Skip to content

GitLab

Commit 3db6b8ee authored by Jason Swails's avatar Jason Swails
Browse files

Merge branch 'master' into gbn-chk

parents 5d8e92ff 3a80b0f1
Expand all Show whitespace changes
Inline Side-by-side
Showing with 804 additions and 347 deletions
platforms/cuda/src/kernels/gbsaObc1.cu
View file @ 3db6b8ee
...@@ -69,7 +69,8 @@ typedef struct { ...@@ -69,7 +69,8 @@ typedef struct {
extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ global_bornSum, const real4* __restrict__ posq, const float2* __restrict__ global_params, extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ global_bornSum, const real4* __restrict__ posq, const float2* __restrict__ global_params,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, const real4* __restrict__ blockCenter, const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, const real4* __restrict__ blockCenter,
const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms,
#else #else
unsigned int numTiles, unsigned int numTiles,
#endif #endif
...@@ -104,9 +105,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa ...@@ -104,9 +105,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
for (unsigned int j = 0; j < TILE_SIZE; j++) { for (unsigned int j = 0; j < TILE_SIZE; j++) {
real3 delta = make_real3(localData[tbx+j].x-posq1.x, localData[tbx+j].y-posq1.y, localData[tbx+j].z-posq1.z); real3 delta = make_real3(localData[tbx+j].x-posq1.x, localData[tbx+j].y-posq1.y, localData[tbx+j].z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -151,9 +150,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa ...@@ -151,9 +150,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z); real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -291,12 +288,8 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa ...@@ -291,12 +288,8 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
// box, then skip having to apply periodic boundary conditions later. // box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
posq1.x -= floor((posq1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(posq1, blockCenterX)
posq1.y -= floor((posq1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y; APPLY_PERIODIC_TO_POS_WITH_CENTER(localData[threadIdx.x], blockCenterX)
posq1.z -= floor((posq1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
localData[threadIdx.x].x -= floor((localData[threadIdx.x].x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
localData[threadIdx.x].y -= floor((localData[threadIdx.x].y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
localData[threadIdx.x].z -= floor((localData[threadIdx.x].z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
unsigned int tj = tgx; unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z); real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z);
...@@ -342,9 +335,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa ...@@ -342,9 +335,7 @@ extern "C" __global__ void computeBornSum(unsigned long long* __restrict__ globa
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z); real3 delta = make_real3(localData[tbx+tj].x-posq1.x, localData[tbx+tj].y-posq1.y, localData[tbx+tj].z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
int atom2 = atomIndices[tbx+tj]; int atom2 = atomIndices[tbx+tj];
...@@ -414,7 +405,8 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo ...@@ -414,7 +405,8 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
real* __restrict__ energyBuffer, const real4* __restrict__ posq, const real* __restrict__ global_bornRadii, real* __restrict__ energyBuffer, const real4* __restrict__ posq, const real* __restrict__ global_bornRadii,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, const real4* __restrict__ blockCenter, const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, const real4* __restrict__ blockCenter,
const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms,
#else #else
unsigned int numTiles, unsigned int numTiles,
#endif #endif
...@@ -451,9 +443,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo ...@@ -451,9 +443,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
real4 posq2 = make_real4(localData[tbx+j].x, localData[tbx+j].y, localData[tbx+j].z, localData[tbx+j].q); real4 posq2 = make_real4(localData[tbx+j].x, localData[tbx+j].y, localData[tbx+j].z, localData[tbx+j].q);
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -508,9 +498,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo ...@@ -508,9 +498,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
real4 posq2 = make_real4(localData[tbx+tj].x, localData[tbx+tj].y, localData[tbx+tj].z, localData[tbx+tj].q); real4 posq2 = make_real4(localData[tbx+tj].x, localData[tbx+tj].y, localData[tbx+tj].z, localData[tbx+tj].q);
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -656,12 +644,8 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo ...@@ -656,12 +644,8 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
// box, then skip having to apply periodic boundary conditions later. // box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
posq1.x -= floor((posq1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(posq1, blockCenterX)
posq1.y -= floor((posq1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y; APPLY_PERIODIC_TO_POS_WITH_CENTER(localData[threadIdx.x], blockCenterX)
posq1.z -= floor((posq1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
localData[threadIdx.x].x -= floor((localData[threadIdx.x].x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
localData[threadIdx.x].y -= floor((localData[threadIdx.x].y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
localData[threadIdx.x].z -= floor((localData[threadIdx.x].z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
unsigned int tj = tgx; unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
int atom2 = atomIndices[tbx+tj]; int atom2 = atomIndices[tbx+tj];
...@@ -713,9 +697,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo ...@@ -713,9 +697,7 @@ extern "C" __global__ void computeGBSAForce1(unsigned long long* __restrict__ fo
real4 posq2 = make_real4(localData[tbx+tj].x, localData[tbx+tj].y, localData[tbx+tj].z, localData[tbx+tj].q); real4 posq2 = make_real4(localData[tbx+tj].x, localData[tbx+tj].y, localData[tbx+tj].z, localData[tbx+tj].q);
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
platforms/cuda/src/kernels/monteCarloBarostat.cu
View file @ 3db6b8ee
...@@ -2,7 +2,8 @@ ...@@ -2,7 +2,8 @@
* Scale the particle positions with each axis independent * Scale the particle positions with each axis independent
*/ */
extern "C" __global__ void scalePositions(float scaleX, float scaleY, float scaleZ, int numMolecules, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4* __restrict__ posq, extern "C" __global__ void scalePositions(float scaleX, float scaleY, float scaleZ, int numMolecules, real4 periodicBoxSize,
real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, real4* __restrict__ posq,
const int* __restrict__ moleculeAtoms, const int* __restrict__ moleculeStartIndex) { const int* __restrict__ moleculeAtoms, const int* __restrict__ moleculeStartIndex) {
for (int index = blockIdx.x*blockDim.x+threadIdx.x; index < numMolecules; index += blockDim.x*gridDim.x) { for (int index = blockIdx.x*blockDim.x+threadIdx.x; index < numMolecules; index += blockDim.x*gridDim.x) {
int first = moleculeStartIndex[index]; int first = moleculeStartIndex[index];
...@@ -25,13 +26,9 @@ extern "C" __global__ void scalePositions(float scaleX, float scaleY, float scal ...@@ -25,13 +26,9 @@ extern "C" __global__ void scalePositions(float scaleX, float scaleY, float scal
// Move it into the first periodic box. // Move it into the first periodic box.
int xcell = (int) floor(center.x*invPeriodicBoxSize.x); real3 oldCenter = center;
int ycell = (int) floor(center.y*invPeriodicBoxSize.y); APPLY_PERIODIC_TO_POS(center)
int zcell = (int) floor(center.z*invPeriodicBoxSize.z); real3 delta = make_real3(oldCenter.x-center.x, oldCenter.y-center.y, oldCenter.z-center.z);
real3 delta = make_real3(xcell*periodicBoxSize.x, ycell*periodicBoxSize.y, zcell*periodicBoxSize.z);
center.x -= delta.x;
center.y -= delta.y;
center.z -= delta.z;
// Now scale the position of the molecule center. // Now scale the position of the molecule center.
......
platforms/cuda/src/kernels/nonbonded.cu
View file @ 3db6b8ee
...@@ -103,8 +103,9 @@ extern "C" __global__ void computeNonbonded( ...@@ -103,8 +103,9 @@ extern "C" __global__ void computeNonbonded(
unsigned long long* __restrict__ forceBuffers, real* __restrict__ energyBuffer, const real4* __restrict__ posq, const tileflags* __restrict__ exclusions, unsigned long long* __restrict__ forceBuffers, real* __restrict__ energyBuffer, const real4* __restrict__ posq, const tileflags* __restrict__ exclusions,
const ushort2* __restrict__ exclusionTiles, unsigned int startTileIndex, unsigned int numTileIndices const ushort2* __restrict__ exclusionTiles, unsigned int startTileIndex, unsigned int numTileIndices
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
, const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, , const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount,real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, const real4* __restrict__ blockCenter, const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, const real4* __restrict__ blockCenter,
const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms
#endif #endif
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
const unsigned int totalWarps = (blockDim.x*gridDim.x)/TILE_SIZE; const unsigned int totalWarps = (blockDim.x*gridDim.x)/TILE_SIZE;
...@@ -155,9 +156,7 @@ extern "C" __global__ void computeNonbonded( ...@@ -155,9 +156,7 @@ extern "C" __global__ void computeNonbonded(
#endif #endif
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
real invR = RSQRT(r2); real invR = RSQRT(r2);
...@@ -223,9 +222,7 @@ extern "C" __global__ void computeNonbonded( ...@@ -223,9 +222,7 @@ extern "C" __global__ void computeNonbonded(
#endif #endif
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
real invR = RSQRT(r2); real invR = RSQRT(r2);
...@@ -412,17 +409,11 @@ extern "C" __global__ void computeNonbonded( ...@@ -412,17 +409,11 @@ extern "C" __global__ void computeNonbonded(
// The box is small enough that we can just translate all the atoms into a single periodic // The box is small enough that we can just translate all the atoms into a single periodic
// box, then skip having to apply periodic boundary conditions later. // box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
posq1.x -= floor((posq1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(posq1, blockCenterX)
posq1.y -= floor((posq1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
posq1.z -= floor((posq1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#ifdef ENABLE_SHUFFLE #ifdef ENABLE_SHUFFLE
shflPosq.x -= floor((shflPosq.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(shflPosq, blockCenterX)
shflPosq.y -= floor((shflPosq.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
shflPosq.z -= floor((shflPosq.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#else #else
localData[threadIdx.x].x -= floor((localData[threadIdx.x].x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(localData[threadIdx.x], blockCenterX)
localData[threadIdx.x].y -= floor((localData[threadIdx.x].y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
localData[threadIdx.x].z -= floor((localData[threadIdx.x].z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
unsigned int tj = tgx; unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
...@@ -499,9 +490,7 @@ extern "C" __global__ void computeNonbonded( ...@@ -499,9 +490,7 @@ extern "C" __global__ void computeNonbonded(
#endif #endif
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z); real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(delta)
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
real invR = RSQRT(r2); real invR = RSQRT(r2);
......
platforms/cuda/src/kernels/pme.cu
View file @ 3db6b8ee
extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int2* __restrict__ pmeAtomGridIndex, extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int2* __restrict__ pmeAtomGridIndex,
real4 periodicBoxSize, real4 invPeriodicBoxSize) { real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ) {
// Compute the index of the grid point each atom is associated with. // Compute the index of the grid point each atom is associated with.
