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

Enable use of IBM XL compilers with openMP Offload support #605

Merged
merged 2 commits into from
Apr 7, 2022
+147 −46
Merged

Enable use of IBM XL compilers with openMP Offload support #605

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3141ea1
Enable use of IBM XL compilers with openMP Offload support
oseikuffuor1 Apr 6, 2022
4bea902
Added build script for Lassen
oseikuffuor1 Apr 6, 2022
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
37 changes: 37 additions & 0 deletions scripts/build_lassen_xl_offload.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
#!/bin/bash

# Make sure all the paths are correct

source setenv_lassen_offload.sh

rm -r build
rm -r install

MY_PATH=$(pwd)

export CC=${CC:=xlc-gpu}
export FC=${FC:=xlf2003-gpu}
export CXX=${CXX:=xlc++-gpu}
export BLAS_VENDOR=${BLAS_VENDOR:=Auto}
export BML_OPENMP=${BML_OPENMP:=yes}
export BML_OMP_OFFLOAD=${BML_OMP_OFFLOAD:=yes}
export BML_CUSPARSE=${BML_CUSPARSE:=yes}
export BML_COMPLEX=${BML_COMPLEX:=no}
export INSTALL_DIR=${INSTALL_DIR:="${MY_PATH}/install"}
export BML_TESTING=${BML_TESTING:=yes}
export CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:=Release}
export EXTRA_CFLAGS=${EXTRA_CFLAGS:=""}
export EXTRA_LINK_FLAGS=${EXTRA_LINK_FLAGS:=""}
#export EXTRA_LINK_FLAGS=${EXTRA_LINK_FLAGS:="-lm -L/usr/tce/packages/xl/xl-2021.03.11/xlC/16.1.1/lib -libmc++"}
export BLAS_LIBRARIES=${BLAS_LIBRARIES:="-L${LAPACK_DIR} -llapack -lblas"}

./build.sh configure

pushd build
make -j
make install
popd




10 changes: 10 additions & 0 deletions scripts/setenv_lassen_offload.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
#!/bin/bash

#module purge
module load cmake
module load xl/2021.03.11-cuda-11.2.0
module load cuda/11.2.0
module load lapack/3.9.0-xl-2020.11.12
#module load essl
export CUDA_TOOLKIT_ROOT_DIR=${CUDA_TOOLKIT_ROOT_DIR="/usr/tce/packages/cuda/cuda-11.2.0"}

23 changes: 21 additions & 2 deletions src/C-interface/ellblock/bml_add_ellblock_typed.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,6 @@ void TYPED_FUNC(

int NB = A->NB;
int MB = A->MB;
int ix[NB], jx[NB];

int *A_nnzb = A->nnzb;
int *A_indexb = A->indexb;
Expand All @@ -55,12 +54,16 @@ void TYPED_FUNC(

int *bsize = A->bsize;

REAL_T *x_ptr[NB];
REAL_T **A_ptr_value = (REAL_T **) A->ptr_value;
REAL_T **B_ptr_value = (REAL_T **) B->ptr_value;

#if !(defined(__IBMC__) || defined(__ibmxl__))
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
memset(jx, 0, NB * sizeof(int));
#endif

int maxbsize = 0;
for (int ib = 0; ib < NB; ib++)
Expand All @@ -74,13 +77,29 @@ void TYPED_FUNC(
REAL_T *x_ptr_storage = calloc(maxbsize2 * NB * nthreads, sizeof(REAL_T));

char xptrset = 0;
#if defined(__IBMC__) || defined(__ibmxl__)
#pragma omp parallel for \
shared(A_indexb, A_ptr_value, A_nnzb) \
shared(B_indexb, B_ptr_value, B_nnzb) \
shared(x_ptr_storage) \
firstprivate(xptrset)
#else
#pragma omp parallel for \
shared(A_indexb, A_ptr_value, A_nnzb) \
shared(B_indexb, B_ptr_value, B_nnzb) \
shared(x_ptr_storage) \
firstprivate(ix, jx, x_ptr, xptrset)
#endif
for (int ib = 0; ib < NB; ib++)
{

