Skip to content

GitLab

Commit 73183c61 authored by ChayaSt's avatar ChayaSt
Browse files

resolved conflict

parents 0e218233 32e08b87
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 328 additions and 201 deletions
platforms/cuda/src/CudaContext.cpp
View file @ 73183c61
......@@ -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) 2009-2015 Stanford University and the Authors. *
* Portions copyright (c) 2009-2016 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -106,7 +106,7 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
useMixedPrecision = false;
}
else
throw OpenMMException("Illegal value for CudaPrecision: "+precision);
throw OpenMMException("Illegal value for Precision: "+precision);
char* cacheVariable = getenv("OPENMM_CACHE_DIR");
cacheDir = (cacheVariable == NULL ? tempDir : string(cacheVariable));
#ifdef WIN32
......@@ -121,7 +121,7 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
string errorMessage = "Error initializing Context";
CHECK_RESULT(cuDeviceGetCount(&numDevices));
if (deviceIndex < -1 || deviceIndex >= numDevices)
throw OpenMMException("Illegal value for CudaDeviceIndex: "+intToString(deviceIndex));
throw OpenMMException("Illegal value for DeviceIndex: "+intToString(deviceIndex));
vector<int> devicePrecedence;
if (deviceIndex == -1) {
......@@ -1131,7 +1131,6 @@ void CudaContext::reorderAtoms() {
reorderAtomsImpl<float, float4, double, double4>();
else
reorderAtomsImpl<float, float4, float, float4>();
nonbonded->updateNeighborListSize();
}
template <class Real, class Real4, class Mixed, class Mixed4>
......
platforms/cuda/src/CudaExpressionUtilities.cpp
View file @ 73183c61
......@@ -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) 2009-2015 Stanford University and the Authors. *
* Portions copyright (c) 2009-2016 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -174,8 +174,8 @@ void CudaExpressionUtilities::processExpression(stringstream& out, const Express
out << "if (x >= " << paramsFloat[2] << " && x <= " << paramsFloat[3] << " && y >= " << paramsFloat[4] << " && y <= " << paramsFloat[5] << ") {\n";
out << "x = (x - " << paramsFloat[2] << ")*" << paramsFloat[6] << ";\n";
out << "y = (y - " << paramsFloat[4] << ")*" << paramsFloat[7] << ";\n";
out << "int s = min((int) floor(x), " << paramsInt[0] << ");\n";
out << "int t = min((int) floor(y), " << paramsInt[1] << ");\n";
out << "int s = min((int) floor(x), " << paramsInt[0] << "-1);\n";
out << "int t = min((int) floor(y), " << paramsInt[1] << "-1);\n";
out << "int coeffIndex = 4*(s+" << paramsInt[0] << "*t);\n";
out << "float4 c[4];\n";
for (int j = 0; j < 4; j++)
......@@ -217,9 +217,9 @@ void CudaExpressionUtilities::processExpression(stringstream& out, const Express
out << "x = (x - " << paramsFloat[3] << ")*" << paramsFloat[9] << ";\n";
out << "y = (y - " << paramsFloat[5] << ")*" << paramsFloat[10] << ";\n";
out << "z = (z - " << paramsFloat[7] << ")*" << paramsFloat[11] << ";\n";
out << "int s = min((int) floor(x), " << paramsInt[0] << ");\n";
out << "int t = min((int) floor(y), " << paramsInt[1] << ");\n";
out << "int u = min((int) floor(z), " << paramsInt[2] << ");\n";
out << "int s = min((int) floor(x), " << paramsInt[0] << "-1);\n";
out << "int t = min((int) floor(y), " << paramsInt[1] << "-1);\n";
out << "int u = min((int) floor(z), " << paramsInt[2] << "-1);\n";
out << "int coeffIndex = 16*(s+" << paramsInt[0] << "*(t+" << paramsInt[1] << "*u));\n";
out << "float4 c[16];\n";
for (int j = 0; j < 16; j++)
......@@ -254,7 +254,7 @@ void CudaExpressionUtilities::processExpression(stringstream& out, const Express
for (int k = 3; k >= 0; k--)
for (int m = 0; m < 4; m++) {
int base = 4*m;
string suffix = suffixes[m];
string suffix = suffixes[k];
out << "derivy[" << m << "] = da*derivy[" << m << "] + (3*c[" << (base+3) << "]" << suffix << "*db + 2*c[" << (base+2) << "]" << suffix << ")*db + c[" << (base+1) << "]" << suffix << ";\n";
}
out << nodeNames[j] << " = derivy[0] + dc*(derivy[1] + dc*(derivy[2] + dc*derivy[3]));\n";
......@@ -271,7 +271,7 @@ void CudaExpressionUtilities::processExpression(stringstream& out, const Express
out << nodeNames[j] << " *= " << paramsFloat[11] << ";\n";
}
else
throw OpenMMException("Unsupported derivative order for Continuous2DFunction");
throw OpenMMException("Unsupported derivative order for Continuous3DFunction");
}
out << "}\n";
}
......
