ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@12588 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2014-10-06 22:59:05 +00:00
parent 4bb43ca885
commit 621fa7d600
122 changed files with 1934 additions and 1929 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
lib/cuda/fix_shake_cuda.cu
View File

@ -41,37 +41,37 @@
__device__ __constant__ int* _shake_atom;
__device__ __constant__ int* _shake_type;
__device__ __constant__ int* _shake_flag;
__device__ __constant__ X_FLOAT3* _xshake;
__device__ __constant__ F_FLOAT _dtfsq;
__device__ __constant__ X_FLOAT* _bond_distance;
__device__ __constant__ X_FLOAT* _angle_distance;
__device__ __constant__ X_CFLOAT3* _xshake;
__device__ __constant__ F_CFLOAT _dtfsq;
__device__ __constant__ X_CFLOAT* _bond_distance;
__device__ __constant__ X_CFLOAT* _angle_distance;
__device__ __constant__ int _max_iter;
__device__ __constant__ X_FLOAT _tolerance;
__device__ __constant__ X_CFLOAT _tolerance;
#include "fix_shake_cuda_kernel.cu"
void Cuda_FixShakeCuda_UpdateNmax(cuda_shared_data* sdata)
{
cudaMemcpyToSymbol(MY_AP(x) , & sdata->atom.x .dev_data, sizeof(X_FLOAT*));
cudaMemcpyToSymbol(MY_AP(v) , & sdata->atom.v .dev_data, sizeof(V_FLOAT*));
cudaMemcpyToSymbol(MY_AP(f) , & sdata->atom.f .dev_data, sizeof(F_FLOAT*));
cudaMemcpyToSymbol(MY_AP(x) , & sdata->atom.x .dev_data, sizeof(X_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(v) , & sdata->atom.v .dev_data, sizeof(V_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(f) , & sdata->atom.f .dev_data, sizeof(F_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(mask) , & sdata->atom.mask .dev_data, sizeof(int*));
cudaMemcpyToSymbol(MY_AP(tag) , & sdata->atom.tag .dev_data, sizeof(int*));
cudaMemcpyToSymbol(MY_AP(nlocal) , & sdata->atom.nlocal , sizeof(int));
cudaMemcpyToSymbol(MY_AP(nmax) , & sdata->atom.nmax , sizeof(int));
cudaMemcpyToSymbol(MY_AP(rmass) , & sdata->atom.rmass.dev_data, sizeof(V_FLOAT*));
cudaMemcpyToSymbol(MY_AP(rmass) , & sdata->atom.rmass.dev_data, sizeof(V_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(type) , & sdata->atom.type .dev_data, sizeof(int*));
cudaMemcpyToSymbol(MY_AP(map_array), & sdata->atom.map_array .dev_data, sizeof(int*));
cudaMemcpyToSymbol(MY_AP(vatom) , & sdata->atom.vatom.dev_data, sizeof(ENERGY_FLOAT*));
cudaMemcpyToSymbol(MY_AP(vatom) , & sdata->atom.vatom.dev_data, sizeof(ENERGY_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(debugdata), & sdata->debugdata , sizeof(int*));
}
void Cuda_FixShakeCuda_UpdateDomain(cuda_shared_data* sdata)
{
cudaMemcpyToSymbol(MY_AP(periodicity), sdata->domain.periodicity , sizeof(int) * 3);
cudaMemcpyToSymbol(MY_AP(prd) , sdata->domain.prd , sizeof(X_FLOAT) * 3);
cudaMemcpyToSymbol(MY_AP(prd) , sdata->domain.prd , sizeof(X_CFLOAT) * 3);
cudaMemcpyToSymbol(MY_AP(triclinic) , &sdata->domain.triclinic , sizeof(int));
cudaMemcpyToSymbol(MY_AP(h) , sdata->domain.h , sizeof(X_FLOAT) * 6);
cudaMemcpyToSymbol(MY_AP(h) , sdata->domain.h , sizeof(X_CFLOAT) * 6);
