Skip to content

GitLab

Commit 19d2885a authored by Lee-Ping's avatar Lee-Ping
Browse files

Merge github.com:SimTk/openmm

parents 99ef4344 57a6768e
Hide whitespace changes
Inline Side-by-side
Showing with 251 additions and 75 deletions
platforms/cpu/tests/TestCpuSettle.cpp 0 → 100644
View file @ 19d2885a
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2008-2013 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
/**
* This tests the CPU implementation of the SETTLE algorithm.
*/
#include "openmm/internal/AssertionUtilities.h"
#include "openmm/Context.h"
#include "CpuPlatform.h"
#include "openmm/NonbondedForce.h"
#include "openmm/System.h"
#include "openmm/LangevinIntegrator.h"
#include "sfmt/SFMT.h"
#include <iostream>
#include <vector>
using namespace OpenMM;
using namespace std;
void testConstraints() {
const int numMolecules = 10;
const int numParticles = numMolecules*3;
const int numConstraints = numMolecules*3;
const double temp = 100.0;
CpuPlatform platform;
System system;
LangevinIntegrator integrator(temp, 2.0, 0.001);
integrator.setConstraintTolerance(1e-5);
NonbondedForce* forceField = new NonbondedForce();
for (int i = 0; i < numMolecules; ++i) {
system.addParticle(16.0);
system.addParticle(1.0);
system.addParticle(1.0);
forceField->addParticle(-0.82, 0.317, 0.65);
forceField->addParticle(0.41, 1.0, 0.0);
forceField->addParticle(0.41, 1.0, 0.0);
system.addConstraint(i*3, i*3+1, 0.1);
system.addConstraint(i*3, i*3+2, 0.1);
system.addConstraint(i*3+1, i*3+2, 0.163);
}
system.addForce(forceField);
Context context(system, integrator, platform);
vector<Vec3> positions(numParticles);
vector<Vec3> velocities(numParticles);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int i = 0; i < numMolecules; ++i) {
positions[i*3] = Vec3((i%4)*0.4, (i/4)*0.4, 0);
positions[i*3+1] = positions[i*3]+Vec3(0.1, 0, 0);
positions[i*3+2] = positions[i*3]+Vec3(-0.03333, 0.09428, 0);
velocities[i*3] = Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5);
velocities[i*3+1] = Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5);
velocities[i*3+2] = Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5);
}
context.setPositions(positions);
context.setVelocities(velocities);
// Simulate it and see whether the constraints remain satisfied.
for (int i = 0; i < 1000; ++i) {
integrator.step(1);
State state = context.getState(State::Positions | State::Forces);
for (int j = 0; j < numConstraints; ++j) {
int particle1, particle2;
double distance;
system.getConstraintParameters(j, particle1, particle2, distance);
Vec3 p1 = state.getPositions()[particle1];
Vec3 p2 = state.getPositions()[particle2];
double dist = std::sqrt((p1[0]-p2[0])*(p1[0]-p2[0])+(p1[1]-p2[1])*(p1[1]-p2[1])+(p1[2]-p2[2])*(p1[2]-p2[2]));
ASSERT_EQUAL_TOL(distance, dist, 1e-5);
}
}
}
int main(int argc, char* argv[]) {
try {
if (!CpuPlatform::isProcessorSupported()) {
cout << "CPU is not supported. Exiting." << endl;
return 0;
}
testConstraints();
}
catch(const exception& e) {
cout << "exception: " << e.what() << endl;
return 1;
}
cout << "Done" << endl;
return 0;
}
platforms/cuda/CMakeLists.txt
View file @ 19d2885a
......@@ -14,10 +14,6 @@
# libOpenMMCUDA_static[_d].a
#----------------------------------------------------
IF (APPLE)
SET (CMAKE_OSX_DEPLOYMENT_TARGET "10.6")
ENDIF (APPLE)
set(OPENMM_BUILD_CUDA_TESTS TRUE CACHE BOOL "Whether to build CUDA test cases")
if(OPENMM_BUILD_CUDA_TESTS)
SUBDIRS (tests)
......@@ -39,9 +35,9 @@ SET(STATIC_TARGET ${OPENMMCUDA_LIBRARY_NAME}_static)
# Ensure that debug libraries have "_d" appended to their names.
# CMake gets this right on Windows automatically with this definition.
IF (${CMAKE_GENERATOR} MATCHES "Visual Studio")
IF (MSVC)
SET(CMAKE_DEBUG_POSTFIX "_d" CACHE INTERNAL "" FORCE)
ENDIF (${CMAKE_GENERATOR} MATCHES "Visual Studio")
ENDIF (MSVC)
# But on Unix or Cygwin we have to add the suffix manually
IF (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
......
