Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "aacefccf9c0326416bc853864db4bb13402b7b1a"
Commit 38f6c8f8 authored by Peter Eastman's avatar Peter Eastman
Browse files

Checked in Cuda code

parent 95d79181
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 9174 additions and 0 deletions
platforms/cuda/src/kernels/cudaKernels.h 0 → 100755
View file @ 38f6c8f8
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include "gputypes.h"
// Initialization
extern void kClearForces(gpuContext gpu);
extern void kCalculateObcGbsaBornSum(gpuContext gpu);
extern void kReduceObcGbsaBornSum(gpuContext gpu);
extern void kGenerateRandoms(gpuContext gpu);
// Main loop
extern void kCalculateCDLJObcGbsaForces1(gpuContext gpu);
extern void kCalculateCDLJObcGbsaForces1_12(gpuContext gpu);
extern void kCalculateCDLJForces(gpuContext gpu);
extern void kCalculateCDLJForces_12(gpuContext gpu);
extern void kCalculateObcGbsaForces1(gpuContext gpu);
extern void kCalculateObcGbsaForces1_12(gpuContext gpu);
extern void kReduceObcGbsaBornForces(gpuContext gpu);
extern void kCalculateObcGbsaForces2(gpuContext gpu);
extern void kCalculateObcGbsaForces2_12(gpuContext gpu);
extern void kCalculateLocalForces(gpuContext gpu);
extern void kCalculateAndersenThermostat(gpuContext gpu);
extern void kReduceBornSumAndForces(gpuContext gpu);
extern void kUpdatePart1(gpuContext gpu);
extern void kApplyFirstShake(gpuContext gpu);
extern void kUpdatePart2(gpuContext gpu);
extern void kApplySecondShake(gpuContext gpu);
extern void kVerletUpdatePart1(gpuContext gpu);
extern void kVerletUpdatePart2(gpuContext gpu);
extern void kBrownianUpdatePart1(gpuContext gpu);
extern void kBrownianUpdatePart2(gpuContext gpu);
// Extras
extern void kReduceForces(gpuContext gpu);
extern void kClearBornForces(gpuContext gpu);
// Initializers
extern void SetCalculateCDLJObcGbsaForces1Sim(gpuContext gpu);
extern void GetCalculateCDLJObcGbsaForces1Sim(gpuContext gpu);
extern void SetCalculateCDLJObcGbsaForces1_12Sim(gpuContext gpu);
extern void GetCalculateCDLJObcGbsaForces1_12Sim(gpuContext gpu);
extern void SetCalculateCDLJForcesSim(gpuContext gpu);
extern void GetCalculateCDLJForcesSim(gpuContext gpu);
extern void SetCalculateCDLJForces_12Sim(gpuContext gpu);
extern void GetCalculateCDLJForces_12Sim(gpuContext gpu);
extern void SetCalculateLocalForcesSim(gpuContext gpu);
extern void GetCalculateLocalForcesSim(gpuContext gpu);
extern void SetCalculateObcGbsaBornSumSim(gpuContext gpu);
extern void GetCalculateObcGbsaBornSumSim(gpuContext gpu);
extern void SetCalculateObcGbsaForces1Sim(gpuContext gpu);
extern void GetCalculateObcGbsaForces1Sim(gpuContext gpu);
extern void SetCalculateObcGbsaForces1_12Sim(gpuContext gpu);
extern void GetCalculateObcGbsaForces1_12Sim(gpuContext gpu);
extern void SetCalculateObcGbsaForces2Sim(gpuContext gpu);
extern void GetCalculateObcGbsaForces2Sim(gpuContext gpu);
extern void SetCalculateObcGbsaForces2_12Sim(gpuContext gpu);
extern void GetCalculateObcGbsaForces2_12Sim(gpuContext gpu);
extern void SetCalculateAndersenThermostatSim(gpuContext gpu);
extern void GetCalculateAndersenThermostatSim(gpuContext gpu);
extern void SetForcesSim(gpuContext gpu);
extern void GetForcesSim(gpuContext gpu);
extern void SetUpdateShakeHSim(gpuContext gpu);
extern void GetUpdateShakeHSim(gpuContext gpu);
extern void SetVerletUpdateSim(gpuContext gpu);
extern void GetVerletUpdateSim(gpuContext gpu);
extern void SetBrownianUpdateSim(gpuContext gpu);
extern void GetBrownianUpdateSim(gpuContext gpu);
extern void SetRandomSim(gpuContext gpu);
extern void GetRandomSim(gpuContext gpu);
platforms/cuda/src/kernels/cudatypes.h 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/gpu.cpp 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/gputypes.h 0 → 100755
View file @ 38f6c8f8
#ifndef __GPUTYPES_H__
#define __GPUTYPES_H__
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include "cudatypes.h"
#include <vector>
