Skip to content

GitLab

Commit 839f9a81 authored by Peter Eastman's avatar Peter Eastman
Browse files

Finished changes to reduce memory use for large systems

parent a8fe9cea
Hide whitespace changes
Inline Side-by-side
Showing with 264 additions and 154 deletions
platforms/opencl/src/OpenCLKernels.cpp
View file @ 839f9a81
...@@ -1691,6 +1691,8 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1691,6 +1691,8 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF
defines["NUM_ATOMS"] = intToString(cl.getNumAtoms()); defines["NUM_ATOMS"] = intToString(cl.getNumAtoms());
defines["PADDED_NUM_ATOMS"] = intToString(cl.getPaddedNumAtoms()); defines["PADDED_NUM_ATOMS"] = intToString(cl.getPaddedNumAtoms());
defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(cl.getNumAtomBlocks()); defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(cl.getNumAtomBlocks());
if (nb.getUseCutoff())
defines["MAX_TILES"] = OpenCLExpressionUtilities::intToString(nb.getInteractingTiles().getSize());
string file = (cl.getSIMDWidth() == 32 ? OpenCLKernelSources::gbsaObc_nvidia : OpenCLKernelSources::gbsaObc_default); string file = (cl.getSIMDWidth() == 32 ? OpenCLKernelSources::gbsaObc_nvidia : OpenCLKernelSources::gbsaObc_default);
cl::Program program = cl.createProgram(file, defines); cl::Program program = cl.createProgram(file, defines);
int index = 0; int index = 0;
...@@ -1702,8 +1704,9 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1702,8 +1704,9 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF
computeBornSumKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float), NULL); computeBornSumKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float), NULL);
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
computeBornSumKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer()); computeBornSumKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer());
computeBornSumKernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
computeBornSumKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer()); computeBornSumKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer());
if (cl.getSIMDWidth() == 32)
computeBornSumKernel.setArg<cl::Buffer>(index+2, nb.getInteractionFlags().getDeviceBuffer());
} }
else else
computeBornSumKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2); computeBornSumKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2);
...@@ -1718,8 +1721,9 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1718,8 +1721,9 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF
force1Kernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(mm_float4), NULL); force1Kernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(mm_float4), NULL);
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
force1Kernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer()); force1Kernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer());
force1Kernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
force1Kernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer()); force1Kernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer());
if (cl.getSIMDWidth() == 32)
force1Kernel.setArg<cl::Buffer>(index+2, nb.getInteractionFlags().getDeviceBuffer());
} }
else else
force1Kernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2); force1Kernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2);
...@@ -1744,10 +1748,10 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1744,10 +1748,10 @@ double OpenCLCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF
reduceBornForceKernel.setArg<cl::Buffer>(6, obcChain->getDeviceBuffer()); reduceBornForceKernel.setArg<cl::Buffer>(6, obcChain->getDeviceBuffer());
} }
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
computeBornSumKernel.setArg<mm_float4>(8, cl.getPeriodicBoxSize()); computeBornSumKernel.setArg<mm_float4>(7, cl.getPeriodicBoxSize());
computeBornSumKernel.setArg<mm_float4>(9, cl.getInvPeriodicBoxSize()); computeBornSumKernel.setArg<mm_float4>(8, cl.getInvPeriodicBoxSize());
force1Kernel.setArg<mm_float4>(10, cl.getPeriodicBoxSize()); force1Kernel.setArg<mm_float4>(9, cl.getPeriodicBoxSize());
force1Kernel.setArg<mm_float4>(11, cl.getInvPeriodicBoxSize()); force1Kernel.setArg<mm_float4>(10, cl.getInvPeriodicBoxSize());
} }
int numTiles = cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2; int numTiles = cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2;
cl.executeKernel(computeBornSumKernel, numTiles*OpenCLContext::TileSize); cl.executeKernel(computeBornSumKernel, numTiles*OpenCLContext::TileSize);
...@@ -2399,9 +2403,11 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include ...@@ -2399,9 +2403,11 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include
pairValueKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float), NULL); pairValueKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float), NULL);
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer()); pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer());
pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer()); pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer());
index += 2; // Periodic box size arguments are set when the kernel is executed. index += 2; // Periodic box size arguments are set when the kernel is executed.