platforms/cuda/src/CudaKernels.cpp
View file @ 73183c61
This diff is collapsed.
platforms/cuda/src/CudaNonbondedUtilities.cpp
View file @ 73183c61
......@@ -65,12 +65,14 @@ private:
CudaNonbondedUtilities::CudaNonbondedUtilities(CudaContext& context) : context(context), useCutoff(false), usePeriodic(false), anyExclusions(false), usePadding(true),
exclusionIndices(NULL), exclusionRowIndices(NULL), exclusionTiles(NULL), exclusions(NULL), interactingTiles(NULL), interactingAtoms(NULL),
interactionCount(NULL), blockCenter(NULL), blockBoundingBox(NULL), sortedBlocks(NULL), sortedBlockCenter(NULL), sortedBlockBoundingBox(NULL),
oldPositions(NULL), rebuildNeighborList(NULL), blockSorter(NULL), forceRebuildNeighborList(true), lastCutoff(0.0), groupFlags(0) {
oldPositions(NULL), rebuildNeighborList(NULL), blockSorter(NULL), pinnedCountBuffer(NULL), forceRebuildNeighborList(true), lastCutoff(0.0), groupFlags(0) {
// Decide how many thread blocks to use.
string errorMessage = "Error initializing nonbonded utilities";
int multiprocessors;
CHECK_RESULT(cuDeviceGetAttribute(&multiprocessors, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, context.getDevice()));
CHECK_RESULT(cuEventCreate(&downloadCountEvent, 0));
CHECK_RESULT(cuMemHostAlloc((void**) &pinnedCountBuffer, sizeof(int), CU_MEMHOSTALLOC_PORTABLE));
numForceThreadBlocks = 4*multiprocessors;
forceThreadBlockSize = (context.getComputeCapability() < 2.0 ? 128 : 256);
}
......@@ -106,6 +108,9 @@ CudaNonbondedUtilities::~CudaNonbondedUtilities() {
delete rebuildNeighborList;
if (blockSorter != NULL)
delete blockSorter;
if (pinnedCountBuffer != NULL)
cuMemFreeHost(pinnedCountBuffer);
cuEventDestroy(downloadCountEvent);
}
void CudaNonbondedUtilities::addInteraction(bool usesCutoff, bool usesPeriodic, bool usesExclusions, double cutoffDistance, const vector<vector<int> >& exclusionList, const string& kernel, int forceGroup) {
......@@ -375,17 +380,14 @@ void CudaNonbondedUtilities::prepareInteractions(int forceGroups) {
if (lastCutoff != kernels.cutoffDistance)
forceRebuildNeighborList = true;
bool rebuild = false;
do {
context.executeKernel(kernels.findBlockBoundsKernel, &findBlockBoundsArgs[0], context.getNumAtoms());
blockSorter->sort(*sortedBlocks);
context.executeKernel(kernels.sortBoxDataKernel, &sortBoxDataArgs[0], context.getNumAtoms());
context.executeKernel(kernels.findInteractingBlocksKernel, &findInteractingBlocksArgs[0], context.getNumAtoms(), 256);
forceRebuildNeighborList = false;
if (context.getComputeForceCount() == 1)
rebuild = updateNeighborListSize(); // This is the first time step, so check whether our initial guess was large enough.
