Added option for setting CUDA host compiler

platforms/cuda/include/CudaContext.h

@@ -75,7 +75,7 @@ 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, CudaPlatform::PlatformData& platformData);
+            const std::string& compiler, const std::string& tempDir, const std::string& hostCompiler, CudaPlatform::PlatformData& platformData);
     ~CudaContext();
     /**
      * This is called to initialize internal data structures after all Forces in the system

platforms/cuda/include/CudaPlatform.h

@@ -95,6 +95,13 @@ public:
         static const std::string key = "CudaCompiler";
         return key;
     }
+    /**
+     * This is the name of the parameter for specifying the host compiler for the CUDA compiler to use.
+     */
+    static const std::string& CudaHostCompiler() {
+        static const std::string key = "CudaHostCompiler";
+        return key;
+    }
     /**
      * This is the name of the parameter for specifying the path to the directory for creating temporary files.
      */
@@ -107,7 +114,7 @@ public:
 class OPENMM_EXPORT_CUDA 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& cpuPmeProperty, const std::string& compilerProperty, const std::string& tempProperty, const std::string& hostCompilerProperty);
     ~PlatformData();
     void initializeContexts(const System& system);
     void syncContexts();

platforms/cuda/src/CudaContext.cpp

@@ -72,9 +72,11 @@ const int CudaContext::TileSize = sizeof(tileflags)*8;
 bool CudaContext::hasInitializedCuda = false;
 
 CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlockingSync, const string& precision, const string& compiler,
-        const string& tempDir, CudaPlatform::PlatformData& platformData) : system(system), compiler(compiler),
+        const string& tempDir, const std::string& hostCompiler, CudaPlatform::PlatformData& platformData) : system(system), compiler(compiler),
         time(0.0), platformData(platformData), stepCount(0), computeForceCount(0), stepsSinceReorder(99999), contextIsValid(false), atomsWereReordered(false), pinnedBuffer(NULL), posq(NULL),
         posqCorrection(NULL), velm(NULL), force(NULL), energyBuffer(NULL), integration(NULL), expression(NULL), bonded(NULL), nonbonded(NULL), thread(NULL) {
+    if (hostCompiler.size() > 0)
+        this->compiler = compiler+" --compiler-bindir "+hostCompiler;
     if (!hasInitializedCuda) {
         CHECK_RESULT2(cuInit(0), "Error initializing CUDA");
         hasInitializedCuda = true;

platforms/cuda/src/CudaPlatform.cpp

@@ -90,6 +90,7 @@ CudaPlatform::CudaPlatform() {
     platformProperties.push_back(CudaUseCpuPme());
     platformProperties.push_back(CudaCompiler());
     platformProperties.push_back(CudaTempDirectory());
+    platformProperties.push_back(CudaHostCompiler());
     setPropertyDefaultValue(CudaDeviceIndex(), "");
     setPropertyDefaultValue(CudaDeviceName(), "");
     setPropertyDefaultValue(CudaUseBlockingSync(), "true");
@@ -114,6 +115,8 @@ CudaPlatform::CudaPlatform() {
     string tmp = (tmpdir == NULL ? string(P_tmpdir) : string(tmpdir));
     setPropertyDefaultValue(CudaTempDirectory(), tmp);
 #endif
+    char* hostCompiler = getenv("CUDA_HOST_COMPILER");
+    setPropertyDefaultValue(CudaHostCompiler(), (hostCompiler == NULL ? "" : string(hostCompiler)));
 }
 
 double CudaPlatform::getSpeed() const {
@@ -149,6 +152,8 @@ void CudaPlatform::contextCreated(ContextImpl& context, const map<string, string
             getPropertyDefaultValue(CudaCompiler()) : properties.find(CudaCompiler())->second);
     const string& tempPropValue = (properties.find(CudaTempDirectory()) == properties.end() ?
             getPropertyDefaultValue(CudaTempDirectory()) : properties.find(CudaTempDirectory())->second);
+    const string& hostCompilerPropValue = (properties.find(CudaHostCompiler()) == properties.end() ?
+            getPropertyDefaultValue(CudaHostCompiler()) : properties.find(CudaHostCompiler())->second);
     transform(blockingPropValue.begin(), blockingPropValue.end(), blockingPropValue.begin(), ::tolower);
     transform(precisionPropValue.begin(), precisionPropValue.end(), precisionPropValue.begin(), ::tolower);
     transform(cpuPmePropValue.begin(), cpuPmePropValue.end(), cpuPmePropValue.begin(), ::tolower);
@@ -156,7 +161,7 @@ void CudaPlatform::contextCreated(ContextImpl& context, const map<string, string
     pmeKernelName.push_back(CalcPmeReciprocalForceKernel::Name());
     if (!supportsKernels(pmeKernelName))
         cpuPmePropValue = "false";
-    context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue));
+    context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue, hostCompilerPropValue));
 }
 
 void CudaPlatform::contextDestroyed(ContextImpl& context) const {
@@ -165,7 +170,7 @@ 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) : context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0)  {
+            const string& cpuPmeProperty, const string& compilerProperty, const string& tempProperty, const string& hostCompilerProperty) : context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0) {
     bool blocking = (blockingProperty == "true");
     vector<string> devices;
     size_t searchPos = 0, nextPos;
@@ -178,11 +183,11 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
         if (devices[i].length() > 0) {
             unsigned int deviceIndex;
             stringstream(devices[i]) >> deviceIndex;
-            contexts.push_back(new CudaContext(system, deviceIndex, blocking, precisionProperty, compilerProperty, tempProperty, *this));
+            contexts.push_back(new CudaContext(system, deviceIndex, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, *this));
         }
     }
     if (contexts.size() == 0)
-        contexts.push_back(new CudaContext(system, -1, blocking, precisionProperty, compilerProperty, tempProperty, *this));
+        contexts.push_back(new CudaContext(system, -1, blocking, precisionProperty, compilerProperty, tempProperty, hostCompilerProperty, *this));
     stringstream deviceIndex, deviceName;
     for (int i = 0; i < (int) contexts.size(); i++) {
         if (i > 0) {
@@ -202,6 +207,7 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
     propertyValues[CudaPlatform::CudaUseCpuPme()] = useCpuPme ? "true" : "false";
     propertyValues[CudaPlatform::CudaCompiler()] = compilerProperty;
     propertyValues[CudaPlatform::CudaTempDirectory()] = tempProperty;
+    propertyValues[CudaPlatform::CudaHostCompiler()] = hostCompilerProperty;
     contextEnergy.resize(contexts.size());
     
     // Determine whether peer-to-peer copying is supported, and enable it if so.

platforms/cuda/tests/TestCudaRandom.cpp

@@ -55,7 +55,8 @@ void testGaussian() {
     for (int i = 0; i < numAtoms; i++)
         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::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()));
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     context.getIntegrationUtilities().initRandomNumberGenerator(0);

platforms/cuda/tests/TestCudaSort.cpp

@@ -65,7 +65,8 @@ void verifySorting(vector<float> array) {
     System system;
     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::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()));
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     CudaArray data(context, array.size(), 4, "sortData");