code qmckl_compute_tmp_c_acc_offload (const qmckl_context context,
             for (int l=0; l<elec_num; l++) {
     
               // Single reduction
-              tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] = 0;
+              tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] = 0.;
               for (int m=0; m<elec_num; m++) {
                 tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] =
                   tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] +
@@ -5967,13 +5963,11 @@ 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; 
 
-  for (int64_t i=0 ; i<size_tmp_c ; ++i)
-      tmp_c[i] = 0.;
 
-#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)
+#pragma omp target teams distribute parallel for collapse(5) \
+            map(to:een_rescaled_e[0:size_e], \
+                   een_rescaled_n[0:size_n]) \
+            map(from:tmp_c[0:size_tmp_c])
     for (int nw=0; nw < walk_num; ++nw) {
       for (int i=0; i<cord_num; ++i){
 
@@ -5983,7 +5977,7 @@ qmckl_compute_tmp_c_omp_offload (const qmckl_context context,
             for (int l=0; l<elec_num; l++) {
     
               // Single reduction
-              tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] = 0;
+              tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] = 0.;
               for (int m=0; m<elec_num; m++) {
                 tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] =
                   tmp_c[l + k*stride_k_c + j*stride_j_c + i*stride_i_c + nw*stride_nw_c] +
@@ -6491,10 +6485,10 @@ qmckl_exit_code qmckl_compute_dtmp_c_omp_offload (
   const int64_t size_e = walk_num*(cord_num+1)*elec_num*4*elec_num;
 
 
-#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)
+#pragma omp target teams distribute parallel for collapse(6) \
+            map(to:een_rescaled_e_deriv_e[0:size_e], \
+                   een_rescaled_n[0:size_n]), \
+            map(tofrom:dtmp_c[0:size_dtmp_c])
       for (int nw=0; nw < walk_num; nw++) {
         for (int i=0; i < cord_num; i++) {