struct gpuAtomType {
string name;
char symbol;
float r;
};
enum SM_VERSION
{
SM_10,
SM_11,
SM_12
};
/* Pointer to this structure will be given
* to gromacs functions*/
struct _gpuContext {
//Cache this here so that it doesn't
//have to be repeatedly passed around
int natoms;
gpuAtomType* gpAtomTable;
int gAtomTypes;
cudaGmxSimulation sim;
unsigned int* pOutputBufferCounter;
unsigned int* pExclusion;
unsigned char* pAtomSymbol;
float iterations;
float epsfac;
float solventDielectric;
float soluteDielectric;
int grid;
bool bCalculateCM;
bool bRemoveCM;
bool bRecalculateBornRadii;
unsigned long seed;
SM_VERSION sm_version;
CUDAStream<float4>* psPosq4;
CUDAStream<float4>* psPosqP4;
CUDAStream<float4>* psOldPosq4;
CUDAStream<float4>* psVelm4;
CUDAStream<float4>* psForce4;
CUDAStream<float4>* psxVector4;
CUDAStream<float4>* psvVector4;
CUDAStream<float2>* psSigEps2;
CUDAStream<float2>* psObcData;
CUDAStream<float>* psObcChain;
CUDAStream<float>* psBornForce;
CUDAStream<float>* psBornRadii;
CUDAStream<float>* psBornSum;
CUDAStream<int4>* psBondID;
CUDAStream<float2>* psBondParameter;
CUDAStream<int4>* psBondAngleID1;
CUDAStream<int2>* psBondAngleID2;
CUDAStream<float2>* psBondAngleParameter;
CUDAStream<int4>* psDihedralID1;
CUDAStream<int4>* psDihedralID2;
CUDAStream<float4>* psDihedralParameter;
CUDAStream<int4>* psRbDihedralID1;
CUDAStream<int4>* psRbDihedralID2;
CUDAStream<float4>* psRbDihedralParameter1;
CUDAStream<float2>* psRbDihedralParameter2;
CUDAStream<int4>* psLJ14ID;
CUDAStream<float4>* psLJ14Parameter;
CUDAStream<int>* psNonShakeID;
CUDAStream<int4>* psShakeID;
CUDAStream<float4>* psShakeParameter;
CUDAStream<unsigned int>* psExclusion;
CUDAStream<unsigned int>* psWorkUnit;
CUDAStream<float4>* psRandom4; // Pointer to sets of 4 random numbers for MD integration
CUDAStream<float2>* psRandom2; // Pointer to sets of 2 random numbers for MD integration
CUDAStream<uint4>* psRandomSeed; // Pointer to each random seed
CUDAStream<int>* psRandomPosition; // Pointer to random number positions
CUDAStream<float4>* psLinearMomentum; // Pointer to total linear momentum per CTA
};
typedef struct _gpuContext *gpuContext;
// Function prototypes
extern "C"
bool gpuIsAvailable();
extern "C"
int gpuReadBondParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetBondParameters(gpuContext gpu, const std::vector<int>& atom1, const std::vector<int>& atom2, const std::vector<float>& length, const std::vector<float>& k);
extern "C"
int gpuReadBondAngleParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetBondAngleParameters(gpuContext gpu, const std::vector<int>& atom1, const std::vector<int>& atom2, const std::vector<int>& atom3,
const std::vector<float>& angle, const std::vector<float>& k);
extern "C"
int gpuReadDihedralParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetDihedralParameters(gpuContext gpu, const std::vector<int>& atom1, const std::vector<int>& atom2, const std::vector<int>& atom3, const std::vector<int>& atom4,
const std::vector<float>& k, const std::vector<float>& phase, const std::vector<int>& periodicity);
extern "C"
int gpuReadRbDihedralParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetRbDihedralParameters(gpuContext gpu, const std::vector<int>& atom1, const std::vector<int>& atom2, const std::vector<int>& atom3, const std::vector<int>& atom4,
const std::vector<float>& c0, const std::vector<float>& c1, const std::vector<float>& c2, const std::vector<float>& c3, const std::vector<float>& c4, const std::vector<float>& c5);
extern "C"
int gpuReadLJ14Parameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetLJ14Parameters(gpuContext gpu, float epsfac, float fudge, const std::vector<int>& atom1, const std::vector<int>& atom2,
const std::vector<float>& c6, const std::vector<float>& c12, const std::vector<float>& q1, const std::vector<float>& q2);
extern "C"
float gpuGetAtomicRadius(gpuContext gpu, string s);
extern "C"
unsigned char gpuGetAtomicSymbol(gpuContext gpu, string s);
