Use command line compiler if it is explicitly specified (#3018)

* Use command line compiler if it is explicitly specified

* Don't set OPENMM_CUDA_COMPILER on Jenkins

devtools/Jenkinsfile

@@ -15,12 +15,10 @@ pipeline {
                     }
                     steps {
                         sh "git clean -fxd && git checkout ."
-                        withEnv(["OPENMM_CUDA_COMPILER=/usr/local/cuda/bin/nvcc"]) {
                         sh "devtools/ci/jenkins/install.sh"
                         sh "devtools/ci/jenkins/test.sh -R 'TestCuda' --parallel 2"
                     }
                 }
-                }
 
                 stage("Build and test OpenCL platform") {
                     agent {

platforms/cuda/include/CudaContext.h

@@ -9,7 +9,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2009-2019 Stanford University and the Authors.      *
+ * Portions copyright (c) 2009-2021 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -75,8 +75,8 @@ public:
     static const int ThreadBlockSize;
     static const int TileSize;
     CudaContext(const System& system, int deviceIndex, bool useBlockingSync, const std::string& precision,
-            const std::string& compiler, const std::string& tempDir, const std::string& hostCompiler, CudaPlatform::PlatformData& platformData,
-            CudaContext* originalContext);
+            const std::string& compiler, const std::string& tempDir, const std::string& hostCompiler, bool allowRuntimeCompiler,
+            CudaPlatform::PlatformData& platformData, CudaContext* originalContext);
     ~CudaContext();
     /**
      * This is called to initialize internal data structures after all Forces in the system

platforms/cuda/include/CudaPlatform.h

@@ -9,7 +9,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2008-2016 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2021 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -131,14 +131,14 @@ class OPENMM_EXPORT_COMMON CudaPlatform::PlatformData {
 public:
     PlatformData(ContextImpl* context, const System& system, const std::string& deviceIndexProperty, const std::string& blockingProperty, const std::string& precisionProperty,
             const std::string& cpuPmeProperty, const std::string& compilerProperty, const std::string& tempProperty, const std::string& hostCompilerProperty,
-            const std::string& pmeStreamProperty, const std::string& deterministicForcesProperty, int numThreads, ContextImpl* originalContext);
+            const std::string& pmeStreamProperty, const std::string& deterministicForcesProperty, int numThreads, bool allowRuntimeCompiler, ContextImpl* originalContext);
     ~PlatformData();
     void initializeContexts(const System& system);
     void syncContexts();
     ContextImpl* context;
     std::vector<CudaContext*> contexts;
     std::vector<double> contextEnergy;
-    bool hasInitializedContexts, removeCM, peerAccessSupported, useCpuPme, disablePmeStream, deterministicForces;
+    bool hasInitializedContexts, removeCM, peerAccessSupported, useCpuPme, disablePmeStream, deterministicForces, allowRuntimeCompiler;
     int cmMotionFrequency;
     int stepCount, computeForceCount;
     double time;

platforms/cuda/src/CudaContext.cpp

@@ -6,7 +6,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2009-2019 Stanford University and the Authors.      *
+ * Portions copyright (c) 2009-2021 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -106,13 +106,13 @@ static int executeInWindows(const string &command) {
 #endif
 
 CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlockingSync, const string& precision, const string& compiler,
-        const string& tempDir, const std::string& hostCompiler, CudaPlatform::PlatformData& platformData, CudaContext* originalContext) : ComputeContext(system), currentStream(0),
-        platformData(platformData), contextIsValid(false), hasAssignedPosqCharges(false),
+        const string& tempDir, const std::string& hostCompiler, bool allowRuntimeCompiler, CudaPlatform::PlatformData& platformData,
+        CudaContext* originalContext) : ComputeContext(system), currentStream(0), platformData(platformData), contextIsValid(false), hasAssignedPosqCharges(false),
         hasCompilerKernel(false), isNvccAvailable(false), pinnedBuffer(NULL), integration(NULL), expression(NULL), bonded(NULL), nonbonded(NULL) {
     // Determine what compiler to use.
     
