From d3c69f763ad309a68d4e4784e95c633be7421f71 Mon Sep 17 00:00:00 2001
From: LwhJesse <256257451+LwhJesse@users.noreply.github.com>
Date: Thu, 30 Apr 2026 16:58:43 +0800
Subject: [PATCH 1/2] Reduce redundant CUDA Jacobian uploads

---
 Common/src/linear_algebra/CSysMatrixGPU.cu         | 1 -
 Common/src/linear_algebra/CSysSolve.cpp            | 6 ++++++
 SU2_CFD/include/integration/CNewtonIntegration.hpp | 3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/Common/src/linear_algebra/CSysMatrixGPU.cu b/Common/src/linear_algebra/CSysMatrixGPU.cu
index 90389264ed3..7e0c81ca54f 100644
--- a/Common/src/linear_algebra/CSysMatrixGPU.cu
+++ b/Common/src/linear_algebra/CSysMatrixGPU.cu
@@ -70,7 +70,6 @@ void CSysMatrix<ScalarType>::GPUMatrixVectorProduct(const CSysVector<ScalarType>
    ScalarType* d_vec = vec.GetDevicePointer();
    ScalarType* d_prod = prod.GetDevicePointer();
 
-   HtDTransfer();
    vec.HtDTransfer();
    prod.GPUSetVal(0.0);
 
diff --git a/Common/src/linear_algebra/CSysSolve.cpp b/Common/src/linear_algebra/CSysSolve.cpp
index 953bed0ef2a..ad57bfb179b 100644
--- a/Common/src/linear_algebra/CSysSolve.cpp
+++ b/Common/src/linear_algebra/CSysSolve.cpp
@@ -1464,6 +1464,9 @@ unsigned long CSysSolve<ScalarType>::Solve(CSysMatrix<ScalarType>& Jacobian, con
 
     HandleTemporariesIn(LinSysRes, LinSysSol);
 
+#ifdef HAVE_CUDA
+    if (config->GetCUDA()) Jacobian.HtDTransfer();
+#endif
     auto mat_vec = CSysMatrixVectorProduct<ScalarType>(Jacobian, geometry, config);
 
     /*--- Build preconditioner. ---*/
@@ -1643,6 +1646,9 @@ unsigned long CSysSolve<ScalarType>::Solve_b(CSysMatrix<ScalarType>& Jacobian, c
 
   /*--- Set up preconditioner and matrix-vector product ---*/
 
+#ifdef HAVE_CUDA
+  if (config->GetCUDA()) Jacobian.HtDTransfer();
+#endif
   auto mat_vec = CSysMatrixVectorProduct<ScalarType>(Jacobian, geometry, config);
 
   const auto kindPrec = static_cast<ENUM_LINEAR_SOLVER_PREC>(KindPrecond);
diff --git a/SU2_CFD/include/integration/CNewtonIntegration.hpp b/SU2_CFD/include/integration/CNewtonIntegration.hpp
index da8ff67930a..fc27d5b77c0 100644
--- a/SU2_CFD/include/integration/CNewtonIntegration.hpp
+++ b/SU2_CFD/include/integration/CNewtonIntegration.hpp
@@ -158,6 +158,9 @@ class CNewtonIntegration final : public CIntegration {
       (*preconditioner)(u, v);
       return 0;
     }
+#ifdef HAVE_CUDA
+    if (config->GetCUDA()) solvers[FLOW_SOL]->Jacobian.HtDTransfer();
+#endif
     auto product = CSysMatrixVectorProduct<MixedScalar>(solvers[FLOW_SOL]->Jacobian, geometry, config);
     v = MixedScalar(0.0);
     MixedScalar eps_t = eps;

From 71e26d09820d9163c053d230f71c6fbe4e680126 Mon Sep 17 00:00:00 2001
From: LwhJesse <256257451+LwhJesse@users.noreply.github.com>
Date: Wed, 6 May 2026 22:14:30 +0800
Subject: [PATCH 2/2] Move CUDA Jacobian upload into CSysMatrixVectorProduct