for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) { for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) {
real4 pos = posq[i]; real4 pos = posq[i];
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x; real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y; pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z; pos.z*recipBoxVecZ.z);
real3 t = make_real3((pos.x*invPeriodicBoxSize.x)*GRID_SIZE_X, t.x = (t.x-floor(t.x))*GRID_SIZE_X;
(pos.y*invPeriodicBoxSize.y)*GRID_SIZE_Y, t.y = (t.y-floor(t.y))*GRID_SIZE_Y;
(pos.z*invPeriodicBoxSize.z)*GRID_SIZE_Z); t.z = (t.z-floor(t.z))*GRID_SIZE_Z;
int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X, int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X,
((int) t.y) % GRID_SIZE_Y, ((int) t.y) % GRID_SIZE_Y,
((int) t.z) % GRID_SIZE_Z); ((int) t.z) % GRID_SIZE_Z);
...@@ -18,7 +18,7 @@ extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int ...@@ -18,7 +18,7 @@ extern "C" __global__ void findAtomGridIndex(const real4* __restrict__ posq, int
} }
extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real* __restrict__ originalPmeGrid, extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real* __restrict__ originalPmeGrid,
real4 periodicBoxSize, real4 invPeriodicBoxSize, const int2* __restrict__ pmeAtomGridIndex) { real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real3 data[PME_ORDER]; real3 data[PME_ORDER];
const real scale = RECIP(PME_ORDER-1); const real scale = RECIP(PME_ORDER-1);
...@@ -27,15 +27,16 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real ...@@ -27,15 +27,16 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real
for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) { for (int i = blockIdx.x*blockDim.x+threadIdx.x; i < NUM_ATOMS; i += blockDim.x*gridDim.x) {
int atom = pmeAtomGridIndex[i].x; int atom = pmeAtomGridIndex[i].x;
real charge = posq[atom].w;
real3 force = make_real3(0);
real4 pos = posq[atom]; real4 pos = posq[atom];
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x; pos.x -= floor(pos.x*recipBoxVecX.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y; pos.y -= floor(pos.y*recipBoxVecY.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z; pos.z -= floor(pos.z*recipBoxVecZ.z)*periodicBoxSize.z;
real3 t = make_real3((pos.x*invPeriodicBoxSize.x)*GRID_SIZE_X, real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
(pos.y*invPeriodicBoxSize.y)*GRID_SIZE_Y, pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
(pos.z*invPeriodicBoxSize.z)*GRID_SIZE_Z); pos.z*recipBoxVecZ.z);
t.x = (t.x-floor(t.x))*GRID_SIZE_X;
t.y = (t.y-floor(t.y))*GRID_SIZE_Y;
t.z = (t.z-floor(t.z))*GRID_SIZE_Z;
int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X, int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X,
((int) t.y) % GRID_SIZE_Y, ((int) t.y) % GRID_SIZE_Y,
((int) t.z) % GRID_SIZE_Z); ((int) t.z) % GRID_SIZE_Z);
...@@ -78,7 +79,7 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real ...@@ -78,7 +79,7 @@ extern "C" __global__ void gridSpreadCharge(const real4* __restrict__ posq, real
zindex -= (zindex >= GRID_SIZE_Z ? GRID_SIZE_Z : 0); zindex -= (zindex >= GRID_SIZE_Z ? GRID_SIZE_Z : 0);
int index = ybase + zindex; int index = ybase + zindex;
real add = charge*dx*dy*data[iz].z; real add = pos.w*dx*dy*data[iz].z;
#ifdef USE_DOUBLE_PRECISION #ifdef USE_DOUBLE_PRECISION
unsigned long long * ulonglong_p = (unsigned long long *) originalPmeGrid; unsigned long long * ulonglong_p = (unsigned long long *) originalPmeGrid;
atomicAdd(&ulonglong_p[index], static_cast<unsigned long long>((long long) (add*0x100000000))); atomicAdd(&ulonglong_p[index], static_cast<unsigned long long>((long long) (add*0x100000000)));
...@@ -115,9 +116,8 @@ extern "C" __global__ void finishSpreadCharge(long long* __restrict__ originalPm ...@@ -115,9 +116,8 @@ extern "C" __global__ void finishSpreadCharge(long long* __restrict__ originalPm
// convolutes on the halfcomplex_pmeGrid, which is of size NX*NY*(NZ/2+1) as F(Q) is conjugate symmetric // convolutes on the halfcomplex_pmeGrid, which is of size NX*NY*(NZ/2+1) as F(Q) is conjugate symmetric
extern "C" __global__ void extern "C" __global__ void
reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ energyBuffer, reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ energyBuffer,
const real* __restrict__ pmeBsplineModuliX, const real* __restrict__ pmeBsplineModuliX, const real* __restrict__ pmeBsplineModuliY, const real* __restrict__ pmeBsplineModuliZ,
const real* __restrict__ pmeBsplineModuliY, const real* __restrict__ pmeBsplineModuliZ, real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ) {
real4 periodicBoxSize, real4 invPeriodicBoxSize) {
// R2C stores into a half complex matrix where the last dimension is cut by half // R2C stores into a half complex matrix where the last dimension is cut by half
const unsigned int gridSize = GRID_SIZE_X*GRID_SIZE_Y*(GRID_SIZE_Z/2+1); const unsigned int gridSize = GRID_SIZE_X*GRID_SIZE_Y*(GRID_SIZE_Z/2+1);
const real recipScaleFactor = RECIP(M_PI*periodicBoxSize.x*periodicBoxSize.y*periodicBoxSize.z); const real recipScaleFactor = RECIP(M_PI*periodicBoxSize.x*periodicBoxSize.y*periodicBoxSize.z);
...@@ -131,9 +131,9 @@ reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict_ ...@@ -131,9 +131,9 @@ reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict_
int mx = (kx < (GRID_SIZE_X+1)/2) ? kx : (kx-GRID_SIZE_X); int mx = (kx < (GRID_SIZE_X+1)/2) ? kx : (kx-GRID_SIZE_X);
int my = (ky < (GRID_SIZE_Y+1)/2) ? ky : (ky-GRID_SIZE_Y); int my = (ky < (GRID_SIZE_Y+1)/2) ? ky : (ky-GRID_SIZE_Y);
int mz = (kz < (GRID_SIZE_Z+1)/2) ? kz : (kz-GRID_SIZE_Z); int mz = (kz < (GRID_SIZE_Z+1)/2) ? kz : (kz-GRID_SIZE_Z);
real mhx = mx*invPeriodicBoxSize.x; real mhx = mx*recipBoxVecX.x;
real mhy = my*invPeriodicBoxSize.y; real mhy = mx*recipBoxVecY.x+my*recipBoxVecY.y;
real mhz = mz*invPeriodicBoxSize.z; real mhz = mx*recipBoxVecZ.x+my*recipBoxVecZ.y+mz*recipBoxVecZ.z;
real bx = pmeBsplineModuliX[kx]; real bx = pmeBsplineModuliX[kx];
real by = pmeBsplineModuliY[ky]; real by = pmeBsplineModuliY[ky];
real bz = pmeBsplineModuliZ[kz]; real bz = pmeBsplineModuliZ[kz];
...@@ -151,9 +151,8 @@ reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict_ ...@@ -151,9 +151,8 @@ reciprocalConvolution(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict_
extern "C" __global__ void extern "C" __global__ void
gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ energyBuffer, gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ energyBuffer,
const real* __restrict__ pmeBsplineModuliX, const real* __restrict__ pmeBsplineModuliX, const real* __restrict__ pmeBsplineModuliY, const real* __restrict__ pmeBsplineModuliZ,
const real* __restrict__ pmeBsplineModuliY, const real* __restrict__ pmeBsplineModuliZ, real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ) {
real4 periodicBoxSize, real4 invPeriodicBoxSize) {
// R2C stores into a half complex matrix where the last dimension is cut by half // R2C stores into a half complex matrix where the last dimension is cut by half
const unsigned int gridSize = GRID_SIZE_X*GRID_SIZE_Y*GRID_SIZE_Z; const unsigned int gridSize = GRID_SIZE_X*GRID_SIZE_Y*GRID_SIZE_Z;
const real recipScaleFactor = RECIP(M_PI*periodicBoxSize.x*periodicBoxSize.y*periodicBoxSize.z); const real recipScaleFactor = RECIP(M_PI*periodicBoxSize.x*periodicBoxSize.y*periodicBoxSize.z);
...@@ -168,9 +167,9 @@ gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ e ...@@ -168,9 +167,9 @@ gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ e
int mx = (kx < (GRID_SIZE_X+1)/2) ? kx : (kx-GRID_SIZE_X); int mx = (kx < (GRID_SIZE_X+1)/2) ? kx : (kx-GRID_SIZE_X);
int my = (ky < (GRID_SIZE_Y+1)/2) ? ky : (ky-GRID_SIZE_Y); int my = (ky < (GRID_SIZE_Y+1)/2) ? ky : (ky-GRID_SIZE_Y);
int mz = (kz < (GRID_SIZE_Z+1)/2) ? kz : (kz-GRID_SIZE_Z); int mz = (kz < (GRID_SIZE_Z+1)/2) ? kz : (kz-GRID_SIZE_Z);
real mhx = mx*invPeriodicBoxSize.x; real mhx = mx*recipBoxVecX.x;
real mhy = my*invPeriodicBoxSize.y; real mhy = mx*recipBoxVecY.x+my*recipBoxVecY.y;
real mhz = mz*invPeriodicBoxSize.z; real mhz = mx*recipBoxVecZ.x+my*recipBoxVecZ.y+mz*recipBoxVecZ.z;
real m2 = mhx*mhx+mhy*mhy+mhz*mhz; real m2 = mhx*mhx+mhy*mhy+mhz*mhz;
real bx = pmeBsplineModuliX[kx]; real bx = pmeBsplineModuliX[kx];
real by = pmeBsplineModuliY[ky]; real by = pmeBsplineModuliY[ky];
...@@ -194,7 +193,7 @@ gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ e ...@@ -194,7 +193,7 @@ gridEvaluateEnergy(real2* __restrict__ halfcomplex_pmeGrid, real* __restrict__ e
extern "C" __global__ extern "C" __global__
void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __restrict__ forceBuffers, const real* __restrict__ originalPmeGrid, void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __restrict__ forceBuffers, const real* __restrict__ originalPmeGrid,
real4 periodicBoxSize, real4 invPeriodicBoxSize, const int2* __restrict__ pmeAtomGridIndex) { real4 periodicBoxSize, real3 recipBoxVecX, real3 recipBoxVecY, real3 recipBoxVecZ, const int2* __restrict__ pmeAtomGridIndex) {
real3 data[PME_ORDER]; real3 data[PME_ORDER];
real3 ddata[PME_ORDER]; real3 ddata[PME_ORDER];
const real scale = RECIP(PME_ORDER-1); const real scale = RECIP(PME_ORDER-1);
...@@ -206,12 +205,15 @@ void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __ ...@@ -206,12 +205,15 @@ void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __
int atom = pmeAtomGridIndex[i].x; int atom = pmeAtomGridIndex[i].x;
real3 force = make_real3(0); real3 force = make_real3(0);
real4 pos = posq[atom]; real4 pos = posq[atom];
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x; pos.x -= floor(pos.x*recipBoxVecX.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y; pos.y -= floor(pos.y*recipBoxVecY.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z; pos.z -= floor(pos.z*recipBoxVecZ.z)*periodicBoxSize.z;
real3 t = make_real3((pos.x*invPeriodicBoxSize.x)*GRID_SIZE_X, real3 t = make_real3(pos.x*recipBoxVecX.x+pos.y*recipBoxVecY.x+pos.z*recipBoxVecZ.x,
(pos.y*invPeriodicBoxSize.y)*GRID_SIZE_Y, pos.y*recipBoxVecY.y+pos.z*recipBoxVecZ.y,
(pos.z*invPeriodicBoxSize.z)*GRID_SIZE_Z); pos.z*recipBoxVecZ.z);
t.x = (t.x-floor(t.x))*GRID_SIZE_X;
t.y = (t.y-floor(t.y))*GRID_SIZE_Y;
t.z = (t.z-floor(t.z))*GRID_SIZE_Z;
int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X, int3 gridIndex = make_int3(((int) t.x) % GRID_SIZE_X,
((int) t.y) % GRID_SIZE_Y, ((int) t.y) % GRID_SIZE_Y,
((int) t.z) % GRID_SIZE_Z); ((int) t.z) % GRID_SIZE_Z);
...@@ -266,9 +268,12 @@ void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __ ...@@ -266,9 +268,12 @@ void gridInterpolateForce(const real4* __restrict__ posq, unsigned long long* __
} }
} }
real q = pos.w*EPSILON_FACTOR; real q = pos.w*EPSILON_FACTOR;
forceBuffers[atom] += static_cast<unsigned long long>((long long) (-q*force.x*GRID_SIZE_X*invPeriodicBoxSize.x*0x100000000)); real forceX = -q*(force.x*GRID_SIZE_X*recipBoxVecX.x);
forceBuffers[atom+PADDED_NUM_ATOMS] += static_cast<unsigned long long>((long long) (-q*force.y*GRID_SIZE_Y*invPeriodicBoxSize.y*0x100000000)); real forceY = -q*(force.x*GRID_SIZE_X*recipBoxVecY.x+force.y*GRID_SIZE_Y*recipBoxVecY.y);
forceBuffers[atom+2*PADDED_NUM_ATOMS] += static_cast<unsigned long long>((long long) (-q*force.z*GRID_SIZE_Z*invPeriodicBoxSize.z*0x100000000)); real forceZ = -q*(force.x*GRID_SIZE_X*recipBoxVecZ.x+force.y*GRID_SIZE_Y*recipBoxVecZ.y+force.z*GRID_SIZE_Z*recipBoxVecZ.z);
forceBuffers[atom] += static_cast<unsigned long long>((long long) (forceX*0x100000000));
forceBuffers[atom+PADDED_NUM_ATOMS] += static_cast<unsigned long long>((long long) (forceY*0x100000000));
forceBuffers[atom+2*PADDED_NUM_ATOMS] += static_cast<unsigned long long>((long long) (forceZ*0x100000000));
} }
} }
......