platforms/cuda/include/CudaContext.h
View file @ 19d2885a
......@@ -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
......@@ -623,6 +623,8 @@ public:
*/
class CudaContext::ForcePreComputation {
public:
virtual ~ForcePreComputation() {
}
/**
* @param includeForce true if forces should be computed
* @param includeEnergy true if potential energy should be computed
......@@ -639,6 +641,8 @@ public:
*/
class CudaContext::ForcePostComputation {
public:
virtual ~ForcePostComputation() {
}
/**
* @param includeForce true if forces should be computed
* @param includeEnergy true if potential energy should be computed
......
platforms/cuda/include/CudaKernels.h
View file @ 19d2885a
......@@ -738,7 +738,7 @@ private:
class CudaCalcCustomGBForceKernel : public CalcCustomGBForceKernel {
public:
CudaCalcCustomGBForceKernel(std::string name, const Platform& platform, CudaContext& cu, const System& system) : CalcCustomGBForceKernel(name, platform),
hasInitializedKernels(false), cu(cu), params(NULL), computedValues(NULL), energyDerivs(NULL), longEnergyDerivs(NULL), globals(NULL),
hasInitializedKernels(false), cu(cu), params(NULL), computedValues(NULL), energyDerivs(NULL), energyDerivChain(NULL), longEnergyDerivs(NULL), globals(NULL),
valueBuffers(NULL), tabulatedFunctionParams(NULL), system(system) {
}
~CudaCalcCustomGBForceKernel();
......@@ -772,6 +772,7 @@ private:
CudaParameterSet* params;
CudaParameterSet* computedValues;
CudaParameterSet* energyDerivs;
CudaParameterSet* energyDerivChain;
CudaArray* longEnergyDerivs;
CudaArray* globals;
CudaArray* valueBuffers;
......
platforms/cuda/include/CudaPlatform.h
View file @ 19d2885a
......@@ -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/sharedTarget/CMakeLists.txt
View file @ 19d2885a
......@@ -19,6 +19,11 @@ ELSE (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
SET(MAIN_OPENMM_LIB ${OPENMM_LIBRARY_NAME})
ENDIF (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${MAIN_OPENMM_LIB} ${CUDA_CUDA_LIBRARY} ${CUDA_cufft_LIBRARY} ${PTHREADS_LIB})
SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES COMPILE_FLAGS "-DOPENMM_CUDA_BUILDING_SHARED_LIBRARY")
SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_CUDA_BUILDING_SHARED_LIBRARY")
IF (APPLE)
SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS} -F/Library/Frameworks -framework CUDA")
ELSE (APPLE)
SET_TARGET_PROPERTIES(${SHARED_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS}")
ENDIF (APPLE)
INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${SHARED_TARGET})
platforms/cuda/src/CudaContext.cpp
View file @ 19d2885a
......@@ -72,9 +72,12 @@ 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),
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) {
this->compiler = "\""+compiler+"\"";
if (hostCompiler.size() > 0)
this->compiler = compiler+" --compiler-bindir "+hostCompiler;
if (!hasInitializedCuda) {
CHECK_RESULT2(cuInit(0), "Error initializing CUDA");
hasInitializedCuda = true;
......@@ -153,9 +156,6 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
CHECK_RESULT(cuDeviceGetAttribute(&multiprocessors, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, device));
int numThreadBlocksPerComputeUnit = 6;
numThreadBlocks = numThreadBlocksPerComputeUnit*multiprocessors;
bonded = new CudaBondedUtilities(*this);
nonbonded = new CudaNonbondedUtilities(*this);
int numEnergyBuffers = max(numThreadBlocks*ThreadBlockSize, nonbonded->getNumEnergyBuffers());
if (useDoublePrecision) {
posq = CudaArray::create<double4>(*this, paddedNumAtoms, "posq");
velm = CudaArray::create<double4>(*this, paddedNumAtoms, "velm");
......@@ -166,9 +166,6 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
compilationDefines["make_mixed2"] = "make_double2";
compilationDefines["make_mixed3"] = "make_double3";
