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

Added cusparse capability for bml_transpose_ellpack. #674

Merged
merged 2 commits into from
Nov 30, 2022
Merged

Added cusparse capability for bml_transpose_ellpack. #674

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
1d4e45a
Added cusparse capability for bml_transpose_ellpack.
oseikuffuor1 Nov 16, 2022
8f50dda
Removed commented out code.
oseikuffuor1 Nov 17, 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
16 changes: 16 additions & 0 deletions src/C-interface/ellpack/bml_transpose_ellpack.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,4 +33,20 @@ void bml_transpose_ellpack_single_complex(
void bml_transpose_ellpack_double_complex(
bml_matrix_ellpack_t * A);

#if defined(BML_USE_CUSPARSE)
void bml_transpose_cusparse_ellpack(
bml_matrix_ellpack_t * A);

void bml_transpose_cusparse_ellpack_single_real(
bml_matrix_ellpack_t * A);

void bml_transpose_cusparse_ellpack_double_real(
bml_matrix_ellpack_t * A);

void bml_transpose_cusparse_ellpack_single_complex(
bml_matrix_ellpack_t * A);

void bml_transpose_cusparse_ellpack_double_complex(
bml_matrix_ellpack_t * A);
#endif
#endif
111 changes: 111 additions & 0 deletions src/C-interface/ellpack/bml_transpose_ellpack_typed.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,11 @@
#include <omp.h>
#endif

#ifdef BML_USE_CUSPARSE
#include <cusparse.h>
#include "bml_copy_ellpack.h"
#endif

#define COMPUTE_ON_HOST

/** Transpose a matrix.
Expand Down Expand Up @@ -50,6 +55,10 @@ bml_matrix_ellpack_t

int myRank = bml_getMyRank();

#if defined(BML_USE_CUSPARSE)
TYPED_FUNC(bml_copy_ellpack) (A, B);
TYPED_FUNC(bml_transpose_cusparse_ellpack) (B);
#else
// Transpose all elements
#ifdef _OPENMP
omp_lock_t *row_lock = (omp_lock_t *) malloc(sizeof(omp_lock_t) * N);
Expand Down Expand Up @@ -94,6 +103,7 @@ bml_matrix_ellpack_t

#if defined (USE_OMP_OFFLOAD) && defined(COMPUTE_ON_HOST)
#pragma omp target update to(B_index[:N*M], B_value[:N*M], B_nnz[:N])
#endif
#endif
return B;
/*
Expand Down Expand Up @@ -146,6 +156,9 @@ void TYPED_FUNC(
int *A_index = A->index;
int *A_nnz = A->nnz;

#if defined(BML_USE_CUSPARSE)
TYPED_FUNC(bml_transpose_cusparse_ellpack) (A);
#else
#if defined(USE_OMP_OFFLOAD)
#ifdef COMPUTE_ON_HOST
#pragma omp target update from(A_index[:N*M], A_value[:N*M], A_nnz[:N])
Expand Down Expand Up @@ -204,5 +217,103 @@ void TYPED_FUNC(
#pragma omp target update to(A_index[:N*M], A_value[:N*M], A_nnz[:N])
#endif
#endif
#endif
}

#if defined(BML_USE_CUSPARSE)
/** cuSPARSE matrix transpose
*
* \ingroup transpose_group
*
* \param A The matrix to be transposed
* \return the transposed A
*/
void TYPED_FUNC(
bml_transpose_cusparse_ellpack) (
bml_matrix_ellpack_t * A)
{
int N = A->N;
int M = A->M;

REAL_T *A_value = (REAL_T *) A->value;
int *csrColIndA = A->csrColInd;
int *csrRowPtrA = A->csrRowPtr;
REAL_T *csrValA = (REAL_T *) A->csrVal;

/* temporary arrays to hold compressed sparse column (CSC) values */
int *cscRowInd = NULL;
int *cscColPtr = NULL;
REAL_T *cscVal = NULL;

cusparseStatus_t status = CUSPARSE_STATUS_SUCCESS;
cudaDataType valType = BML_CUSPARSE_T;

// CUSPARSE APIs
cusparseHandle_t handle = NULL;
void *dBuffer = NULL;
size_t bufferSize = 0;

// convert ellpack to cucsr
TYPED_FUNC(bml_ellpack2cucsr_ellpack) (A);

// Create cusparse matrix A and B in CSR format
// Note: The following update is not necessary since the ellpack2cucsr
// routine updates the csr rowpointers on host and device
#pragma omp target update from(csrRowPtrA[:N+1])
int nnzA = csrRowPtrA[N];

// Allocate memory for result arrays
cscVal =
(REAL_T *) omp_target_alloc(sizeof(REAL_T) * nnzA,
omp_get_default_device());
cscRowInd =
(int *) omp_target_alloc(sizeof(int) * nnzA,
omp_get_default_device());
cscColPtr =
(int *) omp_target_alloc(sizeof(int) * (N + 1),
omp_get_default_device());

BML_CHECK_CUSPARSE(cusparseCreate(&handle));
#pragma omp target data use_device_ptr(csrRowPtrA,csrColIndA,csrValA)
{
// Get storage buffer size
BML_CHECK_CUSPARSE(cusparseCsr2cscEx2_bufferSize
(handle, N, N, nnzA, csrValA, csrRowPtrA,
csrColIndA, cscVal, cscColPtr, cscRowInd, valType,
CUSPARSE_ACTION_NUMERIC, CUSPARSE_INDEX_BASE_ZERO,
CUSPARSE_CSR2CSC_ALG1, &bufferSize));

// Allocate buffer and perform transpose operation
dBuffer =
(char *) omp_target_alloc(bufferSize, omp_get_default_device());

BML_CHECK_CUSPARSE(cusparseCsr2cscEx2
(handle, N, N, nnzA, csrValA, csrRowPtrA,
csrColIndA, cscVal, cscColPtr, cscRowInd, valType,
CUSPARSE_ACTION_NUMERIC, CUSPARSE_INDEX_BASE_ZERO,
CUSPARSE_CSR2CSC_ALG1, dBuffer));

/* Update matA with new result. Note that only device arrays are updated. */
omp_target_memcpy(csrRowPtrA, cscColPtr, (N + 1) * sizeof(int),
0, 0, omp_get_default_device(),
omp_get_default_device());
omp_target_memcpy(csrColIndA, cscRowInd, nnzA * sizeof(int), 0,
0, omp_get_default_device(),
omp_get_default_device());
omp_target_memcpy(csrValA, cscVal, nnzA * sizeof(REAL_T), 0, 0,
omp_get_default_device(), omp_get_default_device());

// deallocate storage buffer
omp_target_free(dBuffer, omp_get_default_device());
}

// Done with matrix transpose operation.
// Update ellpack matrix (on device): copy from csr to ellpack format
TYPED_FUNC(bml_cucsr2ellpack_ellpack) (A);

// device memory deallocation for temp arrays
omp_target_free(cscVal, omp_get_default_device());
omp_target_free(cscRowInd, omp_get_default_device());
omp_target_free(cscColPtr, omp_get_default_device());
}
#endif