extern "C"
int gpuReadAtomicParameters(gpuContext gpu, char* fname);
extern "C"
int gpuReadCoulombParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetCoulombParameters(gpuContext gpu, float epsfac, const std::vector<int>& atom, const std::vector<float>& c6, const std::vector<float>& c12, const std::vector<float>& q,
const std::vector<char>& symbol, const std::vector<vector<int> >& exclusions);
extern "C"
void gpuSetObcParameters(gpuContext gpu, float innerDielectric, float solventDielectric, const std::vector<int>& atom, const std::vector<float>& radius, const std::vector<float>& scale);
extern "C"
int gpuReadShakeParameters(gpuContext gpu, char* fname);
extern "C"
void gpuSetShakeParameters(gpuContext gpu, const std::vector<int>& atom1, const std::vector<int>& atom2, const std::vector<float>& distance,
const std::vector<float>& invMass1, const std::vector<float>& invMass2, float tolerance);
extern "C"
int gpuAllocateInitialBuffers(gpuContext gpu);
extern "C"
void gpuReadCoordinates(gpuContext gpu, char* fname);
extern "C"
void gpuSetPositions(gpuContext gpu, const std::vector<float>& x, const std::vector<float>& y, const std::vector<float>& z);
extern "C"
void gpuSetVelocities(gpuContext gpu, const std::vector<float>& x, const std::vector<float>& y, const std::vector<float>& z);
extern "C"
void gpuSetMass(gpuContext gpu, const std::vector<float>& mass);
extern "C"
void gpuInitializeRandoms(gpuContext gpu);
extern "C"
void* gpuInitFromFile(char* fname);
extern "C"
void* gpuInit(int numAtoms);
extern "C"
void gpuSetIntegrationParameters(gpuContext gpu, float tau, float deltaT, float temperature);
extern "C"
void gpuSetVerletIntegrationParameters(gpuContext gpu, float deltaT);
extern "C"
void gpuSetBrownianIntegrationParameters(gpuContext gpu, float tau, float deltaT, float temperature);
extern "C"
void gpuSetAndersenThermostatParameters(gpuContext gpu, float temperature, float collisionProbability);
extern "C"
void gpuShutDown(gpuContext gpu);
extern "C"
int gpuBuildOutputBuffers(gpuContext gpu);
extern "C"
int gpuBuildThreadBlockWorkList(gpuContext gpu);
extern "C"
int gpuBuildExclusionList(gpuContext gpu);
extern "C"
int gpuSetConstants(gpuContext gpu);
extern "C"
void gpuDumpCoordinates(gpuContext gpu);
extern "C"
void gpuDumpPrimeCoordinates(gpuContext gpu);
extern "C"
void gpuDumpForces(gpuContext gpu);
extern "C"
void gpuDumpAtomData(gpuContext gpu);
extern "C"
bool gpuCheckData(gpuContext gpu);
extern "C"
void gpuSetup(void* pVoid);
extern "C"
void kCPUCalculate14(gpuContext gpu);
extern "C"
void kCPUCalculateLocalForces(gpuContext gpu);
extern "C"
void WriteArrayToFile1( gpuContext gpu, char* fname, int step, CUDAStream<float>* psPos, int numPrint );
extern "C"
void WriteArrayToFile2( gpuContext gpu, char* fname, int step, CUDAStream<float2>* psPos, int numPrint );
extern "C"
void WriteArrayToFile3( gpuContext gpu, char* fname, int step, CUDAStream<float3>* psPos, int numPrint );
extern "C"
void WriteArrayToFile4( gpuContext gpu, char* fname, int step, CUDAStream<float4>* psPos, int numPrint );
extern "C"
void gpuDumpObcInfo(gpuContext gpu);
extern "C"
void gpuDumpObcLoop1(gpuContext gpu);
#endif //__GPUTYPES_H__
platforms/cuda/src/kernels/kBrownianUpdate.cu 0 → 100755
View file @ 38f6c8f8
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include <stdio.h>
#include <cuda.h>
#include <vector_functions.h>
#include <cstdlib>
#include <string>
#include <iostream>
//#include <fstream>
using namespace std;
#include "gputypes.h"
#define DeltaShake
static __constant__ cudaGmxSimulation cSim;
void SetBrownianUpdateSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyToSymbol(cSim, &gpu->sim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyToSymbol: SetSim copy to cSim failed");
}
void GetBrownianUpdateSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyFromSymbol(&gpu->sim, cSim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyFromSymbol: SetSim copy from cSim failed");