platforms/cuda/tests/TestCudaCustomManyParticleForce.cpp
View file @ 3db6b8ee
...@@ -55,7 +55,17 @@ const double TOL = 1e-5; ...@@ -55,7 +55,17 @@ const double TOL = 1e-5;
CudaPlatform platform; CudaPlatform platform;
void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& positions, const vector<const int*>& expectedSets, double boxSize) { Vec3 computeDelta(const Vec3& pos1, const Vec3& pos2, bool periodic, const Vec3* periodicBoxVectors) {
Vec3 diff = pos1-pos2;
if (periodic) {
diff -= periodicBoxVectors[2]*floor(diff[2]/periodicBoxVectors[2][2]+0.5);
diff -= periodicBoxVectors[1]*floor(diff[1]/periodicBoxVectors[1][1]+0.5);
diff -= periodicBoxVectors[0]*floor(diff[0]/periodicBoxVectors[0][0]+0.5);
}
return diff;
}
void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& positions, const vector<const int*>& expectedSets, double boxSize, bool triclinic) {
// Create a System and Context. // Create a System and Context.
int numParticles = force->getNumParticles(); int numParticles = force->getNumParticles();
...@@ -63,7 +73,18 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p ...@@ -63,7 +73,18 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p
System system; System system;
for (int i = 0; i < numParticles; i++) for (int i = 0; i < numParticles; i++)
system.addParticle(1.0); system.addParticle(1.0);
system.setDefaultPeriodicBoxVectors(Vec3(boxSize, 0, 0), Vec3(0, boxSize, 0), Vec3(0, 0, boxSize)); Vec3 boxVectors[3];
if (triclinic) {
boxVectors[0] = Vec3(boxSize, 0, 0);
boxVectors[1] = Vec3(0.2*boxSize, boxSize, 0);
boxVectors[2] = Vec3(-0.3*boxSize, -0.1*boxSize, boxSize);
}
else {
boxVectors[0] = Vec3(boxSize, 0, 0);
boxVectors[1] = Vec3(0, boxSize, 0);
boxVectors[2] = Vec3(0, 0, boxSize);
}
system.setDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
system.addForce(force); system.addForce(force);
VerletIntegrator integrator(0.001); VerletIntegrator integrator(0.001);
Context context(system, integrator, platform); Context context(system, integrator, platform);
...@@ -74,20 +95,14 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p ...@@ -74,20 +95,14 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p
// See if the energy matches the expected value. // See if the energy matches the expected value.
double expectedEnergy = 0; double expectedEnergy = 0;
bool periodic = (nonbondedMethod == CustomManyParticleForce::CutoffPeriodic);
for (int i = 0; i < (int) expectedSets.size(); i++) { for (int i = 0; i < (int) expectedSets.size(); i++) {
int p1 = expectedSets[i][0]; int p1 = expectedSets[i][0];
int p2 = expectedSets[i][1]; int p2 = expectedSets[i][1];
int p3 = expectedSets[i][2]; int p3 = expectedSets[i][2];
Vec3 d12 = positions[p2]-positions[p1]; Vec3 d12 = computeDelta(positions[p2], positions[p1], periodic, boxVectors);
Vec3 d13 = positions[p3]-positions[p1]; Vec3 d13 = computeDelta(positions[p3], positions[p1], periodic, boxVectors);
Vec3 d23 = positions[p3]-positions[p2]; Vec3 d23 = computeDelta(positions[p3], positions[p2], periodic, boxVectors);
if (nonbondedMethod == CustomManyParticleForce::CutoffPeriodic) {
for (int j = 0; j < 3; j++) {
d12[j] -= floor(d12[j]/boxSize+0.5f)*boxSize;
d13[j] -= floor(d13[j]/boxSize+0.5f)*boxSize;
d23[j] -= floor(d23[j]/boxSize+0.5f)*boxSize;
}
}
double r12 = sqrt(d12.dot(d12)); double r12 = sqrt(d12.dot(d12));
double r13 = sqrt(d13.dot(d13)); double r13 = sqrt(d13.dot(d13));
double r23 = sqrt(d23.dot(d23)); double r23 = sqrt(d23.dot(d23));
...@@ -210,7 +225,7 @@ void testNoCutoff() { ...@@ -210,7 +225,7 @@ void testNoCutoff() {
positions.push_back(Vec3(0.4, 0, -0.8)); positions.push_back(Vec3(0.4, 0, -0.8));
int sets[4][3] = {{0,1,2}, {1,2,3}, {2,3,0}, {3,0,1}}; int sets[4][3] = {{0,1,2}, {1,2,3}, {2,3,0}, {3,0,1}};
vector<const int*> expectedSets(&sets[0], &sets[4]); vector<const int*> expectedSets(&sets[0], &sets[4]);
validateAxilrodTeller(force, positions, expectedSets, 2.0); validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
} }
void testCutoff() { void testCutoff() {
...@@ -235,7 +250,7 @@ void testCutoff() { ...@@ -235,7 +250,7 @@ void testCutoff() {
positions.push_back(Vec3(0.2, 0.5, -0.1)); positions.push_back(Vec3(0.2, 0.5, -0.1));
int sets[7][3] = {{0,1,2}, {0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,2,4}, {1,3,4}}; int sets[7][3] = {{0,1,2}, {0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,2,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[7]); vector<const int*> expectedSets(&sets[0], &sets[7]);
validateAxilrodTeller(force, positions, expectedSets, 2.0); validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
} }
void testPeriodic() { void testPeriodic() {
...@@ -261,7 +276,33 @@ void testPeriodic() { ...@@ -261,7 +276,33 @@ void testPeriodic() {
double boxSize = 2.1; double boxSize = 2.1;
int sets[5][3] = {{0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,3,4}}; int sets[5][3] = {{0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[5]); vector<const int*> expectedSets(&sets[0], &sets[5]);
validateAxilrodTeller(force, positions, expectedSets, boxSize); validateAxilrodTeller(force, positions, expectedSets, boxSize, false);
}
void testTriclinic() {
CustomManyParticleForce* force = new CustomManyParticleForce(3,
"C*(1+3*cos(theta1)*cos(theta2)*cos(theta3))/(r12*r13*r23)^3;"
"theta1=angle(p1,p2,p3); theta2=angle(p2,p3,p1); theta3=angle(p3,p1,p2);"
"r12=distance(p1,p2); r13=distance(p1,p3); r23=distance(p2,p3)");
force->addGlobalParameter("C", 1.5);
force->setNonbondedMethod(CustomManyParticleForce::CutoffPeriodic);
force->setCutoffDistance(1.05);
vector<double> params;
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
vector<Vec3> positions;
positions.push_back(Vec3(0, 0, 0));
positions.push_back(Vec3(1, 0, 0));
positions.push_back(Vec3(0, 1.1, 0.3));
positions.push_back(Vec3(0.4, 0, -0.8));
positions.push_back(Vec3(0.2, 0.5, -0.1));
double boxSize = 2.1;
int sets[4][3] = {{0,1,3}, {0,1,4}, {0,3,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[4]);
validateAxilrodTeller(force, positions, expectedSets, boxSize, true);
} }
void testExclusions() { void testExclusions() {
...@@ -286,7 +327,7 @@ void testExclusions() { ...@@ -286,7 +327,7 @@ void testExclusions() {
force->addExclusion(0, 3); force->addExclusion(0, 3);
int sets[5][3] = {{0,1,4}, {1,2,3}, {1,2,4}, {1,3,4}, {2,3,4}}; int sets[5][3] = {{0,1,4}, {1,2,3}, {1,2,4}, {1,3,4}, {2,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[5]); vector<const int*> expectedSets(&sets[0], &sets[5]);
validateAxilrodTeller(force, positions, expectedSets, 2.0); validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
} }
void testAllTerms() { void testAllTerms() {
...@@ -672,6 +713,7 @@ int main(int argc, char* argv[]) { ...@@ -672,6 +713,7 @@ int main(int argc, char* argv[]) {
testNoCutoff(); testNoCutoff();
testCutoff(); testCutoff();
testPeriodic(); testPeriodic();
testTriclinic();
testExclusions(); testExclusions();
testAllTerms(); testAllTerms();
testParameters(); testParameters();
......
platforms/cuda/tests/TestCudaCustomNonbondedForce.cpp
View file @ 3db6b8ee
...@@ -7,7 +7,7 @@ ...@@ -7,7 +7,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2008-2014 Stanford University and the Authors. * * Portions copyright (c) 2008-2015 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -261,6 +261,65 @@ void testPeriodic() { ...@@ -261,6 +261,65 @@ void testPeriodic() {
ASSERT_EQUAL_TOL(1.9+1+0.9, state.getPotentialEnergy(), TOL); ASSERT_EQUAL_TOL(1.9+1+0.9, state.getPotentialEnergy(), TOL);
} }
void testTriclinic() {
System system;
system.addParticle(1.0);
system.addParticle(1.0);
Vec3 a(3.1, 0, 0);
Vec3 b(0.4, 3.5, 0);
Vec3 c(-0.1, -0.5, 4.0);
system.setDefaultPeriodicBoxVectors(a, b, c);
VerletIntegrator integrator(0.01);
CustomNonbondedForce* nonbonded = new CustomNonbondedForce("r");
nonbonded->addParticle(vector<double>());
nonbonded->addParticle(vector<double>());
nonbonded->setNonbondedMethod(CustomNonbondedForce::CutoffPeriodic);
const double cutoff = 1.5;
nonbonded->setCutoffDistance(cutoff);
system.addForce(nonbonded);
Context context(system, integrator, platform);
vector<Vec3> positions(2);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int iteration = 0; iteration < 50; iteration++) {
// Generate random positions for the two particles.
positions[0] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
positions[1] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
context.setPositions(positions);
// Loop over all possible periodic copies and find the nearest one.
Vec3 delta;
double distance2 = 100.0;
for (int i = -1; i < 2; i++)
for (int j = -1; j < 2; j++)
for (int k = -1; k < 2; k++) {
Vec3 d = positions[1]-positions[0]+a*i+b*j+c*k;
if (d.dot(d) < distance2) {
delta = d;
distance2 = d.dot(d);
}
}
double distance = sqrt(distance2);
// See if the force and energy are correct.
