Package jdistlib.accelerator

Interface ComputeBackend

All Superinterfaces:

AutoCloseable, LinearAlgebraBackend, SinglePrecisionLinearAlgebraBackend

All Known Implementing Classes:

CpuComputeBackend

public interface ComputeBackend
extends LinearAlgebraBackend, SinglePrecisionLinearAlgebraBackend, AutoCloseable

Optional backend for vector, dense-linear-algebra, and batched likelihood work.

Method Summary

Modifier and Type	Method	Description
default boolean	automaticRouting()	Whether individual operations may route between CPU and the selected provider.
boolean	available()
default double[]	axpy(double alpha, double[] x, double[] y)
ComputeCapabilities	capabilities()
void	close()
default ComputeDeviceInfo	deviceInfo()	Returns stable backend, runtime, driver, and device provenance where available.
default double	dot(double[] x, double[] y)
String	id()
default LogisticRegressionBatchResult	logisticRegression(double[][] design, double[] outcomes, double[][] states, double priorPrecision)
default double[][]	matrixMultiply(double[][] left, double[][] right)
default ExecutionPlan	plan(LinearAlgebraOperation operation, NumericPrecision precision, int... dimensions)	Predicts execution for the supplied operation and dimensions without running it.
default PreparedCholesky	prepareDpotrf(double[] matrix, int dimension)	Prepares a reusable FP64 Cholesky factor and solve handle.
default PreparedLogisticRegression	prepareLogisticRegression(double[][] design, double[] outcomes)	Keeps reusable data in backend-optimal storage when the backend supports it.
default PreparedFloatCholesky	prepareSpotrf(float[] matrix, int dimension)	Prepares a reusable FP32 Cholesky factor and solve handle.
default PreparedTransposeProduct	prepareTransposeProduct(double[][] matrix)	Keeps a reusable row-by-feature matrix ready for repeated X'v batches.
default String	selectedBackend()	Concrete provider used for accelerated work; differs from id() for AUTO routing.
default double[]	unary(UnaryOperation operation, double[] input)

Methods inherited from interface jdistlib.accelerator.LinearAlgebraBackend

dasum, daxpy, dcopy, dcsrgemm, dcsrmm, dcsrmv, dcsrpotrf, dcsrpotrf, dcsrsv, ddot, dgemm, dgemm, dgemmBatched, dgemv, dgemv, dgeqp3, dger, dgesvd, dgetrf, dgetrfBatched, dnrm2, dpotrf, dpotrfBatched, dscal, dswap, dsyev, dsygvd, dsymm, dsyr, dsyr2, dsyr2k, dsyrk, dsyrk, dsytrf, dtrsm, dtrsm, dtrsv, idamax, prepareDcsr, prepareDcsrpotrf, prepareDcsrpotrf, prepareDge

Methods inherited from interface jdistlib.accelerator.SinglePrecisionLinearAlgebraBackend

isamax, prepareScsr, prepareScsrpotrf, prepareScsrpotrf, prepareSge, sasum, saxpy, scopy, scsrgemm, scsrmm, scsrmv, scsrpotrf, scsrpotrf, scsrsv, sdot, sgemm, sgemm, sgemmBatched, sgemv, sgemv, sgeqp3, sger, sgesvd, sgetrf, sgetrfBatched, snrm2, spotrf, spotrfBatched, sscal, sswap, ssyev, ssygvd, ssymm, ssyr, ssyr2, ssyr2k, ssyrk, ssyrk, ssytrf, strsm, strsm, strsv

Method Details

id

String id()

selectedBackend

default String selectedBackend()

Concrete provider used for accelerated work; differs from id() for AUTO routing.

automaticRouting

default boolean automaticRouting()

Whether individual operations may route between CPU and the selected provider.

available

boolean available()

capabilities

ComputeCapabilities capabilities()

deviceInfo

default ComputeDeviceInfo deviceInfo()

Returns stable backend, runtime, driver, and device provenance where available.

plan

default ExecutionPlan plan(LinearAlgebraOperation operation,
 NumericPrecision precision,
 int... dimensions)

Predicts execution for the supplied operation and dimensions without running it.

prepareDpotrf

default PreparedCholesky prepareDpotrf(double[] matrix,
 int dimension)

Prepares a reusable FP64 Cholesky factor and solve handle.

prepareSpotrf

default PreparedFloatCholesky prepareSpotrf(float[] matrix,
 int dimension)

Prepares a reusable FP32 Cholesky factor and solve handle.

unary

default double[] unary(UnaryOperation operation,
 double[] input)

axpy

default double[] axpy(double alpha,
 double[] x,
 double[] y)

dot

default double dot(double[] x,
 double[] y)

matrixMultiply

default double[][] matrixMultiply(double[][] left,
 double[][] right)

prepareTransposeProduct

default PreparedTransposeProduct prepareTransposeProduct(double[][] matrix)

Keeps a reusable row-by-feature matrix ready for repeated X'v batches.

logisticRegression

default LogisticRegressionBatchResult logisticRegression(double[][] design,
 double[] outcomes,
 double[][] states,
 double priorPrecision)

prepareLogisticRegression

default PreparedLogisticRegression prepareLogisticRegression(double[][] design,
 double[] outcomes)

Keeps reusable data in backend-optimal storage when the backend supports it.

close

void close()

Specified by:

close in interface AutoCloseable