tenzr::txpr::Txpr¶

SSA-style intermediate representation of a computation as typed variables and equations; produced by tracing and consumed by emitters and validators.

#include <txpr.h>

Public Functions¶

Public Functions Documentation¶

function Txpr¶

Txpr() =default

function AddVar¶

VarId AddVar(
    Shape shape,
    DType dtype,
    std::string name =""
)

Allocate a new symbolic variable in the Txpr.

Parameters:

• shape Shape of the variable.
• dtype Dtype of the variable.
• name Optional debug name.

Return: The new variable's id.

Append a new variable definition to the graph.

function AppendInstruction¶

VarId AppendInstruction(
    OpCode op,
    std::vector< VarId > in_vars,
    Shape out_shape,
    DType out_dtype,
    AttributePayload attrs ={}
)

Append a single-output equation to the Txpr.

Parameters:

• op Operation to execute.
• in_vars Input variable ids.
• out_shape Result shape.
• out_dtype Result dtype.
• attrs Operation attributes.

Return: The instruction's output variable id.

Append a single-output instruction and return its result variable.

function AppendMultiOutInstruction¶

std::vector< VarId > AppendMultiOutInstruction(
    OpCode op,
    std::vector< VarId > in_vars,
    const std::vector< Shape > & out_shapes,
    const std::vector< DType > & out_dtypes,
    AttributePayload attrs ={}
)

Append a multi-output equation to the Txpr.

Parameters:

• op Operation to execute.
• in_vars Input variable ids.
• out_shapes Result shapes.
• out_dtypes Result dtypes, matching out_shapes.
• attrs Operation attributes.

Return: The instruction's output variable ids.

Append a multi-output instruction and return its result variables.

function MarkInput¶

void MarkInput(
    VarId var
)

Declare a variable as a function input parameter.

Parameters:

• var Variable to mark as input.

Register var as an input binder of the graph.

function MarkOutput¶

void MarkOutput(
    VarId var
)

Declare a variable as a function output result.

Parameters:

• var Variable to mark as output.

Register var as an output binder of the graph.

function Variables¶

inline const std::vector< VarDef > & Variables() const

Accessors.

function Equations¶

inline const std::vector< TxprEqn > & Equations() const

function Instructions¶

inline const std::vector< TxprEqn > & Instructions() const

function InBinders¶

inline const std::vector< VarId > & InBinders() const

function OutBinders¶

inline const std::vector< VarId > & OutBinders() const

function NumVars¶

inline size_t NumVars() const

function NumEquations¶

inline size_t NumEquations() const

function NumInstructions¶

inline size_t NumInstructions() const

function Var¶

inline const VarDef & Var(
    VarId id
) const

function Variable¶

inline const VarDef & Variable(
    VarId id
) const

function ShapeOf¶

inline const Shape & ShapeOf(
    VarId id
) const

function DtypeOf¶

inline DType DtypeOf(
    VarId id
) const

function Validate¶

bool Validate(
    std::string * error_msg =nullptr
) const

Perform strict SSA and type validation.

Parameters:

• error_msg Optional out-parameter receiving the first failure reason.

Return: true when the graph is well-formed.

Check that binders are unique and every use is SSA-defined before it.

function ToString¶

std::string ToString() const

Render readable Jaxpr-style text representation.

Return: Textual representation of the graph.

Render the graph in the textual lambda; let ... in (...) form.