State state = context.getState(State::Forces | State::Energy);
if (distance >= cutoff) {
ASSERT_EQUAL(0.0, state.getPotentialEnergy());
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[0], 0);
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[1], 0);
}
else {
const Vec3 force = delta/sqrt(delta.dot(delta));
ASSERT_EQUAL_TOL(distance, state.getPotentialEnergy(), TOL);
ASSERT_EQUAL_VEC(force, state.getForces()[0], TOL);
ASSERT_EQUAL_VEC(-force, state.getForces()[1], TOL);
}
}
}
void testContinuous1DFunction() { void testContinuous1DFunction() {
System system; System system;
system.addParticle(1.0); system.addParticle(1.0);
...@@ -925,6 +984,7 @@ int main(int argc, char* argv[]) { ...@@ -925,6 +984,7 @@ int main(int argc, char* argv[]) {
testExclusions(); testExclusions();
testCutoff(); testCutoff();
testPeriodic(); testPeriodic();
testTriclinic();
testContinuous1DFunction(); testContinuous1DFunction();
testContinuous2DFunction(); testContinuous2DFunction();
testContinuous3DFunction(); testContinuous3DFunction();
......
platforms/cuda/tests/TestCudaEwald.cpp
View file @ 3db6b8ee
...@@ -201,6 +201,57 @@ void testEwald2Ions() { ...@@ -201,6 +201,57 @@ void testEwald2Ions() {
ASSERT_EQUAL_TOL(-217.276, state.getPotentialEnergy(), 0.01/*10*TOL*/); ASSERT_EQUAL_TOL(-217.276, state.getPotentialEnergy(), 0.01/*10*TOL*/);
} }
void testTriclinic() {
// Create a triclinic box containing eight particles.
System system;
system.setDefaultPeriodicBoxVectors(Vec3(2.5, 0, 0), Vec3(0.5, 3.0, 0), Vec3(0.7, 0.9, 3.5));
for (int i = 0; i < 8; i++)
system.addParticle(1.0);
NonbondedForce* force = new NonbondedForce();
system.addForce(force);
force->setNonbondedMethod(NonbondedForce::PME);
force->setCutoffDistance(1.0);
force->setPMEParameters(3.45891, 32, 40, 48);
for (int i = 0; i < 4; i++)
force->addParticle(-1, 0.440104, 0.4184); // Cl parameters
for (int i = 0; i < 4; i++)
force->addParticle(1, 0.332840, 0.0115897); // Na parameters
vector<Vec3> positions(8);
positions[0] = Vec3(1.744, 2.788, 3.162);
positions[1] = Vec3(1.048, 0.762, 2.340);
positions[2] = Vec3(2.489, 1.570, 2.817);
positions[3] = Vec3(1.027, 1.893, 3.271);
positions[4] = Vec3(0.937, 0.825, 0.009);
positions[5] = Vec3(2.290, 1.887, 3.352);
positions[6] = Vec3(1.266, 1.111, 2.894);
positions[7] = Vec3(0.933, 1.862, 3.490);
// Compute the forces and energy.
VerletIntegrator integ(0.001);
Context context(system, integ, platform);
context.setPositions(positions);
State state = context.getState(State::Forces | State::Energy);
// Compare them to values computed by Gromacs.
double expectedEnergy = -963.370;
vector<Vec3> expectedForce(8);
expectedForce[0] = Vec3(4.25253e+01, -1.23503e+02, 1.22139e+02);
expectedForce[1] = Vec3(9.74752e+01, 1.68213e+02, 1.93169e+02);
expectedForce[2] = Vec3(-1.50348e+02, 1.29165e+02, 3.70435e+02);
expectedForce[3] = Vec3(9.18644e+02, -3.52571e+00, -1.34772e+03);
expectedForce[4] = Vec3(-1.61193e+02, 9.01528e+01, -7.12904e+01);
expectedForce[5] = Vec3(2.82630e+02, 2.78029e+01, -3.72864e+02);
expectedForce[6] = Vec3(-1.47454e+02, -2.14448e+02, -3.55789e+02);
expectedForce[7] = Vec3(-8.82195e+02, -7.39132e+01, 1.46202e+03);
for (int i = 0; i < 8; i++) {
ASSERT_EQUAL_VEC(expectedForce[i], state.getForces()[i], 1e-4);
}
ASSERT_EQUAL_TOL(expectedEnergy, state.getPotentialEnergy(), 1e-4);
}
void testErrorTolerance(NonbondedForce::NonbondedMethod method) { void testErrorTolerance(NonbondedForce::NonbondedMethod method) {
// Create a cloud of random point charges. // Create a cloud of random point charges.
...@@ -307,6 +358,7 @@ int main(int argc, char* argv[]) { ...@@ -307,6 +358,7 @@ int main(int argc, char* argv[]) {
testEwaldPME(false); testEwaldPME(false);
testEwaldPME(true); testEwaldPME(true);
// testEwald2Ions(); // testEwald2Ions();
testTriclinic();
testErrorTolerance(NonbondedForce::Ewald); testErrorTolerance(NonbondedForce::Ewald);
testErrorTolerance(NonbondedForce::PME); testErrorTolerance(NonbondedForce::PME);
testPMEParameters(); testPMEParameters();
......
platforms/cuda/tests/TestCudaMonteCarloAnisotropicBarostat.cpp
View file @ 3db6b8ee
...@@ -236,6 +236,82 @@ void testRandomSeed() { ...@@ -236,6 +236,82 @@ void testRandomSeed() {
} }
} }
void testTriclinic() {
const int numParticles = 64;
const int frequency = 10;
const int steps = 1000;
const double pressure = 1.5;
const double pressureInMD = pressure*(AVOGADRO*1e-25); // pressure in kJ/mol/nm^3
const double temperature = 300.0;
const double initialVolume = numParticles*BOLTZ*temperature/pressureInMD;
const double initialLength = std::pow(initialVolume, 1.0/3.0);
// Create a gas of noninteracting particles.
System system;
Vec3 initialBox[3];
initialBox[0] = Vec3(initialLength, 0, 0);
initialBox[1] = Vec3(0.2*initialLength, initialLength, 0);
initialBox[2] = Vec3(0.1*initialLength, 0.3*initialLength, initialLength);
system.setDefaultPeriodicBoxVectors(initialBox[0], initialBox[1], initialBox[2]);
vector<Vec3> positions(numParticles);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int i = 0; i < numParticles; ++i) {
system.addParticle(1.0);
positions[i] = Vec3(initialLength*genrand_real2(sfmt), initialLength*genrand_real2(sfmt), initialLength*genrand_real2(sfmt));
}
MonteCarloAnisotropicBarostat* barostat = new MonteCarloAnisotropicBarostat(Vec3(pressure, pressure, pressure), temperature, true, true, true, frequency);
system.addForce(barostat);
// Run a simulation
LangevinIntegrator integrator(temperature, 0.1, 0.01);
Context context(system, integrator, platform);
context.setPositions(positions);
// Let it equilibrate.
integrator.step(10000);
// Now run it for a while and see if the volume is correct.
double volume = 0.0;
for (int j = 0; j < steps; ++j) {
Vec3 box[3];
context.getState(0).getPeriodicBoxVectors(box[0], box[1], box[2]);
volume += box[0][0]*box[1][1]*box[2][2];
integrator.step(frequency);
}
volume /= steps;
double expected = (numParticles+1)*BOLTZ*temperature/pressureInMD;
ASSERT_USUALLY_EQUAL_TOL(expected, volume, 3/std::sqrt((double) steps));
// Make sure the box vectors have been scaled consistently.
State state = context.getState(State::Positions);
Vec3 box[3];
state.getPeriodicBoxVectors(box[0], box[1], box[2]);
double xscale = box[2][0]/(0.1*initialLength);
double yscale = box[2][1]/(0.3*initialLength);
double zscale = box[2][2]/(1.0*initialLength);
for (int i = 0; i < 3; i++) {
ASSERT_EQUAL_VEC(Vec3(xscale*initialBox[i][0], yscale*initialBox[i][1], zscale*initialBox[i][2]), box[i], 1e-5);
}
// The barostat should have put all particles inside the first periodic box. One integration step
// has happened since then, so they may have moved slightly outside it.
for (int i = 0; i < numParticles; i++) {
Vec3 pos = state.getPositions()[i];
ASSERT(pos[2]/box[2][2] > -1 && pos[2]/box[2][2] < 2);
pos -= box[2]*floor(pos[2]/box[2][2]);
ASSERT(pos[1]/box[1][1] > -1 && pos[1]/box[1][1] < 2);
pos -= box[1]*floor(pos[1]/box[1][1]);
ASSERT(pos[0]/box[0][0] > -1 && pos[0]/box[0][0] < 2);
}
}
/** /**
* Run a constant pressure simulation on an anisotropic Einstein crystal * Run a constant pressure simulation on an anisotropic Einstein crystal
* using isotropic and anisotropic barostats. There are a total of 15 simulations: * using isotropic and anisotropic barostats. There are a total of 15 simulations:
...@@ -389,6 +465,7 @@ int main(int argc, char* argv[]) { ...@@ -389,6 +465,7 @@ int main(int argc, char* argv[]) {
testIdealGasAxis(1); testIdealGasAxis(1);
testIdealGasAxis(2); testIdealGasAxis(2);
testRandomSeed(); testRandomSeed();
testTriclinic();
//testEinsteinCrystal(); //testEinsteinCrystal();
} }
catch(const exception& e) { catch(const exception& e) {
......
platforms/cuda/tests/TestCudaNonbondedForce.cpp
View file @ 3db6b8ee
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2008-2013 Stanford University and the Authors. * * Portions copyright (c) 2008-2015 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -355,6 +355,67 @@ void testPeriodic() { ...@@ -355,6 +355,67 @@ void testPeriodic() {
ASSERT_EQUAL_TOL(2*ONE_4PI_EPS0*(1.0)*(1.0+krf*1.0-crf), state.getPotentialEnergy(), TOL); ASSERT_EQUAL_TOL(2*ONE_4PI_EPS0*(1.0)*(1.0+krf*1.0-crf), state.getPotentialEnergy(), TOL);
} }
void testTriclinic() {
System system;
system.addParticle(1.0);
system.addParticle(1.0);
Vec3 a(3.1, 0, 0);
Vec3 b(0.4, 3.5, 0);
Vec3 c(-0.1, -0.5, 4.0);
system.setDefaultPeriodicBoxVectors(a, b, c);
VerletIntegrator integrator(0.01);
NonbondedForce* nonbonded = new NonbondedForce();
nonbonded->addParticle(1.0, 1, 0);
nonbonded->addParticle(1.0, 1, 0);
nonbonded->setNonbondedMethod(NonbondedForce::CutoffPeriodic);
const double cutoff = 1.5;
nonbonded->setCutoffDistance(cutoff);
system.addForce(nonbonded);
Context context(system, integrator, platform);
vector<Vec3> positions(2);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
const double eps = 78.3;
const double krf = (1.0/(cutoff*cutoff*cutoff))*(eps-1.0)/(2.0*eps+1.0);
const double crf = (1.0/cutoff)*(3.0*eps)/(2.0*eps+1.0);
for (int iteration = 0; iteration < 50; iteration++) {
// Generate random positions for the two particles.
positions[0] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
positions[1] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
context.setPositions(positions);
// Loop over all possible periodic copies and find the nearest one.
Vec3 delta;
double distance2 = 100.0;
for (int i = -1; i < 2; i++)
for (int j = -1; j < 2; j++)
for (int k = -1; k < 2; k++) {
Vec3 d = positions[1]-positions[0]+a*i+b*j+c*k;
if (d.dot(d) < distance2) {
delta = d;
distance2 = d.dot(d);
}
}
double distance = sqrt(distance2);
// See if the force and energy are correct.