pairValueKernel.setArg<cl_uint>(index++, nb.getInteractingTiles().getSize());
if (cl.getSIMDWidth() == 32)
pairValueKernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
} }
else else
pairValueKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2); pairValueKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2);
...@@ -2445,9 +2451,11 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include ...@@ -2445,9 +2451,11 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include
pairEnergyKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float4), NULL); pairEnergyKernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float4), NULL);
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer()); pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractingTiles().getDeviceBuffer());
pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer()); pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractionCount().getDeviceBuffer());
index += 2; // Periodic box size arguments are set when the kernel is executed. index += 2; // Periodic box size arguments are set when the kernel is executed.
pairEnergyKernel.setArg<cl_uint>(index++, nb.getInteractingTiles().getSize());
if (cl.getSIMDWidth() == 32)
pairEnergyKernel.setArg<cl::Buffer>(index++, nb.getInteractionFlags().getDeviceBuffer());
} }
else else
pairEnergyKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2); pairEnergyKernel.setArg<cl_uint>(index++, cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2);
...@@ -2518,10 +2526,10 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include ...@@ -2518,10 +2526,10 @@ double OpenCLCalcCustomGBForceKernel::execute(ContextImpl& context, bool include
globals->upload(globalParamValues); globals->upload(globalParamValues);
} }
if (nb.getUseCutoff()) { if (nb.getUseCutoff()) {
pairValueKernel.setArg<mm_float4>(11, cl.getPeriodicBoxSize()); pairValueKernel.setArg<mm_float4>(10, cl.getPeriodicBoxSize());
pairValueKernel.setArg<mm_float4>(12, cl.getInvPeriodicBoxSize()); pairValueKernel.setArg<mm_float4>(11, cl.getInvPeriodicBoxSize());
pairEnergyKernel.setArg<mm_float4>(12, cl.getPeriodicBoxSize()); pairEnergyKernel.setArg<mm_float4>(11, cl.getPeriodicBoxSize());
pairEnergyKernel.setArg<mm_float4>(13, cl.getInvPeriodicBoxSize()); pairEnergyKernel.setArg<mm_float4>(12, cl.getInvPeriodicBoxSize());
} }
int numTiles = cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2; int numTiles = cl.getNumAtomBlocks()*(cl.getNumAtomBlocks()+1)/2;
cl.executeKernel(pairValueKernel, numTiles*OpenCLContext::TileSize); cl.executeKernel(pairValueKernel, numTiles*OpenCLContext::TileSize);
......
platforms/opencl/src/OpenCLNonbondedUtilities.cpp
View file @ 839f9a81
...@@ -200,8 +200,18 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -200,8 +200,18 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
// Create data structures for the neighbor list. // Create data structures for the neighbor list.
if (useCutoff) { if (useCutoff) {
interactingTiles = new OpenCLArray<mm_ushort2>(context, numTiles, "interactingTiles"); // Select a size for the arrays that hold the neighbor list. This estimate is intentionally very
interactionFlags = new OpenCLArray<cl_uint>(context, numTiles, "interactionFlags"); // high, because if it ever is too small, we have to fall back to the N^2 algorithm.
