Crate runmat_accelerate_api

Crate runmat_accelerate_api 

Source

Structs§

ApiDeviceInfo
BindGroupLayoutTelemetry
CorrcoefOptions
Options controlling provider-backed correlation coefficient computation.
CovarianceOptions
Options controlling provider-backed covariance computation.
FspecialRequest
Request dispatched to acceleration providers for fspecial kernels.
GpuTensorHandle
HostLogicalOwned
Host-resident logical output returned by providers.
HostTensorOwned
HostTensorView
ImageNormalizeDescriptor
ImfilterOptions
Options supplied to acceleration providers for imfilter.
IsMemberOptions
Options controlling provider-backed ismember operations.
IsMemberResult
Host-resident outputs returned by provider-backed ismember operations.
KernelAttrTelemetry
KernelLaunchTelemetry
MatmulEpilogue
MeshgridAxisView
Lightweight 1-D axis view used by provider meshgrid hooks.
PagefunRequest
PowerStepEpilogue
ProviderBandwidth
ProviderCholResult
ProviderConv1dOptions
ProviderCumminResult
ProviderDispatchStats
ProviderEigResult
ProviderFindResult
ProviderIirFilterOptions
ProviderIirFilterResult
ProviderInvOptions
ProviderLinsolveOptions
ProviderLinsolveResult
ProviderLuResult
ProviderMeshgridResult
Provider-side meshgrid result containing coordinate tensor handles.
ProviderMoments2
ProviderPinvOptions
ProviderPolyderQuotient
Numerator/denominator payload returned by provider-backed polyder quotient rule.
ProviderPolyfitResult
ProviderPolyvalMu
ProviderPolyvalOptions
ProviderQrOptions
ProviderQrPowerIterResult
ProviderQrResult
ProviderTelemetry
ReduceDimResult
SetdiffOptions
Options controlling provider-backed setdiff operations.
SetdiffResult
Host-resident outputs returned by provider-backed setdiff operations.
SortResult
Host-resident outputs returned by provider-backed sort operations.
SortRowsColumnSpec
ThreadProviderGuard
TransposeInfo
UnionOptions
Options controlling provider-backed union operations.
UnionResult
Host-resident outputs returned by provider-backed union operations.
UniqueOptions
Options controlling provider-backed unique operations.
UniqueResult
Host-resident outputs returned by provider-backed unique operations.

Enums§

CorrcoefNormalization
Normalisation mode for correlation coefficients.
CorrcoefRows
Row-selection strategy for correlation coefficients.
CovNormalization
Normalisation mode used by covariance computations.
CovRows
Row handling strategy for covariance computations.
FindDirection
FspecialFilter
Parameterisation of 2-D filters generated by fspecial.
ImfilterMode
Correlation vs convolution behaviour for imfilter.
ImfilterPadding
Padding strategy used by imfilter.
ImfilterShape
Output sizing mode used by imfilter.
PagefunOp
ProviderCondNorm
Supported norm specifications for the cond builtin.
ProviderConvMode
ProviderConvOrientation
ProviderHermitianKind
ProviderNanMode
NaN handling mode for provider-backed reductions.
ProviderNormOrder
Supported norm orders for the norm builtin.
ProviderPrecision
ProviderQrPivot
ProviderScanDirection
Direction used when computing prefix sums on the device.
ProviderStdNormalization
Normalization strategy used by provider-backed standard deviation reductions.
ProviderSymmetryKind
ReductionFlavor
ReductionTwoPassMode
ScaleOp
Descriptor for GEMM epilogues applied to C = A * B before storing to C.
SetdiffOrder
Ordering applied by provider-backed setdiff operations.
SortComparison
Comparison strategy applied during sorting.
SortOrder
Sort direction used by acceleration providers.
UnionOrder
Ordering applied by provider-backed union operations.
UniqueOccurrence
Occurrence selection for provider-backed unique operations.
UniqueOrder
Ordering applied by provider-backed unique operations.

Traits§

AccelProvider
Device/provider interface that backends implement and register into the runtime layer

Functions§

clear_handle_logical
Convenience helper for clearing logical annotations explicitly.
clear_handle_precision
Clear any recorded precision metadata for a GPU tensor handle.
clear_handle_transpose
clear_provider
Clear the globally registered provider. Intended for tests to ensure deterministic behaviour.
clear_residency
Clear residency metadata for the provided GPU tensor handle, if a backend has registered a handler via register_residency_clear.
handle_is_logical
Returns true when the supplied handle has been marked as logical.
handle_is_transposed
handle_logical_hits
handle_precision
Look up the recorded precision for a GPU tensor handle, if any.
handle_transpose_info
next_device_id
provider
provider_for_device
provider_for_handle
record_handle_transpose
register_provider
Register a global acceleration provider.
register_residency_clear
Register a callback used to clear residency tracking when GPU tensors are gathered back to the host. Backends that maintain residency metadata should install this hook during initialization.
register_sequence_threshold_provider
Register a callback that exposes the current sequence length threshold derived from the auto-offload planner. Array constructors can use this hint to decide when to prefer GPU residency automatically.
sequence_threshold_hint
Query the currently registered sequence threshold hint, if any.
set_handle_logical
Annotate a GPU tensor handle as logically-typed (logical in MATLAB terms) or clear the logical flag when logical is false.
set_handle_precision
Record the precision associated with a GPU tensor handle so host operations can reconstruct the original dtype when gathering back to the CPU.
set_thread_provider
try_elem_add
Convenience: perform elementwise add via provider if possible; otherwise return None
try_elem_atan2
Convenience: perform elementwise atan2 via provider if possible; otherwise return None
try_elem_hypot
Convenience: perform elementwise hypot via provider if possible; otherwise return None
try_elem_max
Convenience: perform elementwise max via provider if possible; otherwise return None
try_elem_min
Convenience: perform elementwise min via provider if possible; otherwise return None

Type Aliases§

ProviderCummaxResult
Result payload returned by provider-side cummax scans.