State state = context.getState(State::Forces | State::Energy);
if (distance >= cutoff) {
ASSERT_EQUAL(0.0, state.getPotentialEnergy());
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[0], 0);
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[1], 0);
}
else {
const Vec3 force = delta*ONE_4PI_EPS0*(-1.0/(distance*distance*distance)+2.0*krf);
ASSERT_EQUAL_TOL(ONE_4PI_EPS0*(1.0/distance+krf*distance*distance-crf), state.getPotentialEnergy(), TOL);
ASSERT_EQUAL_VEC(force, state.getForces()[0], TOL);
ASSERT_EQUAL_VEC(-force, state.getForces()[1], TOL);
}
}
}
void testLargeSystem() { void testLargeSystem() {
const int numMolecules = 600; const int numMolecules = 600;
...@@ -862,6 +923,33 @@ void testSwitchingFunction(NonbondedForce::NonbondedMethod method) { ...@@ -862,6 +923,33 @@ void testSwitchingFunction(NonbondedForce::NonbondedMethod method) {
} }
} }
void testReordering() {
// Check that reordering of atoms doesn't alter their positions.
const int numParticles = 200;
System system;
system.setDefaultPeriodicBoxVectors(Vec3(6, 0, 0), Vec3(2.1, 6, 0), Vec3(-1.5, -0.5, 6));
NonbondedForce *nonbonded = new NonbondedForce();
nonbonded->setNonbondedMethod(NonbondedForce::PME);
system.addForce(nonbonded);
vector<Vec3> positions;
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int i = 0; i < numParticles; i++) {
system.addParticle(1.0);
nonbonded->addParticle(0.0, 0.0, 0.0);
positions.push_back(Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5)*20);
}
VerletIntegrator integrator(0.001);
Context context(system, integrator, platform);
context.setPositions(positions);
integrator.step(1);
State state = context.getState(State::Positions | State::Velocities);
for (int i = 0; i < numParticles; i++) {
ASSERT_EQUAL_VEC(positions[i], state.getPositions()[i], 1e-6);
}
}
int main(int argc, char* argv[]) { int main(int argc, char* argv[]) {
try { try {
if (argc > 1) if (argc > 1)
...@@ -872,6 +960,7 @@ int main(int argc, char* argv[]) { ...@@ -872,6 +960,7 @@ int main(int argc, char* argv[]) {
testCutoff(); testCutoff();
testCutoff14(); testCutoff14();
testPeriodic(); testPeriodic();
testTriclinic();
testLargeSystem(); testLargeSystem();
//testBlockInteractions(false); //testBlockInteractions(false);
//testBlockInteractions(true); //testBlockInteractions(true);
...@@ -882,6 +971,7 @@ int main(int argc, char* argv[]) { ...@@ -882,6 +971,7 @@ int main(int argc, char* argv[]) {
testParallelComputation(NonbondedForce::PME); testParallelComputation(NonbondedForce::PME);
testSwitchingFunction(NonbondedForce::CutoffNonPeriodic); testSwitchingFunction(NonbondedForce::CutoffNonPeriodic);
testSwitchingFunction(NonbondedForce::PME); testSwitchingFunction(NonbondedForce::PME);
testReordering();
} }
catch(const exception& e) { catch(const exception& e) {
cout << "exception: " << e.what() << endl; cout << "exception: " << e.what() << endl;
......
platforms/opencl/include/OpenCLContext.h
View file @ 3db6b8ee
...@@ -9,7 +9,7 @@ ...@@ -9,7 +9,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. * * Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -448,36 +448,65 @@ public: ...@@ -448,36 +448,65 @@ public:
/** /**
* Get whether the device being used supports 64 bit atomic operations on global memory. * Get whether the device being used supports 64 bit atomic operations on global memory.
*/ */
bool getSupports64BitGlobalAtomics() { bool getSupports64BitGlobalAtomics() const {
return supports64BitGlobalAtomics; return supports64BitGlobalAtomics;
} }
/** /**
* Get whether the device being used supports double precision math. * Get whether the device being used supports double precision math.
*/ */
bool getSupportsDoublePrecision() { bool getSupportsDoublePrecision() const {
return supportsDoublePrecision; return supportsDoublePrecision;
} }
/** /**
* Get whether double precision is being used. * Get whether double precision is being used.
*/ */
bool getUseDoublePrecision() { bool getUseDoublePrecision() const {
return useDoublePrecision; return useDoublePrecision;
} }
/** /**
* Get whether mixed precision is being used. * Get whether mixed precision is being used.
*/ */
bool getUseMixedPrecision() { bool getUseMixedPrecision() const {
return useMixedPrecision; return useMixedPrecision;
} }
/**
* Get whether the periodic box is triclinic.
*/
bool getBoxIsTriclinic() const {
return boxIsTriclinic;
}
/** /**
* Convert a number to a string in a format suitable for including in a kernel. * Convert a number to a string in a format suitable for including in a kernel.
* This takes into account whether the context uses single or double precision. * This takes into account whether the context uses single or double precision.
*/ */
std::string doubleToString(double value); std::string doubleToString(double value) const;
/** /**
* Convert a number to a string in a format suitable for including in a kernel. * Convert a number to a string in a format suitable for including in a kernel.
*/ */
std::string intToString(int value); std::string intToString(int value) const;
/**
* Get the vectors defining the periodic box.
*/
void getPeriodicBoxVectors(Vec3& a, Vec3& b, Vec3& c) const {
a = Vec3(periodicBoxVecXDouble.x, periodicBoxVecXDouble.y, periodicBoxVecXDouble.z);
b = Vec3(periodicBoxVecYDouble.x, periodicBoxVecYDouble.y, periodicBoxVecYDouble.z);
c = Vec3(periodicBoxVecZDouble.x, periodicBoxVecZDouble.y, periodicBoxVecZDouble.z);
}
/**
* Set the vectors defining the periodic box.
*/
void setPeriodicBoxVectors(const Vec3& a, const Vec3& b, const Vec3& c) {
periodicBoxVecX = mm_float4((float) a[0], (float) a[1], (float) a[2], 0.0f);
periodicBoxVecY = mm_float4((float) b[0], (float) b[1], (float) b[2], 0.0f);
periodicBoxVecZ = mm_float4((float) c[0], (float) c[1], (float) c[2], 0.0f);
periodicBoxVecXDouble = mm_double4(a[0], a[1], a[2], 0.0);
periodicBoxVecYDouble = mm_double4(b[0], b[1], b[2], 0.0);
periodicBoxVecZDouble = mm_double4(c[0], c[1], c[2], 0.0);
periodicBoxSize = mm_float4((float) a[0], (float) b[1], (float) c[2], 0.0f);
invPeriodicBoxSize = mm_float4(1.0f/(float) a[0], 1.0f/(float) b[1], 1.0f/(float) c[2], 0.0f);
periodicBoxSizeDouble = mm_double4(a[0], b[1], c[2], 0.0);
invPeriodicBoxSizeDouble = mm_double4(1.0/a[0], 1.0/b[1], 1.0/c[2], 0.0);
}
/** /**
* Get the size of the periodic box. * Get the size of the periodic box.
*/ */
...@@ -490,15 +519,6 @@ public: ...@@ -490,15 +519,6 @@ public:
mm_double4 getPeriodicBoxSizeDouble() const { mm_double4 getPeriodicBoxSizeDouble() const {
return periodicBoxSizeDouble; return periodicBoxSizeDouble;
} }
/**
* Set the size of the periodic box.
*/
void setPeriodicBoxSize(double xsize, double ysize, double zsize) {
periodicBoxSize = mm_float4((float) xsize, (float) ysize, (float) zsize, 0);
invPeriodicBoxSize = mm_float4((float) (1.0/xsize), (float) (1.0/ysize), (float) (1.0/zsize), 0);
periodicBoxSizeDouble = mm_double4(xsize, ysize, zsize, 0);
invPeriodicBoxSizeDouble = mm_double4(1.0/xsize, 1.0/ysize, 1.0/zsize, 0);
}
/** /**
* Get the inverse of the size of the periodic box. * Get the inverse of the size of the periodic box.
*/ */
...@@ -511,6 +531,42 @@ public: ...@@ -511,6 +531,42 @@ public:
mm_double4 getInvPeriodicBoxSizeDouble() const { mm_double4 getInvPeriodicBoxSizeDouble() const {
return invPeriodicBoxSizeDouble; return invPeriodicBoxSizeDouble;
} }
/**
* Get the first periodic box vector.
*/
mm_float4 getPeriodicBoxVecX() {
return periodicBoxVecX;
}
/**
* Get the first periodic box vector.
*/
mm_double4 getPeriodicBoxVecXDouble() {
return periodicBoxVecXDouble;
}
/**
* Get the second periodic box vector.
*/
mm_float4 getPeriodicBoxVecY() {
return periodicBoxVecY;
}
/**
* Get the second periodic box vector.
*/
mm_double4 getPeriodicBoxVecYDouble() {
return periodicBoxVecYDouble;
}
/**
* Get the third periodic box vector.
*/
mm_float4 getPeriodicBoxVecZ() {
return periodicBoxVecZ;
}
/**
* Get the third periodic box vector.
*/
mm_double4 getPeriodicBoxVecZDouble() {
return periodicBoxVecZDouble;
}
/** /**
* Get the OpenCLIntegrationUtilities for this context. * Get the OpenCLIntegrationUtilities for this context.
*/ */
...@@ -628,9 +684,9 @@ private: ...@@ -628,9 +684,9 @@ private:
int numThreadBlocks; int numThreadBlocks;
int numForceBuffers; int numForceBuffers;
int simdWidth; int simdWidth;
bool supports64BitGlobalAtomics, supportsDoublePrecision, useDoublePrecision, useMixedPrecision, atomsWereReordered; bool supports64BitGlobalAtomics, supportsDoublePrecision, useDoublePrecision, useMixedPrecision, atomsWereReordered, boxIsTriclinic;
mm_float4 periodicBoxSize, invPeriodicBoxSize; mm_float4 periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ;
mm_double4 periodicBoxSizeDouble, invPeriodicBoxSizeDouble; mm_double4 periodicBoxSizeDouble, invPeriodicBoxSizeDouble, periodicBoxVecXDouble, periodicBoxVecYDouble, periodicBoxVecZDouble;
std::string defaultOptimizationOptions; std::string defaultOptimizationOptions;
std::map<std::string, std::string> compilationDefines; std::map<std::string, std::string> compilationDefines;
cl::Context context; cl::Context context;
......
platforms/opencl/src/OpenCLContext.cpp
View file @ 3db6b8ee
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. * * Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -336,6 +336,54 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device ...@@ -336,6 +336,54 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
compilationDefines["LOG"] = "log"; compilationDefines["LOG"] = "log";
} }
// Set defines for applying periodic boundary conditions.