}
void Cuda_FixShakeCuda_UpdateBuffer(cuda_shared_data* sdata, int size)
@ -89,10 +89,10 @@ void Cuda_FixShakeCuda_UpdateBuffer(cuda_shared_data* sdata, int size)
cudaMemcpyToSymbol(MY_AP(buffer) , & sdata->buffer, sizeof(int*));
}
void Cuda_FixShakeCuda_Init(cuda_shared_data* sdata, X_FLOAT dtv, F_FLOAT dtfsq,
void Cuda_FixShakeCuda_Init(cuda_shared_data* sdata, X_CFLOAT dtv, F_CFLOAT dtfsq,
void* shake_flag, void* shake_atom, void* shake_type, void* xshake,
void* bond_distance, void* angle_distance, void* virial,
int max_iter, X_FLOAT tolerance)
int max_iter, X_CFLOAT tolerance)
{
Cuda_FixShakeCuda_UpdateNmax(sdata);
Cuda_FixShakeCuda_UpdateDomain(sdata);
@ -100,17 +100,17 @@ void Cuda_FixShakeCuda_Init(cuda_shared_data* sdata, X_FLOAT dtv, F_FLOAT dtfsq,
cudaMemcpyToSymbol(MY_AP(shake_type) , & shake_type , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(shake_flag) , & shake_flag , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(xshake) , & xshake , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(dtv) , & dtv , sizeof(X_FLOAT));
cudaMemcpyToSymbol(MY_AP(dtfsq) , & dtfsq , sizeof(F_FLOAT));
cudaMemcpyToSymbol(MY_AP(dtv) , & dtv , sizeof(X_CFLOAT));
cudaMemcpyToSymbol(MY_AP(dtfsq) , & dtfsq , sizeof(F_CFLOAT));
cudaMemcpyToSymbol(MY_AP(bond_distance) , & bond_distance , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(angle_distance) , & angle_distance , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(virial) , & virial , sizeof(void*));
cudaMemcpyToSymbol(MY_AP(flag) , &sdata->flag , sizeof(int*));
cudaMemcpyToSymbol(MY_AP(max_iter) , &max_iter , sizeof(int));
cudaMemcpyToSymbol(MY_AP(tolerance) , &tolerance , sizeof(X_FLOAT));
cudaMemcpyToSymbol(MY_AP(tolerance) , &tolerance , sizeof(X_CFLOAT));
if(sdata->atom.mass_host)
cudaMemcpyToSymbol(MY_AP(mass), & sdata->atom.mass.dev_data , sizeof(V_FLOAT*));
cudaMemcpyToSymbol(MY_AP(mass), & sdata->atom.mass.dev_data , sizeof(V_CFLOAT*));
cudaMemcpyToSymbol(MY_AP(rmass_flag), & sdata->atom.rmass_flag , sizeof(int)); //
@ -149,16 +149,16 @@ void Cuda_FixShakeCuda_Shake(cuda_shared_data* sdata, int vflag, int vflag_atom,
if(sdata->atom.update_nlocal)
cudaMemcpyToSymbol(MY_AP(nlocal) , & sdata->atom.nlocal , sizeof(int));
int3 layout = getgrid(sdata->atom.nlocal, 6 * sizeof(ENERGY_FLOAT), 64);
int3 layout = getgrid(sdata->atom.nlocal, 6 * sizeof(ENERGY_CFLOAT), 64);
dim3 threads(layout.z, 1, 1);
dim3 grid(layout.x, layout.y, 1);
if(sdata->buffer_new)
Cuda_FixShakeCuda_UpdateBuffer(sdata, grid.x * grid.y * 6 * sizeof(ENERGY_FLOAT));
Cuda_FixShakeCuda_UpdateBuffer(sdata, grid.x * grid.y * 6 * sizeof(ENERGY_CFLOAT));
BindXTypeTexture(sdata);
FixShakeCuda_Shake_Kernel <<< grid, threads, 6* threads.x* sizeof(ENERGY_FLOAT)>>> (vflag, vflag_atom, list, nlist);
FixShakeCuda_Shake_Kernel <<< grid, threads, 6* threads.x* sizeof(ENERGY_CFLOAT)>>> (vflag, vflag_atom, list, nlist);
cudaThreadSynchronize();
CUT_CHECK_ERROR("FixShakeCuda_Shake: Kernel execution failed");
