Pipeline GPU PME Spline/Spread with PP Comms

[alexxy/gromacs.git] / src / gromacs / ewald / tests / pmetestcommon.cpp

diff --git a/src/gromacs/ewald/tests/pmetestcommon.cpp b/src/gromacs/ewald/tests/pmetestcommon.cpp
index 63f0cadb114e2c80bfddcff043167835f1572490..0016cd80241e43f304af1595ddd7fab1b694d799 100644 (file)
--- a/src/gromacs/ewald/tests/pmetestcommon.cpp
+++ b/src/gromacs/ewald/tests/pmetestcommon.cpp
@@ -1,7 +1,8 @@
 /*
  * This file is part of the GROMACS molecular simulation package.
  *
- * Copyright (c) 2016,2017,2018,2019,2020, by the GROMACS development team, led by
+ * Copyright (c) 2016,2017,2018,2019,2020 by the GROMACS development team.
+ * Copyright (c) 2021, by the GROMACS development team, led by
  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
  * and including many others, as listed in the AUTHORS file in the
  * top-level source directory and at http://www.gromacs.org.
@@ -69,6 +70,7 @@
 #include "gromacs/utility/gmxassert.h"
 #include "gromacs/utility/logger.h"
 #include "gromacs/utility/stringutil.h"
+#include "gromacs/ewald/pme_coordinate_receiver_gpu.h"
 
 #include "testutils/test_hardware_environment.h"
 #include "testutils/testasserts.h"
@@ -199,7 +201,7 @@ void pmeInitAtoms(gmx_pme_t*               pme,
             atc              = &(pme->atc[0]);
             atc->x           = coordinates;
             atc->coefficient = charges;
-            gmx_pme_reinit_atoms(pme, atomCount, charges.data(), nullptr);
+            gmx_pme_reinit_atoms(pme, atomCount, charges, {});
             /* With decomposition there would be more boilerplate atc code here, e.g. do_redist_pos_coeffs */
             break;
 
@@ -208,11 +210,11 @@ void pmeInitAtoms(gmx_pme_t*               pme,
             atc = &(pme->atc[0]);
             // We need to set atc->n for passing the size in the tests
             atc->setNumAtoms(atomCount);
-            gmx_pme_reinit_atoms(pme, atomCount, charges.data(), nullptr);
+            gmx_pme_reinit_atoms(pme, atomCount, charges, {});
 
             stateGpu->reinit(atomCount, atomCount);
             stateGpu->copyCoordinatesToGpu(arrayRefFromArray(coordinates.data(), coordinates.size()),
-                                           gmx::AtomLocality::All);
+                                           gmx::AtomLocality::Local);
             pme_gpu_set_kernelparam_coordinates(pme->gpu, stateGpu->getCoordinates());
 
             break;
@@ -278,7 +280,7 @@ static void pmeGetGridAndSizesInternal(const gmx_pme_t* /*unused*/,
                                        IVec& /*unused*/)       //NOLINT(google-runtime-references)
 {
     GMX_THROW(InternalError("Deleted function call"));
-    // explicitly deleting general template does not compile in clang/icc, see https://llvm.org/bugs/show_bug.cgi?id=17537
+    // explicitly deleting general template does not compile in clang, see https://llvm.org/bugs/show_bug.cgi?id=17537
 }
 
 //! Getting the PME real grid memory buffer and its sizes
@@ -341,7 +343,18 @@ void pmePerformSplineAndSpread(gmx_pme_t* pme,
             const real lambdaQ = 1.0;
             // no synchronization needed as x is transferred in the PME stream
             GpuEventSynchronizer* xReadyOnDevice = nullptr;
-            pme_gpu_spread(pme->gpu, xReadyOnDevice, fftgrid, computeSplines, spreadCharges, lambdaQ);
+
+            bool                           useGpuDirectComm         = false;
+            gmx::PmeCoordinateReceiverGpu* pmeCoordinateReceiverGpu = nullptr;
+
+            pme_gpu_spread(pme->gpu,
+                           xReadyOnDevice,
+                           fftgrid,
+                           computeSplines,
+                           spreadCharges,
+                           lambdaQ,
+                           useGpuDirectComm,
+                           pmeCoordinateReceiverGpu);
         }
         break;
 #endif
@@ -910,8 +923,7 @@ const char* codePathToString(CodePath codePath)
 PmeTestHardwareContext::PmeTestHardwareContext() : codePath_(CodePath::CPU) {}
 
 PmeTestHardwareContext::PmeTestHardwareContext(TestDevice* testDevice) :
-    codePath_(CodePath::CPU),
-    testDevice_(testDevice)
+    codePath_(CodePath::GPU), testDevice_(testDevice)
 {
     setActiveDevice(testDevice_->deviceInfo());
     pmeGpuProgram_ = buildPmeGpuProgram(testDevice_->deviceContext());