Cleaned up a few details of the Ewald code

platforms/cuda/src/kernels/gpu.cpp

@@ -432,7 +432,7 @@ void gpuSetEwaldParameters(gpuContext gpu)//, float alphaEwald, int kmax )
     gpu->sim.alphaEwald         = alpha;
     gpu->sim.factorEwald        = -1 / (4*alpha*alpha);
     gpu->sim.kmax               = 20+1;
-    gpu->psEwaldEikr            = new CUDAStream<float2>(gpu->sim.paddedNumberOfAtoms*gpu->sim.kmax, 1, "EwaldEikr");
+    gpu->psEwaldEikr            = new CUDAStream<float2>(gpu->sim.atoms*gpu->sim.kmax, 1, "EwaldEikr");
     gpu->sim.pEwaldEikr         = gpu->psEwaldEikr->_pDevStream[0];
     gpu->psEwaldCosSinSum       = new CUDAStream<float2>((gpu->sim.kmax*2-1) * (gpu->sim.kmax*2-1) * (gpu->sim.kmax*2-1), 1, "EwaldCosSinSum");
     gpu->sim.pEwaldCosSinSum    = gpu->psEwaldCosSinSum->_pDevStream[0];

platforms/cuda/src/kernels/kCalculateCDLJForces.cu

@@ -198,18 +198,21 @@ void kCalculateCDLJForces(gpuContext gpu)
             kFindInteractionsWithinBlocksEwaldDirect_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block,
                     sizeof(unsigned int)*gpu->sim.nonbond_threads_per_block>>>(gpu->sim.pInteractingWorkUnit);
             if (gpu->bOutputBufferPerWarp)
-            {
-// Vanilla (N2) Ewald
                 kCalculateCDLJEwaldDirectByWarpForces_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block,
                         (sizeof(Atom)+sizeof(float3))*gpu->sim.nonbond_threads_per_block>>>(gpu->sim.pInteractingWorkUnit);
+            else
+                kCalculateCDLJEwaldDirectForces_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block,
+                        (sizeof(Atom)+sizeof(float3))*gpu->sim.nonbond_threads_per_block>>>(gpu->sim.pInteractingWorkUnit);
+            LAUNCHERROR("kCalculateCDLJEwaldDirectForces");
+            if (false)
+            {
+                // O(N2) Ewald summation
                 kCalculateCDLJEwaldReciprocalForces_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
                 LAUNCHERROR("kCalculateCDLJEwaldReciprocalForces");
             }
             else
             {
-                kCalculateCDLJEwaldDirectForces_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block,
-                        (sizeof(Atom)+sizeof(float3))*gpu->sim.nonbond_threads_per_block>>>(gpu->sim.pInteractingWorkUnit);
-                LAUNCHERROR("kCalculateCDLJEwaldDirectForces");
+                // O(N3/2) Ewald summation
                 kCalculateEwaldFastEikr_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
                 LAUNCHERROR("kCalculateEwaldFastEikr");
                 kCalculateEwaldFastCosSinSums_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();