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

Fix omp offload without cuda #3514

Merged
merged 3 commits into from
Oct 6, 2021
Merged

Fix omp offload without cuda #3514

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3191d09
Revert integration with DiracMatrixComputeOMPTarget
ye-luo Oct 6, 2021
4b2ef72
Mark all the RefVector const.
ye-luo Oct 6, 2021
3a1fc75
Merge branch 'develop' into fix-omp-nocuda
ye-luo Oct 6, 2021
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
4 changes: 2 additions & 2 deletions src/QMCWaveFunctions/Fermion/DiracDeterminantBatched.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,8 +54,8 @@ void DiracDeterminantBatched<DET_ENGINE>::invertPsiM(const DualMatrix<Value>& ps

template<typename DET_ENGINE>
void DiracDeterminantBatched<DET_ENGINE>::mw_invertPsiM(const RefVectorWithLeader<WaveFunctionComponent>& wfc_list,
RefVector<const DualMatrix<Value>>& logdetT_list,
RefVector<DualMatrix<Value>>& a_inv_list,
const RefVector<const DualMatrix<Value>>& logdetT_list,
const RefVector<DualMatrix<Value>>& a_inv_list,
const std::vector<bool>& compute_mask) const
{
auto& wfc_leader = wfc_list.getCastedLeader<DiracDeterminantBatched<DET_ENGINE>>();
Expand Down
4 changes: 2 additions & 2 deletions src/QMCWaveFunctions/Fermion/DiracDeterminantBatched.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -279,8 +279,8 @@ class DiracDeterminantBatched : public DiracDeterminantBase
* the compute mask. See future PR for those changes, or drop of compute_mask argument.
*/
void mw_invertPsiM(const RefVectorWithLeader<WaveFunctionComponent>& wfc_list,
RefVector<const DualMatrix<Value>>& logdetT_list,
RefVector<DualMatrix<Value>>& a_inv_list,
const RefVector<const DualMatrix<Value>>& logdetT_list,
const RefVector<DualMatrix<Value>>& a_inv_list,
const std::vector<bool>& compute_mask) const;

/** Does a Phi->mw_evaluate_notranspose then mw_invertPsiM over a set of
Expand Down
22 changes: 16 additions & 6 deletions src/QMCWaveFunctions/Fermion/DiracMatrix.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -180,9 +180,14 @@ class DiracMatrix
* @tparam TMAT matrix value type
* @tparam TREAL real type
*/
template<typename TMAT, typename TREAL>
inline std::enable_if_t<std::is_same<T_FP, TMAT>::value> invert_transpose(const Matrix<TMAT>& amat,
Matrix<TMAT>& invMat,
template<typename TMAT,
typename ALLOC1,
typename ALLOC2,
typename TREAL,
typename = std::enable_if_t<qmc_allocator_traits<ALLOC1>::is_host_accessible>,
typename = std::enable_if_t<qmc_allocator_traits<ALLOC2>::is_host_accessible>>
inline std::enable_if_t<std::is_same<T_FP, TMAT>::value> invert_transpose(const Matrix<TMAT, ALLOC1>& amat,
Matrix<TMAT, ALLOC2>& invMat,
std::complex<TREAL>& LogDet)
{
const int n = invMat.rows();
Expand All @@ -196,9 +201,14 @@ class DiracMatrix
* @tparam TMAT matrix value type
* @tparam TREAL real type
*/
template<typename TMAT, typename TREAL>
inline std::enable_if_t<!std::is_same<T_FP, TMAT>::value> invert_transpose(const Matrix<TMAT>& amat,
Matrix<TMAT>& invMat,
template<typename TMAT,
typename ALLOC1,
typename ALLOC2,
typename TREAL,
typename = std::enable_if_t<qmc_allocator_traits<ALLOC1>::is_host_accessible>,
typename = std::enable_if_t<qmc_allocator_traits<ALLOC2>::is_host_accessible>>
inline std::enable_if_t<!std::is_same<T_FP, TMAT>::value> invert_transpose(const Matrix<TMAT, ALLOC1>& amat,
Matrix<TMAT, ALLOC2>& invMat,
std::complex<TREAL>& LogDet)
{
const int n = invMat.rows();
Expand Down
12 changes: 6 additions & 6 deletions src/QMCWaveFunctions/Fermion/DiracMatrixComputeCUDA.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,8 +87,8 @@ class DiracMatrixComputeCUDA : public Resource
*/
template<typename TMAT>
inline void mw_computeInvertAndLog(CUDALinearAlgebraHandles& cuda_handles,
RefVector<DualMatrix<TMAT>>& a_mats,
RefVector<DualMatrix<TMAT>>& inv_a_mats,
const RefVector<DualMatrix<TMAT>>& a_mats,
const RefVector<DualMatrix<TMAT>>& inv_a_mats,
const int n,
const int lda,
DualVector<LogValue>& log_values)
Expand Down Expand Up @@ -276,8 +276,8 @@ class DiracMatrixComputeCUDA : public Resource
template<typename TMAT>
inline std::enable_if_t<!std::is_same<VALUE_FP, TMAT>::value> mw_invertTranspose(
CUDALinearAlgebraHandles& cuda_handles,
RefVector<DualMatrix<TMAT>>& a_mats,
RefVector<DualMatrix<TMAT>>& inv_a_mats,
const RefVector<DualMatrix<TMAT>>& a_mats,
const RefVector<DualMatrix<TMAT>>& inv_a_mats,
DualVector<LogValue>& log_values,
const std::vector<bool>& compute_mask)
{
Expand Down Expand Up @@ -319,8 +319,8 @@ class DiracMatrixComputeCUDA : public Resource
template<typename TMAT>
inline std::enable_if_t<std::is_same<VALUE_FP, TMAT>::value> mw_invertTranspose(
CUDALinearAlgebraHandles& cuda_handles,
RefVector<DualMatrix<TMAT>>& a_mats,
RefVector<DualMatrix<TMAT>>& inv_a_mats,
const RefVector<DualMatrix<TMAT>>& a_mats,
const RefVector<DualMatrix<TMAT>>& inv_a_mats,
DualVector<LogValue>& log_values,
const std::vector<bool>& compute_mask)
{
Expand Down
4 changes: 2 additions & 2 deletions src/QMCWaveFunctions/Fermion/DiracMatrixComputeOMPTarget.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -224,8 +224,8 @@ class DiracMatrixComputeOMPTarget : public Resource
*/
template<typename TMAT>
inline void mw_invertTranspose(HandleResource& resource,
RefVector<const OffloadPinnedMatrix<TMAT>>& a_mats,
RefVector<OffloadPinnedMatrix<TMAT>>& inv_a_mats,
const RefVector<const OffloadPinnedMatrix<TMAT>>& a_mats,
const RefVector<OffloadPinnedMatrix<TMAT>>& inv_a_mats,
OffloadPinnedVector<LogValue>& log_values,
const std::vector<bool>& recompute)
{
Expand Down
4 changes: 2 additions & 2 deletions src/QMCWaveFunctions/Fermion/MatrixDelayedUpdateCUDA.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,8 +412,8 @@ class MatrixDelayedUpdateCUDA
}

