fhelium.backend.rns.operations

Native implementations of logical RNS operations.

NativeMontgomeryAccumulateImplementation

class View source

NativeMontgomeryAccumulateImplementation(name: str = 'native-montgomery-accumulate', supports_in_place: bool = False, operation_types: tuple[type[Operation], ...] = (rns.AddMontgomeryLazyOp,))

Bases: _TensorOperandImplementation

Accumulate equal-layout Montgomery bundles modulo twice each prime.

Attributes

Name	Type	Default/value
name	str	'native-montgomery-accumulate'
supports_in_place	bool	False
operation_types	tuple[type[Operation], ...]	(rns.AddMontgomeryLazyOp,)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeMontgomeryMultiplyImplementation

class View source

NativeMontgomeryMultiplyImplementation(name: str = 'native-montgomery-multiply', supports_in_place: bool = False, operation_types: tuple[type[Operation], ...] = (rns.MontgomeryMultiplyOp,))

Bases: _TensorOperandImplementation

Multiply two logical NTT/Montgomery polynomial bundles.

Attributes

Name	Type	Default/value
name	str	'native-montgomery-multiply'
supports_in_place	bool	False
operation_types	tuple[type[Operation], ...]	(rns.MontgomeryMultiplyOp,)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeMontgomeryWeightedSumImplementation

class View source

NativeMontgomeryWeightedSumImplementation(name: str = 'native-montgomery-weighted-sum', supports_in_place: bool = False, operation_types: tuple[type[Operation], ...] = (rns.MontgomeryWeightedSumOp, rns.MontgomeryWeightedSumsOp))

Bases: _TensorOperandImplementation

Compute one or more $\sum _t{c}_t{p}_t$ results without term products.

Attributes

Name	Type	Default/value
name	str	'native-montgomery-weighted-sum'
supports_in_place	bool	False
operation_types	tuple[type[Operation], ...]	(rns.MontgomeryWeightedSumOp, rns.MontgomeryWeightedSumsOp)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativePlaintextArithmeticImplementation

class View source

NativePlaintextArithmeticImplementation(name: str = 'native-plaintext-arithmetic', supports_in_place: bool = True, operation_types: tuple[type[Operation], ...] = (rns.AddPlaintextOp, rns.MultiplyPlaintextOp))

Bases: _TensorOperandImplementation

Execute prepared plaintext addition and multiplication.

Attributes

Name	Type	Default/value
name	str	'native-plaintext-arithmetic'
supports_in_place	bool	True
operation_types	tuple[type[Operation], ...]	(rns.AddPlaintextOp, rns.MultiplyPlaintextOp)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeRnsLinearImplementation

class View source

NativeRnsLinearImplementation(name: str = 'native-rns-linear', supports_in_place: bool = True, operation_types: tuple[type[Operation], ...] = (rns.AddStandardOp, rns.SubtractStandardOp, rns.NegateStandardOp))

Bases: _TensorOperandImplementation

Execute standard-range RNS add, subtract, and negate operations.

Attributes

Name	Type	Default/value
name	str	'native-rns-linear'
supports_in_place	bool	True
operation_types	tuple[type[Operation], ...]	(rns.AddStandardOp, rns.SubtractStandardOp, rns.NegateStandardOp)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeRnsStructureImplementation

class View source

NativeRnsStructureImplementation(name: str = 'native-rns-structure', supports_in_place: bool = False, operation_types: tuple[type[Operation], ...] = (rns.ExtractComponentOp, rns.PackTwoComponentsOp, rns.PackThreeComponentsOp))

Bases: _TensorOperandImplementation

Extract and assemble logical RNS component bundles.

Attributes

Name	Type	Default/value
name	str	'native-rns-structure'
supports_in_place	bool	False
operation_types	tuple[type[Operation], ...]	(rns.ExtractComponentOp, rns.PackTwoComponentsOp, rns.PackThreeComponentsOp)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...

NativeRnsTransitionImplementation

class View source

NativeRnsTransitionImplementation(name: str = 'native-rns-transition', supports_in_place: bool = True, operation_types: tuple[type[Operation], ...] = (rns.StandardToMontgomeryOp, rns.MontgomeryToStandardOp, rns.RestrictDepthOp, rns.ReinterpretScaleOp))

Bases: _TensorOperandImplementation

Execute residue conversion, depth restriction, and scale metadata ops.

Attributes

Name	Type	Default/value
name	str	'native-rns-transition'
supports_in_place	bool	True
operation_types	tuple[type[Operation], ...]	(rns.StandardToMontgomeryOp, rns.MontgomeryToStandardOp, rns.RestrictDepthOp, rns.ReinterpretScaleOp)

execute

method

def execute(invocation: OperationInvocation, values: tuple[torch.Tensor, ...], resources: tuple[BoundResource, ...], *, in_place: bool) -> tuple[torch.Tensor, ...]: ...