Vec3 boxVectors[3];
system.getDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
boxIsTriclinic = (boxVectors[0][1] != 0.0 || boxVectors[0][2] != 0.0 ||
boxVectors[1][0] != 0.0 || boxVectors[1][2] != 0.0 ||
boxVectors[2][0] != 0.0 || boxVectors[2][1] != 0.0);
if (boxIsTriclinic) {
compilationDefines["APPLY_PERIODIC_TO_DELTA(delta)"] =
"{"
"real scale3 = floor(delta.z*invPeriodicBoxSize.z+0.5f); \\\n"
"delta.xyz -= scale3*periodicBoxVecZ.xyz; \\\n"
"real scale2 = floor(delta.y*invPeriodicBoxSize.y+0.5f); \\\n"
"delta.xy -= scale2*periodicBoxVecY.xy; \\\n"
"real scale1 = floor(delta.x*invPeriodicBoxSize.x+0.5f); \\\n"
"delta.x -= scale1*periodicBoxVecX.x;}";
compilationDefines["APPLY_PERIODIC_TO_POS(pos)"] =
"{"
"real scale3 = floor(pos.z*invPeriodicBoxSize.z); \\\n"
"pos.xyz -= scale3*periodicBoxVecZ.xyz; \\\n"
"real scale2 = floor(pos.y*invPeriodicBoxSize.y); \\\n"
"pos.xy -= scale2*periodicBoxVecY.xy; \\\n"
"real scale1 = floor(pos.x*invPeriodicBoxSize.x); \\\n"
"pos.x -= scale1*periodicBoxVecX.x;}";
compilationDefines["APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)"] =
"{"
"real scale3 = floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f); \\\n"
"pos.x -= scale3*periodicBoxVecZ.x; \\\n"
"pos.y -= scale3*periodicBoxVecZ.y; \\\n"
"pos.z -= scale3*periodicBoxVecZ.z; \\\n"
"real scale2 = floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f); \\\n"
"pos.x -= scale2*periodicBoxVecY.x; \\\n"
"pos.y -= scale2*periodicBoxVecY.y; \\\n"
"real scale1 = floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f); \\\n"
"pos.x -= scale1*periodicBoxVecX.x;}";
}
else {
compilationDefines["APPLY_PERIODIC_TO_DELTA(delta)"] =
"delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz;";
compilationDefines["APPLY_PERIODIC_TO_POS(pos)"] =
"pos.xyz -= floor(pos.xyz*invPeriodicBoxSize.xyz)*periodicBoxSize.xyz;";
compilationDefines["APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)"] =
"{"
"pos.x -= floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; \\\n"
"pos.y -= floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y; \\\n"
"pos.z -= floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;}";
}
// Create the work thread used for parallelization when running on multiple devices. // Create the work thread used for parallelization when running on multiple devices.
thread = new WorkThread(); thread = new WorkThread();
...@@ -527,7 +575,7 @@ void OpenCLContext::restoreDefaultQueue() { ...@@ -527,7 +575,7 @@ void OpenCLContext::restoreDefaultQueue() {
currentQueue = defaultQueue; currentQueue = defaultQueue;
} }
string OpenCLContext::doubleToString(double value) { string OpenCLContext::doubleToString(double value) const {
stringstream s; stringstream s;
s.precision(useDoublePrecision ? 16 : 8); s.precision(useDoublePrecision ? 16 : 8);
s << scientific << value; s << scientific << value;
...@@ -536,7 +584,7 @@ string OpenCLContext::doubleToString(double value) { ...@@ -536,7 +584,7 @@ string OpenCLContext::doubleToString(double value) {
return s.str(); return s.str();
} }
string OpenCLContext::intToString(int value) { string OpenCLContext::intToString(int value) const {
stringstream s; stringstream s;
s << value; s << value;
return s.str(); return s.str();
...@@ -1039,16 +1087,21 @@ void OpenCLContext::reorderAtomsImpl() { ...@@ -1039,16 +1087,21 @@ void OpenCLContext::reorderAtomsImpl() {
// Move each molecule position into the same box. // Move each molecule position into the same box.
for (int i = 0; i < numMolecules; i++) { for (int i = 0; i < numMolecules; i++) {
int xcell = (int) floor(molPos[i].x*invPeriodicBoxSizeDouble.x); Real4 center = molPos[i];
int ycell = (int) floor(molPos[i].y*invPeriodicBoxSizeDouble.y); int zcell = (int) floor(center.z*invPeriodicBoxSize.z);
int zcell = (int) floor(molPos[i].z*invPeriodicBoxSizeDouble.z); center.x -= zcell*periodicBoxVecZ.x;
Real dx = xcell*periodicBoxSizeDouble.x; center.y -= zcell*periodicBoxVecZ.y;
Real dy = ycell*periodicBoxSizeDouble.y; center.z -= zcell*periodicBoxVecZ.z;
Real dz = zcell*periodicBoxSizeDouble.z; int ycell = (int) floor(center.y*invPeriodicBoxSize.y);
if (dx != 0.0f || dy != 0.0f || dz != 0.0f) { center.x -= ycell*periodicBoxVecY.x;
molPos[i].x -= dx; center.y -= ycell*periodicBoxVecY.y;
molPos[i].y -= dy; int xcell = (int) floor(center.x*invPeriodicBoxSize.x);
molPos[i].z -= dz; center.x -= xcell*periodicBoxVecX.x;
if (xcell != 0 || ycell != 0 || zcell != 0) {
Real dx = molPos[i].x-center.x;
Real dy = molPos[i].y-center.y;
Real dz = molPos[i].z-center.z;
molPos[i] = center;
for (int j = 0; j < (int) atoms.size(); j++) { for (int j = 0; j < (int) atoms.size(); j++) {
int atom = atoms[j]+mol.offsets[i]; int atom = atoms[j]+mol.offsets[i];
Real4 p = oldPosq[atom]; Real4 p = oldPosq[atom];
......
platforms/opencl/src/OpenCLKernels.cpp
View file @ 3db6b8ee
This diff is collapsed.
platforms/opencl/src/OpenCLNonbondedUtilities.cpp
View file @ 3db6b8ee
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. * * Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -325,11 +325,11 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -325,11 +325,11 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
cl::Program interactingBlocksProgram = context.createProgram(file, defines); cl::Program interactingBlocksProgram = context.createProgram(file, defines);
findBlockBoundsKernel = cl::Kernel(interactingBlocksProgram, "findBlockBounds"); findBlockBoundsKernel = cl::Kernel(interactingBlocksProgram, "findBlockBounds");
findBlockBoundsKernel.setArg<cl_int>(0, context.getNumAtoms()); findBlockBoundsKernel.setArg<cl_int>(0, context.getNumAtoms());
findBlockBoundsKernel.setArg<cl::Buffer>(3, context.getPosq().getDeviceBuffer()); findBlockBoundsKernel.setArg<cl::Buffer>(6, context.getPosq().getDeviceBuffer());
findBlockBoundsKernel.setArg<cl::Buffer>(4, blockCenter->getDeviceBuffer()); findBlockBoundsKernel.setArg<cl::Buffer>(7, blockCenter->getDeviceBuffer());
findBlockBoundsKernel.setArg<cl::Buffer>(5, blockBoundingBox->getDeviceBuffer()); findBlockBoundsKernel.setArg<cl::Buffer>(8, blockBoundingBox->getDeviceBuffer());
findBlockBoundsKernel.setArg<cl::Buffer>(6, rebuildNeighborList->getDeviceBuffer()); findBlockBoundsKernel.setArg<cl::Buffer>(9, rebuildNeighborList->getDeviceBuffer());
findBlockBoundsKernel.setArg<cl::Buffer>(7, sortedBlocks->getDeviceBuffer()); findBlockBoundsKernel.setArg<cl::Buffer>(10, sortedBlocks->getDeviceBuffer());
sortBoxDataKernel = cl::Kernel(interactingBlocksProgram, "sortBoxData"); sortBoxDataKernel = cl::Kernel(interactingBlocksProgram, "sortBoxData");
sortBoxDataKernel.setArg<cl::Buffer>(0, sortedBlocks->getDeviceBuffer()); sortBoxDataKernel.setArg<cl::Buffer>(0, sortedBlocks->getDeviceBuffer());
sortBoxDataKernel.setArg<cl::Buffer>(1, blockCenter->getDeviceBuffer()); sortBoxDataKernel.setArg<cl::Buffer>(1, blockCenter->getDeviceBuffer());
...@@ -341,20 +341,20 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -341,20 +341,20 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
sortBoxDataKernel.setArg<cl::Buffer>(7, interactionCount->getDeviceBuffer()); sortBoxDataKernel.setArg<cl::Buffer>(7, interactionCount->getDeviceBuffer());
sortBoxDataKernel.setArg<cl::Buffer>(8, rebuildNeighborList->getDeviceBuffer()); sortBoxDataKernel.setArg<cl::Buffer>(8, rebuildNeighborList->getDeviceBuffer());
findInteractingBlocksKernel = cl::Kernel(interactingBlocksProgram, "findBlocksWithInteractions"); findInteractingBlocksKernel = cl::Kernel(interactingBlocksProgram, "findBlocksWithInteractions");
findInteractingBlocksKernel.setArg<cl::Buffer>(2, interactionCount->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(5, interactionCount->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(3, interactingTiles->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(6, interactingTiles->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(4, interactingAtoms->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(7, interactingAtoms->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(5, context.getPosq().getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(8, context.getPosq().getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl_uint>(6, interactingTiles->getSize()); findInteractingBlocksKernel.setArg<cl_uint>(9, interactingTiles->getSize());
findInteractingBlocksKernel.setArg<cl_uint>(7, startBlockIndex); findInteractingBlocksKernel.setArg<cl_uint>(10, startBlockIndex);
findInteractingBlocksKernel.setArg<cl_uint>(8, numBlocks); findInteractingBlocksKernel.setArg<cl_uint>(11, numBlocks);
findInteractingBlocksKernel.setArg<cl::Buffer>(9, sortedBlocks->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(12, sortedBlocks->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(10, sortedBlockCenter->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(13, sortedBlockCenter->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(11, sortedBlockBoundingBox->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(14, sortedBlockBoundingBox->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(12, exclusionIndices->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(15, exclusionIndices->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(13, exclusionRowIndices->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(16, exclusionRowIndices->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(14, oldPositions->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(17, oldPositions->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(15, rebuildNeighborList->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(18, rebuildNeighborList->getDeviceBuffer());
if (findInteractingBlocksKernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(context.getDevice()) < groupSize) { if (findInteractingBlocksKernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(context.getDevice()) < groupSize) {
// The device can't handle this block size, so reduce it. // The device can't handle this block size, so reduce it.
...@@ -369,18 +369,21 @@ void OpenCLNonbondedUtilities::initialize(const System& system) { ...@@ -369,18 +369,21 @@ void OpenCLNonbondedUtilities::initialize(const System& system) {
} }
} }
static void setPeriodicBoxSizeArg(OpenCLContext& cl, cl::Kernel& kernel, int index) { static void setPeriodicBoxArgs(OpenCLContext& cl, cl::Kernel& kernel, int index) {
if (cl.getUseDoublePrecision()) if (cl.getUseDoublePrecision()) {
kernel.setArg<mm_double4>(index, cl.getPeriodicBoxSizeDouble()); kernel.setArg<mm_double4>(index++, cl.getPeriodicBoxSizeDouble());
else kernel.setArg<mm_double4>(index++, cl.getInvPeriodicBoxSizeDouble());
kernel.setArg<mm_float4>(index, cl.getPeriodicBoxSize()); kernel.setArg<mm_double4>(index++, cl.getPeriodicBoxVecXDouble());
} kernel.setArg<mm_double4>(index++, cl.getPeriodicBoxVecYDouble());
kernel.setArg<mm_double4>(index, cl.getPeriodicBoxVecZDouble());
static void setInvPeriodicBoxSizeArg(OpenCLContext& cl, cl::Kernel& kernel, int index) { }
if (cl.getUseDoublePrecision()) else {
kernel.setArg<mm_double4>(index, cl.getInvPeriodicBoxSizeDouble()); kernel.setArg<mm_float4>(index++, cl.getPeriodicBoxSize());
else kernel.setArg<mm_float4>(index++, cl.getInvPeriodicBoxSize());
kernel.setArg<mm_float4>(index, cl.getInvPeriodicBoxSize()); kernel.setArg<mm_float4>(index++, cl.getPeriodicBoxVecX());
kernel.setArg<mm_float4>(index++, cl.getPeriodicBoxVecY());
kernel.setArg<mm_float4>(index, cl.getPeriodicBoxVecZ());
}
} }
void OpenCLNonbondedUtilities::prepareInteractions() { void OpenCLNonbondedUtilities::prepareInteractions() {
...@@ -397,22 +400,18 @@ void OpenCLNonbondedUtilities::prepareInteractions() { ...@@ -397,22 +400,18 @@ void OpenCLNonbondedUtilities::prepareInteractions() {
// Compute the neighbor list. // Compute the neighbor list.
setPeriodicBoxSizeArg(context, findBlockBoundsKernel, 1); setPeriodicBoxArgs(context, findBlockBoundsKernel, 1);
setInvPeriodicBoxSizeArg(context, findBlockBoundsKernel, 2);
context.executeKernel(findBlockBoundsKernel, context.getNumAtoms()); context.executeKernel(findBlockBoundsKernel, context.getNumAtoms());
blockSorter->sort(*sortedBlocks); blockSorter->sort(*sortedBlocks);
context.executeKernel(sortBoxDataKernel, context.getNumAtoms()); context.executeKernel(sortBoxDataKernel, context.getNumAtoms());
setPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 0); setPeriodicBoxArgs(context, findInteractingBlocksKernel, 0);
setInvPeriodicBoxSizeArg(context, findInteractingBlocksKernel, 1);
context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms(), interactingBlocksThreadBlockSize); context.executeKernel(findInteractingBlocksKernel, context.getNumAtoms(), interactingBlocksThreadBlockSize);
} }
void OpenCLNonbondedUtilities::computeInteractions() { void OpenCLNonbondedUtilities::computeInteractions() {
if (kernelSource.size() > 0) { if (kernelSource.size() > 0) {
if (useCutoff) { if (useCutoff)
setPeriodicBoxSizeArg(context, forceKernel, 9); setPeriodicBoxArgs(context, forceKernel, 9);
setInvPeriodicBoxSizeArg(context, forceKernel, 10);
}
context.executeKernel(forceKernel, numForceThreadBlocks*forceThreadBlockSize, forceThreadBlockSize); context.executeKernel(forceKernel, numForceThreadBlocks*forceThreadBlockSize, forceThreadBlockSize);
if (context.getComputeForceCount() == 1) if (context.getComputeForceCount() == 1)
updateNeighborListSize(); // This is the first time step, so check whether our initial guess was large enough. updateNeighborListSize(); // This is the first time step, so check whether our initial guess was large enough.
