Skip to content

GitLab

Commit c9d720bb authored by Peter Eastman's avatar Peter Eastman
Browse files

Bug fixes

parent 53fec1af
Show whitespace changes
Inline Side-by-side
Showing with 5 additions and 3 deletions
plugins/amoeba/platforms/cuda/src/AmoebaCudaKernels.cpp
View file @ c9d720bb
...@@ -846,7 +846,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const ...@@ -846,7 +846,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
double fixedThreadMemory = 19*elementSize+2*sizeof(float)+3*sizeof(int)/(double) cu.TileSize; double fixedThreadMemory = 19*elementSize+2*sizeof(float)+3*sizeof(int)/(double) cu.TileSize;
double inducedThreadMemory = 15*elementSize+2*sizeof(float); double inducedThreadMemory = 15*elementSize+2*sizeof(float);
double electrostaticsThreadMemory = 0; double electrostaticsThreadMemory = 0;
if (useShuffle) if (!useShuffle)
fixedThreadMemory += 3*elementSize; fixedThreadMemory += 3*elementSize;
map<string, string> defines; map<string, string> defines;
defines["NUM_ATOMS"] = cu.intToString(numMultipoles); defines["NUM_ATOMS"] = cu.intToString(numMultipoles);
...@@ -855,6 +855,8 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const ...@@ -855,6 +855,8 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
defines["ENERGY_SCALE_FACTOR"] = cu.doubleToString(138.9354558456/innerDielectric); defines["ENERGY_SCALE_FACTOR"] = cu.doubleToString(138.9354558456/innerDielectric);
if (force.getPolarizationType() == AmoebaMultipoleForce::Direct) if (force.getPolarizationType() == AmoebaMultipoleForce::Direct)
defines["DIRECT_POLARIZATION"] = ""; defines["DIRECT_POLARIZATION"] = "";
if (useShuffle)
defines["USE_SHUFFLE"] = "";
double alpha = force.getAEwald(); double alpha = force.getAEwald();
int gridSizeX, gridSizeY, gridSizeZ; int gridSizeX, gridSizeY, gridSizeZ;
if (usePME) { if (usePME) {
...@@ -915,7 +917,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const ...@@ -915,7 +917,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
electrostaticsSource << "#define APPLY_SCALE\n"; electrostaticsSource << "#define APPLY_SCALE\n";
electrostaticsSource << CudaAmoebaKernelSources::pmeElectrostaticPairForce; electrostaticsSource << CudaAmoebaKernelSources::pmeElectrostaticPairForce;
electrostaticsThreadMemory = 24*elementSize+3*sizeof(float)+3*sizeof(int)/(double) cu.TileSize; electrostaticsThreadMemory = 24*elementSize+3*sizeof(float)+3*sizeof(int)/(double) cu.TileSize;
if (useShuffle) if (!useShuffle)
electrostaticsThreadMemory += 3*elementSize; electrostaticsThreadMemory += 3*elementSize;
} }
else { else {
...@@ -930,7 +932,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const ...@@ -930,7 +932,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
electrostaticsSource << "#define T3\n"; electrostaticsSource << "#define T3\n";
electrostaticsSource << CudaAmoebaKernelSources::electrostaticPairForce; electrostaticsSource << CudaAmoebaKernelSources::electrostaticPairForce;
electrostaticsThreadMemory = 21*elementSize+2*sizeof(float)+3*sizeof(int)/(double) cu.TileSize; electrostaticsThreadMemory = 21*elementSize+2*sizeof(float)+3*sizeof(int)/(double) cu.TileSize;
if (useShuffle) if (!useShuffle)
electrostaticsThreadMemory += 3*elementSize; electrostaticsThreadMemory += 3*elementSize;
if (gk != NULL) if (gk != NULL)
electrostaticsThreadMemory += 4*elementSize; electrostaticsThreadMemory += 4*elementSize;
......
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