} while(rebuild);
context.executeKernel(kernels.findBlockBoundsKernel, &findBlockBoundsArgs[0], context.getNumAtoms());
blockSorter->sort(*sortedBlocks);
context.executeKernel(kernels.sortBoxDataKernel, &sortBoxDataArgs[0], context.getNumAtoms());
context.executeKernel(kernels.findInteractingBlocksKernel, &findInteractingBlocksArgs[0], context.getNumAtoms(), 256);
forceRebuildNeighborList = false;
lastCutoff = kernels.cutoffDistance;
interactionCount->download(pinnedCountBuffer, false);
cuEventRecord(downloadCountEvent, context.getCurrentStream());
}
void CudaNonbondedUtilities::computeInteractions(int forceGroups, bool includeForces, bool includeEnergy) {
......@@ -398,20 +400,22 @@ void CudaNonbondedUtilities::computeInteractions(int forceGroups, bool includeFo
kernel = createInteractionKernel(kernels.source, parameters, arguments, true, true, forceGroups, includeForces, includeEnergy);
context.executeKernel(kernel, &forceArgs[0], numForceThreadBlocks*forceThreadBlockSize, forceThreadBlockSize);
}
if (useCutoff && numTiles > 0) {
cuEventSynchronize(downloadCountEvent);
updateNeighborListSize();
}
}
bool CudaNonbondedUtilities::updateNeighborListSize() {
if (!useCutoff)
return false;
unsigned int* pinnedInteractionCount = (unsigned int*) context.getPinnedBuffer();
interactionCount->download(pinnedInteractionCount);
if (pinnedInteractionCount[0] <= (unsigned int) maxTiles)
if (pinnedCountBuffer[0] <= (unsigned int) maxTiles)
return false;
// The most recent timestep had too many interactions to fit in the arrays. Make the arrays bigger to prevent
// this from happening in the future.
maxTiles = (int) (1.2*pinnedInteractionCount[0]);
maxTiles = (int) (1.2*pinnedCountBuffer[0]);
int totalTiles = context.getNumAtomBlocks()*(context.getNumAtomBlocks()+1)/2;
if (maxTiles > totalTiles)
maxTiles = totalTiles;
......@@ -428,6 +432,7 @@ bool CudaNonbondedUtilities::updateNeighborListSize() {
forceArgs[17] = &interactingAtoms->getDevicePointer();
findInteractingBlocksArgs[7] = &interactingAtoms->getDevicePointer();
forceRebuildNeighborList = true;
context.setForcesValid(false);
return true;
}
......