...@@ -441,11 +440,11 @@ void OpenCLNonbondedUtilities::updateNeighborListSize() { ...@@ -441,11 +440,11 @@ void OpenCLNonbondedUtilities::updateNeighborListSize() {
interactingTiles = OpenCLArray::create<cl_int>(context, maxTiles, "interactingTiles"); interactingTiles = OpenCLArray::create<cl_int>(context, maxTiles, "interactingTiles");
interactingAtoms = OpenCLArray::create<cl_int>(context, OpenCLContext::TileSize*maxTiles, "interactingAtoms"); interactingAtoms = OpenCLArray::create<cl_int>(context, OpenCLContext::TileSize*maxTiles, "interactingAtoms");
forceKernel.setArg<cl::Buffer>(7, interactingTiles->getDeviceBuffer()); forceKernel.setArg<cl::Buffer>(7, interactingTiles->getDeviceBuffer());
forceKernel.setArg<cl_uint>(11, maxTiles); forceKernel.setArg<cl_uint>(14, maxTiles);
forceKernel.setArg<cl::Buffer>(14, interactingAtoms->getDeviceBuffer()); forceKernel.setArg<cl::Buffer>(17, interactingAtoms->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(3, interactingTiles->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(6, interactingTiles->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl::Buffer>(4, interactingAtoms->getDeviceBuffer()); findInteractingBlocksKernel.setArg<cl::Buffer>(7, interactingAtoms->getDeviceBuffer());
findInteractingBlocksKernel.setArg<cl_uint>(6, maxTiles); findInteractingBlocksKernel.setArg<cl_uint>(9, maxTiles);
int numAtoms = context.getNumAtoms(); int numAtoms = context.getNumAtoms();
if (context.getUseDoublePrecision()) { if (context.getUseDoublePrecision()) {
vector<mm_double4> oldPositionsVec(numAtoms, mm_double4(1e30, 1e30, 1e30, 0)); vector<mm_double4> oldPositionsVec(numAtoms, mm_double4(1e30, 1e30, 1e30, 0));
...@@ -473,8 +472,8 @@ void OpenCLNonbondedUtilities::setAtomBlockRange(double startFraction, double en ...@@ -473,8 +472,8 @@ void OpenCLNonbondedUtilities::setAtomBlockRange(double startFraction, double en
forceKernel.setArg<cl_uint>(5, startTileIndex); forceKernel.setArg<cl_uint>(5, startTileIndex);
forceKernel.setArg<cl_uint>(6, numTiles); forceKernel.setArg<cl_uint>(6, numTiles);
findInteractingBlocksKernel.setArg<cl_uint>(7, startBlockIndex); findInteractingBlocksKernel.setArg<cl_uint>(10, startBlockIndex);
findInteractingBlocksKernel.setArg<cl_uint>(8, numBlocks); findInteractingBlocksKernel.setArg<cl_uint>(11, numBlocks);
} }
} }
...@@ -617,7 +616,7 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc ...@@ -617,7 +616,7 @@ cl::Kernel OpenCLNonbondedUtilities::createInteractionKernel(const string& sourc
if (useCutoff) { if (useCutoff) {
kernel.setArg<cl::Buffer>(index++, interactingTiles->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, interactingTiles->getDeviceBuffer());
kernel.setArg<cl::Buffer>(index++, interactionCount->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, interactionCount->getDeviceBuffer());
index += 2; // The periodic box size arguments are set when the kernel is executed. index += 5; // The periodic box size arguments are set when the kernel is executed.
kernel.setArg<cl_uint>(index++, interactingTiles->getSize()); kernel.setArg<cl_uint>(index++, interactingTiles->getSize());
kernel.setArg<cl::Buffer>(index++, blockCenter->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, blockCenter->getDeviceBuffer());
kernel.setArg<cl::Buffer>(index++, blockBoundingBox->getDeviceBuffer()); kernel.setArg<cl::Buffer>(index++, blockBoundingBox->getDeviceBuffer());
......
platforms/opencl/src/kernels/customGBEnergyN2.cl
View file @ 3db6b8ee
...@@ -21,7 +21,8 @@ __kernel void computeN2Energy( ...@@ -21,7 +21,8 @@ __kernel void computeN2Energy(
__global const ushort2* exclusionTiles, __global const ushort2* exclusionTiles,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, __global const real4* restrict blockCenter, __global const real4* restrict blockSize, __global const int* restrict interactingAtoms real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, __global const real4* restrict blockCenter,
__global const real4* restrict blockSize, __global const int* restrict interactingAtoms
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
...@@ -60,7 +61,7 @@ __kernel void computeN2Energy( ...@@ -60,7 +61,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -110,7 +111,7 @@ __kernel void computeN2Energy( ...@@ -110,7 +111,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -266,10 +267,8 @@ __kernel void computeN2Energy( ...@@ -266,10 +267,8 @@ __kernel void computeN2Energy(
// box, then skip having to apply periodic boundary conditions later. // box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
posq1.xyz -= floor((posq1.xyz-blockCenterX.xyz)*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_POS_WITH_CENTER(posq1, blockCenterX)
local_posq[get_local_id(0)].x -= floor((local_posq[get_local_id(0)].x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(local_posq[get_local_id(0)], blockCenterX)
local_posq[get_local_id(0)].y -= floor((local_posq[get_local_id(0)].y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
local_posq[get_local_id(0)].z -= floor((local_posq[get_local_id(0)].z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
SYNC_WARPS; SYNC_WARPS;
unsigned int tj = tgx; unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
...@@ -310,7 +309,7 @@ __kernel void computeN2Energy( ...@@ -310,7 +309,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
platforms/opencl/src/kernels/customGBEnergyN2_cpu.cl
View file @ 3db6b8ee
...@@ -21,7 +21,8 @@ __kernel void computeN2Energy( ...@@ -21,7 +21,8 @@ __kernel void computeN2Energy(
__global const ushort2* exclusionTiles, __global const ushort2* exclusionTiles,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, __global const real4* restrict blockCenter, __global const real4* restrict blockSize, __global const int* restrict interactingAtoms real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, __global const real4* restrict blockCenter,
__global const real4* restrict blockSize, __global const int* restrict interactingAtoms
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
...@@ -60,7 +61,7 @@ __kernel void computeN2Energy( ...@@ -60,7 +61,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -126,7 +127,7 @@ __kernel void computeN2Energy( ...@@ -126,7 +127,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -272,7 +273,7 @@ __kernel void computeN2Energy( ...@@ -272,7 +273,7 @@ __kernel void computeN2Energy(
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++)
local_posq[tgx].xyz -= floor((local_posq[tgx].xyz-blockCenterX.xyz)*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_POS_WITH_CENTER(local_posq[tgx], blockCenterX)
for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) { for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) {
unsigned int atom1 = x*TILE_SIZE+tgx; unsigned int atom1 = x*TILE_SIZE+tgx;
real4 force = 0; real4 force = 0;
...@@ -332,7 +333,7 @@ __kernel void computeN2Energy( ...@@ -332,7 +333,7 @@ __kernel void computeN2Energy(
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
platforms/opencl/src/kernels/customGBValueN2.cl
View file @ 3db6b8ee
...@@ -15,7 +15,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -15,7 +15,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
__local real* restrict local_value, __local real* restrict local_value,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, __global const real4* restrict blockCenter, __global const real4* restrict blockSize, __global const int* restrict interactingAtoms real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, __global const real4* restrict blockCenter,
__global const real4* restrict blockSize, __global const int* restrict interactingAtoms
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
...@@ -52,7 +53,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -52,7 +53,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -100,7 +101,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -100,7 +101,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -239,10 +240,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -239,10 +240,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
// box, then skip having to apply periodic boundary conditions later. // box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
posq1.xyz -= floor((posq1.xyz-blockCenterX.xyz)*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_POS_WITH_CENTER(posq1, blockCenterX)
local_posq[get_local_id(0)].x -= floor((local_posq[get_local_id(0)].x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(local_posq[get_local_id(0)], blockCenterX)
local_posq[get_local_id(0)].y -= floor((local_posq[get_local_id(0)].y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
local_posq[get_local_id(0)].z -= floor((local_posq[get_local_id(0)].z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
SYNC_WARPS; SYNC_WARPS;
unsigned int tj = tgx; unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) { for (j = 0; j < TILE_SIZE; j++) {
...@@ -278,7 +277,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -278,7 +277,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[atom2]; real4 posq2 = local_posq[atom2];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
platforms/opencl/src/kernels/customGBValueN2_cpu.cl
View file @ 3db6b8ee
...@@ -15,7 +15,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -15,7 +15,8 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
__local real* restrict local_value, __local real* restrict local_value,
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
__global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const int* restrict tiles, __global const unsigned int* restrict interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, __global const real4* restrict blockCenter, __global const real4* restrict blockSize, __global const int* restrict interactingAtoms real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, __global const real4* restrict blockCenter,
__global const real4* restrict blockSize, __global const int* restrict interactingAtoms
#else #else
unsigned int numTiles unsigned int numTiles
#endif #endif
...@@ -52,7 +53,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -52,7 +53,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -109,7 +110,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -109,7 +110,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
...@@ -239,7 +240,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -239,7 +240,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 blockCenterX = blockCenter[x]; real4 blockCenterX = blockCenter[x];
for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++)
local_posq[tgx].xyz -= floor((local_posq[tgx].xyz-blockCenterX.xyz)*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_POS_WITH_CENTER(local_posq[tgx], blockCenterX)
for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) { for (unsigned int tgx = 0; tgx < TILE_SIZE; tgx++) {
unsigned int atom1 = x*TILE_SIZE+tgx; unsigned int atom1 = x*TILE_SIZE+tgx;
real value = 0; real value = 0;
...@@ -287,7 +288,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4* ...@@ -287,7 +288,7 @@ __kernel void computeN2Value(__global const real4* restrict posq, __local real4*
real4 posq2 = local_posq[j]; real4 posq2 = local_posq[j];
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = dot(delta.xyz, delta.xyz); real r2 = dot(delta.xyz, delta.xyz);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
platforms/opencl/src/kernels/customHbondForce.cl
View file @ 3db6b8ee
...@@ -11,12 +11,10 @@ real4 delta(real4 vec1, real4 vec2) { ...@@ -11,12 +11,10 @@ real4 delta(real4 vec1, real4 vec2) {
* Compute the difference between two vectors, taking periodic boundary conditions into account * Compute the difference between two vectors, taking periodic boundary conditions into account
* and setting the fourth component to the squared magnitude. * and setting the fourth component to the squared magnitude.
