Skip to content

Fix low efficiency of stress_kin in DCU #5636

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

Merged
merged 4 commits into from
Nov 28, 2024
Merged

Fix low efficiency of stress_kin in DCU #5636

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c76845a
fix stuck in out_chg
Qianruipku Nov 25, 2024
308c983
Merge branch 'develop' into develop
Qianruipku Nov 25, 2024
94e1c9a
Merge branch 'develop' of https://github.com/deepmodeling/abacus-deve…
Qianruipku Nov 28, 2024
e85e43b
fix DCU low efficiency
Qianruipku Nov 28, 2024
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
22 changes: 19 additions & 3 deletions source/module_hamilt_pw/hamilt_pwdft/kernels/cuda/stress_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,9 +185,25 @@ __global__ void cal_multi_dot(const int npw,
const thrust::complex<FPTYPE>* psi,
FPTYPE* sum)
{
int idx = threadIdx.x + blockIdx.x * blockDim.x;
if (idx < npw) {
atomicAdd(sum, fac * gk1[idx] * gk2[idx] * d_kfac[idx] * thrust::norm(psi[idx]));
__shared__ FPTYPE s_sum[THREADS_PER_BLOCK];
int tid = threadIdx.x + blockIdx.x * blockDim.x;
int cacheid = threadIdx.x;
FPTYPE local_sum = 0;
while (tid < npw) {
local_sum += fac * gk1[tid] * gk2[tid] * d_kfac[tid] * thrust::norm(psi[tid]);
tid += blockDim.x * gridDim.x;
}
s_sum[cacheid] = local_sum;
__syncthreads();

for (int s = blockDim.x / 2; s > 0; s >>= 1) {
if (cacheid < s) {
s_sum[cacheid] += s_sum[cacheid + s];
}
__syncthreads();
}
if (cacheid == 0) {
atomicAdd(sum, s_sum[0]);
}
}

Expand Down
22 changes: 19 additions & 3 deletions source/module_hamilt_pw/hamilt_pwdft/kernels/rocm/stress_op.hip.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,9 +342,25 @@ __global__ void cal_multi_dot(const int npw,
const thrust::complex<FPTYPE>* psi,
FPTYPE* sum)
{
int idx = threadIdx.x + blockIdx.x * blockDim.x;
if (idx < npw) {
atomicAdd(sum, fac * gk1[idx] * gk2[idx] * d_kfac[idx] * thrust::norm(psi[idx]));
__shared__ FPTYPE s_sum[THREADS_PER_BLOCK];
int tid = threadIdx.x + blockIdx.x * blockDim.x;
int cacheid = threadIdx.x;
FPTYPE local_sum = 0;
while (tid < npw) {
local_sum += fac * gk1[tid] * gk2[tid] * d_kfac[tid] * thrust::norm(psi[tid]);
tid += blockDim.x * gridDim.x;
}
s_sum[cacheid] = local_sum;
__syncthreads();

for (int s = blockDim.x / 2; s > 0; s >>= 1) {
if (cacheid < s) {
s_sum[cacheid] += s_sum[cacheid + s];
}
__syncthreads();
}
if (cacheid == 0) {
atomicAdd(sum, s_sum[0]);
}
}

Expand Down