mm_float4 boxSize = context.getPeriodicBoxSize();
int maxInteractingTiles = (int) (numTiles*(cutoff/boxSize.x+cutoff/boxSize.y+cutoff/boxSize.z));
if (maxInteractingTiles < 1)
maxInteractingTiles = 1;
if (maxInteractingTiles > numTiles)
maxInteractingTiles = numTiles;
interactingTiles = new OpenCLArray<mm_ushort2>(context, maxInteractingTiles, "interactingTiles");
if (context.getSIMDWidth() == 32)
interactionFlags = new OpenCLArray<cl_uint>(context, maxInteractingTiles, "interactionFlags");
interactionCount = new OpenCLArray<cl_uint>(context, 1, "interactionCount"); interactionCount = new OpenCLArray<cl_uint>(context, 1, "interactionCount");
blockCenter = new OpenCLArray<mm_float4>(context, numAtomBlocks, "blockCenter"); blockCenter = new OpenCLArray<mm_float4>(context, numAtomBlocks, "blockCenter");
blockBoundingBox = new OpenCLArray<mm_float4>(context, numAtomBlocks, "blockBoundingBox"); blockBoundingBox = new OpenCLArray<mm_float4>(context, numAtomBlocks, "blockBoundingBox");
...@@ -213,6 +223,7 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -213,6 +223,7 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
if (useCutoff) { if (useCutoff) {
map<string, string> defines; map<string, string> defines;
defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(context.getNumAtomBlocks()); defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(context.getNumAtomBlocks());
defines["MAX_TILES"] = OpenCLExpressionUtilities::intToString(interactingTiles->getSize());
if (forceBufferPerAtomBlock) if (forceBufferPerAtomBlock)
defines["USE_OUTPUT_BUFFER_PER_BLOCK"] = "1"; defines["USE_OUTPUT_BUFFER_PER_BLOCK"] = "1";
if (usePeriodic) if (usePeriodic)
...@@ -231,15 +242,17 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -231,15 +242,17 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
findInteractingBlocksKernel.setArg<cl::Buffer>(5, interactionCount->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(5, interactionCount->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(6, interactingTiles->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(6, interactingTiles->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(7, context.getPosq().getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(7, context.getPosq().getDeviceBuffer());
findInteractionsWithinBlocksKernel = cl::Kernel(interactingBlocksProgram, "findInteractionsWithinBlocks"); if (context.getSIMDWidth() == 32) {
findInteractionsWithinBlocksKernel.setArg<cl_float>(0, (cl_float) (cutoff*cutoff)); findInteractionsWithinBlocksKernel = cl::Kernel(interactingBlocksProgram, "findInteractionsWithinBlocks");
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(3, context.getPosq().getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl_float>(0, (cl_float) (cutoff*cutoff));
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(4, interactingTiles->getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(3, context.getPosq().getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(5, blockCenter->getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(4, interactingTiles->getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(6, blockBoundingBox->getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(5, blockCenter->getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(7, interactionFlags->getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(6, blockBoundingBox->getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(8, interactionCount->getDeviceBuffer()); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(7, interactionFlags->getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg(9, OpenCLContext::ThreadBlockSize*sizeof(cl_uint), NULL); findInteractionsWithinBlocksKernel.setArg<cl::Buffer>(8, interactionCount->getDeviceBuffer());
findInteractionsWithinBlocksKernel.setArg(9, OpenCLContext::ThreadBlockSize*sizeof(cl_uint), NULL);
}
} }
} }
...@@ -264,6 +277,8 @@ void OpenCLNonbondedUtilities::prepareInteractions() { ...@@ -264,6 +277,8 @@ void OpenCLNonbondedUtilities::prepareInteractions() {
findInteractingBlocksKernel.setArg<mm_float4>(1, context.getPeriodicBoxSize()); findInteractingBlocksKernel.setArg<mm_float4>(1, context.getPeriodicBoxSize());
findInteractingBlocksKernel.setArg<mm_float4>(2, context.getInvPeriodicBoxSize()); findInteractingBlocksKernel.setArg<mm_float4>(2, context.getInvPeriodicBoxSize());
context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms()); context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms());
vector<cl_uint> count;
interactionCount->download(count);
if (context.getSIMDWidth() == 32) { if (context.getSIMDWidth() == 32) {
findInteractionsWithinBlocksKernel.setArg<mm_float4>(1, context.getPeriodicBoxSize()); findInteractionsWithinBlocksKernel.setArg<mm_float4>(1, context.getPeriodicBoxSize());
findInteractionsWithinBlocksKernel.setArg<mm_float4>(2, context.getInvPeriodicBoxSize()); findInteractionsWithinBlocksKernel.setArg<mm_float4>(2, context.getInvPeriodicBoxSize());