platforms/cuda/src/CudaPlatform.cpp
View file @ 73183c61
......@@ -62,6 +62,14 @@ extern "C" OPENMM_EXPORT_CUDA void registerPlatforms() {
#endif
CudaPlatform::CudaPlatform() {
deprecatedPropertyReplacements["CudaDeviceIndex"] = CudaDeviceIndex();
deprecatedPropertyReplacements["CudaDeviceName"] = CudaDeviceName();
deprecatedPropertyReplacements["CudaUseBlockingSync"] = CudaUseBlockingSync();
deprecatedPropertyReplacements["CudaPrecision"] = CudaPrecision();
deprecatedPropertyReplacements["CudaUseCpuPme"] = CudaUseCpuPme();
deprecatedPropertyReplacements["CudaTempDirectory"] = CudaTempDirectory();
deprecatedPropertyReplacements["CudaDisablePmeStream"] = CudaDisablePmeStream();
deprecatedPropertyReplacements["CudaDeterministicForces"] = CudaDeterministicForces();
CudaKernelFactory* factory = new CudaKernelFactory();
registerKernelFactory(CalcForcesAndEnergyKernel::Name(), factory);
registerKernelFactory(UpdateStateDataKernel::Name(), factory);
......@@ -102,12 +110,14 @@ CudaPlatform::CudaPlatform() {
platformProperties.push_back(CudaTempDirectory());
platformProperties.push_back(CudaHostCompiler());
platformProperties.push_back(CudaDisablePmeStream());
platformProperties.push_back(CudaDeterministicForces());
setPropertyDefaultValue(CudaDeviceIndex(), "");
setPropertyDefaultValue(CudaDeviceName(), "");
setPropertyDefaultValue(CudaUseBlockingSync(), "true");
setPropertyDefaultValue(CudaPrecision(), "single");
setPropertyDefaultValue(CudaUseCpuPme(), "false");
setPropertyDefaultValue(CudaDisablePmeStream(), "false");
setPropertyDefaultValue(CudaDeterministicForces(), "false");
#ifdef _MSC_VER
char* bindir = getenv("CUDA_BIN_PATH");
string nvcc = (bindir == NULL ? "nvcc.exe" : string(bindir)+"\\nvcc.exe");
......@@ -142,7 +152,10 @@ bool CudaPlatform::supportsDoublePrecision() const {
const string& CudaPlatform::getPropertyValue(const Context& context, const string& property) const {
const ContextImpl& impl = getContextImpl(context);
const PlatformData* data = reinterpret_cast<const PlatformData*>(impl.getPlatformData());
map<string, string>::const_iterator value = data->propertyValues.find(property);
string propertyName = property;
if (deprecatedPropertyReplacements.find(property) != deprecatedPropertyReplacements.end())
propertyName = deprecatedPropertyReplacements.find(property)->second;
map<string, string>::const_iterator value = data->propertyValues.find(propertyName);
if (value != data->propertyValues.end())
return value->second;
return Platform::getPropertyValue(context, property);
......@@ -168,10 +181,13 @@ void CudaPlatform::contextCreated(ContextImpl& context, const map<string, string
getPropertyDefaultValue(CudaHostCompiler()) : properties.find(CudaHostCompiler())->second);
string pmeStreamPropValue = (properties.find(CudaDisablePmeStream()) == properties.end() ?
getPropertyDefaultValue(CudaDisablePmeStream()) : properties.find(CudaDisablePmeStream())->second);
string deterministicForcesValue = (properties.find(CudaDeterministicForces()) == properties.end() ?
getPropertyDefaultValue(CudaDeterministicForces()) : properties.find(CudaDeterministicForces())->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);
transform(pmeStreamPropValue.begin(), pmeStreamPropValue.end(), pmeStreamPropValue.begin(), ::tolower);
transform(deterministicForcesValue.begin(), deterministicForcesValue.end(), deterministicForcesValue.begin(), ::tolower);
vector<string> pmeKernelName;
pmeKernelName.push_back(CalcPmeReciprocalForceKernel::Name());
if (!supportsKernels(pmeKernelName))
......@@ -180,7 +196,8 @@ void CudaPlatform::contextCreated(ContextImpl& context, const map<string, string
char* threadsEnv = getenv("OPENMM_CPU_THREADS");
if (threadsEnv != NULL)
stringstream(threadsEnv) >> threads;
