Skip to content

GitLab

Commit 01afee8b authored by Peter Eastman's avatar Peter Eastman
Browse files

Changed workgroup size to work around NVIDIA bug. This also improves performance slightly.

parent 83284ce6
Hide whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
platforms/opencl/src/OpenCLNonbondedUtilities.cpp
View file @ 01afee8b
...@@ -317,7 +317,7 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -317,7 +317,7 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
for (int i = 0; i < (int) exclusionBlocksForBlock.size(); i++) for (int i = 0; i < (int) exclusionBlocksForBlock.size(); i++)
maxExclusions = (maxExclusions > exclusionBlocksForBlock[i].size() ? maxExclusions : exclusionBlocksForBlock[i].size()); maxExclusions = (maxExclusions > exclusionBlocksForBlock[i].size() ? maxExclusions : exclusionBlocksForBlock[i].size());
defines["MAX_EXCLUSIONS"] = context.intToString(maxExclusions); defines["MAX_EXCLUSIONS"] = context.intToString(maxExclusions);
defines["GROUP_SIZE"] = (deviceIsCpu ? "32" : "256"); defines["GROUP_SIZE"] = (deviceIsCpu ? "32" : "128");
defines["BUFFER_GROUPS"] = (deviceIsCpu ? "4" : "2"); defines["BUFFER_GROUPS"] = (deviceIsCpu ? "4" : "2");
string file = (deviceIsCpu ? OpenCLKernelSources::findInteractingBlocks_cpu : OpenCLKernelSources::findInteractingBlocks); string file = (deviceIsCpu ? OpenCLKernelSources::findInteractingBlocks_cpu : OpenCLKernelSources::findInteractingBlocks);
cl::Program interactingBlocksProgram = context.createProgram(file, defines); cl::Program interactingBlocksProgram = context.createProgram(file, defines);
...@@ -391,7 +391,7 @@ void OpenCLNonbondedUtilities::prepareInteractions() { ...@@ -391,7 +391,7 @@ void OpenCLNonbondedUtilities::prepareInteractions() {
context.executeKernel(sortBoxDataKernel, context.getNumAtoms()); context.executeKernel(sortBoxDataKernel, context.getNumAtoms());
setPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 0); setPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 0);
setInvPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 1); setInvPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 1);
context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms(), deviceIsCpu ? 1 : 256); context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms(), deviceIsCpu ? 1 : 128);
} }
void OpenCLNonbondedUtilities::computeInteractions() { void OpenCLNonbondedUtilities::computeInteractions() {
......
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