}
__global__ void kBrownianUpdatePart1_kernel()
{
unsigned int pos = threadIdx.x + blockIdx.x * blockDim.x;
unsigned int rpos = cSim.pRandomPosition[blockIdx.x];
__syncthreads();
while (pos < cSim.atoms)
{
float4 random4a = cSim.pRandom4a[rpos + pos];
float4 apos = cSim.pPosq[pos];
float4 force = cSim.pForce4[pos];
cSim.pOldPosq[pos] = apos;
#ifndef DeltaShake
apos.x += force.x*cSim.GDT + random4a.x;
apos.y += force.y*cSim.GDT + random4a.y;
apos.z += force.z*cSim.GDT + random4a.z;
#else
apos.x = force.x*cSim.GDT + random4a.x;
apos.y = force.y*cSim.GDT + random4a.y;
apos.z = force.z*cSim.GDT + random4a.z;
#endif
cSim.pPosqP[pos] = apos;
pos += blockDim.x * gridDim.x;
}
}
void kBrownianUpdatePart1(gpuContext gpu)
{
// printf("kBrownianUpdatePart1\n");
kBrownianUpdatePart1_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
LAUNCHERROR("kBrownianUpdatePart1");
}
__global__ void kBrownianUpdatePart2_kernel()
{
unsigned int pos = threadIdx.x + blockIdx.x * blockDim.x;
unsigned int rpos = cSim.pRandomPosition[blockIdx.x];
__syncthreads();
while (pos < cSim.atoms)
{
float4 velocity = cSim.pVelm4[pos];
float4 apos = cSim.pPosq[pos];
float4 xPrime = cSim.pPosqP[pos];
#ifndef DeltaShake
velocity.x = cSim.oneOverDeltaT*(xPrime.x-apos.x);
velocity.y = cSim.oneOverDeltaT*(xPrime.y-apos.y);
velocity.z = cSim.oneOverDeltaT*(xPrime.z-apos.z);
#else
velocity.x = cSim.oneOverDeltaT*(xPrime.x);
velocity.y = cSim.oneOverDeltaT*(xPrime.y);
velocity.z = cSim.oneOverDeltaT*(xPrime.z);
xPrime.x += apos.x;
xPrime.y += apos.y;
xPrime.z += apos.z;
#endif
cSim.pPosq[pos] = xPrime;
cSim.pVelm4[pos] = velocity;
pos += blockDim.x * gridDim.x;
}
// Update random position pointer
if (threadIdx.x == 0)
{
rpos += cSim.paddedNumberOfAtoms;
if (rpos > cSim.randoms)
rpos -= cSim.randoms;
cSim.pRandomPosition[blockIdx.x] = rpos;
}
}
extern void kGenerateRandoms(gpuContext gpu);
void kBrownianUpdatePart2(gpuContext gpu)
{
// printf("kBrownianUpdatePart2\n");
kBrownianUpdatePart2_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
LAUNCHERROR("kBrownianUpdatePart2");
// Update randoms if necessary
static int iteration = 0;
iteration++;
if (iteration == gpu->sim.randomIterations)
{
kGenerateRandoms(gpu);
iteration = 0;
}
}
platforms/cuda/src/kernels/kCalculateAndersenThermostat.cu 0 → 100755
View file @ 38f6c8f8
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include <stdio.h>
#include <cuda.h>
#include <vector_functions.h>
#include <cstdlib>
#include <string>
#include <iostream>
//#include <fstream>
using namespace std;
#include "gputypes.h"
static __constant__ cudaGmxSimulation cSim;
void SetCalculateAndersenThermostatSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyToSymbol(cSim, &gpu->sim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyToSymbol: SetSim copy to cSim failed");
}
void GetCalculateAndersenThermostatSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyFromSymbol(&gpu->sim, cSim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyFromSymbol: SetSim copy from cSim failed");
}
__global__ void kCalculateAndersenThermostat_kernel()
{
unsigned int pos = threadIdx.x + blockIdx.x * blockDim.x;
unsigned int rpos = cSim.pRandomPosition[blockIdx.x];
__syncthreads();
while (pos < cSim.atoms)
{
float4 velocity = cSim.pVelm4[pos];
float4 random4a = cSim.pRandom4a[rpos + pos];
float scale = (random4a.w < cSim.collisionProbability ? 0.0 : 1.0);
float add = (1.0-scale)*sqrt(cSim.kT*velocity.w);
velocity.x = scale*velocity.x + add*random4a.x;
velocity.y = scale*velocity.y + add*random4a.y;
velocity.z = scale*velocity.z + add*random4a.z;
cSim.pVelm4[pos] = velocity;
pos += blockDim.x * gridDim.x;
}
// Update random position pointer
if (threadIdx.x == 0)
{
rpos += cSim.paddedNumberOfAtoms;
if (rpos > cSim.randoms)
rpos -= cSim.randoms;
cSim.pRandomPosition[blockIdx.x] = rpos;
}
}
extern void kGenerateRandoms(gpuContext gpu);
void kCalculateAndersenThermostat(gpuContext gpu)