context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue, hostCompilerPropValue, pmeStreamPropValue, threads));
context.setPlatformData(new PlatformData(&context, context.getSystem(), devicePropValue, blockingPropValue, precisionPropValue, cpuPmePropValue, compilerPropValue, tempPropValue,
hostCompilerPropValue, pmeStreamPropValue, deterministicForcesValue, threads));
}
void CudaPlatform::contextDestroyed(ContextImpl& context) const {
......@@ -189,7 +206,8 @@ 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, int numThreads) :
const string& cpuPmeProperty, const string& compilerProperty, const string& tempProperty, const string& hostCompilerProperty, const string& pmeStreamProperty,
const string& deterministicForcesProperty, int numThreads) :
context(context), removeCM(false), stepCount(0), computeForceCount(0), time(0.0), hasInitializedContexts(false), threads(numThreads) {
bool blocking = (blockingProperty == "true");
vector<string> devices;
......@@ -230,6 +248,7 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
}
useCpuPme = (cpuPmeProperty == "true" && !contexts[0]->getUseDoublePrecision());
disablePmeStream = (pmeStreamProperty == "true");
deterministicForces = (deterministicForcesProperty == "true");
propertyValues[CudaPlatform::CudaDeviceIndex()] = deviceIndex.str();
propertyValues[CudaPlatform::CudaDeviceName()] = deviceName.str();
propertyValues[CudaPlatform::CudaUseBlockingSync()] = blocking ? "true" : "false";
......@@ -239,6 +258,7 @@ CudaPlatform::PlatformData::PlatformData(ContextImpl* context, const System& sys
propertyValues[CudaPlatform::CudaTempDirectory()] = tempProperty;
propertyValues[CudaPlatform::CudaHostCompiler()] = hostCompilerProperty;
propertyValues[CudaPlatform::CudaDisablePmeStream()] = disablePmeStream ? "true" : "false";
propertyValues[CudaPlatform::CudaDeterministicForces()] = deterministicForces ? "true" : "false";
contextEnergy.resize(contexts.size());
// Determine whether peer-to-peer copying is supported, and enable it if so.
......
platforms/cuda/src/kernels/angleForce.cu
View file @ 73183c61
real3 v0 = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
real3 v1 = make_real3(pos2.x-pos3.x, pos2.y-pos3.y, pos2.z-pos3.z);
#if APPLY_PERIODIC
APPLY_PERIODIC_TO_DELTA(v0)
APPLY_PERIODIC_TO_DELTA(v1)
#endif
real3 cp = cross(v0, v1);
real rp = cp.x*cp.x + cp.y*cp.y + cp.z*cp.z;
rp = max(SQRT(rp), (real) 1.0e-06f);
......
platforms/cuda/src/kernels/bondForce.cu
View file @ 73183c61
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#if APPLY_PERIODIC
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r = SQRT(delta.x*delta.x + delta.y*delta.y + delta.z*delta.z);
COMPUTE_FORCE
dEdR = (r > 0) ? (dEdR / r) : 0;
......
platforms/cuda/src/kernels/cmapTorsionForce.cu
View file @ 73183c61
......@@ -5,6 +5,11 @@ const real PI = (real) 3.14159265358979323846;
real3 v0a = make_real3(pos1.x-pos2.x, pos1.y-pos2.y, pos1.z-pos2.z);
real3 v1a = make_real3(pos3.x-pos2.x, pos3.y-pos2.y, pos3.z-pos2.z);
real3 v2a = make_real3(pos3.x-pos4.x, pos3.y-pos4.y, pos3.z-pos4.z);
#if APPLY_PERIODIC
APPLY_PERIODIC_TO_DELTA(v0a)
APPLY_PERIODIC_TO_DELTA(v1a)
APPLY_PERIODIC_TO_DELTA(v2a)
#endif
real3 cp0a = cross(v0a, v1a);
real3 cp1a = cross(v1a, v2a);
real cosangle = dot(normalize(cp0a), normalize(cp1a));
......@@ -28,6 +33,11 @@ angleA = fmod(angleA+2.0f*PI, 2.0f*PI);
real3 v0b = make_real3(pos5.x-pos6.x, pos5.y-pos6.y, pos5.z-pos6.z);
real3 v1b = make_real3(pos7.x-pos6.x, pos7.y-pos6.y, pos7.z-pos6.z);
real3 v2b = make_real3(pos7.x-pos8.x, pos7.y-pos8.y, pos7.z-pos8.z);
#if APPLY_PERIODIC
APPLY_PERIODIC_TO_DELTA(v0b)
APPLY_PERIODIC_TO_DELTA(v1b)
APPLY_PERIODIC_TO_DELTA(v2b)
#endif
real3 cp0b = cross(v0b, v1b);
real3 cp1b = cross(v1b, v2b);
cosangle = dot(normalize(cp0b), normalize(cp1b));
......