     this->compiler = "\""+compiler+"\"";
-    if (platformData.context != NULL) {
+    if (allowRuntimeCompiler && platformData.context != NULL) {
         try {
             compilerKernel = platformData.context->getPlatform().createKernel(CudaCompilerKernel::Name(), *platformData.context);
             hasCompilerKernel = true;

platforms/cuda/src/CudaPlatform.cpp

@@ -6,7 +6,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2008-2019 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2021 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -201,8 +201,10 @@ void CudaPlatform::contextCreated(ContextImpl& context, const map<string, string
     char* threadsEnv = getenv("OPENMM_CPU_THREADS");
     if (threadsEnv != NULL)
         stringstream(threadsEnv) >> threads;
+    char* compilerEnv = getenv("OPENMM_CUDA_COMPILER");
+    bool allowRuntimeCompiler = (compilerEnv == NULL && properties.find(CudaCompiler()) == properties.end());
     context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue,
-            hostCompilerPropValue, pmeStreamPropValue, deterministicForcesValue, threads, NULL));
+            hostCompilerPropValue, pmeStreamPropValue, deterministicForcesValue, threads, allowRuntimeCompiler, NULL));
 }
 
 void CudaPlatform::linkedContextCreated(ContextImpl& context, ContextImpl& originalContext) const {
@@ -217,8 +219,9 @@ void CudaPlatform::linkedContextCreated(ContextImpl& context, ContextImpl& origi
     string pmeStreamPropValue = platform.getPropertyValue(originalContext.getOwner(), CudaDisablePmeStream());
     string deterministicForcesValue = platform.getPropertyValue(originalContext.getOwner(), CudaDeterministicForces());
     int threads = reinterpret_cast<PlatformData*>(originalContext.getPlatformData())->threads.getNumThreads();
+    bool allowRuntimeCompiler = reinterpret_cast<PlatformData*>(originalContext.getPlatformData())->allowRuntimeCompiler;
     context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue,
-            hostCompilerPropValue, pmeStreamPropValue, deterministicForcesValue, threads, &originalContext));
+            hostCompilerPropValue, pmeStreamPropValue, deterministicForcesValue, threads, allowRuntimeCompiler, &originalContext));
 }
 
 void CudaPlatform::contextDestroyed(ContextImpl& context) const {
@@ -228,8 +231,9 @@ void CudaPlatform::contextDestroyed(ContextImpl& context) const {
 
 CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& system, const string& deviceIndexProperty, const string& blockingProperty, const string& precisionProperty,
             const string& cpuPmeProperty, const string& compilerProperty, const string& tempProperty, const string& hostCompilerProperty, const string& pmeStreamProperty,
-            const string& deterministicForcesProperty, int numThreads, ContextImpl* originalContext) :
-                context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0), hasInitializedContexts(false), threads(numThreads) {
+            const string& deterministicForcesProperty, int numThreads, bool allowRuntimeCompiler, ContextImpl* originalContext) :
+                context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0), hasInitializedContexts(false),
+                threads(numThreads), allowRuntimeCompiler(allowRuntimeCompiler) {
     bool blocking = (blockingProperty == "true");
     vector<string> devices;
     size_t searchPos = 0, nextPos;
@@ -246,11 +250,11 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
             if (devices[i].length() > 0) {
                 int deviceIndex;
                 stringstream(devices[i]) >> deviceIndex;
-                contexts.push_back(new CudaContext(system, deviceIndex, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, *this, (originalData == NULL ? NULL : originalData->contexts[i])));
+                contexts.push_back(new CudaContext(system, deviceIndex, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, allowRuntimeCompiler, *this, (originalData == NULL ? NULL : originalData->contexts[i])));
             }
         }
         if (contexts.size() == 0)
-            contexts.push_back(new CudaContext(system, -1, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, *this, (originalData == NULL ? NULL : originalData->contexts[0])));
+            contexts.push_back(new CudaContext(system, -1, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, allowRuntimeCompiler, *this, (originalData == NULL ? NULL : originalData->contexts[0])));
     }
     catch (...) {
         // If an exception was thrown, do our best to clean up memory.

platforms/cuda/tests/TestCudaFFT3D.cpp

@@ -56,7 +56,7 @@ void testTransform(bool realToComplex, int xsize, int ysize, int zsize) {
     system.addParticle(0.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     OpenMM_SFMT::SFMT sfmt;

platforms/cuda/tests/TestCudaRandom.cpp

@@ -56,7 +56,7 @@ void testGaussian() {
         system.addParticle(1.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     context.getIntegrationUtilities().initRandomNumberGenerator(0);

platforms/cuda/tests/TestCudaSort.cpp

@@ -66,7 +66,7 @@ void verifySorting(vector<float> array) {
     system.addParticle(0.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1, NULL);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     CudaArray data(context, array.size(), 4, "sortData");