compilationDefines["make_mixed4"] = "make_double4";
energyBuffer = CudaArray::create<double>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*4, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(double), 0));
}
else if (useMixedPrecision) {
posq = CudaArray::create<float4>(*this, paddedNumAtoms, "posq");
......@@ -181,9 +178,6 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
compilationDefines["make_mixed2"] = "make_double2";
compilationDefines["make_mixed3"] = "make_double3";
compilationDefines["make_mixed4"] = "make_double4";
energyBuffer = CudaArray::create<float>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*4, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(double), 0));
}
else {
posq = CudaArray::create<float4>(*this, paddedNumAtoms, "posq");
......@@ -194,9 +188,6 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
compilationDefines["make_mixed2"] = "make_float2";
compilationDefines["make_mixed3"] = "make_float3";
compilationDefines["make_mixed4"] = "make_float4";
energyBuffer = CudaArray::create<float>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*6, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(float), 0));
}
posCellOffsets.resize(paddedNumAtoms, make_int4(0, 0, 0, 0));
......@@ -233,6 +224,8 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
// Create utilities objects.
bonded = new CudaBondedUtilities(*this);
nonbonded = new CudaNonbondedUtilities(*this);
integration = new CudaIntegrationUtilities(*this, system);
expression = new CudaExpressionUtilities(*this);
}
......@@ -280,6 +273,22 @@ CudaContext::~CudaContext() {
void CudaContext::initialize() {
cuCtxSetCurrent(context);
string errorMessage = "Error initializing Context";
int numEnergyBuffers = max(numThreadBlocks*ThreadBlockSize, nonbonded->getNumEnergyBuffers());
if (useDoublePrecision) {
energyBuffer = CudaArray::create<double>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*4, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(double), 0));
}
else if (useMixedPrecision) {
energyBuffer = CudaArray::create<float>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*4, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(double), 0));
}
else {
energyBuffer = CudaArray::create<float>(*this, numEnergyBuffers, "energyBuffer");
int pinnedBufferSize = max(paddedNumAtoms*6, numEnergyBuffers);
CHECK_RESULT(cuMemHostAlloc(&pinnedBuffer, pinnedBufferSize*sizeof(float), 0));
}
for (int i = 0; i < numAtoms; i++) {
double mass = system.getParticleMass(i);
if (useDoublePrecision || useMixedPrecision)
......@@ -441,13 +450,13 @@ CUmodule CudaContext::createModule(const string source, const map<string, string
out.close();
#ifdef WIN32
#ifdef _DEBUG
string command = "\""+compiler+"\" --ptx -G -g --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o "+outputFile+" "+options+" "+inputFile+" 2> "+logFile;
string command = compiler+" --ptx -G -g --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o "+outputFile+" "+options+" "+inputFile+" 2> "+logFile;
#else
string command = "\""+compiler+"\" --ptx -lineinfo --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o "+outputFile+" "+options+" "+inputFile+" 2> "+logFile;
string command = compiler+" --ptx -lineinfo --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o "+outputFile+" "+options+" "+inputFile+" 2> "+logFile;
#endif
int res = compileInWindows(command);
#else
string command = "\""+compiler+"\" --ptx --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o \""+outputFile+"\" "+options+" \""+inputFile+"\" 2> \""+logFile+"\"";
string command = compiler+" --ptx --machine "+bits+" -arch=sm_"+gpuArchitecture+" -o \""+outputFile+"\" "+options+" \""+inputFile+"\" 2> \""+logFile+"\"";
int res = std::system(command.c_str());
#endif
try {
......