platforms/cuda/src/kernels/customCentroidBond.cu
View file @ 73183c61
......@@ -66,8 +66,11 @@ inline __device__ real3 trim(real4 v) {
/**
* Compute the difference between two vectors, setting the fourth component to the squared magnitude.
*/
inline __device__ real4 delta(real4 vec1, real4 vec2) {
inline __device__ real4 delta(real4 vec1, real4 vec2, bool periodic, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = make_real4(vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0);
if (periodic)
APPLY_PERIODIC_TO_DELTA(result);
result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result;
}
......@@ -105,7 +108,7 @@ inline __device__ real4 computeCross(real4 vec1, real4 vec2) {
* Compute the forces on groups based on the bonds.
*/
extern "C" __global__ void computeGroupForces(unsigned long long* __restrict__ groupForce, mixed* __restrict__ energyBuffer, const real4* __restrict__ centerPositions,
const int* __restrict__ bondGroups
const int* __restrict__ bondGroups, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
EXTRA_ARGS) {
mixed energy = 0;
for (int index = blockIdx.x*blockDim.x+threadIdx.x; index < NUM_BONDS; index += blockDim.x*gridDim.x) {
......
platforms/cuda/src/kernels/customCompoundBond.cu
View file @ 73183c61
......@@ -8,8 +8,11 @@ inline __device__ real3 ccb_trim(real4 v) {
/**
* Compute the difference between two vectors, setting the fourth component to the squared magnitude.
*/
inline __device__ real4 ccb_delta(real4 vec1, real4 vec2) {
inline __device__ real4 ccb_delta(real4 vec1, real4 vec2, bool periodic, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = make_real4(vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0);
if (periodic)
APPLY_PERIODIC_TO_DELTA(result);
result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result;
}
......
platforms/cuda/src/kernels/customGBEnergyN2.cu
View file @ 73183c61
......@@ -168,6 +168,8 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (warp*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
int end = (int) ((warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
#else
......@@ -191,39 +193,35 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
int x, y;
bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
}
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
#else
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
// Skip over tiles that have exclusions, since they were already processed.
// Skip over tiles that have exclusions, since they were already processed.
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
#endif
if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx;
......
platforms/cuda/src/kernels/customGBValueN2.cu
View file @ 73183c61
......@@ -146,6 +146,8 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (warp*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
int end = (int) ((warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
#else
......@@ -167,39 +169,35 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
int x, y;
bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
}
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
#else
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
// Skip over tiles that have exclusions, since they were already processed.
// Skip over tiles that have exclusions, since they were already processed.
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
#endif
if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx;
......
platforms/cuda/src/kernels/gbsaObc1.cu
View file @ 73183c61
......@@ -204,6 +204,8 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (warp*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
int end = (int) ((warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
#else
......@@ -225,39 +227,35 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
int x, y;
bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
}
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
#else
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
// Skip over tiles that have exclusions, since they were already processed.
// Skip over tiles that have exclusions, since they were already processed.
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
#endif
if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx;
......@@ -559,6 +557,8 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (warp*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
int end = (int) ((warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*((long long)NUM_BLOCKS+1)/2 : (long)numTiles)/totalWarps);
#else
......@@ -580,39 +580,35 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
int x, y;
bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
}
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= CUTOFF);
#else
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
// Skip over tiles that have exclusions, since they were already processed.
// Skip over tiles that have exclusions, since they were already processed.