{
// printf("kCalculateAndersenThermostat\n");
kCalculateAndersenThermostat_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
LAUNCHERROR("kCalculateAndersenThermostat");
// Update randoms if necessary
static int iteration = 0;
iteration++;
if (iteration == gpu->sim.randomIterations)
{
kGenerateRandoms(gpu);
iteration = 0;
}
}
platforms/cuda/src/kernels/kCalculateCDLJForces.cu 0 → 100755
View file @ 38f6c8f8
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include <stdio.h>
#include <cuda.h>
#include <vector_functions.h>
#include <cstdlib>
#include <string>
#include <iostream>
#include <fstream>
using namespace std;
#include "gputypes.h"
#include "cudatypes.h"
#define UNROLLXX 0
#define UNROLLXY 0
struct Atom {
float x;
float y;
float z;
float q;
float sig;
float eps;
float fx;
float fy;
float fz;
float eps2;
float sig2;
};
__shared__ Atom sA[G8X_NONBOND_THREADS_PER_BLOCK];
__shared__ unsigned int sWorkUnit[G8X_NONBOND_WORKUNITS_PER_SM];
__shared__ unsigned int sNext[GRID];
static __constant__ cudaGmxSimulation cSim;
void SetCalculateCDLJForcesSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyToSymbol(cSim, &gpu->sim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyToSymbol: SetSim copy to cSim failed");
}
void GetCalculateCDLJForcesSim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyFromSymbol(&gpu->sim, cSim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyFromSymbol: SetSim copy from cSim failed");
}
__global__ void kCalculateCDLJForces_kernel()
{
// Read queue of work blocks once so the remainder of
// kernel can run asynchronously
int pos = cSim.nbWorkUnitsPerBlock * blockIdx.x + min(blockIdx.x, cSim.nbWorkUnitsPerBlockRemainder);
int end = cSim.nbWorkUnitsPerBlock * (blockIdx.x + 1) + min((blockIdx.x + 1), cSim.nbWorkUnitsPerBlockRemainder);
if (threadIdx.x < end - pos)
{
sWorkUnit[threadIdx.x] = cSim.pWorkUnit[pos + threadIdx.x];
}
if (threadIdx.x < GRID)
{
sNext[threadIdx.x] = (threadIdx.x + 1) & (GRID - 1);
}
__syncthreads();
// Now change pos and end to reflect work queue just read
// into shared memory
end = end - pos;
pos = end - (threadIdx.x >> GRIDBITS) - 1;
while (pos >= 0)
{
// Extract cell coordinates from appropriate work unit
unsigned int x = sWorkUnit[pos];
unsigned int y = ((x >> 2) & 0x7fff) << GRIDBITS;
bool bExclusionFlag = (x & 0x1);
x = (x >> 17) << GRIDBITS;
float4 apos; // Local atom x, y, z, q
float3 af; // Local atom fx, fy, fz
float dx;
float dy;
float dz;
float r2;
float invR;
float sig;
float sig2;
float sig6;
float eps;
float dEdR;
unsigned int tgx = threadIdx.x & (GRID - 1);
unsigned int tbx = threadIdx.x - tgx;
int tj = tgx;
Atom* psA = &sA[tbx];
if (!bExclusionFlag)
{
if (x == y) // Handle diagonals uniquely at 50% efficiency
{
// Read fixed atom data into registers and GRF
unsigned int i = x + tgx;
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = apos.x;
sA[threadIdx.x].y = apos.y;
sA[threadIdx.x].z = apos.z;
sA[threadIdx.x].q = apos.w;
sA[threadIdx.x].sig = a.x;
sA[threadIdx.x].eps = a.y;
af.x = 0.0f;
af.y = 0.0f;
af.z = 0.0f;
apos.w *= cSim.epsfac;
for (unsigned int j = 0; j < GRID; j++)
{
dx = psA[j].x - apos.x;
dy = psA[j].y - apos.y;
dz = psA[j].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[j].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[j].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[j].q * invR;
dEdR *= invR * invR;
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
else // 100% utilization
{
// Read fixed atom data into registers and GRF
int j = y + tgx;
unsigned int i = x + tgx;
float4 temp = cSim.pPosq[j];
float2 temp1 = cSim.pAttr[j];
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = temp.x;
sA[threadIdx.x].y = temp.y;
sA[threadIdx.x].z = temp.z;
sA[threadIdx.x].q = temp.w;
sA[threadIdx.x].sig = temp1.x;
sA[threadIdx.x].eps = temp1.y;
sA[threadIdx.x].fx = af.x = 0.0f;
sA[threadIdx.x].fy = af.y = 0.0f;
sA[threadIdx.x].fz = af.z = 0.0f;