...@@ -274,8 +289,8 @@ void OpenCLNonbondedUtilities::prepareInteractions() { ...@@ -274,8 +289,8 @@ void OpenCLNonbondedUtilities::prepareInteractions() {
void OpenCLNonbondedUtilities::computeInteractions() { void OpenCLNonbondedUtilities::computeInteractions() {
if (cutoff != -1.0) { if (cutoff != -1.0) {
if (useCutoff) { if (useCutoff) {
forceKernel.setArg<mm_float4>(11, context.getPeriodicBoxSize()); forceKernel.setArg<mm_float4>(10, context.getPeriodicBoxSize());
forceKernel.setArg<mm_float4>(12, context.getInvPeriodicBoxSize()); forceKernel.setArg<mm_float4>(11, context.getInvPeriodicBoxSize());
} }
context.executeKernel(forceKernel, (context.getNumAtomBlocks()*(context.getNumAtomBlocks()+1)/2)*OpenCLContext::TileSize); context.executeKernel(forceKernel, (context.getNumAtomBlocks()*(context.getNumAtomBlocks()+1)/2)*OpenCLContext::TileSize);
} }
...@@ -388,6 +403,8 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc ...@@ -388,6 +403,8 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc
defines["NUM_ATOMS"] = OpenCLExpressionUtilities::intToString(context.getNumAtoms()); defines["NUM_ATOMS"] = OpenCLExpressionUtilities::intToString(context.getNumAtoms());
defines["PADDED_NUM_ATOMS"] = OpenCLExpressionUtilities::intToString(context.getPaddedNumAtoms()); defines["PADDED_NUM_ATOMS"] = OpenCLExpressionUtilities::intToString(context.getPaddedNumAtoms());
defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(context.getNumAtomBlocks()); defines["NUM_BLOCKS"] = OpenCLExpressionUtilities::intToString(context.getNumAtomBlocks());
if (useCutoff)
defines["MAX_TILES"] = OpenCLExpressionUtilities::intToString(interactingTiles->getSize());
string file = (context.getSIMDWidth() == 32 ? OpenCLKernelSources::nonbonded_nvidia : OpenCLKernelSources::nonbonded_default); string file = (context.getSIMDWidth() == 32 ? OpenCLKernelSources::nonbonded_nvidia : OpenCLKernelSources::nonbonded_default);
cl::Program program = context.createProgram(context.replaceStrings(file, replacements), defines); cl::Program program = context.createProgram(context.replaceStrings(file, replacements), defines);
cl::Kernel kernel(program, "computeNonbonded"); cl::Kernel kernel(program, "computeNonbonded");
...@@ -405,9 +422,10 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc ...@@ -405,9 +422,10 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc
kernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float4), NULL); kernel.setArg(index++, OpenCLContext::ThreadBlockSize*sizeof(cl_float4), NULL);
if (useCutoff) { if (useCutoff) {
kernel.setArg<cl::Buffer>(index++, interactingTiles->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, interactingTiles->getDeviceBuffer());
kernel.setArg<cl::Buffer>(index++, interactionFlags->getDeviceBuffer());
kernel.setArg<cl::Buffer>(index++, interactionCount->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, interactionCount->getDeviceBuffer());
index += 2; // The periodic box size arguments are set when the kernel is executed. index += 2; // The periodic box size arguments are set when the kernel is executed.
if (context.getSIMDWidth() == 32)
kernel.setArg<cl::Buffer>(index++, interactionFlags->getDeviceBuffer());
} }
else { else {
kernel.setArg<cl_uint>(index++, context.getNumAtomBlocks()*(context.getNumAtomBlocks()+1)/2); kernel.setArg<cl_uint>(index++, context.getNumAtomBlocks()*(context.getNumAtomBlocks()+1)/2);
......
platforms/opencl/src/kernels/customGBEnergyN2_default.cl
View file @ 839f9a81
...@@ -11,16 +11,19 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer ...@@ -11,16 +11,19 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer
__global float4* posq, __local float4* local_posq, __global unsigned int* exclusions, __global unsigned int* exclusionIndices, __global float4* posq, __local float4* local_posq, __global unsigned int* exclusions, __global unsigned int* exclusionIndices,
__global unsigned int* exclusionRowIndices, __local float4* tempForceBuffer, __global unsigned int* exclusionRowIndices, __local float4* tempForceBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, unsigned int maxTiles
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
#endif unsigned int pos = get_group_id(0)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
#else
unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0); unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0); unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0);
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[2]; __local unsigned int exclusionRange[2];
...@@ -29,18 +32,23 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer ...@@ -29,18 +32,23 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= maxTiles) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2); unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2);
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE); unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE);
......
