Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

ProdEnvMatAMixOp: move filter_ftype out of nsamples loop #2604

Merged
merged 1 commit into from
Jun 25, 2023
Merged

ProdEnvMatAMixOp: move filter_ftype out of nsamples loop #2604

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 21 additions & 21 deletions source/op/prod_env_mat_multi_device.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1135,6 +1135,12 @@ class ProdEnvMatAMixOp : public OpKernel {
context->allocate_output(context_output_index++, nmask_shape,
&nmask_tensor));

Tensor fake_type_tensor; // all zeros
TensorShape fake_type_shape;
fake_type_shape.AddDim(nsamples * nall);
OP_REQUIRES_OK(context, context->allocate_temp(DT_INT32, fake_type_shape,
&fake_type_tensor));

FPTYPE* p_em = descrpt_tensor->flat<FPTYPE>().data();
FPTYPE* p_em_deriv = descrpt_deriv_tensor->flat<FPTYPE>().data();
FPTYPE* p_rij = rij_tensor->flat<FPTYPE>().data();
Expand All @@ -1146,6 +1152,20 @@ class ProdEnvMatAMixOp : public OpKernel {
const FPTYPE* avg = avg_tensor.flat<FPTYPE>().data();
const FPTYPE* std = std_tensor.flat<FPTYPE>().data();
const int* p_type = type_tensor.flat<int>().data();
int* p_f_type = fake_type_tensor.flat<int>().data();

if (device == "GPU") {
#if GOOGLE_CUDA
deepmd::filter_ftype_gpu_cuda(p_f_type, p_type, nsamples * nall);
#endif
#if TENSORFLOW_USE_ROCM
deepmd::filter_ftype_gpu_rocm(p_f_type, p_type, nsamples * nall);
#endif
} else if (device == "CPU") {
for (int ii = 0; ii < nsamples * nall; ii++) {
p_f_type[ii] = (p_type[ii] < 0) ? -1 : 0;
}
}

// loop over samples
for (int_64 ff = 0; ff < nsamples; ++ff) {
Expand All @@ -1158,6 +1178,7 @@ class ProdEnvMatAMixOp : public OpKernel {
const FPTYPE* coord = p_coord + ff * nall * 3;
const FPTYPE* box = p_box + ff * 9;
const int* type = p_type + ff * nall;
const int* f_type = p_f_type + ff * nall;

if (device == "GPU") {
#if GOOGLE_CUDA
Expand All @@ -1171,13 +1192,6 @@ class ProdEnvMatAMixOp : public OpKernel {
int frame_nall = nall;
int mesh_tensor_size = static_cast<int>(mesh_tensor.NumElements());
std::vector<Tensor> tensor_list(7);
Tensor fake_type; // all zeros
TensorShape fake_type_shape;
fake_type_shape.AddDim(nall);
OP_REQUIRES_OK(context, context->allocate_temp(
DT_INT32, fake_type_shape, &fake_type));
deepmd::filter_ftype_gpu_cuda(fake_type.flat<int>().data(), type, nall);
const int* f_type = fake_type.flat<int>().data();
// prepare coord and nlist
_prepare_coord_nlist_gpu<FPTYPE>(
context, &tensor_list[0], &coord, coord_cpy, &f_type, type_cpy,
Expand Down Expand Up @@ -1222,13 +1236,6 @@ class ProdEnvMatAMixOp : public OpKernel {
int frame_nall = nall;
int mesh_tensor_size = static_cast<int>(mesh_tensor.NumElements());
std::vector<Tensor> tensor_list(7);
Tensor fake_type; // all zeros
TensorShape fake_type_shape;
fake_type_shape.AddDim(nall);
OP_REQUIRES_OK(context, context->allocate_temp(
DT_INT32, fake_type_shape, &fake_type));
deepmd::filter_ftype_gpu_rocm(fake_type.flat<int>().data(), type, nall);
const int* f_type = fake_type.flat<int>().data();
// prepare coord and nlist
_prepare_coord_nlist_gpu_rocm<FPTYPE>(
context, &tensor_list[0], &coord, coord_cpy, &f_type, type_cpy,
Expand Down Expand Up @@ -1271,13 +1278,6 @@ class ProdEnvMatAMixOp : public OpKernel {
std::vector<FPTYPE> coord_cpy;
std::vector<int> type_cpy;
int frame_nall = nall;
std::vector<int> fake_type(nall, 0);
for (int ii = 0; ii < nall; ii++) {
if (type[ii] < 0) {
fake_type[ii] = -1;
}
}
const int* f_type = &fake_type[0];
// prepare coord and nlist
_prepare_coord_nlist_cpu<FPTYPE>(
context, &coord, coord_cpy, &f_type, type_cpy, idx_mapping, inlist,
Expand Down