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
#endif
if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx;
......
platforms/cuda/src/kernels/nonbonded.cu
View file @ 73183c61
......@@ -117,7 +117,9 @@ extern "C" __global__ void computeNonbonded(
#ifndef ENABLE_SHUFFLE
__shared__ AtomData localData[THREAD_BLOCK_SIZE];
#endif
// First loop: process tiles that contain exclusions.
const unsigned int firstExclusionTile = FIRST_EXCLUSION_TILE+warp*(LAST_EXCLUSION_TILE-FIRST_EXCLUSION_TILE)/totalWarps;
const unsigned int lastExclusionTile = FIRST_EXCLUSION_TILE+(warp+1)*(LAST_EXCLUSION_TILE-FIRST_EXCLUSION_TILE)/totalWarps;
for (int pos = firstExclusionTile; pos < lastExclusionTile; pos++) {
......@@ -309,8 +311,11 @@ extern "C" __global__ void computeNonbonded(
// Second loop: tiles without exclusions, either from the neighbor list (with cutoff) or just enumerating all
// of them (no cutoff).
#ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps);
#else
......@@ -335,39 +340,35 @@ extern "C" __global__ void computeNonbonded(
int x, y;
bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF);
}
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = tiles[pos];
real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF);
#else
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y += (x < y ? -1 : 1);
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
// Skip over tiles that have exclusions, since they were already processed.
// Skip over tiles that have exclusions, since they were already processed.
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
while (skipTiles[tbx+TILE_SIZE-1] < pos) {
if (skipBase+tgx < NUM_TILES_WITH_EXCLUSIONS) {
ushort2 tile = exclusionTiles[skipBase+tgx];
skipTiles[threadIdx.x] = tile.x + tile.y*NUM_BLOCKS - tile.y*(tile.y+1)/2;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
else
skipTiles[threadIdx.x] = end;
skipBase += TILE_SIZE;
currentSkipIndex = tbx;
}
while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos);
#endif
if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx;
// Load atom data for this tile.
......
platforms/cuda/src/kernels/pme.cu
View file @ 73183c61
extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int2* __restrict__ pmeAtomGridIndex,
real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ) {
real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ) {
// Compute the index of the grid point each atom is associated with.
for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) {
real4 pos = posq[i];
APPLY_PERIODIC_TO_POS(pos)
real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
pos.z*recipBoxVecZ.z);
......@@ -18,7 +20,8 @@ extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int
}
extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real* __restrict__ originalPmeGrid,
real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real3 data[PME_ORDER];
const real scale = RECIP(PME_ORDER-1);
......@@ -28,9 +31,7 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real
for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) {
int atom = pmeAtomGridIndex[i].x;
real4 pos = posq[atom];
pos.x -= floor(pos.x*recipBoxVecX.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*recipBoxVecY.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*recipBoxVecZ.z)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS(pos)
real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
pos.z*recipBoxVecZ.z);
......@@ -83,7 +84,7 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real
#ifdef USE_DOUBLE_PRECISION
unsigned long long * ulonglong_p = (unsigned long long *) originalPmeGrid;
atomicAdd(&ulonglong_p[index], static_cast<unsigned long long>((long long) (add*0x100000000)));
#elif __CUDA_ARCH__ < 200
#elif __CUDA_ARCH__ < 200 || defined(USE_DETERMINISTIC_FORCES)
unsigned long long * ulonglong_p = (unsigned long long *) originalPmeGrid;
int gridIndex = index;
gridIndex = (gridIndex%2 == 0 ? gridIndex/2 : (gridIndex+GRID_SIZE_X*GRID_SIZE_Y*GRID_SIZE_Z)/2);
......@@ -197,7 +198,8 @@ gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, mixed* __restrict__
extern "C" __global__
void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __restrict__ forceBuffers, const real* __restrict__ originalPmeGrid,
real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real3 data[PME_ORDER];
real3 ddata[PME_ORDER];
const real scale = RECIP(PME_ORDER-1);
......@@ -209,9 +211,7 @@ void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __
int atom = pmeAtomGridIndex[i].x;
real3 force = make_real3(0);
real4 pos = posq[atom];
pos.x -= floor(pos.x*recipBoxVecX.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*recipBoxVecY.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*recipBoxVecZ.z)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS(pos)
real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
pos.z*recipBoxVecZ.z);
......