sA[threadIdx.x].sig2 = a.x;
sA[threadIdx.x].eps2 = a.y;
apos.w *= cSim.epsfac;
for (j = 0; j < GRID; j++)
{
dx = psA[tj].x - apos.x;
dy = psA[tj].y - apos.y;
dz = psA[tj].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[tj].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[tj].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[tj].q * invR;
dEdR *= invR * invR;
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
psA[tj].fx += dx;
psA[tj].fy += dy;
psA[tj].fz += dz;
tj = sNext[tj];
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (y >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
of.x = sA[threadIdx.x].fx;
of.y = sA[threadIdx.x].fy;
of.z = sA[threadIdx.x].fz;
offset = y + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
}
else // bExclusion
{
// Read exclusion data
if (x == y) // Handle diagonals uniquely at 50% efficiency
{
// Read fixed atom data into registers and GRF
unsigned int excl = cSim.pExclusion[x * cSim.exclusionStride + y + tgx];
unsigned int i = x + tgx;
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = apos.x;
sA[threadIdx.x].y = apos.y;
sA[threadIdx.x].z = apos.z;
sA[threadIdx.x].q = apos.w;
sA[threadIdx.x].sig = a.x;
sA[threadIdx.x].eps = a.y;
af.x = 0.0f;
af.y = 0.0f;
af.z = 0.0f;
sA[threadIdx.x].sig2 = a.x;
sA[threadIdx.x].eps2 = a.y;
apos.w *= cSim.epsfac;
for (unsigned int j = 0; j < GRID; j++)
{
dx = psA[j].x - apos.x;
dy = psA[j].y - apos.y;
dz = psA[j].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = psA[tgx].sig2 + psA[j].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = psA[tgx].eps2 * psA[j].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[j].q * invR;
dEdR *= invR * invR;
if (!(excl & 0x1))
{
dEdR = 0.0f;
}
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
excl >>= 1;
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
else // 100% utilization
{
// Read fixed atom data into registers and GRF
unsigned int excl = cSim.pExclusion[x * cSim.exclusionStride + y + tgx];
excl = (excl >> tgx) | (excl << (GRID - tgx));
int j = y + tgx;
unsigned int i = x + tgx;
float4 temp = cSim.pPosq[j];
float2 temp1 = cSim.pAttr[j];
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = temp.x;
sA[threadIdx.x].y = temp.y;
sA[threadIdx.x].z = temp.z;
sA[threadIdx.x].q = temp.w;
sA[threadIdx.x].sig = temp1.x;
sA[threadIdx.x].eps = temp1.y;
sA[threadIdx.x].fx = af.x = 0.0f;
sA[threadIdx.x].fy = af.y = 0.0f;
sA[threadIdx.x].fz = af.z = 0.0f;
sA[threadIdx.x].sig2 = a.x;
sA[threadIdx.x].eps2 = a.y;
apos.w *= cSim.epsfac;
for (j = 0; j < GRID; j++)
{
dx = psA[tj].x - apos.x;
dy = psA[tj].y - apos.y;
dz = psA[tj].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = psA[tgx].sig2 + psA[tj].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = psA[tgx].eps2 * psA[tj].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[tj].q * invR;
dEdR *= invR * invR;
if (!(excl & 0x1))
{
dEdR = 0.0f;
}
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
psA[tj].fx += dx;
psA[tj].fy += dy;
psA[tj].fz += dz;
excl >>= 1;
tj = sNext[tj];
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (y >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
of.x = sA[threadIdx.x].fx;
of.y = sA[threadIdx.x].fy;
of.z = sA[threadIdx.x].fz;
offset = y + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
}
pos -= cSim.nonbond_workBlock;
}
}
__global__ extern void kCalculateCDLJForces_12_kernel();
void kCalculateCDLJForces(gpuContext gpu)
{
// printf("kCalculateCDLJForces\n");
if (gpu->sm_version < SM_12)
kCalculateCDLJForces_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block>>>();
else
kCalculateCDLJForces_12_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block>>>();
LAUNCHERROR("kCalculateCDLJForces");
}
\ No newline at end of file
platforms/cuda/src/kernels/kCalculateCDLJForces_12.cu 0 → 100755
View file @ 38f6c8f8
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Scott Le Grand, Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include <stdio.h>
