single precision
[Level-3 BLAS]

Functions

magma_int_t magma_strsm_m (magma_int_t nrgpu, magma_side_t side, magma_uplo_t uplo, magma_trans_t transa, magma_diag_t diag, magma_int_t m, magma_int_t n, float alpha, float *A, magma_int_t lda, float *B, magma_int_t ldb)
 STRSM solves one of the matrix equations op( A )*X = alpha*B, or X*op( A ) = alpha*B, where alpha is a scalar, X and B are m by n matrices, A is a unit, or non-unit, upper or lower triangular matrix and op( A ) is one of.
void magma_sgemm (magma_trans_t transA, magma_trans_t transB, magma_int_t m, magma_int_t n, magma_int_t k, float alpha, const float *dA, magma_int_t ldda, const float *dB, magma_int_t lddb, float beta, float *dC, magma_int_t lddc)
 Perform matrix-matrix product, $ C = \alpha op(A) op(B) + \beta C $.
void magma_ssymm (magma_side_t side, magma_uplo_t uplo, magma_int_t m, magma_int_t n, float alpha, const float *dA, magma_int_t ldda, const float *dB, magma_int_t lddb, float beta, float *dC, magma_int_t lddc)
 Perform symmetric matrix-matrix product.
void magma_ssyrk (magma_uplo_t uplo, magma_trans_t trans, magma_int_t n, magma_int_t k, float alpha, const float *dA, magma_int_t ldda, float beta, float *dC, magma_int_t lddc)
 Perform symmetric rank-k update.
void magma_ssyr2k (magma_uplo_t uplo, magma_trans_t trans, magma_int_t n, magma_int_t k, float alpha, const float *dA, magma_int_t ldda, const float *dB, magma_int_t lddb, float beta, float *dC, magma_int_t lddc)
 Perform symmetric rank-2k update.
void magma_strmm (magma_side_t side, magma_uplo_t uplo, magma_trans_t trans, magma_diag_t diag, magma_int_t m, magma_int_t n, float alpha, const float *dA, magma_int_t ldda, float *dB, magma_int_t lddb)
 Perform triangular matrix-matrix product.
void magma_strsm (magma_side_t side, magma_uplo_t uplo, magma_trans_t trans, magma_diag_t diag, magma_int_t m, magma_int_t n, float alpha, const float *dA, magma_int_t ldda, float *dB, magma_int_t lddb)
 Solve triangular matrix-matrix system (multiple right-hand sides).
void magmablas_sgemm (magma_trans_t TRANSA, magma_trans_t TRANSB, magma_int_t m, magma_int_t n, magma_int_t k, float alpha, const float *d_A, magma_int_t lda, const float *d_B, magma_int_t ldb, float beta, float *d_C, magma_int_t ldc)
 SGEMM performs one of the matrix-matrix operations.
void magmablas_sgemm_reduce (magma_int_t m, magma_int_t n, magma_int_t k, float alpha, const float *d_A, magma_int_t lda, const float *d_B, magma_int_t ldb, float beta, float *d_C, magma_int_t ldc)
 SGEMM_REDUCE performs one of the matrix-matrix operations.
void magmablas_sgemm_tesla (magma_trans_t transA, magma_trans_t transB, magma_int_t m, magma_int_t n, magma_int_t k, float alpha, const float *A, magma_int_t lda, const float *B, magma_int_t ldb, float beta, float *C, magma_int_t ldc)
 SGEMM performs one of the matrix-matrix operations.
__global__ void sgemm_kernel_N_N_64_16_16_16_4 (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A*B + beta * C.
__global__ void sgemm_kernel_N_N_64_16_16_16_4_special (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A*B + beta * C.
__global__ void sgemm_kernel_N_T_64_16_4_16_4 (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A*B^T + beta * C.
__global__ void sgemm_kernel_T_N_32_32_8_8_8 (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A^T*B + beta * C.
__global__ void sgemm_kernel_T_T_64_16_16_16_4 (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A^T*B^T + beta * C.
__global__ void sgemm_kernel_T_T_64_16_16_16_4_special (float *__restrict__ C, const float *__restrict__ A, const float *__restrict__ B, int m, int n, int k, int lda, int ldb, int ldc, float alpha, float beta)
 Purpose: -------- This routine computes C = alpha * A^T*B^T + beta * C.
void magmablas_ssyr2k_mgpu2 (magma_uplo_t uplo, magma_trans_t trans, magma_int_t n, magma_int_t k, float alpha, float *dA[], magma_int_t lda, magma_int_t aoffset, float *dB[], magma_int_t ldb, magma_int_t boffset, float beta, float *dC[], magma_int_t ldc, magma_int_t coffset, magma_int_t ngpu, magma_int_t nb, magma_queue_t streams[][20], magma_int_t nstream)
 SSYR2K performs one of the symmetric rank 2k operations.
void magmablas_ssyr2k_mgpu_spec (magma_uplo_t uplo, magma_trans_t trans, magma_int_t n, magma_int_t k, float alpha, float *dA[], magma_int_t lda, magma_int_t aoffset, float *dB[], magma_int_t ldb, magma_int_t boffset, float beta, float *dC[], magma_int_t ldc, magma_int_t coffset, magma_int_t ngpu, magma_int_t nb, magma_queue_t streams[][20], magma_int_t nstream)
 SSYR2K performs one of the symmetric rank 2k operations.
void magmablas_strsm_work (magma_side_t side, magma_uplo_t uplo, magma_trans_t transA, magma_diag_t diag, magma_int_t m, magma_int_t n, float alpha, const float *dA, magma_int_t ldda, float *dB, magma_int_t lddb, magma_int_t flag, float *d_dinvA, float *dX)
 strsm_work solves one of the matrix equations on gpu
void magmablas_strsm (magma_side_t side, magma_uplo_t uplo, magma_trans_t transA, magma_diag_t diag, magma_int_t m, magma_int_t n, float alpha, const float *dA, magma_int_t ldda, float *dB, magma_int_t lddb)
void magmablas_strtri_diag_q (magma_uplo_t uplo, magma_diag_t diag, magma_int_t n, const float *dA, magma_int_t ldda, float *d_dinvA, magma_queue_t queue)
 Inverts the NB x NB diagonal blocks of a triangular matrix.
void magmablas_strtri_diag (magma_uplo_t uplo, magma_diag_t diag, magma_int_t n, const float *dA, magma_int_t ldda, float *d_dinvA)

Function Documentation

void magma_sgemm ( magma_trans_t  transA,
magma_trans_t  transB,
magma_int_t  m,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  dA,
magma_int_t  ldda,
const float *  dB,
magma_int_t  lddb,
float  beta,
float *  dC,
magma_int_t  lddc 
)

Perform matrix-matrix product, $ C = \alpha op(A) op(B) + \beta C $.

Parameters:
[in] transA Operation op(A) to perform on matrix A.
[in] transB Operation op(B) to perform on matrix B.
[in] m Number of rows of C and op(A). m >= 0.
[in] n Number of columns of C and op(B). n >= 0.
[in] k Number of columns of op(A) and rows of op(B). k >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If transA == MagmaNoTrans, the m-by-k matrix A of dimension (ldda,k), ldda >= max(1,m);
otherwise, the k-by-m matrix A of dimension (ldda,m), ldda >= max(1,k).
[in] ldda Leading dimension of dA.
[in] dB REAL array on GPU device. If transB == MagmaNoTrans, the k-by-n matrix B of dimension (lddb,n), lddb >= max(1,k);
otherwise, the n-by-k matrix B of dimension (lddb,k), lddb >= max(1,n).
[in] lddb Leading dimension of dB.
[in] beta Scalar $ \beta $
[in,out] dC REAL array on GPU device. The m-by-n matrix C of dimension (lddc,n), lddc >= max(1,m).
[in] lddc Leading dimension of dC.
void magma_ssymm ( magma_side_t  side,
magma_uplo_t  uplo,
magma_int_t  m,
magma_int_t  n,
float  alpha,
const float *  dA,
magma_int_t  ldda,
const float *  dB,
magma_int_t  lddb,
float  beta,
float *  dC,
magma_int_t  lddc 
)

Perform symmetric matrix-matrix product.

$ C = \alpha A B + \beta C $ (side == MagmaLeft), or
$ C = \alpha B A + \beta C $ (side == MagmaRight),
where $ A $ is symmetric.

Parameters:
[in] side Whether A is on the left or right.
[in] uplo Whether the upper or lower triangle of A is referenced.
[in] m Number of rows of C. m >= 0.
[in] n Number of columns of C. n >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If side == MagmaLeft, the m-by-m symmetric matrix A of dimension (ldda,m), ldda >= max(1,m);
otherwise, the n-by-n symmetric matrix A of dimension (ldda,n), ldda >= max(1,n).
[in] ldda Leading dimension of dA.
[in] dB REAL array on GPU device. The m-by-n matrix B of dimension (lddb,n), lddb >= max(1,m).
[in] lddb Leading dimension of dB.
[in] beta Scalar $ \beta $
[in,out] dC REAL array on GPU device. The m-by-n matrix C of dimension (lddc,n), lddc >= max(1,m).
[in] lddc Leading dimension of dC.
void magma_ssyr2k ( magma_uplo_t  uplo,
magma_trans_t  trans,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  dA,
magma_int_t  ldda,
const float *  dB,
magma_int_t  lddb,
float  beta,
float *  dC,
magma_int_t  lddc 
)

Perform symmetric rank-2k update.

$ C = \alpha A B^T + \alpha B A^T \beta C $ (trans == MagmaNoTrans), or
$ C = \alpha A^T B + \alpha B^T A \beta C $ (trans == MagmaTrans),
where $ C $ is symmetric.

Parameters:
[in] uplo Whether the upper or lower triangle of C is referenced.
[in] trans Operation to perform on A and B.
[in] n Number of rows and columns of C. n >= 0.
[in] k Number of columns of A and B (for MagmaNoTrans) or rows of A and B (for MagmaTrans). k >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If trans == MagmaNoTrans, the n-by-k matrix A of dimension (ldda,k), ldda >= max(1,n);
otherwise, the k-by-n matrix A of dimension (ldda,n), ldda >= max(1,k).
[in] ldda Leading dimension of dA.
[in] dB REAL array on GPU device. If trans == MagmaNoTrans, the n-by-k matrix B of dimension (lddb,k), lddb >= max(1,n);
otherwise, the k-by-n matrix B of dimension (lddb,n), lddb >= max(1,k).
[in] lddb Leading dimension of dB.
[in] beta Scalar $ \beta $
[in,out] dC REAL array on GPU device. The n-by-n symmetric matrix C of dimension (lddc,n), lddc >= max(1,n).
[in] lddc Leading dimension of dC.
void magma_ssyrk ( magma_uplo_t  uplo,
magma_trans_t  trans,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  dA,
magma_int_t  ldda,
float  beta,
float *  dC,
magma_int_t  lddc 
)

Perform symmetric rank-k update.

$ C = \alpha A A^T + \beta C $ (trans == MagmaNoTrans), or
$ C = \alpha A^T A + \beta C $ (trans == MagmaTrans),
where $ C $ is symmetric.

Parameters:
[in] uplo Whether the upper or lower triangle of C is referenced.
[in] trans Operation to perform on A.
[in] n Number of rows and columns of C. n >= 0.
[in] k Number of columns of A (for MagmaNoTrans) or rows of A (for MagmaTrans). k >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If trans == MagmaNoTrans, the n-by-k matrix A of dimension (ldda,k), ldda >= max(1,n);
otherwise, the k-by-n matrix A of dimension (ldda,n), ldda >= max(1,k).
[in] ldda Leading dimension of dA.
[in] beta Scalar $ \beta $
[in,out] dC REAL array on GPU device. The n-by-n symmetric matrix C of dimension (lddc,n), lddc >= max(1,n).
[in] lddc Leading dimension of dC.
void magma_strmm ( magma_side_t  side,
magma_uplo_t  uplo,
magma_trans_t  trans,
magma_diag_t  diag,
magma_int_t  m,
magma_int_t  n,
float  alpha,
const float *  dA,
magma_int_t  ldda,
float *  dB,
magma_int_t  lddb 
)

Perform triangular matrix-matrix product.

$ B = \alpha op(A) B $ (side == MagmaLeft), or
$ B = \alpha B op(A) $ (side == MagmaRight),
where $ A $ is triangular.

Parameters:
[in] side Whether A is on the left or right.
[in] uplo Whether A is upper or lower triangular.
[in] trans Operation to perform on A.
[in] diag Whether the diagonal of A is assumed to be unit or non-unit.
[in] m Number of rows of B. m >= 0.
[in] n Number of columns of B. n >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If side == MagmaLeft, the n-by-n triangular matrix A of dimension (ldda,n), ldda >= max(1,n);
otherwise, the m-by-m triangular matrix A of dimension (ldda,m), ldda >= max(1,m).
[in] ldda Leading dimension of dA.
[in] dB REAL array on GPU device. The m-by-n matrix B of dimension (lddb,n), lddb >= max(1,m).
[in] lddb Leading dimension of dB.
void magma_strsm ( magma_side_t  side,
magma_uplo_t  uplo,
magma_trans_t  trans,
magma_diag_t  diag,
magma_int_t  m,
magma_int_t  n,
float  alpha,
const float *  dA,
magma_int_t  ldda,
float *  dB,
magma_int_t  lddb 
)

Solve triangular matrix-matrix system (multiple right-hand sides).

$ op(A) X = \alpha B $ (side == MagmaLeft), or
$ X op(A) = \alpha B $ (side == MagmaRight),
where $ A $ is triangular.

Parameters:
[in] side Whether A is on the left or right.
[in] uplo Whether A is upper or lower triangular.
[in] trans Operation to perform on A.
[in] diag Whether the diagonal of A is assumed to be unit or non-unit.
[in] m Number of rows of B. m >= 0.
[in] n Number of columns of B. n >= 0.
[in] alpha Scalar $ \alpha $
[in] dA REAL array on GPU device. If side == MagmaLeft, the m-by-m triangular matrix A of dimension (ldda,m), ldda >= max(1,m);
otherwise, the n-by-n triangular matrix A of dimension (ldda,n), ldda >= max(1,n).
[in] ldda Leading dimension of dA.
[in,out] dB REAL array on GPU device. On entry, m-by-n matrix B of dimension (lddb,n), lddb >= max(1,m). On exit, overwritten with the solution matrix X.
[in] lddb Leading dimension of dB.
magma_int_t magma_strsm_m ( magma_int_t  nrgpu,
magma_side_t  side,
magma_uplo_t  uplo,
magma_trans_t  transa,
magma_diag_t  diag,
magma_int_t  m,
magma_int_t  n,
float  alpha,
float *  A,
magma_int_t  lda,
float *  B,
magma_int_t  ldb 
)

STRSM solves one of the matrix equations op( A )*X = alpha*B, or X*op( A ) = alpha*B, where alpha is a scalar, X and B are m by n matrices, A is a unit, or non-unit, upper or lower triangular matrix and op( A ) is one of.

op( A ) = A or op( A ) = A**T or op( A ) = A**H.

The matrix X is overwritten on B.

Parameters:
[in] nrgpu INTEGER Number of GPUs to use.
[in] side magma_side_t. On entry, SIDE specifies whether op( A ) appears on the left or right of X as follows:

  • = MagmaLeft: op( A )*X = alpha*B.
  • = MagmaRight: X*op( A ) = alpha*B.
[in] uplo magma_uplo_t. On entry, UPLO specifies whether the matrix A is an upper or lower triangular matrix as follows:

  • = MagmaUpper: A is an upper triangular matrix.
  • = MagmaLower: A is a lower triangular matrix.
[in] transa magma_trans_t. On entry, TRANSA specifies the form of op( A ) to be used in the matrix multiplication as follows:

  • = MagmaNoTrans: op( A ) = A.
  • = MagmaTrans: op( A ) = A**T.
  • = MagmaConjTrans: op( A ) = A**H.
[in] diag magma_diag_t. On entry, DIAG specifies whether or not A is unit triangular as follows:

  • = MagmaUnit: A is assumed to be unit triangular.
  • = MagmaNonUnit: A is not assumed to be unit triangular.
[in] m INTEGER. On entry, M specifies the number of rows of B. M must be at least zero.
[in] n INTEGER. On entry, N specifies the number of columns of B. N must be at least zero.
[in] alpha REAL. On entry, ALPHA specifies the scalar alpha. When alpha is zero then A is not referenced and B need not be set before entry.
[in] A REAL array of DIMENSION ( LDA, k ), where k is m when SIDE = MagmaLeft and is n when SIDE = MagmaRight. Before entry with UPLO = MagmaUpper, the leading k by k upper triangular part of the array A must contain the upper triangular matrix and the strictly lower triangular part of A is not referenced. Before entry with UPLO = MagmaLower, the leading k by k lower triangular part of the array A must contain the lower triangular matrix and the strictly upper triangular part of A is not referenced. Note that when DIAG = MagmaUnit, the diagonal elements of A are not referenced either, but are assumed to be unity.
[in] lda INTEGER. On entry, LDA specifies the first dimension of A as declared in the calling (sub) program. When SIDE = MagmaLeft then LDA >= max( 1, m ), when SIDE = MagmaRight then LDA >= max( 1, n ).
[in,out] B REAL array of DIMENSION ( LDB, n ). Before entry, the leading m by n part of the array B must contain the right-hand side matrix B, and on exit is overwritten by the solution matrix X.
[in] ldb INTEGER. On entry, LDB specifies the first dimension of B as declared in the calling (sub) program. LDB must be at least max( 1, m ).
void magmablas_sgemm ( magma_trans_t  TRANSA,
magma_trans_t  TRANSB,
magma_int_t  m,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  d_A,
magma_int_t  lda,
const float *  d_B,
magma_int_t  ldb,
float  beta,
float *  d_C,
magma_int_t  ldc 
)

SGEMM performs one of the matrix-matrix operations.

C = alpha*op( A )*op( B ) + beta*C,

where op( X ) is one of

op( X ) = X or op( X ) = X**T or op( X ) = X**H,

alpha and beta are scalars, and A, B and C are matrices, with op( A ) an m by k matrix, op( B ) a k by n matrix and C an m by n matrix.

Parameters ----------

Parameters:
[in] TRANSA CHARACTER*1. On entry, TRANSA specifies the form of op( A ) to be used in the matrix multiplication as follows:

  • = 'N': op( A ) = A.
  • = 'T': op( A ) = A**T.
  • = 'C': op( A ) = A**H.
[in] TRANSB CHARACTER*1. On entry, TRANSB specifies the form of op( B ) to be used in the matrix multiplication as follows:

  • = 'N': op( B ) = B.
  • = 'T': op( B ) = B**T.
  • = 'C': op( B ) = B**H.
[in] m INTEGER. On entry, M specifies the number of rows of the matrix op( d_A ) and of the matrix d_C. M must be at least zero.
[in] n INTEGER. On entry, N specifies the number of columns of the matrix op( d_B ) and the number of columns of the matrix d_C. N must be at least zero.
[in] k INTEGER. On entry, K specifies the number of columns of the matrix op( d_A ) and the number of rows of the matrix op( d_B ). K must be at least zero.
[in] alpha REAL On entry, ALPHA specifies the scalar alpha.
[in] d_A REAL array of DIMENSION ( LDA, ka ), where ka is k when TRANSA = MagmaNoTrans, and is m otherwise. Before entry with TRANSA = MagmaNoTrans, the leading m by k part of the array d_A must contain the matrix d_A, otherwise the leading k by m part of the array d_A must contain the matrix d_A.
[in] lda INTEGER. On entry, LDA specifies the first dimension of A as declared in the calling (sub) program. When TRANSA = MagmaNoTrans then LDA must be at least max( 1, m ), otherwise LDA must be at least max( 1, k ).
[in] d_B REAL array of DIMENSION ( LDB, kb ), where kb is n when TRANSB = MagmaNoTrans, and is k otherwise. Before entry with TRANSB = MagmaNoTrans, the leading k by n part of the array d_B must contain the matrix d_B, otherwise the leading n by k part of the array d_B must contain the matrix d_B.
[in] ldb INTEGER. On entry, LDB specifies the first dimension of d_B as declared in the calling (sub) program. When TRANSB = MagmaNoTrans then LDB must be at least max( 1, k ), otherwise LDB must be at least max( 1, n ).
[in] beta REAL. On entry, BETA specifies the scalar beta. When BETA is supplied as zero then d_C need not be set on input.
[in,out] d_C REAL array of DIMENSION ( LDC, n ). Before entry, the leading m by n part of the array d_C must contain the matrix d_C, except when beta is zero, in which case d_C need not be set on entry. On exit, the array d_C is overwritten by the m by n matrix ( alpha*op( d_A )*op( d_B ) + beta*d_C ).
[in] ldc INTEGER. On entry, LDC specifies the first dimension of d_C as declared in the calling (sub) program. LDC must be at least max( 1, m ).
void magmablas_sgemm_reduce ( magma_int_t  m,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  d_A,
magma_int_t  lda,
const float *  d_B,
magma_int_t  ldb,
float  beta,
float *  d_C,
magma_int_t  ldc 
)

SGEMM_REDUCE performs one of the matrix-matrix operations.

C := alpha*A^T*B + beta*C,

where alpha and beta are scalars, and A, B and C are matrices, with A a k-by-m matrix, B a k-by-n matrix, and C an m-by-n matrix.

This routine is tuned for m, n << k. Typically, m and n are expected to be less than 128.

void magmablas_sgemm_tesla ( magma_trans_t  transA,
magma_trans_t  transB,
magma_int_t  m,
magma_int_t  n,
magma_int_t  k,
float  alpha,
const float *  A,
magma_int_t  lda,
const float *  B,
magma_int_t  ldb,
float  beta,
float *  C,
magma_int_t  ldc 
)

SGEMM performs one of the matrix-matrix operations.

C = alpha*op( A )*op( B ) + beta*C,

where op( X ) is one of

op( X ) = X or op( X ) = X**T,

alpha and beta are scalars, and A, B and C are matrices, with op( A ) an m by k matrix, op( B ) a k by n matrix and C an m by n matrix.

Parameters ----------

Parameters:
[in] transA magma_trans_t. On entry, transA specifies the form of op( A ) to be used in the matrix multiplication as follows:

  • = MagmaNoTrans: op( A ) = A.
  • = MagmaTrans: op( A ) = A**T.
  • = MagmaConjTrans: op( A ) = A**T.
[in] transB magma_trans_t. On entry, transB specifies the form of op( B ) to be used in the matrix multiplication as follows:

  • = MagmaNoTrans: op( B ) = B.
  • = MagmaTrans: op( B ) = B**T.
  • = MagmaConjTrans: op( B ) = B**T.
[in] m INTEGER. On entry, M specifies the number of rows of the matrix op( A ) and of the matrix C. M must be at least zero.
[in] n INTEGER. On entry, N specifies the number of columns of the matrix op( B ) and the number of columns of the matrix C. N must be at least zero.
[in] k INTEGER. On entry, K specifies the number of columns of the matrix op( A ) and the number of rows of the matrix op( B ). K must be at least zero.
[in] alpha REAL. On entry, ALPHA specifies the scalar alpha.
[in] A REAL array of DIMENSION ( LDA, ka ), where ka is k when transA = MagmaNoTrans, and is m otherwise. Before entry with transA = MagmaNoTrans, the leading m by k part of the array A must contain the matrix A, otherwise the leading k by m part of the array A must contain the matrix A.
[in] lda INTEGER. On entry, LDA specifies the first dimension of A as declared in the calling (sub) program. When transA = MagmaNoTrans then LDA must be at least max( 1, m ), otherwise LDA must be at least max( 1, k ).
[in] B REAL array of DIMENSION ( LDB, kb ), where kb is n when transB = MagmaNoTrans, and is k otherwise. Before entry with transB = MagmaNoTrans, the leading k by n part of the array B must contain the matrix B, otherwise the leading n by k part of the array B must contain the matrix B.
[in] ldb INTEGER. On entry, LDB specifies the first dimension of B as declared in the calling (sub) program. When transB = MagmaNoTrans then LDB must be at least max( 1, k ), otherwise LDB must be at least max( 1, n ).
[in] beta REAL. On entry, BETA specifies the scalar beta. When BETA is supplied as zero then C need not be set on input.
[in,out] C REAL array of DIMENSION ( LDC, n ). Before entry, the leading m by n part of the array C must contain the matrix C, except when beta is zero, in which case C need not be set on entry. On exit, the array C is overwritten by the m by n matrix ( alpha*op( A )*op( B ) + beta*C ).
[in] ldc INTEGER. On entry, LDC specifies the first dimension of C as declared in the calling (sub) program. LDC must be at least max( 1, m ).
void magmablas_ssyr2k_mgpu2 ( magma_uplo_t  uplo,
magma_trans_t  trans,
magma_int_t  n,
magma_int_t  k,
float  alpha,
float *  dA[],
magma_int_t  lda,
magma_int_t  aoffset,
float *  dB[],
magma_int_t  ldb,
magma_int_t  boffset,
float  beta,
float *  dC[],
magma_int_t  ldc,
magma_int_t  coffset,
magma_int_t  ngpu,
magma_int_t  nb,
magma_queue_t  streams[][20],
magma_int_t  nstream 
)

SSYR2K performs one of the symmetric rank 2k operations.

C := alpha*A*B**H + conjg( alpha )*B*A**H + beta*C,

or

C := alpha*A**H*B + conjg( alpha )*B**H*A + beta*C,

where alpha and beta are scalars with beta real, C is an n by n symmetric matrix and A and B are n by k matrices in the first case and k by n matrices in the second case.

Parameters:
[in] uplo magma_uplo_t. On entry, UPLO specifies whether the upper or lower triangular part of the array C is to be referenced as follows:

  • = MagmaUpper: Only the upper triangular part of C is to be referenced.
  • = MagmaLower: Only the lower triangular part of C is to be referenced.

current only Lower case is implemented.

Parameters:
[in] trans magma_trans_t. On entry, TRANS specifies the operation to be performed as follows:

  • = MagmaNoTrans: C := alpha*A*B**H + conj( alpha )*B*A**H + beta*C.
  • = MagmaTrans: C := alpha*A**H*B + conj( alpha )*B**H*A + beta*C.

current only NoTrans case is implemented.

Parameters:
[in] n INTEGER. On entry, N specifies the order of the matrix C. N must be at least zero.
[in] k INTEGER. On entry with TRANS = MagmaNoTrans, K specifies the number of columns of the matrices A and B, and on entry with TRANS = MagmaTrans, K specifies the number of rows of the matrices A and B. K must be at least zero.
[in] alpha REAL. On entry, ALPHA specifies the scalar alpha.
[in] dA REAL array of DIMENSION ( LDA, ka ), where ka is k when TRANS = MagmaNoTrans, and is n otherwise. Before entry with TRANS = MagmaNoTrans, the leading n by k part of the array A must contain the matrix A, otherwise the leading k by n part of the array A must contain the matrix A.

[TODO: describe distribution: duplicated on all GPUs.]

Parameters:
[in] lda INTEGER. On entry, LDA specifies the first dimension of A as declared in the calling (sub) program. When TRANS = MagmaNoTrans then LDA must be at least max( 1, n ), otherwise LDA must be at least max( 1, k ).
[in] aoffset INTEGER Row offset to start sub-matrix of dA. Uses dA(aoffset:aoffset+n, :). 0 <= aoffset < lda.
[in] dB REAL array of DIMENSION ( LDB, kb ), where kb is k when TRANS = MagmaNoTrans, and is n otherwise. Before entry with TRANS = MagmaNoTrans, the leading n by k part of the array B must contain the matrix B, otherwise the leading k by n part of the array B must contain the matrix B.

[TODO: describe distribution: duplicated on all GPUs.]

Parameters:
[in] ldb INTEGER. On entry, LDB specifies the first dimension of B as declared in the calling (sub) program. When TRANS = MagmaNoTrans then LDB must be at least max( 1, n ), otherwise LDB must be at least max( 1, k ).
[in] boffset INTEGER Row offset to start sub-matrix of dB. Uses dB(boffset:boffset+n, :). 0 <= boffset < ldb.
[in] beta REAL. On entry, BETA specifies the scalar beta.
[in,out] dC REAL array of DIMENSION ( LDC, n ). Before entry with UPLO = MagmaUpper, the leading n by n upper triangular part of the array C must contain the upper triangular part of the symmetric matrix and the strictly lower triangular part of C is not referenced. On exit, the upper triangular part of the array C is overwritten by the upper triangular part of the updated matrix.
Before entry with UPLO = MagmaLower, the leading n by n lower triangular part of the array C must contain the lower triangular part of the symmetric matrix and the strictly upper triangular part of C is not referenced. On exit, the lower triangular part of the array C is overwritten by the lower triangular part of the updated matrix.
Note that the imaginary parts of the diagonal elements need not be set, they are assumed to be zero, and on exit they are set to zero. [TODO: verify]

[TODO: describe distribution: 1D column block-cyclic across GPUs.]

Parameters:
[in] ldc INTEGER. On entry, LDC specifies the first dimension of C as declared in the calling (sub) program. LDC must be at least max( 1, n ).
[in] coffset INTEGER. Row and column offset to start sub-matrix of dC. Uses dC(coffset:coffset+n, coffset:coffset+n). 0 <= coffset < ldc.
[in] ngpu INTEGER. Number of GPUs over which matrix C is distributed.
[in] nb INTEGER. Block size used for distribution of C.
[in] streams array of CUDA streams, of dimension NGPU by 20. Streams to use for running multiple GEMMs in parallel. Only up to NSTREAM streams are used on each GPU.
[in] nstream INTEGER. Number of streams to use on each device
void magmablas_ssyr2k_mgpu_spec ( magma_uplo_t  uplo,
magma_trans_t  trans,
magma_int_t  n,
magma_int_t  k,
float  alpha,
float *  dA[],
magma_int_t  lda,
magma_int_t  aoffset,
float *  dB[],
magma_int_t  ldb,
magma_int_t  boffset,
float  beta,
float *  dC[],
magma_int_t  ldc,
magma_int_t  coffset,
magma_int_t  ngpu,
magma_int_t  nb,
magma_queue_t  streams[][20],
magma_int_t  nstream 
)

SSYR2K performs one of the symmetric rank 2k operations.

C := alpha*A*B**H + conjg( alpha )*B*A**H + beta*C,

or

C := alpha*A**H*B + conjg( alpha )*B**H*A + beta*C,

where alpha and beta are scalars with beta real, C is an n by n symmetric matrix and A and B are n by k matrices in the first case and k by n matrices in the second case.

This version assumes C has been symmetrized, so both upper and lower are stored, and it maintains the symmetry, doing twice the operations.

Parameters:
[in] uplo magma_uplo_t. On entry, UPLO specifies whether the upper or lower triangular part of the array C is to be referenced as follows:

  • = MagmaUpper: Only the upper triangular part of C is to be referenced.
  • = MagmaLower: Only the lower triangular part of C is to be referenced.

current only Lower case is implemented.

Parameters:
[in] trans magma_trans_t. On entry, TRANS specifies the operation to be performed as follows:

  • = MagmaNoTrans: C := alpha*A*B**H + conj( alpha )*B*A**H + beta*C.
  • = MagmaTrans: C := alpha*A**H*B + conj( alpha )*B**H*A + beta*C.

current only NoTrans case is implemented.

Parameters:
[in] n INTEGER. On entry, N specifies the order of the matrix C. N must be at least zero.
[in] k INTEGER. On entry with TRANS = MagmaNoTrans, K specifies the number of columns of the matrices A and B, and on entry with TRANS = MagmaTrans, K specifies the number of rows of the matrices A and B. K must be at least zero.
[in] alpha REAL. On entry, ALPHA specifies the scalar alpha.
[in] dA REAL array of DIMENSION ( LDA, ka ), where ka is k when TRANS = MagmaNoTrans, and is n otherwise. Before entry with TRANS = MagmaNoTrans, the leading n by k part of the array A must contain the matrix A, otherwise the leading k by n part of the array A must contain the matrix A.

[TODO: describe distribution: duplicated on all GPUs.]

Parameters:
[in] lda INTEGER. On entry, LDA specifies the first dimension of A as declared in the calling (sub) program. When TRANS = MagmaNoTrans then LDA must be at least max( 1, n ), otherwise LDA must be at least max( 1, k ).
[in] aoffset INTEGER Row offset to start sub-matrix of dA. Uses dA(aoffset:aoffset+n, :). 0 <= aoffset < lda.
[in] dB REAL array of DIMENSION ( LDB, kb ), where kb is k when TRANS = MagmaNoTrans, and is n otherwise. Before entry with TRANS = MagmaNoTrans, the leading n by k part of the array B must contain the matrix B, otherwise the leading k by n part of the array B must contain the matrix B.

[TODO: describe distribution: duplicated on all GPUs.]

Parameters:
[in] ldb INTEGER. On entry, LDB specifies the first dimension of B as declared in the calling (sub) program. When TRANS = MagmaNoTrans then LDB must be at least max( 1, n ), otherwise LDB must be at least max( 1, k ).
[in] boffset INTEGER Row offset to start sub-matrix of dB. Uses dB(boffset:boffset+n, :). 0 <= boffset < ldb.
[in] beta REAL. On entry, BETA specifies the scalar beta.
[in,out] dC REAL array of DIMENSION ( LDC, n ). Before entry with UPLO = MagmaUpper, the leading n by n upper triangular part of the array C must contain the upper triangular part of the symmetric matrix and the strictly lower triangular part of C is not referenced. On exit, the upper triangular part of the array C is overwritten by the upper triangular part of the updated matrix.
Before entry with UPLO = MagmaLower, the leading n by n lower triangular part of the array C must contain the lower triangular part of the symmetric matrix and the strictly upper triangular part of C is not referenced. On exit, the lower triangular part of the array C is overwritten by the lower triangular part of the updated matrix.
Note that the imaginary parts of the diagonal elements need not be set, they are assumed to be zero, and on exit they are set to zero. [TODO: verify]

[TODO: describe distribution: 1D column block-cyclic across GPUs.]

Parameters:
[in] ldc INTEGER. On entry, LDC specifies the first dimension of C as declared in the calling (sub) program. LDC must be at least max( 1, n ).
[in] coffset INTEGER. Row and column offset to start sub-matrix of dC. Uses dC(coffset:coffset+n, coffset:coffset+n). 0 <= coffset < ldc.
[in] ngpu INTEGER. Number of GPUs over which matrix C is distributed.
[in] nb INTEGER. Block size used for distribution of C.
[in] streams array of CUDA streams, of dimension NGPU by 20. Streams to use for running multiple GEMMs in parallel. Only up to NSTREAM streams are used on each GPU.
[in] nstream INTEGER. Number of streams to use on each device
void magmablas_strsm ( magma_side_t  side,
magma_uplo_t  uplo,
magma_trans_t  transA,
magma_diag_t  diag,
magma_int_t  m,
magma_int_t  n,
float  alpha,
const float *  dA,
magma_int_t  ldda,
float *  dB,
magma_int_t  lddb 
)
void magmablas_strsm_work ( magma_side_t  side,
magma_uplo_t  uplo,
magma_trans_t  transA,
magma_diag_t  diag,
magma_int_t  m,
magma_int_t  n,
float  alpha,
const float *  dA,
magma_int_t  ldda,
float *  dB,
magma_int_t  lddb,
magma_int_t  flag,
float *  d_dinvA,
float *  dX 
)

strsm_work solves one of the matrix equations on gpu

op(A)*X = alpha*B, or X*op(A) = alpha*B,

where alpha is a scalar, X and B are m by n matrices, A is a unit, or non-unit, upper or lower triangular matrix and op(A) is one of

op(A) = A, or op(A) = A^T, or op(A) = A^H.

The matrix X is overwritten on B.

This is an asynchronous version of magmablas_strsm with flag, d_dinvA and dX workspaces as arguments.

Parameters:
[in] side magma_side_t. On entry, side specifies whether op(A) appears on the left or right of X as follows:

  • = MagmaLeft: op(A)*X = alpha*B.
  • = MagmaRight: X*op(A) = alpha*B.
[in] uplo magma_uplo_t. On entry, uplo specifies whether the matrix A is an upper or lower triangular matrix as follows:

  • = MagmaUpper: A is an upper triangular matrix.
  • = MagmaLower: A is a lower triangular matrix.
[in] transA magma_trans_t. On entry, transA specifies the form of op(A) to be used in the matrix multiplication as follows:

  • = MagmaNoTrans: op(A) = A.
  • = MagmaTrans: op(A) = A^T.
  • = MagmaConjTrans: op(A) = A^H.
[in] diag magma_diag_t. On entry, diag specifies whether or not A is unit triangular as follows:

  • = MagmaUnit: A is assumed to be unit triangular.
  • = MagmaNonUnit: A is not assumed to be unit triangular.
[in] m INTEGER. On entry, m specifies the number of rows of B. m >= 0.
[in] n INTEGER. On entry, n specifies the number of columns of B. n >= 0.
[in] alpha REAL. On entry, alpha specifies the scalar alpha. When alpha is zero then A is not referenced and B need not be set before entry.
[in] dA REAL array of dimension ( ldda, k ), where k is m when side = MagmaLeft and is n when side = MagmaRight. Before entry with uplo = MagmaUpper, the leading k by k upper triangular part of the array A must contain the upper triangular matrix and the strictly lower triangular part of A is not referenced. Before entry with uplo = MagmaLower, the leading k by k lower triangular part of the array A must contain the lower triangular matrix and the strictly upper triangular part of A is not referenced. Note that when diag = MagmaUnit, the diagonal elements of A are not referenced either, but are assumed to be unity.
[in] ldda INTEGER. On entry, ldda specifies the first dimension of A. When side = MagmaLeft, ldda >= max( 1, m ), when side = MagmaRight, ldda >= max( 1, n ).
[in,out] dB REAL array of dimension ( lddb, n ). Before entry, the leading m by n part of the array B must contain the right-hand side matrix B, and on exit is overwritten by the solution matrix X.
[in] lddb INTEGER. On entry, lddb specifies the first dimension of B. lddb >= max( 1, m ).
[in] flag BOOLEAN. If flag is true, invert diagonal blocks. If flag is false, assume diagonal blocks (stored in d_dinvA) are already inverted.
d_dinvA (workspace) on device. If side == MagmaLeft, d_dinvA must be of size >= ((m+NB-1)/NB)*NB*NB, If side == MagmaRight, d_dinvA must be of size >= ((n+NB-1)/NB)*NB*NB, where NB = 128.
dX (workspace) size m*n, on device.
void magmablas_strtri_diag ( magma_uplo_t  uplo,
magma_diag_t  diag,
magma_int_t  n,
const float *  dA,
magma_int_t  ldda,
float *  d_dinvA 
)
void magmablas_strtri_diag_q ( magma_uplo_t  uplo,
magma_diag_t  diag,
magma_int_t  n,
const float *  dA,
magma_int_t  ldda,
float *  d_dinvA,
magma_queue_t  queue 
)

Inverts the NB x NB diagonal blocks of a triangular matrix.

This routine is used in strsm.

Same as strtri_diag, but adds queue argument. strtri_diag inverts the NB x NB diagonal blocks of A.

Parameters:
[in] uplo magma_uplo_t. On entry, uplo specifies whether the matrix A is an upper or lower triangular matrix as follows:

  • = MagmaUpper: A is an upper triangular matrix.
  • = MagmaLower: A is a lower triangular matrix.
[in] diag magma_diag_t. On entry, diag specifies whether or not A is unit triangular as follows:

  • = MagmaUnit: A is assumed to be unit triangular.
  • = MagmaNonUnit: A is not assumed to be unit triangular.
[in] n INTEGER. On entry, n specifies the order of the matrix A. N >= 0.
[in] dA REAL array of dimension ( ldda, n ) The triangular matrix A.
If UPLO = 'U', the leading N-by-N upper triangular part of A contains the upper triangular matrix, and the strictly lower triangular part of A is not referenced.
If UPLO = 'L', the leading N-by-N lower triangular part of A contains the lower triangular matrix, and the strictly upper triangular part of A is not referenced.
If DIAG = 'U', the diagonal elements of A are also not referenced and are assumed to be 1.
[in] ldda INTEGER. The leading dimension of the array A. LDDA >= max(1,N).
[out] d_dinvA REAL array of dimension (NB, ((n+NB-1)/NB)*NB), where NB = 128. On exit, contains inverses of the NB-by-NB diagonal blocks of A.
[in] queue magma_queue_t Queue to execute in.
__global__ void sgemm_kernel_N_N_64_16_16_16_4 ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A*B + beta * C.

B is put into shared memory Parameters Used: blk_M=64 blk_N=16 blk_K=16 nthd_x=16 nthd_y=4

This code should run for any matrix size. This kernel outperforms cuda-2.2 when m, n, k >= 512

__global__ void sgemm_kernel_N_N_64_16_16_16_4_special ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A*B + beta * C.

B is put into shared memory Parameters Used: blk_M=64 blk_N=16 blk_K=16 nthd_x=16 nthd_y=4

This kernel is for matrices divisible by the corresponding blocking sizes.

__global__ void sgemm_kernel_N_T_64_16_4_16_4 ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A*B^T + beta * C.

B is put into shared memory Parameters Used: blk_M=64 blk_N=16 blk_K=4 nthd_x=16 nthd_y=4

This code should run for any matrix size.

__global__ void sgemm_kernel_T_N_32_32_8_8_8 ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A^T*B + beta * C.

B is put into shared memory Parameters Used: blk_M=32 blk_N=32 blk_K=8 nthd_x=8 nthd_y=8

This code should run for any matrix size.

__global__ void sgemm_kernel_T_T_64_16_16_16_4 ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A^T*B^T + beta * C.

B is put into shared memory Parameters Used: blk_M=64 blk_N=16 blk_K=16 nthd_x=16 nthd_y=4

This code should run for any matrix size. This kernel outperforms cuda-2.2 when m, n, k >= 512

__global__ void sgemm_kernel_T_T_64_16_16_16_4_special ( float *__restrict__  C,
const float *__restrict__  A,
const float *__restrict__  B,
int  m,
int  n,
int  k,
int  lda,
int  ldb,
int  ldc,
float  alpha,
float  beta 
)

Purpose: -------- This routine computes C = alpha * A^T*B^T + beta * C.

B is put into shared memory Parameters Used: blk_M=64 blk_N=16 blk_K=16 nthd_x=16 nthd_y=4

This kernel is for matrices divisible by the corresponding blocking sizes.


Generated on 17 Sep 2014 for MAGMA by  doxygen 1.6.1