From 8abf589c909911813454155c8ef51ecdfc659688 Mon Sep 17 00:00:00 2001
From: Anthony Scemama <scemama@irsamc.ups-tlse.fr>
Date: Sun, 16 Jul 2023 13:16:11 +0200
Subject: [PATCH] Working

---
 devel/ccsd_gpu/gpu.c | 39 ++++++---------------------------------
 1 file changed, 6 insertions(+), 33 deletions(-)

diff --git a/devel/ccsd_gpu/gpu.c b/devel/ccsd_gpu/gpu.c
index 7cee7f7..ec8a100 100644
--- a/devel/ccsd_gpu/gpu.c
+++ b/devel/ccsd_gpu/gpu.c
@@ -94,7 +94,6 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
         {
            double* d_tmp_cc_ = &(d_tmp_cc[gam*nV*cholesky_mo_num]);
            double* d_cc_space_v_vv_chol_ = &(d_cc_space_v_vv_chol[gam*nV*cholesky_mo_num]);
-//           double* cc_space_v_vv_chol_ = &(cc_space_v_vv_chol[gam*nV*cholesky_mo_num]);
 
            alpha = 1.0;
            beta = -1.0;
@@ -103,13 +102,6 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
            C = d_tmp_cc2 ; ldc = cholesky_mo_num;
            cublasDgeam(handle, CUBLAS_OP_N, CUBLAS_OP_N, cholesky_mo_num, nV, &alpha, A, lda, &beta, B, ldb, C, ldc);
 
-/*
-           for (size_t l=0 ; l<nV*cholesky_mo_num ; ++l) {
-             tmp_cc2[l] = cc_space_v_vv_chol_[l] - tmp_cc_[l];
-           }
-           cublasSetMatrix(cholesky_mo_num, nV, sizeof(double), tmp_cc2, cholesky_mo_num, d_tmp_cc2, cholesky_mo_num);
-*/
-
            for (size_t iblock=0 ; iblock<nV ; iblock += BLOCK_SIZE)
            {
                 const size_t mbs = BLOCK_SIZE < nV-iblock ? BLOCK_SIZE : nV-iblock;
@@ -122,16 +114,6 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
                 C=d_tmpB1 ; ldc=nV*BLOCK_SIZE;
                 cublasDgemm(handle, CUBLAS_OP_T, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, lda, &beta, C, ldc);
 
-/*
-                A=&(tmp_cc[iblock*cholesky_mo_num*nV]); lda=cholesky_mo_num;
-                B=cc_space_v_vv_chol_; ldb=cholesky_mo_num;
-                C=tmpB1 ; ldc=nV*BLOCK_SIZE;
-                dgemm_("T","N", &m, &n, &k, &alpha, A, &lda, B, &lda, &beta, C, &ldc);
-*/
-
-
-
-
                 alpha=1.0; beta=1.0;
                 m=nV*mbs;  n=nV;  k=cholesky_mo_num;
 
@@ -139,25 +121,22 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
                 B=d_tmp_cc2; ldb=cholesky_mo_num;
                 C=d_tmpB1 ; ldc=nV*BLOCK_SIZE;
                 cublasDgemm(handle, CUBLAS_OP_T, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, lda, &beta, C, ldc);
-                cublasGetMatrix(m, n, sizeof(double), d_tmpB1, ldc, tmpB1, ldc);
-/*
-                A=&(cc_space_v_vv_chol[iblock*cholesky_mo_num*nV]); lda=cholesky_mo_num;
-                B=tmp_cc2; ldb=cholesky_mo_num;
-                C=tmpB1 ; ldc=nV*BLOCK_SIZE;
-                dgemm_("T","N", &m, &n, &k, &alpha, A, &lda, B, &ldb, &beta, C, &ldc);
-*/
-
+ //               cublasGetMatrix(m, n, sizeof(double), d_tmpB1, ldc, tmpB1, ldc);
 
 
                 for (size_t beta=iblock ; beta<(nV < iblock+BLOCK_SIZE ? nV : iblock+BLOCK_SIZE) ; ++beta) {
                   for (size_t b=0 ; b<nV ; ++b) {
+                    cublasDcopy(handle, nV, &(d_tmpB1[nV*(beta-iblock + BLOCK_SIZE*b)]), 1,
+                       &(d_B1[nV*(b + nV*(beta-iblock))]),1);
+/*
                     for (size_t a=0 ; a<nV ; ++a) {
                        B1[a + nV*(b + nV*(beta-iblock))] =
                            tmpB1[a + nV*(beta-iblock + BLOCK_SIZE*b)];
                     }
+*/
                   }
                 }
-                cublasSetMatrix(nV*nV, mbs, sizeof(double), B1, nV*nV, d_B1, nV*nV);
+//                cublasSetMatrix(nV*nV, mbs, sizeof(double), B1, nV*nV, d_B1, nV*nV);
 
                 alpha=1.0; beta=1.0;
                 m=nO*nO;  n=mbs;  k=nV*nV;
@@ -166,12 +145,6 @@ void compute_r2_space_chol_gpu(const int nO, const int nV, const int cholesky_mo
                 B=d_B1 ; ldb=nV*nV;
                 C=&(d_r2[nO*nO*(iblock + nV*gam)]); ldc=nO*nO;
                 cublasDgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, &alpha, A, lda, B, ldb, &beta, C, ldc);
-/*
-                A=tau; lda=nO*nO;
-                B=B1 ; ldb=nV*nV;
-                C=&(r2[nO*nO*(iblock + nV*gam)]); ldc=nO*nO;
-                dgemm_("N","N", &m, &n, &k, &alpha, A, &lda, B, &ldb, &beta, C, &ldc);
-*/
 
            }
         }