platforms/cuda/src/kernels/torsionForce.cu
View file @ 73183c61
......@@ -2,6 +2,11 @@ const real PI = (real) 3.14159265358979323846;
real3 v0 = make_real3(pos1.x-pos2.x, pos1.y-pos2.y, pos1.z-pos2.z);
real3 v1 = make_real3(pos3.x-pos2.x, pos3.y-pos2.y, pos3.z-pos2.z);
real3 v2 = make_real3(pos3.x-pos4.x, pos3.y-pos4.y, pos3.z-pos4.z);
#if APPLY_PERIODIC
APPLY_PERIODIC_TO_DELTA(v0)
APPLY_PERIODIC_TO_DELTA(v1)
APPLY_PERIODIC_TO_DELTA(v2)
#endif
real3 cp0 = cross(v0, v1);
real3 cp1 = cross(v1, v2);
real cosangle = dot(normalize(cp0), normalize(cp1));
......
platforms/cuda/tests/CudaTests.h
View file @ 73183c61
......@@ -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) 2015 Stanford University and the Authors. *
* Portions copyright (c) 2015-2016 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -39,5 +39,5 @@ OpenMM::CudaPlatform platform;
void initializeTests(int argc, char* argv[]) {
if (argc > 1)
platform.setPropertyDefaultValue("CudaPrecision", std::string(argv[1]));
platform.setPropertyDefaultValue("Precision", std::string(argv[1]));
}
platforms/cuda/tests/TestCudaFFT3D.cpp
View file @ 73183c61
......@@ -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()), 1);
platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1);
CudaContext& context = *platformData.contexts[0];
context.initialize();
OpenMM_SFMT::SFMT sfmt;
......
platforms/cuda/tests/TestCudaNonbondedForce.cpp
View file @ 73183c61
......@@ -118,9 +118,44 @@ void testReordering() {
}
}
void testDeterministicForces() {
// Check that the CudaDeterministicForces property works correctly.
const int numParticles = 1000;
System system;
system.setDefaultPeriodicBoxVectors(Vec3(6, 0, 0), Vec3(2.1, 6, 0), Vec3(-1.5, -0.5, 6));
NonbondedForce *nonbonded = new NonbondedForce();
nonbonded->setNonbondedMethod(NonbondedForce::PME);
system.addForce(nonbonded);
vector<Vec3> positions;
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int i = 0; i < numParticles; i++) {
system.addParticle(1.0);
nonbonded->addParticle(i%2 == 0 ? 1 : -1, 1, 0);
positions.push_back(Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5)*6);
}
VerletIntegrator integrator(0.001);
map<string, string> properties;
properties[CudaPlatform::CudaDeterministicForces()] = "true";
Context context(system, integrator, platform, properties);
context.setPositions(positions);
State state1 = context.getState(State::Forces);
State state2 = context.getState(State::Forces);
// All forces should be *exactly* equal.
for (int i = 0; i < numParticles; i++) {
ASSERT_EQUAL(state1.getForces()[i][0], state2.getForces()[i][0]);
ASSERT_EQUAL(state1.getForces()[i][1], state2.getForces()[i][1]);
ASSERT_EQUAL(state1.getForces()[i][2], state2.getForces()[i][2]);
}
}
void runPlatformTests() {
testParallelComputation(NonbondedForce::NoCutoff);
testParallelComputation(NonbondedForce::Ewald);
testParallelComputation(NonbondedForce::PME);
testReordering();
testDeterministicForces();
}
platforms/cuda/tests/TestCudaRandom.cpp
View file @ 73183c61
......@@ -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()), 1);
platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1);
CudaContext& context = *platformData.contexts[0];
context.initialize();
context.getIntegrationUtilities().initRandomNumberGenerator(0);
......
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