#if defined(__IBMC__) || defined(__ibmxl__)
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
#endif

if (!xptrset)
{
#ifdef _OPENMP
Expand Down
45 changes: 37 additions & 8 deletions src/C-interface/ellblock/bml_multiply_ellblock_typed.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -255,18 +255,20 @@ void *TYPED_FUNC(
int *X2_indexb = X2->indexb;
int *X2_nnzb = X2->nnzb;

int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

REAL_T traceX = 0.0;
REAL_T traceX2 = 0.0;
REAL_T **X_ptr_value = (REAL_T **) X->ptr_value;
REAL_T **X2_ptr_value = (REAL_T **) X2->ptr_value;

double *trace = bml_allocate_memory(sizeof(double) * 2);

#if !(defined(__IBMC__) || defined(__ibmxl__))
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
memset(jx, 0, NB * sizeof(int));
#endif

int maxbsize = 0;
for (int ib = 0; ib < NB; ib++)
Expand Down Expand Up @@ -297,14 +299,26 @@ void *TYPED_FUNC(
TYPED_FUNC(bml_multiply_block4),
TYPED_FUNC(bml_multiply_block5), TYPED_FUNC(bml_multiply_block6)};


#if defined(__IBMC__) || defined(__ibmxl__)
#pragma omp parallel for \
firstprivate(xptrset) \
reduction(+: traceX, traceX2)
#else
#pragma omp parallel for \
firstprivate(ix,jx, x_ptr, xptrset) \
reduction(+: traceX, traceX2)

#endif
//loop over row blocks
for (int ib = 0; ib < NB; ib++)
{

#if defined(__IBMC__) || defined(__ibmxl__)
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
#endif

int lb = 0;
if (!xptrset)
{
Expand Down Expand Up @@ -456,15 +470,17 @@ void TYPED_FUNC(
int *C_nnzb = C->nnzb;
int *C_indexb = C->indexb;

int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

REAL_T **A_ptr_value = (REAL_T **) A->ptr_value;
REAL_T **B_ptr_value = (REAL_T **) B->ptr_value;
REAL_T **C_ptr_value = (REAL_T **) C->ptr_value;

#if !(defined(__IBMC__) || defined(__ibmxl__))
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
memset(jx, 0, NB * sizeof(int));
#endif

int maxbsize = 0;
for (int ib = 0; ib < NB; ib++)
Expand Down Expand Up @@ -495,11 +511,24 @@ void TYPED_FUNC(
TYPED_FUNC(bml_multiply_block5), TYPED_FUNC(bml_multiply_block6)};

//loop over row blocks
#if defined(__IBMC__) || defined(__ibmxl__)
#pragma omp parallel for \
firstprivate( xptrset)
#else
#pragma omp parallel for \
firstprivate(ix, jx, x_ptr, xptrset)
#endif

for (int ib = 0; ib < NB; ib++)
{

#if defined(__IBMC__) || defined(__ibmxl__)
int ix[NB], jx[NB];
REAL_T *x_ptr[NB];

memset(ix, 0, NB * sizeof(int));
#endif

int lb = 0;
if (!xptrset)
{
Expand Down
37 changes: 19 additions & 18 deletions src/C-interface/ellpack/bml_add_ellpack_typed.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,7 +66,7 @@ void TYPED_FUNC(
memset(x, 0.0, N * sizeof(REAL_T));
#endif

#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
int num_chunks = MIN(OFFLOAD_NUM_CHUNKS, rowMax - rowMin + 1);

int all_ix[N * num_chunks], all_jx[N * num_chunks];
Expand All @@ -81,7 +81,7 @@ void TYPED_FUNC(
#endif