platforms/opencl/src/kernels/customGBEnergyN2_nvidia.cl
View file @ 839f9a81
...@@ -11,18 +11,21 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer ...@@ -11,18 +11,21 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer
__global float4* posq, __local float4* local_posq, __global unsigned int* exclusions, __global unsigned int* exclusionIndices, __global float4* posq, __local float4* local_posq, __global unsigned int* exclusions, __global unsigned int* exclusionIndices,
__global unsigned int* exclusionRowIndices, __local float4* tempBuffer, __global unsigned int* exclusionRowIndices, __local float4* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, unsigned int maxTiles, __global unsigned int* interactionFlags
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
#endif
unsigned int totalWarps = get_global_size(0)/TILE_SIZE; unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
unsigned int warp = get_global_id(0)/TILE_SIZE; unsigned int warp = get_global_id(0)/TILE_SIZE;
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
unsigned int pos = warp*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
unsigned int end = (warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
#else
unsigned int pos = warp*numTiles/totalWarps; unsigned int pos = warp*numTiles/totalWarps;
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[4]; __local unsigned int exclusionRange[4];
...@@ -30,18 +33,23 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer ...@@ -30,18 +33,23 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= maxTiles) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int tbx = get_local_id(0) - tgx; unsigned int tbx = get_local_id(0) - tgx;
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
...@@ -128,7 +136,7 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer ...@@ -128,7 +136,7 @@ void computeN2Energy(__global float4* forceBuffers, __global float* energyBuffer
local_force[get_local_id(0)] = 0.0f; local_force[get_local_id(0)] = 0.0f;
CLEAR_LOCAL_DERIVATIVES CLEAR_LOCAL_DERIVATIVES
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int flags = interactionFlags[pos]; unsigned int flags = (numTiles <= maxTiles ? interactionFlags[pos] : 0xFFFFFFFF);
if (!hasExclusions && flags == 0) { if (!hasExclusions && flags == 0) {
// No interactions in this tile. // No interactions in this tile.
} }
......
platforms/opencl/src/kernels/customGBValueN2_default.cl
View file @ 839f9a81
...@@ -9,16 +9,19 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global ...@@ -9,16 +9,19 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global
__global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __global float* global_value, __local float* local_value, __global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __global float* global_value, __local float* local_value,
__local float* tempBuffer, __local float* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, unsigned int maxTiles
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
#endif unsigned int pos = get_group_id(0)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
#else
unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0); unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0); unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0);
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[2]; __local unsigned int exclusionRange[2];
...@@ -26,18 +29,23 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global ...@@ -26,18 +29,23 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= maxTiles) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2); unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2);
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int valueBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE); unsigned int valueBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE);
......
platforms/opencl/src/kernels/customGBValueN2_nvidia.cl
View file @ 839f9a81
...@@ -9,18 +9,21 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global ...@@ -9,18 +9,21 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global
__global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __global float* global_value, __local float* local_value, __global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __global float* global_value, __local float* local_value,
__local float* tempBuffer, __local float* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, unsigned int maxTiles, __global unsigned int* interactionFlags
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
#endif
unsigned int totalWarps = get_global_size(0)/TILE_SIZE; unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
unsigned int warp = get_global_id(0)/TILE_SIZE; unsigned int warp = get_global_id(0)/TILE_SIZE;
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
unsigned int pos = warp*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
unsigned int end = (warp+1)*(numTiles > maxTiles ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
#else
unsigned int pos = warp*numTiles/totalWarps; unsigned int pos = warp*numTiles/totalWarps;
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[4]; __local unsigned int exclusionRange[4];
...@@ -28,18 +31,23 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global ...@@ -28,18 +31,23 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= maxTiles) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int tbx = get_local_id(0) - tgx; unsigned int tbx = get_local_id(0) - tgx;
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
...@@ -125,7 +133,7 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global ...@@ -125,7 +133,7 @@ void computeN2Value(__global float4* posq, __local float4* local_posq, __global
} }
local_value[get_local_id(0)] = 0.0f; local_value[get_local_id(0)] = 0.0f;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int flags = interactionFlags[pos]; unsigned int flags = (numTiles <= maxTiles ? interactionFlags[pos] : 0xFFFFFFFF);
if (!hasExclusions && flags != 0xFFFFFFFF) { if (!hasExclusions && flags != 0xFFFFFFFF) {
if (flags == 0) { if (flags == 0) {
// No interactions in this tile. // No interactions in this tile.