platforms/cuda/src/CudaKernels.cpp
View file @ 19d2885a
......@@ -1460,8 +1460,9 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
int numParticles = force.getNumParticles();
sigmaEpsilon = CudaArray::create<float2>(cu, cu.getPaddedNumAtoms(), "sigmaEpsilon");
CudaArray& posq = cu.getPosq();
float4* posqf = (float4*) cu.getPinnedBuffer();
double4* posqd = (double4*) cu.getPinnedBuffer();
vector<double4> temp(posq.getSize());
float4* posqf = (float4*) &temp[0];
double4* posqd = (double4*) &temp[0];
vector<float2> sigmaEpsilonVector(cu.getPaddedNumAtoms(), make_float2(0, 0));
vector<vector<int> > exclusionList(numParticles);
double sumSquaredCharges = 0.0;
......@@ -1486,7 +1487,7 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
exclusionList[exclusions[i].first].push_back(exclusions[i].second);
exclusionList[exclusions[i].second].push_back(exclusions[i].first);
}
posq.upload(cu.getPinnedBuffer());
posq.upload(&temp[0]);
sigmaEpsilon->upload(sigmaEpsilonVector);
bool useCutoff = (force.getNonbondedMethod() != NonbondedForce::NoCutoff);
bool usePeriodic = (force.getNonbondedMethod() != NonbondedForce::NoCutoff && force.getNonbondedMethod() != NonbondedForce::CutoffNonPeriodic);
......@@ -2410,8 +2411,9 @@ void CudaCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOBCF
cu.addAutoclearBuffer(*bornSum);
cu.addAutoclearBuffer(*bornForce);
CudaArray& posq = cu.getPosq();
float4* posqf = (float4*) cu.getPinnedBuffer();
double4* posqd = (double4*) cu.getPinnedBuffer();
vector<double4> temp(posq.getSize());
float4* posqf = (float4*) &temp[0];
double4* posqd = (double4*) &temp[0];
vector<float2> paramsVector(cu.getPaddedNumAtoms(), make_float2(1, 1));
const double dielectricOffset = 0.009;
for (int i = 0; i < force.getNumParticles(); i++) {
......@@ -2424,7 +2426,7 @@ void CudaCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOBCF
else
posqf[i] = make_float4(0, 0, 0, (float) charge);
}
posq.upload(cu.getPinnedBuffer());
posq.upload(&temp[0]);
params->upload(paramsVector);
prefactor = -ONE_4PI_EPS0*((1.0/force.getSoluteDielectric())-(1.0/force.getSolventDielectric()));
bool useCutoff = (force.getNonbondedMethod() != GBSAOBCForce::NoCutoff);
......@@ -2600,6 +2602,8 @@ CudaCalcCustomGBForceKernel::~CudaCalcCustomGBForceKernel() {
delete computedValues;
if (energyDerivs != NULL)
delete energyDerivs;
if (energyDerivChain != NULL)
delete energyDerivChain;
if (longEnergyDerivs != NULL)
delete longEnergyDerivs;
if (globals != NULL)
......@@ -2743,6 +2747,7 @@ void CudaCalcCustomGBForceKernel::initialize(const System& system, const CustomG
}
longEnergyDerivs = CudaArray::create<long long>(cu, force.getNumComputedValues()*cu.getPaddedNumAtoms(), "customGBLongEnergyDerivatives");
energyDerivs = new CudaParameterSet(cu, force.getNumComputedValues(), cu.getPaddedNumAtoms(), "customGBEnergyDerivatives", true);
energyDerivChain = new CudaParameterSet(cu, force.getNumComputedValues(), cu.getPaddedNumAtoms(), "customGBEnergyDerivativeChain", true);
// Create the kernels.
......@@ -3009,6 +3014,11 @@ void CudaCalcCustomGBForceKernel::initialize(const System& system, const CustomG
extraArgs << ", " << buffer.getType() << "* __restrict__ derivBuffers" << index;
compute << buffer.getType() << " deriv" << index << " = derivBuffers" << index << "[index];\n";
}
for (int i = 0; i < (int) energyDerivChain->getBuffers().size(); i++) {
CudaNonbondedUtilities::ParameterInfo& buffer = energyDerivChain->getBuffers()[i];
string index = cu.intToString(i+1);
extraArgs << ", " << buffer.getType() << "* __restrict__ derivChain" << index;
}
extraArgs << ", const long long* __restrict__ derivBuffersIn";
for (int i = 0; i < energyDerivs->getNumParameters(); ++i)
load << "derivBuffers" << energyDerivs->getParameterSuffix(i, "[index]") <<
......@@ -3054,6 +3064,10 @@ void CudaCalcCustomGBForceKernel::initialize(const System& system, const CustomG
// Record values.