#include <cuda.h>
#include <vector_functions.h>
#include <cstdlib>
#include <string>
#include <iostream>
#include <fstream>
using namespace std;
#include "gputypes.h"
#include "cudatypes.h"
#define UNROLLXX 0
#define UNROLLXY 0
struct Atom {
float x;
float y;
float z;
float q;
float sig;
float eps;
float fx;
float fy;
float fz;
};
__shared__ Atom sA[GT2XX_NONBOND_THREADS_PER_BLOCK];
__shared__ unsigned int sWorkUnit[GT2XX_NONBOND_WORKUNITS_PER_SM];
__shared__ unsigned int sNext[GRID];
static __constant__ cudaGmxSimulation cSim;
void SetCalculateCDLJForces_12Sim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyToSymbol(cSim, &gpu->sim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyToSymbol: SetSim copy to cSim failed");
}
void GetCalculateCDLJForces_12Sim(gpuContext gpu)
{
cudaError_t status;
status = cudaMemcpyFromSymbol(&gpu->sim, cSim, sizeof(cudaGmxSimulation));
RTERROR(status, "cudaMemcpyFromSymbol: SetSim copy from cSim failed");
}
__global__ void kCalculateCDLJForces_12_kernel()
{
// Read queue of work blocks once so the remainder of
// kernel can run asynchronously
int pos = cSim.nbWorkUnitsPerBlock * blockIdx.x + min(blockIdx.x, cSim.nbWorkUnitsPerBlockRemainder);
int end = cSim.nbWorkUnitsPerBlock * (blockIdx.x + 1) + min((blockIdx.x + 1), cSim.nbWorkUnitsPerBlockRemainder);
if (threadIdx.x < end - pos)
{
sWorkUnit[threadIdx.x] = cSim.pWorkUnit[pos + threadIdx.x];
}
if (threadIdx.x < GRID)
{
sNext[threadIdx.x] = (threadIdx.x + 1) & (GRID - 1);
}
__syncthreads();
// Now change pos and end to reflect work queue just read
// into shared memory
end = end - pos;
pos = end - (threadIdx.x >> GRIDBITS) - 1;
while (pos >= 0)
{
// Extract cell coordinates from appropriate work unit
unsigned int x = sWorkUnit[pos];
unsigned int y = ((x >> 2) & 0x7fff) << GRIDBITS;
bool bExclusionFlag = (x & 0x1);
x = (x >> 17) << GRIDBITS;
float4 apos; // Local atom x, y, z, q
float3 af; // Local atom fx, fy, fz
float dx;
float dy;
float dz;
float r2;
float invR;
float sig;
float sig2;
float sig6;
float eps;
float dEdR;
unsigned int tgx = threadIdx.x & (GRID - 1);
unsigned int tbx = threadIdx.x - tgx;
int tj = tgx;
Atom* psA = &sA[tbx];
if (!bExclusionFlag)
{
if (x == y) // Handle diagonals uniquely at 50% efficiency
{
// Read fixed atom data into registers and GRF
unsigned int i = x + tgx;
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = apos.x;
sA[threadIdx.x].y = apos.y;
sA[threadIdx.x].z = apos.z;
sA[threadIdx.x].q = apos.w;
sA[threadIdx.x].sig = a.x;
sA[threadIdx.x].eps = a.y;
af.x = 0.0f;
af.y = 0.0f;
af.z = 0.0f;
apos.w *= cSim.epsfac;
for (unsigned int j = 0; j < GRID; j++)
{
dx = psA[j].x - apos.x;
dy = psA[j].y - apos.y;
dz = psA[j].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[j].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[j].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[j].q * invR;
dEdR *= invR * invR;
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
else // 100% utilization
{
// Read fixed atom data into registers and GRF
int j = y + tgx;
unsigned int i = x + tgx;
float4 temp = cSim.pPosq[j];
float2 temp1 = cSim.pAttr[j];
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = temp.x;
sA[threadIdx.x].y = temp.y;
sA[threadIdx.x].z = temp.z;
sA[threadIdx.x].q = temp.w;
sA[threadIdx.x].sig = temp1.x;
sA[threadIdx.x].eps = temp1.y;
sA[threadIdx.x].fx = af.x = 0.0f;
sA[threadIdx.x].fy = af.y = 0.0f;
sA[threadIdx.x].fz = af.z = 0.0f;
apos.w *= cSim.epsfac;
for (j = 0; j < GRID; j++)
{
dx = psA[tj].x - apos.x;
dy = psA[tj].y - apos.y;
dz = psA[tj].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[tj].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[tj].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[tj].q * invR;
dEdR *= invR * invR;
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
psA[tj].fx += dx;
psA[tj].fy += dy;
psA[tj].fz += dz;
tj = sNext[tj];