......
platforms/opencl/src/kernels/findInteractingBlocks.cl
View file @ 839f9a81
...@@ -147,8 +147,9 @@ void storeInteractionData(__local ushort2* buffer, __local int* valid, __local s ...@@ -147,8 +147,9 @@ void storeInteractionData(__local ushort2* buffer, __local int* valid, __local s
if (get_local_id(0) == 0) if (get_local_id(0) == 0)
*baseIndex = atom_add(interactionCount, numValid); *baseIndex = atom_add(interactionCount, numValid);
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
for (int i = get_local_id(0); i < numValid; i += GROUP_SIZE) if (*baseIndex+numValid <= MAX_TILES)
interactingTiles[*baseIndex+i] = temp[i]; for (int i = get_local_id(0); i < numValid; i += GROUP_SIZE)
interactingTiles[*baseIndex+i] = temp[i];
barrier(CLK_LOCAL_MEM_FENCE); barrier(CLK_LOCAL_MEM_FENCE);
} }
...@@ -232,6 +233,8 @@ __kernel void findInteractionsWithinBlocks(float cutoffSquared, float4 periodicB ...@@ -232,6 +233,8 @@ __kernel void findInteractionsWithinBlocks(float cutoffSquared, float4 periodicB
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
unsigned int index = get_local_id(0) & (TILE_SIZE - 1); unsigned int index = get_local_id(0) & (TILE_SIZE - 1);
if (numTiles > MAX_TILES)
return;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
float4 apos; float4 apos;
while (pos < end) { while (pos < end) {
......
platforms/opencl/src/kernels/gbsaObc_default.cl
View file @ 839f9a81
...@@ -17,32 +17,40 @@ typedef struct { ...@@ -17,32 +17,40 @@ typedef struct {
__kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1))) __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1)))
void computeBornSum(__global float* global_bornSum, __global float4* posq, __global float2* global_params, __local AtomData* localData, __local float* tempBuffer, void computeBornSum(__global float* global_bornSum, __global float4* posq, __global float2* global_params, __local AtomData* localData, __local float* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) { __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) {
#else #else
unsigned int numTiles) { unsigned int numTiles) {
#endif #endif
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
#endif unsigned int pos = get_group_id(0)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
#else
unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0); unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0); unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0);
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2); unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2);
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE); unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE);
...@@ -196,32 +204,40 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff ...@@ -196,32 +204,40 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff
__global float4* posq, __global float* global_bornRadii, __global float4* posq, __global float* global_bornRadii,
__global float* global_bornForce, __local AtomData* localData, __local float4* tempBuffer, __global float* global_bornForce, __local AtomData* localData, __local float4* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) { __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) {
#else #else
unsigned int numTiles) { unsigned int numTiles) {
#endif #endif
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
#endif unsigned int pos = get_group_id(0)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
#else
unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0); unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0); unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0);
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2); unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2);
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE); unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE);
......
platforms/opencl/src/kernels/gbsaObc_nvidia.cl
View file @ 839f9a81
...@@ -17,34 +17,42 @@ typedef struct { ...@@ -17,34 +17,42 @@ typedef struct {
__kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1))) __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1)))
void computeBornSum(__global float* global_bornSum, __global float4* posq, __global float2* global_params, __local AtomData* localData, __local float* tempBuffer, void computeBornSum(__global float* global_bornSum, __global float4* posq, __global float2* global_params, __local AtomData* localData, __local float* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) { __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, __global unsigned int* interactionFlags) {
#else #else
unsigned int numTiles) { unsigned int numTiles) {
#endif
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
#endif #endif
unsigned int totalWarps = get_global_size(0)/TILE_SIZE; unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
unsigned int warp = get_global_id(0)/TILE_SIZE; unsigned int warp = get_global_id(0)/TILE_SIZE;
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
unsigned int pos = warp*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
unsigned int end = (warp+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
#else
unsigned int pos = warp*numTiles/totalWarps; unsigned int pos = warp*numTiles/totalWarps;
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int tbx = get_local_id(0) - tgx; unsigned int tbx = get_local_id(0) - tgx;
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
...@@ -115,7 +123,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo ...@@ -115,7 +123,7 @@ void computeBornSum(__global float* global_bornSum, __global float4* posq, __glo
} }
localData[get_local_id(0)].bornSum = 0.0f; localData[get_local_id(0)].bornSum = 0.0f;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int flags = interactionFlags[pos]; unsigned int flags = (numTiles <= MAX_TILES ? interactionFlags[pos] : 0xFFFFFFFF);
if (flags != 0xFFFFFFFF && false) { // TODO: Fix this: should be checking for exclusions if (flags != 0xFFFFFFFF && false) { // TODO: Fix this: should be checking for exclusions
if (flags == 0) { if (flags == 0) {
// No interactions in this tile. // No interactions in this tile.