for (int i = 0; i < (int) energyDerivs->getBuffers().size(); i++) {
string index = cu.intToString(i+1);
compute << "derivBuffers" << index << "[index] = deriv" << index << ";\n";
}
compute << "forceBuffers[index] += (long long) (force.x*0x100000000);\n";
compute << "forceBuffers[index+PADDED_NUM_ATOMS] += (long long) (force.y*0x100000000);\n";
compute << "forceBuffers[index+PADDED_NUM_ATOMS*2] += (long long) (force.z*0x100000000);\n";
......@@ -3066,7 +3080,7 @@ void CudaCalcCustomGBForceKernel::initialize(const System& system, const CustomG
}
for (int i = 0; i < (int) energyDerivs->getBuffers().size(); i++) {
string index = cu.intToString(i+1);
compute << "derivBuffers" << index << "[index] = deriv" << index << ";\n";
compute << "derivChain" << index << "[index] = deriv" << index << ";\n";
}
map<string, string> replacements;
replacements["PARAMETER_ARGUMENTS"] = extraArgs.str()+tableArgs.str();
......@@ -3204,9 +3218,9 @@ void CudaCalcCustomGBForceKernel::initialize(const System& system, const CustomG
if (chainStr.find(paramName+"1") != chainStr.npos || chainStr.find(paramName+"2") != chainStr.npos)
parameters.push_back(CudaNonbondedUtilities::ParameterInfo(paramName, buffer.getComponentType(), buffer.getNumComponents(), buffer.getSize(), buffer.getMemory()));
}
for (int i = 0; i < (int) energyDerivs->getBuffers().size(); i++) {
for (int i = 0; i < (int) energyDerivChain->getBuffers().size(); i++) {
if (needChainForValue[i]) {
CudaNonbondedUtilities::ParameterInfo& buffer = energyDerivs->getBuffers()[i];
CudaNonbondedUtilities::ParameterInfo& buffer = energyDerivChain->getBuffers()[i];
string paramName = prefix+"dEdV"+cu.intToString(i+1);
parameters.push_back(CudaNonbondedUtilities::ParameterInfo(paramName, buffer.getComponentType(), buffer.getNumComponents(), buffer.getSize(), buffer.getMemory()));
}
......@@ -3352,6 +3366,8 @@ double CudaCalcCustomGBForceKernel::execute(ContextImpl& context, bool includeFo
perParticleEnergyArgs.push_back(&computedValues->getBuffers()[i].getMemory());
for (int i = 0; i < (int) energyDerivs->getBuffers().size(); i++)
perParticleEnergyArgs.push_back(&energyDerivs->getBuffers()[i].getMemory());
for (int i = 0; i < (int) energyDerivChain->getBuffers().size(); i++)
perParticleEnergyArgs.push_back(&energyDerivChain->getBuffers()[i].getMemory());
perParticleEnergyArgs.push_back(&longEnergyDerivs->getDevicePointer());
if (tabulatedFunctionParams != NULL) {
for (int i = 0; i < (int) tabulatedFunctions.size(); i++)
......
platforms/cuda/src/CudaPlatform.cpp
View file @ 19d2885a
......@@ -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;
......@@ -174,15 +179,24 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
searchPos = nextPos+1;
}
devices.push_back(deviceIndexProperty.substr(searchPos));
for (int i = 0; i < (int) devices.size(); i++) {
if (devices[i].length() > 0) {
unsigned int deviceIndex;
stringstream(devices[i]) >> deviceIndex;
contexts.push_back(new CudaContext(system, deviceIndex, blocking, precisionProperty, compilerProperty, tempProperty, *this));
try {
for (int i = 0; i < (int) devices.size(); i++) {
if (devices[i].length() > 0) {
unsigned int deviceIndex;
stringstream(devices[i]) >> deviceIndex;
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, hostCompilerProperty, *this));
}
catch (...) {
// If an exception was thrown, do our best to clean up memory.
