Bug fixes to RPMD

plugins/rpmd/CMakeLists.txt

@@ -151,11 +151,6 @@ ENDIF(OPENMM_BUILD_STATIC_LIB)
 ADD_SUBDIRECTORY(platforms/reference/tests)
 
 # Which hardware platforms to build
-IF(CUDA_FOUND)
-    SET(OPENMM_BUILD_RPMD_CUDA_LIB ON CACHE BOOL "Build OpenMMRPMDCuda library for Nvidia GPUs")
-ELSE(CUDA_FOUND)
-    SET(OPENMM_BUILD_RPMD_CUDA_LIB OFF CACHE BOOL "Build OpenMMRPMDCuda library for Nvidia GPUs")
-ENDIF(CUDA_FOUND)
 
 #SET(OPENMM_BUILD_RPMD_PATH)
 #SET(OPENMM_BUILD_RPMD_CUDA_PATH)
@@ -166,10 +161,10 @@ ENDIF(CUDA_FOUND)
 #    SET(OPENMM_RPMD_CUDA_SOURCE_SUBDIRS . openmmapi olla platforms/cuda)
 #ENDIF(OPENMM_BUILD_RPMD_CUDA_LIB)
 
-#SET(OPENMM_BUILD_OPENCL_LIB OFF CACHE BOOL "Build OpenMMOpenCL library for Nvidia GPUs")
-#IF(OPENMM_BUILD_OPENCL_LIB)
-#    ADD_SUBDIRECTORY(platforms/opencl)
-#ENDIF(OPENMM_BUILD_OPENCL_LIB)
+SET(OPENMM_BUILD_RPMD_OPENCL_LIB ON CACHE BOOL "Build RPMD implementation for OpenCL")
+IF(OPENMM_BUILD_RPMD_OPENCL_LIB)
+    ADD_SUBDIRECTORY(platforms/opencl)
+ENDIF(OPENMM_BUILD_RPMD_OPENCL_LIB)
 
 INSTALL_TARGETS(/lib RUNTIME_DIRECTORY /lib ${SHARED_RPMD_TARGET})
 IF( CREATE_SERIALIZABLE_OPENMM_RPMD )
@@ -204,4 +199,3 @@ ENDIF (EXECUTABLE_OUTPUT_PATH)
 #INCLUDE(ApiDoxygen.cmake)
 
 #ADD_SUBDIRECTORY(tests)
-ADD_SUBDIRECTORY(platforms/opencl)

plugins/rpmd/platforms/opencl/CMakeLists.txt

@@ -112,6 +112,7 @@ ELSE (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
 ENDIF (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
 TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${MAIN_OPENMM_LIB}  ${OPENCL_LIBRARIES} ${PTHREADS_LIB})
 TARGET_LINK_LIBRARIES(${SHARED_TARGET} debug ${OPENMM_LIBRARY_NAME}OpenCL_d optimized ${OPENMM_LIBRARY_NAME}OpenCL)
+TARGET_LINK_LIBRARIES(${SHARED_TARGET} debug ${SHARED_RPMD_TARGET} optimized ${SHARED_RPMD_TARGET})
 SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES COMPILE_FLAGS "-DOPENMM_BUILDING_SHARED_LIBRARY")
 
 INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${SHARED_TARGET})

plugins/rpmd/platforms/opencl/src/OpenCLRpmdKernels.cpp

@@ -53,7 +53,7 @@ void OpenCLIntegrateRPMDStepKernel::initialize(const System& system, const RPMDI
     numCopies = integrator.getNumCopies();
     numParticles = system.getNumParticles();
     workgroupSize = numCopies;
-    while (workgroupSize < 128-numCopies)
+    while (workgroupSize <= 128-numCopies)
         workgroupSize += numCopies;
     if (numCopies != OpenCLFFT3D::findLegalDimension(numCopies))
         throw OpenMMException("RPMDIntegrator: the number of copies must be a multiple of powers of 2, 3, and 5.");
@@ -68,6 +68,7 @@ void OpenCLIntegrateRPMDStepKernel::initialize(const System& system, const RPMDI
     vector<mm_float4> temp(positions->getSize());
     for (int i = 0; i < positions->getSize(); i++)
         temp[i] = mm_float4(0, 0, 0, 0);
+    positions->upload(temp);
     for (int i = 0; i < velocities->getSize(); i++)
         temp[i] = mm_float4(0, 0, 0, 1);
     velocities->upload(temp);
@@ -120,10 +121,10 @@ void OpenCLIntegrateRPMDStepKernel::execute(ContextImpl& context, const RPMDInte
     if (!forcesAreValid) {
         for (int i = 0; i < numCopies; i++) {
             cl.getQueue().enqueueCopyBuffer(positions->getDeviceBuffer(), cl.getPosq().getDeviceBuffer(),
-                    i*paddedParticles, 0, paddedParticles*sizeof(mm_float4));
+                    i*paddedParticles*sizeof(mm_float4), 0, paddedParticles*sizeof(mm_float4));
             context.calcForcesAndEnergy(true, false);
             cl.getQueue().enqueueCopyBuffer(cl.getForce().getDeviceBuffer(), forces->getDeviceBuffer(),
-                    0, i*paddedParticles, paddedParticles*sizeof(mm_float4));
+                    0, i*paddedParticles*sizeof(mm_float4), paddedParticles*sizeof(mm_float4));
         }
     }
     
@@ -146,10 +147,10 @@ void OpenCLIntegrateRPMDStepKernel::execute(ContextImpl& context, const RPMDInte
     
     for (int i = 0; i < numCopies; i++) {
         cl.getQueue().enqueueCopyBuffer(positions->getDeviceBuffer(), cl.getPosq().getDeviceBuffer(),
-                i*paddedParticles, 0, paddedParticles*sizeof(mm_float4));
+                i*paddedParticles*sizeof(mm_float4), 0, paddedParticles*sizeof(mm_float4));
         context.calcForcesAndEnergy(true, false);
         cl.getQueue().enqueueCopyBuffer(cl.getForce().getDeviceBuffer(), forces->getDeviceBuffer(),
-                0, i*paddedParticles, paddedParticles*sizeof(mm_float4));
+                0, i*paddedParticles*sizeof(mm_float4), paddedParticles*sizeof(mm_float4));
     }
     
     // Update velocities.

plugins/rpmd/platforms/opencl/tests/CMakeLists.txt

@@ -6,8 +6,6 @@ ENABLE_TESTING()
 
 INCLUDE_DIRECTORIES(${OPENCL_INCLUDE_DIR})
 
-SET(SHARED_OPENMM_RPMD_TARGET OpenMMRPMD)
-
 # Automatically create tests using files named "Test*.cpp"
 FILE(GLOB TEST_PROGS "*Test*.cpp")
 FOREACH(TEST_PROG ${TEST_PROGS})
@@ -15,7 +13,7 @@ FOREACH(TEST_PROG ${TEST_PROGS})
 
     # Link with shared library
     ADD_EXECUTABLE(${TEST_ROOT} ${TEST_PROG})
-    TARGET_LINK_LIBRARIES(${TEST_ROOT} ${SHARED_TARGET} ${SHARED_OPENMM_RPMD_TARGET})
+    TARGET_LINK_LIBRARIES(${TEST_ROOT} ${SHARED_TARGET} ${SHARED_RPMD_TARGET})
     ADD_TEST(${TEST_ROOT} ${EXECUTABLE_OUTPUT_PATH}/${TEST_ROOT})
 
 ENDFOREACH(TEST_PROG ${TEST_PROGS})