static void mw_invertTranspose(const RefVectorWithLeader<This_t>& engines,
RefVector<const DualMatrix<Value>>& psiM_list,
RefVector<DualMatrix<Value>>& psiMinv_list,
const RefVector<const DualMatrix<Value>>& psiM_list,
const RefVector<DualMatrix<Value>>& psiMinv_list,
DualVector<LogValue>& log_values,
const std::vector<bool>& compute_mask)
{
Expand Down
20 changes: 11 additions & 9 deletions src/QMCWaveFunctions/Fermion/MatrixUpdateOMPTarget.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,24 +178,26 @@ class MatrixUpdateOMPTarget
}

static void mw_invertTranspose(const RefVectorWithLeader<This_t>& engines,
RefVector<const OffloadMatrix<Value>>& logdetT_list,
RefVector<OffloadMatrix<Value>>& a_inv_refs,
const RefVector<const OffloadMatrix<Value>>& psiM_list,
const RefVector<OffloadMatrix<Value>>& a_inv_refs,
OffloadVector<LogValue>& log_values,
const std::vector<bool>& compute_mask)
{
auto& engine_leader = engines.getLeader();
auto& det_inverter = engine_leader.get_det_inverter();

a_inv_refs.reserve(engines.size());

for (int iw = 0; iw < engines.size(); iw++)
{
a_inv_refs.emplace_back(engines[iw].get_ref_psiMinv());
const Value* a_inv_ptr = a_inv_refs.back().get().data();
PRAGMA_OFFLOAD("omp target update to(a_inv_ptr[:a_inv_refs.back().get().size()])")
auto& Ainv = a_inv_refs[iw].get();
engine_leader.detEng.invert_transpose(psiM_list[iw].get(), Ainv, log_values[iw]);
Value* Ainv_ptr = Ainv.data();
PRAGMA_OFFLOAD("omp target update to(Ainv_ptr[:Ainv.size()])")
}
typename DetInverter::HandleResource dummy;
det_inverter.mw_invertTranspose(dummy, logdetT_list, a_inv_refs, log_values, compute_mask);
PRAGMA_OFFLOAD("omp taskwait")

//FIXME DiracMatrixComputeOMPTarget is either broken or connected incorrectly
//typename DetInverter::HandleResource dummy;
//det_inverter.mw_invertTranspose(dummy, psiM_list, a_inv_refs, log_values, compute_mask);
}

template<typename GT>
Expand Down