Merge pull request #705 from peastman/multigpu

Further improvements to multi-GPU scaling

platforms/cuda/src/CudaParallelKernels.cpp

@@ -98,6 +98,12 @@ public:
         // Execute the kernel, then download forces.
         
         energy += kernel.finishComputation(context, includeForce, includeEnergy, groups);
+        if (cu.getComputeForceCount() < 200) {
+            // Record timing information for load balancing.  Since this takes time, only do it at the start of the simulation.
+
+            CHECK_RESULT(cuCtxSynchronize(), "Error synchronizing CUDA context");
+            completionTime = getTime();
+        }
         if (includeForce) {
             if (cu.getContextIndex() > 0) {
                 int numAtoms = cu.getPaddedNumAtoms();
@@ -110,13 +116,8 @@ public:
                 else
                     cu.getForce().download(&pinnedMemory[(cu.getContextIndex()-1)*numAtoms*3]);
             }
-            else {
-                // In principle this should make the load balancing more accurate, but in practice it just seems to make things slower.
-                //CHECK_RESULT(cuCtxSynchronize(), "Error synchronizing CUDA context");
         }
     }
-        completionTime = getTime();
-    }
 private:
     ContextImpl& context;
     CudaContext& cu;
@@ -192,6 +193,7 @@ void CudaParallelCalcForcesAndEnergyKernel::beginComputation(ContextImpl& contex
     }
 }
 
+#include <cstdio>
 double CudaParallelCalcForcesAndEnergyKernel::finishComputation(ContextImpl& context, bool includeForce, bool includeEnergy, int groups) {
     for (int i = 0; i < (int) data.contexts.size(); i++) {
         CudaContext& cu = *data.contexts[i];
@@ -216,6 +218,7 @@ double CudaParallelCalcForcesAndEnergyKernel::finishComputation(ContextImpl& con
         // Balance work between the contexts by transferring a little nonbonded work from the context that
         // finished last to the one that finished first.
         
+        if (cu.getComputeForceCount() < 200) {
             int firstIndex = 0, lastIndex = 0;
             for (int i = 0; i < (int) completionTimes.size(); i++) {
                 if (completionTimes[i] < completionTimes[firstIndex])
@@ -223,7 +226,7 @@ double CudaParallelCalcForcesAndEnergyKernel::finishComputation(ContextImpl& con
                 if (completionTimes[i] > completionTimes[lastIndex])
                     lastIndex = i;
             }
-        double fractionToTransfer = min(0.001, contextNonbondedFractions[lastIndex]);
+            double fractionToTransfer = min(0.01, contextNonbondedFractions[lastIndex]);
             contextNonbondedFractions[firstIndex] += fractionToTransfer;
             contextNonbondedFractions[lastIndex] -= fractionToTransfer;
             double startFraction = 0.0;
@@ -235,6 +238,7 @@ double CudaParallelCalcForcesAndEnergyKernel::finishComputation(ContextImpl& con
                 startFraction = endFraction;
             }
 	}
+    }
     return energy;
 }