Skip to content

[Fix] Eliminate unnecessary vkb CPU allocation on GPU path #7296

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
mohanchen merged 5 commits into from
Apr 29, 2026
Merged

[Fix] Eliminate unnecessary vkb CPU allocation on GPU path #7296

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
cba00f5
[Fix] Skip unnecessary vkb CPU allocation on GPU path in pseudopot_ce…
Cstandardlib Apr 28, 2026
a0f43dc
Update source/source_pw/module_pwdft/vnl_pw_grad.cpp
Cstandardlib Apr 28, 2026
b71e3fe
Update source/source_pw/module_pwdft/vnl_pw_grad.cpp
Cstandardlib Apr 28, 2026
b963e3b
Revert "Update source/source_pw/module_pwdft/vnl_pw_grad.cpp"
Cstandardlib Apr 29, 2026
0999a78
Revert "Update source/source_pw/module_pwdft/vnl_pw_grad.cpp"
Cstandardlib Apr 29, 2026
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
8 changes: 4 additions & 4 deletions source/source_pw/module_pwdft/hamilt_pw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,7 +272,7 @@ void HamiltPW<T, Device>::sPsi(const T* psi_in, // psi
this->ppcell->nkb,
&one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
psi_in,
inc,
&zero,
Expand All @@ -288,7 +288,7 @@ void HamiltPW<T, Device>::sPsi(const T* psi_in, // psi
npw,
&one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
psi_in,
nrow,
&zero,
Expand Down Expand Up @@ -360,7 +360,7 @@ void HamiltPW<T, Device>::sPsi(const T* psi_in, // psi
this->ppcell->nkb,
&one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
ps,
inc,
&one,
Expand All @@ -376,7 +376,7 @@ void HamiltPW<T, Device>::sPsi(const T* psi_in, // psi
this->ppcell->nkb,
&one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
ps,
this->ppcell->nkb,
&one,
Expand Down
8 changes: 4 additions & 4 deletions source/source_pw/module_pwdft/op_pw_nl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,7 @@ void Nonlocal<OperatorPW<T, Device>>::add_nonlocal_pp(T *hpsi_in, const T *becp,
this->ppcell->nkb,
&this->one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
this->ps,
inc,
&this->one,
Expand All @@ -197,7 +197,7 @@ void Nonlocal<OperatorPW<T, Device>>::add_nonlocal_pp(T *hpsi_in, const T *becp,
this->ppcell->nkb,
&this->one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
this->ps,
npm,
&this->one,
Expand Down Expand Up @@ -251,7 +251,7 @@ void Nonlocal<OperatorPW<T, Device>>::act(
nkb,
&this->one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
tmpsi_in,
inc,
&this->zero,
Expand All @@ -276,7 +276,7 @@ void Nonlocal<OperatorPW<T, Device>>::act(
this->npw,
&this->one,
this->vkb,
this->ppcell->vkb.nc,
this->ppcell->vkbnc,
tmpsi_in,
max_npw,
&this->zero,
Expand Down
11 changes: 9 additions & 2 deletions source/source_pw/module_pwdft/vnl_pw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,10 +214,17 @@ void pseudopot_cell_vnl::init(const UnitCell& ucell,
// dq+4)*cell_factor;
this->lmaxq = 2 * this->lmaxkb + 1;
int npwx = this->wfcpw->npwk_max;
this->vkbnc = npwx;
if (nkb > 0 && allocate_vkb)
{
vkb.create(nkb, npwx);
ModuleBase::Memory::record("VNL::vkb", nkb * npwx * sizeof(std::complex<double>));
if (!this->use_gpu_)
{
vkb.create(nkb, npwx);
ModuleBase::Memory::record("VNL::vkb", nkb * npwx * sizeof(std::complex<double>));
}
// GPU path: vkb ComplexMatrix is not allocated.
// Column dimension is stored in vkbnc for gemm/gemv leading dimension.
// Actual GPU buffers (c_vkb/z_vkb) are allocated below.
}

// this->nqx = 10000; // calculted in allocate_nlpot.f90
Expand Down
4 changes: 4 additions & 0 deletions source/source_pw/module_pwdft/vnl_pw.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,6 +108,10 @@ class pseudopot_cell_vnl
std::complex<double>*** vkb_alpha;
Structure_Factor* psf = nullptr;

// Column dimension of vkb matrix (= npwx), used as leading dimension in gemm/gemv.
// On GPU path vkb ComplexMatrix is not allocated to save CPU memory; this stores the dimension.
int vkbnc = 0;

// other variables
std::complex<double> Cal_C(int alpha, int lu, int mu, int L, int M);

Expand Down
5 changes: 5 additions & 0 deletions source/source_pw/module_pwdft/vnl_pw_grad.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,6 +91,11 @@ void pseudopot_cell_vnl::getgradq_vnl(const UnitCell& ucell,

ModuleBase::YlmReal::grad_Ylm_Real(x1, npw, gk, ylm, dylm[0], dylm[1], dylm[2]);

// GPU path skips vkb allocation in init(); allocate now if needed
if (this->vkb.nc == 0 && this->nkb > 0 && this->vkbnc > 0) {
this->vkb.create(this->nkb, this->vkbnc);
Comment thread
Cstandardlib marked this conversation as resolved.
}

int jkb = 0;
for(int it = 0;it < ucell.ntype;it++)
{
Expand Down
Loading