Skip to content

Commit

Permalink
Merge branch 'add_LGPUMPI' into add_py_LGPUMPI
Browse files Browse the repository at this point in the history
  • Loading branch information
@vincentmr
vincentmr authored Oct 23, 2023
2 parents 6ad1c7c + 84e2eb6 commit d3af819
Show file tree
Hide file tree
Showing 15 changed files with 103 additions and 22 deletions.
6 changes: 6 additions & 0 deletions pennylane_lightning/core/src/algorithms/tests/mpi/Test_AdjointJacobianMPI.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,7 @@ template <typename TypeList> void testAdjointJacobian() {
using ComplexT = typename StateVectorT::ComplexT;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

const std::vector<PrecisionT> param{-M_PI / 7, M_PI / 5, 2 * M_PI / 3};

Expand All @@ -81,6 +82,7 @@ template <typename TypeList> void testAdjointJacobian() {

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -131,6 +133,7 @@ template <typename TypeList> void testAdjointJacobian() {

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -181,6 +184,7 @@ template <typename TypeList> void testAdjointJacobian() {

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -293,6 +297,7 @@ template <typename TypeList> void testAdjointJacobian() {

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -344,6 +349,7 @@ template <typename TypeList> void testAdjointJacobian() {

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down
12 changes: 12 additions & 0 deletions pennylane_lightning/core/src/measurements/tests/mpi/Test_MeasurementsBaseMPI.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,6 +68,7 @@ template <typename TypeList> void testProbabilities() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -77,6 +78,7 @@ template <typename TypeList> void testProbabilities() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -127,6 +129,7 @@ template <typename TypeList> void testNamedObsExpval() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -136,6 +139,7 @@ template <typename TypeList> void testNamedObsExpval() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -197,6 +201,7 @@ template <typename TypeList> void testHermitianObsExpval() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -206,6 +211,7 @@ template <typename TypeList> void testHermitianObsExpval() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -294,6 +300,7 @@ template <typename TypeList> void testNamedObsVar() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -303,6 +310,7 @@ template <typename TypeList> void testNamedObsVar() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -365,6 +373,7 @@ template <typename TypeList> void testHermitianObsVar() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -374,6 +383,7 @@ template <typename TypeList> void testHermitianObsVar() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -470,6 +480,7 @@ template <typename TypeList> void testSamples() {
size_t num_qubits = 3;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -479,6 +490,7 @@ template <typename TypeList> void testSamples() {

int nDevices = 0;
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down
14 changes: 0 additions & 14 deletions pennylane_lightning/core/src/observables/tests/mpi/Test_ObservablesMPI.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -422,20 +422,6 @@ template <typename TypeList> void testHamiltonianBase() {

REQUIRE(ham1->getWires() == std::vector<size_t>{0, 5, 9});
}

/*
DYNAMIC_SECTION("applyInPlace must fail - "
<< StateVectorMPIToName<StateVectorT>::name) {
auto ham =
HamiltonianT::create({PrecisionT{1.0}, h, h}, {zz, x1, x2});
auto st_data = createZeroState<ComplexT>(2);
StateVectorT state_vector(st_data.data(), st_data.size());
REQUIRE_THROWS_AS(ham->applyInPlace(state_vector),
LightningException);
}
*/
}
testHamiltonianBase<typename TypeList::Next>();
}
Expand Down
1 change: 1 addition & 0 deletions pennylane_lightning/core/src/simulators/lightning_gpu/StateVectorCudaMPI.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1867,6 +1867,7 @@ class StateVectorCudaMPI
// LCOV_EXCL_STOP

cuDoubleComplex expect_;

// compute expectation
PL_CUSTATEVEC_IS_SUCCESS(custatevecComputeExpectation(
/* custatevecHandle_t */ handle_.get(),
Expand Down
16 changes: 16 additions & 0 deletions ...ing/core/src/simulators/lightning_gpu/algorithms/tests/mpi/Test_AdjointJacobianGPUMPI.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobianMPI Op=RX, Obs=[Z,Z]",
using StateVectorT = StateVectorCudaMPI<double>;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

AdjointJacobianMPI<StateVectorT> adj;
std::vector<double> param{-M_PI / 7, M_PI / 5, 2 * M_PI / 3};
Expand All @@ -66,6 +67,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobianMPI Op=RX, Obs=[Z,Z]",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -104,6 +106,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobianMPI Op=[QubitStateVector, "
using StateVectorT = StateVectorCudaMPI<double>;

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

AdjointJacobianMPI<StateVectorT> adj;
std::vector<double> param{-M_PI / 7, M_PI / 5, 2 * M_PI / 3};
Expand All @@ -124,6 +127,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobianMPI Op=[QubitStateVector, "

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -167,6 +171,7 @@ TEST_CASE(
std::vector<double> jacobian_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -177,6 +182,7 @@ TEST_CASE(

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -234,6 +240,7 @@ TEST_CASE(
std::vector<double> jacobian_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -244,6 +251,7 @@ TEST_CASE(