...@@ -262,34 +270,42 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff ...@@ -262,34 +270,42 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff
__global float4* posq, __global float* global_bornRadii, __global float4* posq, __global float* global_bornRadii,
__global float* global_bornForce, __local AtomData* localData, __local float4* tempBuffer, __global float* global_bornForce, __local AtomData* localData, __local float4* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize) { __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, __global unsigned int* interactionFlags) {
#else #else
unsigned int numTiles) { unsigned int numTiles) {
#endif
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
#endif #endif
unsigned int totalWarps = get_global_size(0)/TILE_SIZE; unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
unsigned int warp = get_global_id(0)/TILE_SIZE; unsigned int warp = get_global_id(0)/TILE_SIZE;
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
unsigned int pos = warp*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
unsigned int end = (warp+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
#else
unsigned int pos = warp*numTiles/totalWarps; unsigned int pos = warp*numTiles/totalWarps;
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int tbx = get_local_id(0) - tgx; unsigned int tbx = get_local_id(0) - tgx;
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
...@@ -365,7 +381,7 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff ...@@ -365,7 +381,7 @@ void computeGBSAForce1(__global float4* forceBuffers, __global float* energyBuff
localData[get_local_id(0)].fz = 0.0f; localData[get_local_id(0)].fz = 0.0f;
localData[get_local_id(0)].fw = 0.0f; localData[get_local_id(0)].fw = 0.0f;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int flags = interactionFlags[pos]; unsigned int flags = (numTiles <= MAX_TILES ? interactionFlags[pos] : 0xFFFFFFFF);
if (flags != 0xFFFFFFFF && false) { // TODO: Fix this: should be checking for exclusions if (flags != 0xFFFFFFFF && false) { // TODO: Fix this: should be checking for exclusions
if (flags == 0) { if (flags == 0) {
// No interactions in this tile. // No interactions in this tile.
......
platforms/opencl/src/kernels/nonbonded_default.cl
View file @ 839f9a81
...@@ -15,16 +15,19 @@ __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1))) ...@@ -15,16 +15,19 @@ __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1)))
void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffer, __global float4* posq, __global unsigned int* exclusions, void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffer, __global float4* posq, __global unsigned int* exclusions,
__global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __local AtomData* localData, __local float4* tempBuffer, __global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __local AtomData* localData, __local float4* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
#endif unsigned int pos = get_group_id(0)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/get_num_groups(0);
#else
unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0); unsigned int pos = get_group_id(0)*numTiles/get_num_groups(0);
unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0); unsigned int end = (get_group_id(0)+1)*numTiles/get_num_groups(0);
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[2]; __local unsigned int exclusionRange[2];
...@@ -32,18 +35,23 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe ...@@ -32,18 +35,23 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2); unsigned int baseLocalAtom = (get_local_id(0) < TILE_SIZE ? 0 : TILE_SIZE/2);
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE); unsigned int forceBufferOffset = (tgx < TILE_SIZE/2 ? 0 : TILE_SIZE);
......
