Commit 97239ca6 authored by one's avatar one
Browse files

Tune HIP LangevinMiddle launch block size

Use explicit 128-thread launches for the three LangevinMiddle integration kernels to improve HIP throughput while
preserving the existing PME launch heuristics.
parent 20e4b551
......@@ -3187,11 +3187,11 @@ void CommonIntegrateLangevinMiddleStepKernel::execute(ContextImpl& context, cons
// Perform the integration.
kernel2->setArg(7, integration.prepareRandomNumbers(cc.getPaddedNumAtoms()));
kernel1->execute(numAtoms);
kernel1->execute(numAtoms, 128);
integration.applyVelocityConstraints(integrator.getConstraintTolerance());
kernel2->execute(numAtoms);
kernel2->execute(numAtoms, 128);
integration.applyConstraints(integrator.getConstraintTolerance());
kernel3->execute(numAtoms);
kernel3->execute(numAtoms, 128);
integration.computeVirtualSites();
// Update the time and step count.
......
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