Increased number of thread blocks

45c23132 · Peter Eastman · 00b5aa4d · 45c23132
Commit 45c23132 authored Mar 05, 2010 by Peter Eastman
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

platforms/opencl/src/OpenCLContext.cpp platforms/opencl/src/OpenCLContext.cpp +1 -1

No files found.
--- a/platforms/opencl/src/OpenCLContext.cpp
+++ b/platforms/opencl/src/OpenCLContext.cpp
@@ -79,7 +79,7 @@ OpenCLContext::OpenCLContext(int numParticles, int deviceIndex) : time(0.0), ste
        numAtoms = numParticles;
        paddedNumAtoms = TileSize*((numParticles+TileSize-1)/TileSize);
        numAtomBlocks = (paddedNumAtoms+(TileSize-1))/TileSize;
-        numThreadBlocks = device.getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>()[0]/ThreadBlockSize;
+        numThreadBlocks = 4*device.getInfo<CL_DEVICE_MAX_COMPUTE_UNITS>()*device.getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>()[0]/ThreadBlockSize;
        nonbonded = new OpenCLNonbondedUtilities(*this);
        posq = new OpenCLArray<mm_float4>(*this, paddedNumAtoms, "posq", true);
        velm = new OpenCLArray<mm_float4>(*this, paddedNumAtoms, "velm", true);