for (int i = 0; i < (int) contexts.size(); i++)
delete contexts[i];
throw;
}
if (contexts.size() == 0)
contexts.push_back(new CudaContext(system, -1, blocking, precisionProperty, compilerProperty, tempProperty, *this));
stringstream deviceIndex, deviceName;
for (int i = 0; i < (int) contexts.size(); i++) {
if (i > 0) {
......@@ -202,6 +216,7 @@ 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/src/kernels/customGBEnergyN2.cu
View file @ 19d2885a
......@@ -65,7 +65,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+j;
real dEdR = 0;
......@@ -117,7 +117,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+tj;
real dEdR = 0;
......@@ -268,7 +268,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
real dEdR = 0;
......@@ -313,7 +313,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
real dEdR = 0;
......
platforms/cuda/src/kernels/customGBValueN2.cu
View file @ 19d2885a
......@@ -60,7 +60,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+j;
real tempValue1 = 0;
......@@ -108,7 +108,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+tj;
real tempValue1 = 0;
......@@ -241,7 +241,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
if (r2 < CUTOFF_SQUARED) {
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
real tempValue1 = 0;
......@@ -275,7 +275,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
real tempValue1 = 0;
......
platforms/cuda/src/kernels/customNonbondedGroups.cu
View file @ 19d2885a
......@@ -58,7 +58,7 @@ extern "C" __global__ void computeInteractionGroups(
if (!isExcluded && r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
real dEdR = 0.0f;
real tempEnergy = 0.0f;
......
platforms/cuda/src/kernels/gbsaObc1.cu
View file @ 19d2885a
......@@ -116,7 +116,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
if (atom1 < NUM_ATOMS && y*TILE_SIZE+j < NUM_ATOMS) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
float2 params2 = make_float2(localData[tbx+j].radius, localData[tbx+j].scaledRadius);
real rScaledRadiusJ = r+params2.y;
if ((j != tgx) && (params1.x < rScaledRadiusJ)) {
......@@ -163,7 +163,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
if (atom1 < NUM_ATOMS && y*TILE_SIZE+tj < NUM_ATOMS) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
float2 params2 = make_float2(localData[tbx+tj].radius, localData[tbx+tj].scaledRadius);
real rScaledRadiusJ = r+params2.y;
if (params1.x < rScaledRadiusJ) {
......@@ -305,7 +305,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
int atom2 = atomIndices[tbx+tj];
if (atom1 < NUM_ATOMS && atom2 < NUM_ATOMS && r2 < CUTOFF_SQUARED) {
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
float2 params2 = make_float2(localData[tbx+tj].radius, localData[tbx+tj].scaledRadius);
real rScaledRadiusJ = r+params2.y;
if (params1.x < rScaledRadiusJ) {
......@@ -355,7 +355,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
if (atom1 < NUM_ATOMS && atom2 < NUM_ATOMS) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
float2 params2 = make_float2(localData[tbx+tj].radius, localData[tbx+tj].scaledRadius);
real rScaledRadiusJ = r+params2.y;
if (params1.x < rScaledRadiusJ) {
......@@ -461,7 +461,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
real bornRadius2 = localData[tbx+j].bornRadius;
real alpha2_ij = bornRadius1*bornRadius2;
real D_ij = r2*RECIP(4.0f*alpha2_ij);
......@@ -518,7 +518,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
real bornRadius2 = localData[tbx+tj].bornRadius;
real alpha2_ij = bornRadius1*bornRadius2;
real D_ij = r2*RECIP(4.0f*alpha2_ij);
......@@ -672,7 +672,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
if (r2 < CUTOFF_SQUARED) {
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
real bornRadius2 = localData[tbx+tj].bornRadius;
real alpha2_ij = bornRadius1*bornRadius2;
real D_ij = r2*RECIP(4.0f*alpha2_ij);
......@@ -723,7 +723,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
real bornRadius2 = localData[tbx+tj].bornRadius;
real alpha2_ij = bornRadius1*bornRadius2;
real D_ij = r2*RECIP(4.0f*alpha2_ij);
......
platforms/cuda/src/kernels/nonbonded.cu
View file @ 19d2885a
......@@ -161,7 +161,7 @@ extern "C" __global__ void computeNonbonded(
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+j;
#ifdef USE_SYMMETRIC
......@@ -232,7 +232,7 @@ extern "C" __global__ void computeNonbonded(
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = y*TILE_SIZE+tj;
#ifdef USE_SYMMETRIC
......@@ -433,7 +433,7 @@ extern "C" __global__ void computeNonbonded(
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
if (r2 < CUTOFF_SQUARED) {
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
#ifdef USE_SYMMETRIC
......@@ -507,7 +507,7 @@ extern "C" __global__ void computeNonbonded(
if (r2 < CUTOFF_SQUARED) {
#endif
real invR = RSQRT(r2);
real r = RECIP(invR);
real r = r2*invR;
LOAD_ATOM2_PARAMETERS
atom2 = atomIndices[tbx+tj];
#ifdef USE_SYMMETRIC
......