*/ */
real4 deltaPeriodic(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize) { real4 deltaPeriodic(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0); real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
result.x -= floor(result.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(result)
result.y -= floor(result.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
result.z -= floor(result.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
result.w = result.x*result.x + result.y*result.y + result.z*result.z; result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result; return result;
...@@ -56,7 +54,8 @@ real4 computeCross(real4 vec1, real4 vec2) { ...@@ -56,7 +54,8 @@ real4 computeCross(real4 vec1, real4 vec2) {
* Compute forces on donors. * Compute forces on donors.
*/ */
__kernel void computeDonorForces(__global real4* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq, __global const int4* restrict exclusions, __kernel void computeDonorForces(__global real4* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq, __global const int4* restrict exclusions,
__global const int4* restrict donorAtoms, __global const int4* restrict acceptorAtoms, __global const int4* restrict donorBufferIndices, __local real4* posBuffer, real4 periodicBoxSize, real4 invPeriodicBoxSize __global const int4* restrict donorAtoms, __global const int4* restrict acceptorAtoms, __global const int4* restrict donorBufferIndices, __local real4* posBuffer, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
real energy = 0; real energy = 0;
real4 f1 = (real4) 0; real4 f1 = (real4) 0;
...@@ -102,7 +101,7 @@ __kernel void computeDonorForces(__global real4* restrict forceBuffers, __global ...@@ -102,7 +101,7 @@ __kernel void computeDonorForces(__global real4* restrict forceBuffers, __global
real4 a1 = posBuffer[3*index]; real4 a1 = posBuffer[3*index];
real4 a2 = posBuffer[3*index+1]; real4 a2 = posBuffer[3*index+1];
real4 a3 = posBuffer[3*index+2]; real4 a3 = posBuffer[3*index+2];
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize); real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (deltaD1A1.w < CUTOFF_SQUARED) { if (deltaD1A1.w < CUTOFF_SQUARED) {
#endif #endif
...@@ -144,7 +143,8 @@ __kernel void computeDonorForces(__global real4* restrict forceBuffers, __global ...@@ -144,7 +143,8 @@ __kernel void computeDonorForces(__global real4* restrict forceBuffers, __global
* Compute forces on acceptors. * Compute forces on acceptors.
*/ */
__kernel void computeAcceptorForces(__global real4* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq, __global const int4* restrict exclusions, __kernel void computeAcceptorForces(__global real4* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq, __global const int4* restrict exclusions,
__global const int4* restrict donorAtoms, __global const int4* restrict acceptorAtoms, __global const int4* restrict acceptorBufferIndices, __local real4* restrict posBuffer, real4 periodicBoxSize, real4 invPeriodicBoxSize __global const int4* restrict donorAtoms, __global const int4* restrict acceptorAtoms, __global const int4* restrict acceptorBufferIndices, __local real4* restrict posBuffer, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
real4 f1 = (real4) 0; real4 f1 = (real4) 0;
real4 f2 = (real4) 0; real4 f2 = (real4) 0;
...@@ -189,7 +189,7 @@ __kernel void computeAcceptorForces(__global real4* restrict forceBuffers, __glo ...@@ -189,7 +189,7 @@ __kernel void computeAcceptorForces(__global real4* restrict forceBuffers, __glo
real4 d1 = posBuffer[3*index]; real4 d1 = posBuffer[3*index];
real4 d2 = posBuffer[3*index+1]; real4 d2 = posBuffer[3*index+1];
real4 d3 = posBuffer[3*index+2]; real4 d3 = posBuffer[3*index+2];
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize); real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (deltaD1A1.w < CUTOFF_SQUARED) { if (deltaD1A1.w < CUTOFF_SQUARED) {
#endif #endif
......
platforms/opencl/src/kernels/customManyParticle.cl
View file @ 3db6b8ee
...@@ -14,12 +14,10 @@ inline void storeForce(int atom, real4 force, __global long* restrict forceBuffe ...@@ -14,12 +14,10 @@ inline void storeForce(int atom, real4 force, __global long* restrict forceBuffe
* Compute the difference between two vectors, taking periodic boundary conditions into account * Compute the difference between two vectors, taking periodic boundary conditions into account
* and setting the fourth component to the squared magnitude. * and setting the fourth component to the squared magnitude.
*/ */
inline real4 delta(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize) { inline real4 delta(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0.0f); real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0.0f);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
result.x -= floor(result.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(result)
result.y -= floor(result.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
result.z -= floor(result.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
result.w = result.x*result.x + result.y*result.y + result.z*result.z; result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result; return result;
...@@ -75,7 +73,7 @@ inline bool isInteractionExcluded(int atom1, int atom2, __global int* restrict e ...@@ -75,7 +73,7 @@ inline bool isInteractionExcluded(int atom1, int atom2, __global int* restrict e
*/ */
__kernel void computeInteraction( __kernel void computeInteraction(
__global long* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq, __global long* restrict forceBuffers, __global real* restrict energyBuffer, __global const real4* restrict posq,
real4 periodicBoxSize, real4 invPeriodicBoxSize real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
, __global const int* restrict neighbors, __global const int* restrict neighborStartIndex , __global const int* restrict neighbors, __global const int* restrict neighborStartIndex
#endif #endif
...@@ -138,16 +136,14 @@ __kernel void computeInteraction( ...@@ -138,16 +136,14 @@ __kernel void computeInteraction(
/** /**
* Find a bounding box for the atoms in each block. * Find a bounding box for the atoms in each block.
*/ */
__kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const real4* restrict posq, __kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
__global real4* restrict blockCenter, __global real4* restrict blockBoundingBox, __global int* restrict numNeighborPairs) { __global const real4* restrict posq, __global real4* restrict blockCenter, __global real4* restrict blockBoundingBox, __global int* restrict numNeighborPairs) {
int index = get_global_id(0); int index = get_global_id(0);
int base = index*TILE_SIZE; int base = index*TILE_SIZE;
while (base < NUM_ATOMS) { while (base < NUM_ATOMS) {
real4 pos = posq[base]; real4 pos = posq[base];
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS(pos)
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z;
#endif #endif
real4 minPos = pos; real4 minPos = pos;
real4 maxPos = pos; real4 maxPos = pos;
...@@ -156,9 +152,7 @@ __kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, _ ...@@ -156,9 +152,7 @@ __kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, _
pos = posq[i]; pos = posq[i];
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
real4 center = 0.5f*(maxPos+minPos); real4 center = 0.5f*(maxPos+minPos);
pos.x -= floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)
pos.y -= floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos.z -= floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
minPos = (real4) (min(minPos.x,pos.x), min(minPos.y,pos.y), min(minPos.z,pos.z), 0); minPos = (real4) (min(minPos.x,pos.x), min(minPos.y,pos.y), min(minPos.z,pos.z), 0);
maxPos = (real4) (max(maxPos.x,pos.x), max(maxPos.y,pos.y), max(maxPos.z,pos.z), 0); maxPos = (real4) (max(maxPos.x,pos.x), max(maxPos.y,pos.y), max(maxPos.z,pos.z), 0);
...@@ -176,8 +170,8 @@ __kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, _ ...@@ -176,8 +170,8 @@ __kernel void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, _
/** /**
* Find a list of neighbors for each atom. * Find a list of neighbors for each atom.
*/ */
__kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, __global const real4* restrict posq, __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
__global const real4* restrict blockCenter, __global const real4* restrict blockBoundingBox, __global int2* restrict neighborPairs, __global const real4* restrict posq, __global const real4* restrict blockCenter, __global const real4* restrict blockBoundingBox, __global int2* restrict neighborPairs,
__global int* restrict numNeighborPairs, __global int* restrict numNeighborsForAtom, int maxNeighborPairs __global int* restrict numNeighborPairs, __global int* restrict numNeighborsForAtom, int maxNeighborPairs
#ifdef USE_EXCLUSIONS #ifdef USE_EXCLUSIONS
, __global int* restrict exclusions, __global int* restrict exclusionStartIndex , __global int* restrict exclusions, __global int* restrict exclusionStartIndex
...@@ -212,9 +206,7 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, __g ...@@ -212,9 +206,7 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, __g
real4 blockSize2 = blockBoundingBox[block2]; real4 blockSize2 = blockBoundingBox[block2];
real4 blockDelta = blockCenter1-blockCenter2; real4 blockDelta = blockCenter1-blockCenter2;
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
blockDelta.x -= floor(blockDelta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; APPLY_PERIODIC_TO_DELTA(blockDelta)
blockDelta.y -= floor(blockDelta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
blockDelta.z -= floor(blockDelta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
#endif #endif
blockDelta.x = max((real) 0, fabs(blockDelta.x)-blockSize1.x-blockSize2.x); blockDelta.x = max((real) 0, fabs(blockDelta.x)-blockSize1.x-blockSize2.x);
blockDelta.y = max((real) 0, fabs(blockDelta.y)-blockSize1.y-blockSize2.y); blockDelta.y = max((real) 0, fabs(blockDelta.y)-blockSize1.y-blockSize2.y);
...@@ -243,7 +235,7 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, __g ...@@ -243,7 +235,7 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, __g
// Decide whether to include this atom pair in the neighbor list. // Decide whether to include this atom pair in the neighbor list.
real4 atomDelta = delta(pos1, pos2, periodicBoxSize, invPeriodicBoxSize); real4 atomDelta = delta(pos1, pos2, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CENTRAL_PARTICLE #ifdef USE_CENTRAL_PARTICLE
bool includeAtom = (atom2 != atom1 && atom2 < NUM_ATOMS && atomDelta.w < CUTOFF_SQUARED); bool includeAtom = (atom2 != atom1 && atom2 < NUM_ATOMS && atomDelta.w < CUTOFF_SQUARED);
#else #else
......
platforms/opencl/src/kernels/customNonbondedGroups.cl
View file @ 3db6b8ee
...@@ -42,8 +42,8 @@ __kernel void computeInteractionGroups( ...@@ -42,8 +42,8 @@ __kernel void computeInteractionGroups(
#else #else
__global real4* restrict forceBuffers, __global real4* restrict forceBuffers,
#endif #endif
__global real* restrict energyBuffer, __global const real4* restrict posq, __global real* restrict energyBuffer, __global const real4* restrict posq, __global const int4* restrict groupData,
__global const int4* restrict groupData, real4 periodicBoxSize, real4 invPeriodicBoxSize real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) { PARAMETER_ARGUMENTS) {
const unsigned int totalWarps = get_global_size(0)/TILE_SIZE; const unsigned int totalWarps = get_global_size(0)/TILE_SIZE;
const unsigned int warp = get_global_id(0)/TILE_SIZE; // global warpIndex const unsigned int warp = get_global_id(0)/TILE_SIZE; // global warpIndex
...@@ -82,7 +82,7 @@ __kernel void computeInteractionGroups( ...@@ -82,7 +82,7 @@ __kernel void computeInteractionGroups(
posq2 = (real4) (localData[localIndex].x, localData[localIndex].y, localData[localIndex].z, localData[localIndex].q); posq2 = (real4) (localData[localIndex].x, localData[localIndex].y, localData[localIndex].z, localData[localIndex].q);
real4 delta = (real4) (posq2.xyz - posq1.xyz, 0); real4 delta = (real4) (posq2.xyz - posq1.xyz, 0);
#ifdef USE_PERIODIC #ifdef USE_PERIODIC
delta.xyz -= floor(delta.xyz*invPeriodicBoxSize.xyz+0.5f)*periodicBoxSize.xyz; APPLY_PERIODIC_TO_DELTA(delta)
#endif #endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z; real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
......
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