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -297,6 +305,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobian Op=[RX,RX,RX], Obs=[ZZZ]",
std::vector<double> jacobian_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -307,6 +316,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobian Op=[RX,RX,RX], Obs=[ZZZ]",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -358,6 +368,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobian Op=Mixed, Obs=[XXX]",
std::vector<double> jacobian_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -368,6 +379,7 @@ TEST_CASE("AdjointJacobianGPUMPI::adjointJacobian Op=Mixed, Obs=[XXX]",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -436,6 +448,7 @@ TEST_CASE("AdjointJacobianGPU::AdjointJacobianGPUMPI Op=[RX,RX,RX], "
std::vector<double> jacobian_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -446,6 +459,7 @@ TEST_CASE("AdjointJacobianGPU::AdjointJacobianGPUMPI Op=[RX,RX,RX], "

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -504,6 +518,7 @@ TEST_CASE("AdjointJacobianGPU::AdjointJacobianGPU Test HermitianObs",
std::vector<double> jacobian2_serial(num_obs * tp.size(), 0);

MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand All @@ -514,6 +529,7 @@ TEST_CASE("AdjointJacobianGPU::AdjointJacobianGPU Test HermitianObs",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down
2 changes: 2 additions & 0 deletions .../core/src/simulators/lightning_gpu/gates/tests/mpi/Test_StateVectorCudaMPI_Generators.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,7 @@ using namespace Pennylane::LightningGPU::MPI;
using cp_t = std::complex<TestType>; \
using PrecisionT = TestType; \
MPIManager mpi_manager(MPI_COMM_WORLD); \
CHECK(mpi_manager.getSize() == 2); \
size_t mpi_buffersize = 1; \
size_t nGlobalIndexBits = \
std::bit_width(static_cast<size_t>(mpi_manager.getSize())) - 1; \
Expand All @@ -82,6 +83,7 @@ using namespace Pennylane::LightningGPU::MPI;
mpi_manager.Barrier(); \
int nDevices = 0; \
cudaGetDeviceCount(&nDevices); \
CHECK(nDevices >= 2); \
int deviceId = mpi_manager.getRank() % nDevices; \
cudaSetDevice(deviceId); \
DevTag<int> dt_local(deviceId, 0); \
Expand Down
6 changes: 6 additions & 0 deletions ...ng/core/src/simulators/lightning_gpu/gates/tests/mpi/Test_StateVectorCudaMPI_NonParam.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetStateVector",
using PrecisionT = TestType;
using cp_t = std::complex<PrecisionT>;
MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand Down Expand Up @@ -132,6 +133,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetStateVector",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand Down Expand Up @@ -165,6 +167,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetIthStates",
using PrecisionT = TestType;
using cp_t = std::complex<PrecisionT>;
MPIManager mpi_manager(MPI_COMM_WORLD);
CHECK(mpi_manager.getSize() == 2);

size_t mpi_buffersize = 1;

Expand Down Expand Up @@ -193,6 +196,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetIthStates",

int nDevices = 0; // Number of GPU devices per node
cudaGetDeviceCount(&nDevices);
CHECK(nDevices >= 2);
int deviceId = mpi_manager.getRank() % nDevices;
cudaSetDevice(deviceId);
DevTag<int> dt_local(deviceId, 0);
Expand All @@ -219,6 +223,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetIthStates",
using cp_t = std::complex<TestType>; \
using PrecisionT = TestType; \
MPIManager mpi_manager(MPI_COMM_WORLD); \
CHECK(mpi_manager.getSize() == 2); \
size_t mpi_buffersize = 1; \
size_t nGlobalIndexBits = \
std::bit_width(static_cast<size_t>(mpi_manager.getSize())) - 1; \
Expand All @@ -237,6 +242,7 @@ TEMPLATE_TEST_CASE("StateVectorCudaMPI::SetIthStates",
mpi_manager.Barrier(); \
int nDevices = 0; \
cudaGetDeviceCount(&nDevices); \
CHECK(nDevices >= 2); \
int deviceId = mpi_manager.getRank() % nDevices; \
cudaSetDevice(deviceId); \
DevTag<int> dt_local(deviceId, 0); \
Expand Down
2 changes: 2 additions & 0 deletions ...tning/core/src/simulators/lightning_gpu/gates/tests/mpi/Test_StateVectorCudaMPI_Param.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,7 @@ using namespace Pennylane::LightningGPU::MPI;
using cp_t = std::complex<TestType>; \
using PrecisionT = TestType; \
MPIManager mpi_manager(MPI_COMM_WORLD); \
CHECK(mpi_manager.getSize() == 2); \
size_t mpi_buffersize = 1; \
size_t nGlobalIndexBits = \
std::bit_width(static_cast<size_t>(mpi_manager.getSize())) - 1; \
Expand All @@ -81,6 +82,7 @@ using namespace Pennylane::LightningGPU::MPI;
mpi_manager.Barrier(); \
int nDevices = 0; \
cudaGetDeviceCount(&nDevices); \
CHECK(nDevices >= 2); \
int deviceId = mpi_manager.getRank() % nDevices; \
cudaSetDevice(deviceId); \
DevTag<int> dt_local(deviceId, 0); \
Expand Down
Loading

0 comments on commit d3af819

Please sign in to comment.