#if defined (USE_OMP_OFFLOAD)
#if defined(INTEL_SDK) || defined(CRAY_SDK)
#if defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__)
#pragma omp teams distribute parallel for \
shared(rowMin, rowMax) \
shared(A_index, A_value, A_nnz) \
Expand Down Expand Up @@ -116,19 +116,20 @@ void TYPED_FUNC(
firstprivate(ix, jx, x)
#endif
#endif
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
for (int i = rowMin + chunk; i < rowMax; i = i + num_chunks)
{
#else
for (int i = rowMin; i < rowMax; i++)
#endif
{

#if defined(__IBMC__) || defined(__ibmxl__)
int ix[N], jx[N];
REAL_T x[N];

memset(ix, 0, N * sizeof(int));
#endif

#endif
int l = 0;
if (alpha > (double) 0.0 || alpha < (double) 0.0)
for (int jp = 0; jp < A_nnz[i]; jp++)
Expand Down Expand Up @@ -175,7 +176,7 @@ void TYPED_FUNC(
}
A_nnz[i] = ll;
}
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
}
#endif
}
Expand Down Expand Up @@ -234,7 +235,7 @@ double TYPED_FUNC(
memset(y, 0.0, N * sizeof(REAL_T));
#endif

#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
int num_chunks = MIN(OFFLOAD_NUM_CHUNKS, rowMax - rowMin + 1);

int all_ix[N * num_chunks], all_jx[N * num_chunks];
Expand All @@ -250,7 +251,7 @@ double TYPED_FUNC(
#endif

#if defined (USE_OMP_OFFLOAD)
#if defined(INTEL_SDK) || defined(CRAY_SDK)
#if defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__)
#pragma omp teams distribute parallel for \
shared(rowMin, rowMax) \
shared(A_index, A_value, A_nnz) \
Expand Down Expand Up @@ -291,11 +292,11 @@ double TYPED_FUNC(
reduction(+:trnorm)
#endif
#endif
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
for (int i = rowMin + chunk; i < rowMax; i = i + num_chunks)
{
#else
for (int i = rowMin; i < rowMax; i++)
#endif
{

#if defined(__IBMC__) || defined(__ibmxl__)
Expand All @@ -305,7 +306,7 @@ double TYPED_FUNC(

memset(ix, 0, N * sizeof(int));
#endif

#endif
int l = 0;
for (int jp = 0; jp < A_nnz[i]; jp++)
{
Expand Down Expand Up @@ -359,7 +360,7 @@ double TYPED_FUNC(
}
A_nnz[i] = ll;
}
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
}
#endif

Expand Down Expand Up @@ -397,7 +398,7 @@ void TYPED_FUNC(
memset(x, 0.0, A_M * sizeof(REAL_T));
#endif

#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
int num_chunks = MIN(OFFLOAD_NUM_CHUNKS, N);

int all_jx[N * num_chunks];
Expand All @@ -411,7 +412,7 @@ void TYPED_FUNC(
#endif

#if defined (USE_OMP_OFFLOAD)
#if defined(INTEL_SDK) || defined(CRAY_SDK)
#if defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__)
#pragma omp teams distribute parallel for \
shared(N, A_M) \
shared(A_index, A_value, A_nnz)
Expand Down Expand Up @@ -441,18 +442,18 @@ void TYPED_FUNC(
firstprivate(jx, x)
#endif
#endif
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
for (int i = chunk; i < N; i = i + num_chunks)
{
#else
for (int i = 0; i < N; i++)
#endif
{

#if defined(__IBMC__) || defined(__ibmxl__)
int jx[A_M];
REAL_T x[A_M];
#endif

#endif
int l = 0;
int diag = -1;

Expand Down Expand Up @@ -495,7 +496,7 @@ void TYPED_FUNC(
}
A_nnz[i] = ll;
}
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK))
#if defined(USE_OMP_OFFLOAD) && (defined(INTEL_SDK) || defined(CRAY_SDK) || defined(__IBMC__) || defined(__ibmxl__))
}
#endif
}
Expand Down
Loading