---
 Common/include/linear_algebra/CMatrixVectorProduct.hpp | 6 +++++-
 Common/src/linear_algebra/CSysSolve.cpp                | 6 ------
 SU2_CFD/include/integration/CNewtonIntegration.hpp     | 3 ---
 3 files changed, 5 insertions(+), 10 deletions(-)

diff --git a/Common/include/linear_algebra/CMatrixVectorProduct.hpp b/Common/include/linear_algebra/CMatrixVectorProduct.hpp
index 878bb132b98..e6baca2acf1 100644
--- a/Common/include/linear_algebra/CMatrixVectorProduct.hpp
+++ b/Common/include/linear_algebra/CMatrixVectorProduct.hpp
@@ -82,7 +82,11 @@ class CSysMatrixVectorProduct final : public CMatrixVectorProduct<ScalarType> {
    */
   inline CSysMatrixVectorProduct(const CSysMatrix<ScalarType>& matrix_ref, CGeometry* geometry_ref,
                                  const CConfig* config_ref)
-      : matrix(matrix_ref), geometry(geometry_ref), config(config_ref) {}
+      : matrix(matrix_ref), geometry(geometry_ref), config(config_ref) {
+#ifdef HAVE_CUDA
+    if (config->GetCUDA()) matrix.HtDTransfer();
+#endif
+  }
 
   /*!
    * \note This class cannot be default constructed as that would leave us with invalid pointers.
diff --git a/Common/src/linear_algebra/CSysSolve.cpp b/Common/src/linear_algebra/CSysSolve.cpp
index ad57bfb179b..953bed0ef2a 100644
--- a/Common/src/linear_algebra/CSysSolve.cpp
+++ b/Common/src/linear_algebra/CSysSolve.cpp
@@ -1464,9 +1464,6 @@ unsigned long CSysSolve<ScalarType>::Solve(CSysMatrix<ScalarType>& Jacobian, con
 
     HandleTemporariesIn(LinSysRes, LinSysSol);
 
-#ifdef HAVE_CUDA
-    if (config->GetCUDA()) Jacobian.HtDTransfer();
-#endif
     auto mat_vec = CSysMatrixVectorProduct<ScalarType>(Jacobian, geometry, config);
 
     /*--- Build preconditioner. ---*/
@@ -1646,9 +1643,6 @@ unsigned long CSysSolve<ScalarType>::Solve_b(CSysMatrix<ScalarType>& Jacobian, c
 
   /*--- Set up preconditioner and matrix-vector product ---*/
 
-#ifdef HAVE_CUDA
-  if (config->GetCUDA()) Jacobian.HtDTransfer();
-#endif
   auto mat_vec = CSysMatrixVectorProduct<ScalarType>(Jacobian, geometry, config);
 
   const auto kindPrec = static_cast<ENUM_LINEAR_SOLVER_PREC>(KindPrecond);
diff --git a/SU2_CFD/include/integration/CNewtonIntegration.hpp b/SU2_CFD/include/integration/CNewtonIntegration.hpp
index fc27d5b77c0..da8ff67930a 100644
--- a/SU2_CFD/include/integration/CNewtonIntegration.hpp
+++ b/SU2_CFD/include/integration/CNewtonIntegration.hpp
@@ -158,9 +158,6 @@ class CNewtonIntegration final : public CIntegration {
       (*preconditioner)(u, v);
       return 0;
     }
-#ifdef HAVE_CUDA
-    if (config->GetCUDA()) solvers[FLOW_SOL]->Jacobian.HtDTransfer();
-#endif
     auto product = CSysMatrixVectorProduct<MixedScalar>(solvers[FLOW_SOL]->Jacobian, geometry, config);
     v = MixedScalar(0.0);
     MixedScalar eps_t = eps;