platforms/cuda/tests/CMakeLists.txt
View file @ 19d2885a
......@@ -25,6 +25,11 @@ FOREACH(TEST_PROG ${TEST_PROGS})
# Link with shared library
ADD_EXECUTABLE(${TEST_ROOT} ${TEST_PROG})
TARGET_LINK_LIBRARIES(${TEST_ROOT} ${SHARED_TARGET})
IF (APPLE)
SET_TARGET_PROPERTIES(${TEST_ROOT} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS} -F/Library/Frameworks -framework CUDA" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS}")
ELSE (APPLE)
SET_TARGET_PROPERTIES(${TEST_ROOT} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS}" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS}")
ENDIF (APPLE)
IF( ${TEST_ROOT} STREQUAL "TestCUDAGBSAOBCForce2" )
......
platforms/cuda/tests/TestCudaCustomGBForce.cpp
View file @ 19d2885a
......@@ -248,7 +248,7 @@ void testMembrane() {
for (int i = 0; i < (int) forces.size(); ++i)
norm += forces[i].dot(forces[i]);
norm = std::sqrt(norm);
const double stepSize = 1e-3;
const double stepSize = 1e-2;
double step = 0.5*stepSize/norm;
vector<Vec3> positions2(numParticles), positions3(numParticles);
for (int i = 0; i < (int) positions.size(); ++i) {
......
platforms/cuda/tests/TestCudaGBSAOBCForce.cpp
View file @ 19d2885a
......@@ -71,7 +71,7 @@ void testSingleParticle() {
double bornRadius = 0.15-0.009; // dielectric offset
double eps0 = EPSILON0;
double bornEnergy = (-0.5*0.5/(8*PI_M*eps0))*(1.0/gbsa->getSoluteDielectric()-1.0/gbsa->getSolventDielectric())/bornRadius;
double extendedRadius = bornRadius+0.14; // probe radius
double extendedRadius = 0.15+0.14; // probe radius
double nonpolarEnergy = CAL2JOULE*PI_M*0.0216*(10*extendedRadius)*(10*extendedRadius)*std::pow(0.15/bornRadius, 6.0); // Where did this formula come from? Just copied it from CpuImplicitSolvent.cpp
ASSERT_EQUAL_TOL((bornEnergy+nonpolarEnergy), state.getPotentialEnergy(), 0.01);
......
platforms/cuda/tests/TestCudaRandom.cpp
View file @ 19d2885a
......@@ -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
View file @ 19d2885a
......@@ -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");
......
platforms/opencl/CMakeLists.txt
View file @ 19d2885a
......@@ -14,10 +14,6 @@
# libOpenMMOpenCL_static[_d].a
#----------------------------------------------------
IF (APPLE)
SET (CMAKE_OSX_DEPLOYMENT_TARGET "10.6")
ENDIF (APPLE)
set(OPENMM_BUILD_OPENCL_TESTS TRUE CACHE BOOL "Whether to build OpenCL test cases")
if(OPENMM_BUILD_OPENCL_TESTS)
SUBDIRS (tests)
......@@ -39,9 +35,9 @@ SET(STATIC_TARGET ${OPENMMOPENCL_LIBRARY_NAME}_static)
# Ensure that debug libraries have "_d" appended to their names.
# CMake gets this right on Windows automatically with this definition.
IF (${CMAKE_GENERATOR} MATCHES "Visual Studio")
IF (MSVC)
SET(CMAKE_DEBUG_POSTFIX "_d" CACHE INTERNAL "" FORCE)
ENDIF (${CMAKE_GENERATOR} MATCHES "Visual Studio")
ENDIF (MSVC)
# But on Unix or Cygwin we have to add the suffix manually
IF (UNIX AND CMAKE_BUILD_TYPE MATCHES Debug)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment