gemm#
Computes a matrix-matrix product with general matrices.
Description#
The gemm routines compute a scalar-matrix-matrix product and add the result to a scalar-matrix product, with general matrices.
The operation is defined as:
where:
op(
X) is one of op(X) =X, or op(X) =XT, or op(X) =XHalphaandbetaare scalarsA,BandCare matricesop(
A) ismxkmatrixop(
B) iskxnmatrixCismxnmatrix
gemm supports the following precisions:
Ta
(A matrix)
|
Tb
(B matrix)
|
Tc
(C matrix)
|
Ts
(alpha/beta)
|
|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
gemm (Buffer Version)#
Syntax#
namespace oneapi::mkl::blas::column_major {
void gemm(sycl::queue &queue,
oneapi::mkl::transpose transa,
oneapi::mkl::transpose transb,
std::int64_t m,
std::int64_t n,
std::int64_t k,
Ts alpha,
sycl::buffer<Ta,1> &a,
std::int64_t lda,
sycl::buffer<Tb,1> &b,
std::int64_t ldb,
Ts beta,
sycl::buffer<Tc,1> &c,
std::int64_t ldc,
compute_mode mode = compute_mode::unset)
}
namespace oneapi::mkl::blas::row_major {
void gemm(sycl::queue &queue,
oneapi::mkl::transpose transa,
oneapi::mkl::transpose transb,
std::int64_t m,
std::int64_t n,
std::int64_t k,
Ts alpha,
sycl::buffer<Ta,1> &a,
std::int64_t lda,
sycl::buffer<Tb,1> &b,
std::int64_t ldb,
Ts beta,
sycl::buffer<Tc,1> &c,
std::int64_t ldc,
compute_mode mode = compute_mode::unset)
}
Input Parameters#
- queue
The queue where the routine should be executed.
- transa
Specifies op(
A), the transposition operation applied to matrixA. See Data Types for more details.- transb
Specifies op(
B), the transposition operation applied to matrixB. See Data Types for more details.- m
Number of rows of matrix op(
A) and matrixC. Must be at least zero.- n
Number of columns of matrix op(
B) and matrixC. Must be at least zero.- k
Number of columns of matrix op(
A) and rows of matrix op(B). Must be at least zero.- alpha
Scaling factor for matrix-matrix product.
- a
Buffer holding input matrix
A. See Matrix Storage for more details.transa=transpose::nontranstransa=transpose::transortrans=transpose::conjtransColumn major
Aismxkmatrix. Size of arrayamust be at leastlda*kAiskxmmatrix. Size of arrayamust be at leastlda*mRow major
Aismxkmatrix. Size of arrayamust be at leastlda*mAiskxmmatrix. Size of arrayamust be at leastlda*k- lda
Leading dimension of matrix
A. Must be positive.transa=transpose::nontranstransa=transpose::transortrans=transpose::conjtransColumn major
Must be at least
mMust be at least
kRow major
Must be at least
kMust be at least
m- b
Buffer holding input matrix
B. See Matrix Storage for more details.transb=transpose::nontranstransb=transpose::transortrans=transpose::conjtransColumn major
Biskxnmatrix. Size of arraybmust be at leastldb*nBisnxkmatrix. Size of arraybmust be at leastldb*kRow major
Biskxnmatrix. Size of arraybmust be at leastldb*kBisnxkmatrix. Size of arraybmust be at leastldb*n- ldb
Leading dimension of matrix
B. Must be positive.transb=transpose::nontranstransb=transpose::transortrans=transpose::conjtransColumn major
Must be at least
kMust be at least
nRow major
Must be at least
nMust be at least
k- beta
Scaling factor for matrix
C.- c
Buffer holding input/output matrix
C. See Matrix Storage for more details.Column major
Cismxnmatrix. Size of arraycmust be at leastldc*nRow major
Cismxnmatrix. Size of arraycmust be at leastldc*m- ldc
Leading dimension of matrix
C. Must be positive.Column major
Must be at least
mRow major
Must be at least
n- mode
Optional. Compute mode settings. See Compute Modes for more details.
Output Parameters#
- c
Output buffer overwritten by
alpha* op(A)*op(B) +beta*C.
Note
If beta = 0, matrix C does not need to be initialized before calling gemm.
Examples#
An example of how to use buffer version of gemm can be found in oneMKL installation directory, under:
share/doc/mkl/examples/sycl/blas/source/gemm.cpp
gemm (USM Version)#
Syntax#
namespace oneapi::mkl::blas::column_major {
sycl::event gemm(sycl::queue &queue,
oneapi::mkl::transpose transa,
oneapi::mkl::transpose transb,
std::int64_t m,
std::int64_t n,
std::int64_t k,
oneapi::mkl::value_or_pointer<Ts> alpha,
const Ta *a,
std::int64_t lda,
const Tb *b,
std::int64_t ldb,
oneapi::mkl::value_or_pointer<Ts> beta,
Tc *c,
std::int64_t ldc,
compute_mode mode = compute_mode::unset,
const std::vector<sycl::event> &dependencies = {})
}
namespace oneapi::mkl::blas::row_major {
sycl::event gemm(sycl::queue &queue,
oneapi::mkl::transpose transa,
oneapi::mkl::transpose transb,
std::int64_t m,
std::int64_t n,
std::int64_t k,
oneapi::mkl::value_or_pointer<Ts> alpha,
const Ta *a,
std::int64_t lda,
const Tb *b,
std::int64_t ldb,
oneapi::mkl::value_or_pointer<Ts> beta,
Tc *c,
std::int64_t ldc,
compute_mode mode = compute_mode::unset,
const std::vector<sycl::event> &dependencies = {})
}
Input Parameters#
- queue
The queue where the routine should be executed.
- transa
Specifies op(
A), the transposition operation applied to matrixA. See Data Types for more details.- transb
Specifies op(
B), the transposition operation applied to matrixB. See Data Types for more details.- m
Number of rows of matrix op(
A) and matrixC. Must be at least zero.- n
Number of columns of matrix op(
B) and matrixC. Must be at least zero.- k
Number of columns of matrix op(
A) and rows of matrix op(B). Must be at least zero.- alpha
Scaling factor for matrix-matrix product. See Scalar Arguments for more information on the
value_or_pointerdata type.- a
Pointer to input matrix
A. See Matrix Storage for more details.Anot transposedAtransposedColumn major
Aismxkmatrix. Size of arrayamust be at leastlda*kAiskxmmatrix. Size of arrayamust be at leastlda*mRow major
Aismxkmatrix. Size of arrayamust be at leastlda*mAiskxmmatrix. Size of arrayamust be at leastlda*k- lda
Leading dimension of matrix
A. Must be positive.Anot transposedAtransposedColumn major
Must be at least
mMust be at least
kRow major
Must be at least
kMust be at least
m- b
Pointer to input matrix
B. See Matrix Storage for more details.Bnot transposedBtransposedColumn major
Biskxnmatrix. Size of arraybmust be at leastldb*nBisnxkmatrix. Size of arraybmust be at leastldb*kRow major
Biskxnmatrix. Size of arraybmust be at leastldb*kBisnxkmatrix. Size of arraybmust be at leastldb*n- ldb
Leading dimension of matrix
B. Must be positive.Bnot transposedBtransposedColumn major
Must be at least
kMust be at least
nRow major
Must be at least
nMust be at least
k- beta
Scaling factor for matrix
C. See Scalar Arguments for more information on thevalue_or_pointerdata type.- c
Pointer to input/output matrix
C. See Matrix Storage for more details.Column major
Cismxnmatrix. Size of arraycmust be at leastldc*nRow major
Cismxnmatrix. Size of arraycmust be at leastldc*m- ldc
Leading dimension of matrix
C. Must be positive.Column major
Must be at least
mRow major
Must be at least
n- mode
Optional. Compute mode settings. See Compute Modes for more details.
- dependencies
Optional. List of events to wait for before starting computation, if any. If omitted, defaults to no dependencies.
modeanddependenciesmay be omitted independently; it is not necessary to specifymodein order to provide dependencies.
Output Parameters#
- c
Pointer to output matrix overwritten by
alpha* op(A)*op(B) +beta*C.
Note
If beta = 0, matrix C does not need to be initialized before calling gemm.
Return Values#
Output event to wait on to ensure computation is complete.
Examples#
An example of how to use USM version of gemm can be found in oneMKL installation directory, under:
share/doc/mkl/examples/sycl/blas/source/gemm_usm.cpp