platforms/opencl/src/kernels/nonbonded_nvidia.cl
View file @ 839f9a81
...@@ -15,18 +15,21 @@ __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1))) ...@@ -15,18 +15,21 @@ __kernel __attribute__((reqd_work_group_size(WORK_GROUP_SIZE, 1, 1)))
void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffer, __global float4* posq, __global unsigned int* exclusions, void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffer, __global float4* posq, __global unsigned int* exclusions,
__global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __local AtomData* localData, __local float4* tempBuffer, __global unsigned int* exclusionIndices, __global unsigned int* exclusionRowIndices, __local AtomData* localData, __local float4* tempBuffer,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global ushort2* tiles, __global unsigned int* interactionFlags, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize __global ushort2* tiles, __global unsigned int* interactionCount, float4 periodicBoxSize, float4 invPeriodicBoxSize, __global unsigned int* interactionFlags
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
#endif
unsigned int totalWarps = get_global_size(0)/TILE_SIZE; unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
unsigned int warp = get_global_id(0)/TILE_SIZE; unsigned int warp = get_global_id(0)/TILE_SIZE;
#ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0];
unsigned int pos = warp*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
unsigned int end = (warp+1)*(numTiles > MAX_TILES ? NUM_BLOCKS*(NUM_BLOCKS+1)/2 : numTiles)/totalWarps;
#else
unsigned int pos = warp*numTiles/totalWarps; unsigned int pos = warp*numTiles/totalWarps;
unsigned int end = (warp+1)*numTiles/totalWarps; unsigned int end = (warp+1)*numTiles/totalWarps;
#endif
float energy = 0.0f; float energy = 0.0f;
unsigned int lasty = 0xFFFFFFFF; unsigned int lasty = 0xFFFFFFFF;
__local unsigned int exclusionRange[4]; __local unsigned int exclusionRange[4];
...@@ -34,18 +37,23 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe ...@@ -34,18 +37,23 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe
while (pos < end) { while (pos < end) {
// Extract the coordinates of this tile // Extract the coordinates of this tile
unsigned int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
ushort2 tileIndices = tiles[pos]; if (numTiles <= MAX_TILES) {
unsigned int x = tileIndices.x; ushort2 tileIndices = tiles[pos];
unsigned int y = tileIndices.y; x = tileIndices.x;
#else y = tileIndices.y;
unsigned int y = (unsigned int) floor(NUM_BLOCKS+0.5f-sqrt((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
unsigned int x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
} }
else
#endif #endif
{
y = (unsigned 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 >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
y++;
x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
}
}
unsigned int tgx = get_local_id(0) & (TILE_SIZE-1); unsigned int tgx = get_local_id(0) & (TILE_SIZE-1);
unsigned int tbx = get_local_id(0) - tgx; unsigned int tbx = get_local_id(0) - tgx;
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
...@@ -137,7 +145,7 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe ...@@ -137,7 +145,7 @@ void computeNonbonded(__global float4* forceBuffers, __global float* energyBuffe
localData[get_local_id(0)].fy = 0.0f; localData[get_local_id(0)].fy = 0.0f;
localData[get_local_id(0)].fz = 0.0f; localData[get_local_id(0)].fz = 0.0f;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int flags = interactionFlags[pos]; unsigned int flags = (numTiles <= MAX_TILES ? interactionFlags[pos] : 0xFFFFFFFF);
if (!hasExclusions && flags != 0xFFFFFFFF) { if (!hasExclusions && flags != 0xFFFFFFFF) {
if (flags == 0) { if (flags == 0) {
// No interactions in this tile. // No interactions in this tile.
......
platforms/opencl/tests/TestOpenCLNonbondedForce.cpp
View file @ 839f9a81
...@@ -523,7 +523,8 @@ void testBlockInteractions(bool periodic) { ...@@ -523,7 +523,8 @@ void testBlockInteractions(bool periodic) {
int numWithInteractions = interactionCount[0]; int numWithInteractions = interactionCount[0];
vector<bool> hasInteractions(numBlocks*(numBlocks+1)/2, false); vector<bool> hasInteractions(numBlocks*(numBlocks+1)/2, false);
nb.getInteractingTiles().download(interactingTiles); nb.getInteractingTiles().download(interactingTiles);
nb.getInteractionFlags().download(interactionFlags); if (clcontext.getSIMDWidth() == 32)
nb.getInteractionFlags().download(interactionFlags);
const unsigned int atoms = clcontext.getPaddedNumAtoms(); const unsigned int atoms = clcontext.getPaddedNumAtoms();
const unsigned int grid = OpenCLContext::TileSize; const unsigned int grid = OpenCLContext::TileSize;
const unsigned int dim = clcontext.getNumAtomBlocks(); const unsigned int dim = clcontext.getNumAtomBlocks();
......
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