@ -168,7 +168,7 @@ void Cuda_FixShakeCuda_Shake(cuda_shared_data* sdata, int vflag, int vflag_atom,
grid.x = 6;
grid.y = 1;
threads.x = 256;
MY_AP(PairVirialCompute_reduce) <<< grid, threads, threads.x* sizeof(ENERGY_FLOAT)>>>(n);
MY_AP(PairVirialCompute_reduce) <<< grid, threads, threads.x* sizeof(ENERGY_CFLOAT)>>>(n);
cudaThreadSynchronize();
CUT_CHECK_ERROR("Cuda_FixShakeCuda: (no binning) virial compute Kernel execution failed");
}
@ -183,14 +183,14 @@ int Cuda_FixShakeCuda_PackComm(cuda_shared_data* sdata, int n, int iswap, void*
if(sdata->atom.update_nlocal)
cudaMemcpyToSymbol(MY_AP(nlocal) , & sdata->atom.nlocal , sizeof(int));
int size = n * 3 * sizeof(X_FLOAT);
int size = n * 3 * sizeof(X_CFLOAT);
if(sdata->buffer_new or (size > sdata->buffersize))
Cuda_FixShakeCuda_UpdateBuffer(sdata, size);
X_FLOAT dx = 0.0;
X_FLOAT dy = 0.0;
X_FLOAT dz = 0.0;
X_CFLOAT dx = 0.0;
X_CFLOAT dy = 0.0;
X_CFLOAT dz = 0.0;
if(pbc_flag != 0) {
if(sdata->domain.triclinic == 0) {
@ -212,7 +212,7 @@ int Cuda_FixShakeCuda_PackComm(cuda_shared_data* sdata, int n, int iswap, void*
cudaMemset(sdata->flag, 0, sizeof(int));
FixShakeCuda_PackComm_Kernel <<< grid, threads, 0>>>((int*) sdata->comm.sendlist.dev_data, n, sdata->comm.maxlistlength, iswap, dx, dy, dz);
cudaThreadSynchronize();
cudaMemcpy(buf_send, sdata->buffer, n * 3 * sizeof(X_FLOAT), cudaMemcpyDeviceToHost);
cudaMemcpy(buf_send, sdata->buffer, n * 3 * sizeof(X_CFLOAT), cudaMemcpyDeviceToHost);
int aflag;
cudaMemcpy(&aflag, sdata->flag, sizeof(int), cudaMemcpyDeviceToHost);
@ -232,16 +232,16 @@ int Cuda_FixShakeCuda_PackComm_Self(cuda_shared_data* sdata, int n, int iswap, i
if(sdata->atom.update_nlocal)
cudaMemcpyToSymbol(MY_AP(nlocal) , & sdata->atom.nlocal , sizeof(int));
int size = n * 3 * sizeof(X_FLOAT);
int size = n * 3 * sizeof(X_CFLOAT);
if(sdata->buffer_new or (size > sdata->buffersize))
Cuda_FixShakeCuda_UpdateBuffer(sdata, size);
static int count = -1;
count++;
X_FLOAT dx = 0.0;
X_FLOAT dy = 0.0;
X_FLOAT dz = 0.0;
X_CFLOAT dx = 0.0;
X_CFLOAT dy = 0.0;
X_CFLOAT dz = 0.0;
if(pbc_flag != 0) {
if(sdata->domain.triclinic == 0) {
@ -278,7 +278,7 @@ void Cuda_FixShakeCuda_UnpackComm(cuda_shared_data* sdata, int n, int first, voi
if(sdata->atom.update_nlocal)
cudaMemcpyToSymbol(MY_AP(nlocal) , & sdata->atom.nlocal , sizeof(int));
int size = n * 3 * sizeof(X_FLOAT);
int size = n * 3 * sizeof(X_CFLOAT);
if(sdata->buffer_new or (size > sdata->buffersize))
Cuda_FixShakeCuda_UpdateBuffer(sdata, size);
@ -288,7 +288,7 @@ void Cuda_FixShakeCuda_UnpackComm(cuda_shared_data* sdata, int n, int first, voi
dim3 grid(layout.x, layout.y, 1);
if(sdata->atom.nlocal > 0) {
cudaMemcpy(sdata->buffer, (void*)buf_recv, n * 3 * sizeof(X_FLOAT), cudaMemcpyHostToDevice);
cudaMemcpy(sdata->buffer, (void*)buf_recv, n * 3 * sizeof(X_CFLOAT), cudaMemcpyHostToDevice);
FixShakeCuda_UnpackComm_Kernel <<< grid, threads, 0>>>(n, first);
cudaThreadSynchronize();
CUT_CHECK_ERROR("Cuda_FixShakeCuda_UnpackComm: Kernel execution failed");