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (y >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
of.x = sA[threadIdx.x].fx;
of.y = sA[threadIdx.x].fy;
of.z = sA[threadIdx.x].fz;
offset = y + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
}
else // bExclusion
{
// Read exclusion data
if (x == y) // Handle diagonals uniquely at 50% efficiency
{
// Read fixed atom data into registers and GRF
unsigned int excl = cSim.pExclusion[x * cSim.exclusionStride + y + tgx];
unsigned int i = x + tgx;
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = apos.x;
sA[threadIdx.x].y = apos.y;
sA[threadIdx.x].z = apos.z;
sA[threadIdx.x].q = apos.w;
sA[threadIdx.x].sig = a.x;
sA[threadIdx.x].eps = a.y;
af.x = 0.0f;
af.y = 0.0f;
af.z = 0.0f;
apos.w *= cSim.epsfac;
for (unsigned int j = 0; j < GRID; j++)
{
dx = psA[j].x - apos.x;
dy = psA[j].y - apos.y;
dz = psA[j].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[j].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[j].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[j].q * invR;
dEdR *= invR * invR;
if (!(excl & 0x1))
{
dEdR = 0.0f;
}
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
excl >>= 1;
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
else // 100% utilization
{
// Read fixed atom data into registers and GRF
unsigned int excl = cSim.pExclusion[x * cSim.exclusionStride + y + tgx];
excl = (excl >> tgx) | (excl << (GRID - tgx));
int j = y + tgx;
unsigned int i = x + tgx;
float4 temp = cSim.pPosq[j];
float2 temp1 = cSim.pAttr[j];
apos = cSim.pPosq[i];
float2 a = cSim.pAttr[i];
sA[threadIdx.x].x = temp.x;
sA[threadIdx.x].y = temp.y;
sA[threadIdx.x].z = temp.z;
sA[threadIdx.x].q = temp.w;
sA[threadIdx.x].sig = temp1.x;
sA[threadIdx.x].eps = temp1.y;
sA[threadIdx.x].fx = af.x = 0.0f;
sA[threadIdx.x].fy = af.y = 0.0f;
sA[threadIdx.x].fz = af.z = 0.0f;
apos.w *= cSim.epsfac;
for (j = 0; j < GRID; j++)
{
dx = psA[tj].x - apos.x;
dy = psA[tj].y - apos.y;
dz = psA[tj].z - apos.z;
r2 = dx * dx + dy * dy + dz * dz;
invR = 1.0f / sqrt(r2);
sig = a.x + psA[tj].sig;
sig2 = invR * sig;
sig2 *= sig2;
sig6 = sig2 * sig2 * sig2;
eps = a.y * psA[tj].eps;
dEdR = eps * (12.0f * sig6 - 6.0f) * sig6;
dEdR += apos.w * psA[tj].q * invR;
dEdR *= invR * invR;
if (!(excl & 0x1))
{
dEdR = 0.0f;
}
dx *= dEdR;
dy *= dEdR;
dz *= dEdR;
af.x -= dx;
af.y -= dy;
af.z -= dz;
psA[tj].fx += dx;
psA[tj].fy += dy;
psA[tj].fz += dz;
excl >>= 1;
tj = sNext[tj];
}
// Write results
float4 of;
of.x = af.x;
of.y = af.y;
of.z = af.z;
of.w = 0.0f;
int offset = x + tgx + (y >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
of.x = sA[threadIdx.x].fx;
of.y = sA[threadIdx.x].fy;
of.z = sA[threadIdx.x].fz;
offset = y + tgx + (x >> GRIDBITS) * cSim.stride;
cSim.pForce4a[offset] = of;
}
}
pos -= cSim.nonbond_workBlock;
}
}
void kCalculateCDLJForces_12(gpuContext gpu)
{
// printf("kCalculateCDLJForces_12\n");
kCalculateCDLJForces_12_kernel<<<gpu->sim.nonbond_blocks, gpu->sim.nonbond_threads_per_block>>>();
LAUNCHERROR("kCalculateCDLJForces_12");
}
platforms/cuda/src/kernels/kCalculateCDLJObcGbsaForces1.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateCDLJObcGbsaForces1_12.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateLocalForces.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateObcGbsaBornSum.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateObcGbsaForces1.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateObcGbsaForces1_12.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateObcGbsaForces2.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kCalculateObcGbsaForces2_12.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kForces.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kRandom.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kUpdateShakeH.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
platforms/cuda/src/kernels/kVerletUpdate.cu 0 → 100755
View file @ 38f6c8f8
This diff is collapsed.
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