Inheritance diagram for q3::execution::gpu::blas::SYCLGPUBLAS:

Public Member Functions
	SYCLGPUBLAS (std::shared_ptr< device::GPUDeviceBackend > device_backend)

	~SYCLGPUBLAS () override

void	axpy (int number_of_elements, Complex scaling_factor, const Complex vector_x, Complex vector_y) const override

void	dgemm (char transpose_a, char transpose_b, int m, int n, int k, Real alpha, const Real matrix_a, int leading_dimension_a, const Real matrix_b, int leading_dimension_b, Real beta, Real *matrix_c, int leading_dimension_c) const override

Complex	dot_product (int number_of_elements, const Complex vector_x, const Complex vector_y) const override

ComplexSingle	dot_product_single (int number_of_elements, const ComplexSingle vector_x, const ComplexSingle vector_y) const override

void	gemm (char transpose_a, char transpose_b, int m, int n, int k, Complex alpha, const Complex matrix_a, int leading_dimension_a, const Complex matrix_b, int leading_dimension_b, Complex beta, Complex *matrix_c, int leading_dimension_c) const override

void	gemm_single (char transpose_a, char transpose_b, int m, int n, int k, ComplexSingle alpha, const ComplexSingle matrix_a, int leading_dimension_a, const ComplexSingle matrix_b, int leading_dimension_b, ComplexSingle beta, ComplexSingle *matrix_c, int leading_dimension_c) const override

double	norm (int number_of_elements, const Complex *vector_x) const override

float	norm_single (int number_of_elements, const ComplexSingle *vector_x) const override

void	scale (int number_of_elements, Complex scaling_factor, Complex *vector_x) const override

Public Member Functions inherited from q3::execution::gpu::blas::GPUBLAS
	GPUBLAS (std::shared_ptr< device::GPUDeviceBackend > device_backend)

virtual	~GPUBLAS ()=default

device::GPUDeviceBackend &	device_backend ()

const device::GPUDeviceBackend &	device_backend () const

Additional Inherited Members
Protected Attributes inherited from q3::execution::gpu::blas::GPUBLAS
std::shared_ptr< device::GPUDeviceBackend >	device_backend_

Detailed Description

Definition at line 54 of file sycl.cpp.

Constructor & Destructor Documentation

◆ SYCLGPUBLAS()

q3::execution::gpu::blas::SYCLGPUBLAS::SYCLGPUBLAS ( std::shared_ptr< device::GPUDeviceBackend > device_backend )

inlineexplicit

Construct a oneMKL BLAS wrapper.

Obtains the SYCL queue from the device backend via dynamic_cast. Pre-allocates persistent shared-memory buffers for scalar results to avoid per-call allocation overhead. Throws if the device backend is not a SYCLGPUDeviceBackend.

Definition at line 64 of file sycl.cpp.

References q3::execution::gpu::blas::GPUBLAS::device_backend_.

◆ ~SYCLGPUBLAS()

q3::execution::gpu::blas::SYCLGPUBLAS::~SYCLGPUBLAS ( )

inlineoverride

Definition at line 83 of file sycl.cpp.

Member Function Documentation

◆ axpy()

void q3::execution::gpu::blas::SYCLGPUBLAS::axpy	(	int	number_of_elements,
		Complex	scaling_factor,
		const Complex *	vector_x,
		Complex *	vector_y
	)		const

inlineoverridevirtual

Scaled vector addition (ZAXPY) via oneMKL axpy.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 185 of file sycl.cpp.

◆ dgemm()

void q3::execution::gpu::blas::SYCLGPUBLAS::dgemm	(	char	transpose_a,
		char	transpose_b,
		int	m,
		int	n,
		int	k,
		Real	alpha,
		const Real *	matrix_a,
		int	leading_dimension_a,
		const Real *	matrix_b,
		int	leading_dimension_b,
		Real	beta,
		Real *	matrix_c,
		int	leading_dimension_c
	)		const

inlineoverridevirtual

Real double-precision DGEMM via oneMKL gemm.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 275 of file sycl.cpp.

◆ dot_product()

Complex q3::execution::gpu::blas::SYCLGPUBLAS::dot_product	(	int	number_of_elements,
		const Complex *	vector_x,
		const Complex *	vector_y
	)		const

inlineoverridevirtual

Complex double-precision conjugated dot product via oneMKL dotc.

Uses a pre-allocated shared-memory scalar buffer to avoid per-call allocation overhead in the Davidson hot loop.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 139 of file sycl.cpp.

◆ dot_product_single()

ComplexSingle q3::execution::gpu::blas::SYCLGPUBLAS::dot_product_single	(	int	number_of_elements,
		const ComplexSingle *	vector_x,
		const ComplexSingle *	vector_y
	)		const

inlineoverridevirtual

Complex single-precision conjugated dot product via oneMKL dotc.

Uses a pre-allocated shared-memory scalar buffer.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 239 of file sycl.cpp.

◆ gemm()

void q3::execution::gpu::blas::SYCLGPUBLAS::gemm	(	char	transpose_a,
		char	transpose_b,
		int	m,
		int	n,
		int	k,
		Complex	alpha,
		const Complex *	matrix_a,
		int	leading_dimension_a,
		const Complex *	matrix_b,
		int	leading_dimension_b,
		Complex	beta,
		Complex *	matrix_c,
		int	leading_dimension_c
	)		const

inlineoverridevirtual

Complex double-precision ZGEMM via oneMKL gemm.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 103 of file sycl.cpp.

◆ gemm_single()

void q3::execution::gpu::blas::SYCLGPUBLAS::gemm_single	(	char	transpose_a,
		char	transpose_b,
		int	m,
		int	n,
		int	k,
		ComplexSingle	alpha,
		const ComplexSingle *	matrix_a,
		int	leading_dimension_a,
		const ComplexSingle *	matrix_b,
		int	leading_dimension_b,
		ComplexSingle	beta,
		ComplexSingle *	matrix_c,
		int	leading_dimension_c
	)		const

inlineoverridevirtual

Complex single-precision CGEMM via oneMKL gemm.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 204 of file sycl.cpp.

◆ norm()

double q3::execution::gpu::blas::SYCLGPUBLAS::norm	(	int	number_of_elements,
		const Complex *	vector_x
	)		const

inlineoverridevirtual

Euclidean norm of a complex double-precision vector via oneMKL nrm2.

Uses a pre-allocated shared-memory scalar buffer.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 159 of file sycl.cpp.

◆ norm_single()

float q3::execution::gpu::blas::SYCLGPUBLAS::norm_single	(	int	number_of_elements,
		const ComplexSingle *	vector_x
	)		const

inlineoverridevirtual

Euclidean norm of a complex single-precision vector via oneMKL nrm2.

Uses a pre-allocated shared-memory scalar buffer.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 259 of file sycl.cpp.

◆ scale()

void q3::execution::gpu::blas::SYCLGPUBLAS::scale	(	int	number_of_elements,
		Complex	scaling_factor,
		Complex *	vector_x
	)		const

inlineoverridevirtual

Scale a complex double-precision vector in-place via oneMKL scal.

Implements q3::execution::gpu::blas::GPUBLAS.

Definition at line 173 of file sycl.cpp.

The documentation for this class was generated from the following file:

src/layer_0_core/math/linear_algebra/gpu/blas/sycl.cpp

Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ SYCLGPUBLAS()

◆ ~SYCLGPUBLAS()

Member Function Documentation

◆ axpy()

◆ dgemm()

◆ dot_product()

◆ dot_product_single()

◆ gemm()

◆ gemm_single()

◆ norm()

◆ norm_single()

◆ scale()