nvjitlink

Note

The nvjitlink bindings are not supported on nvJitLink installations <12.3. Ensure the installed CUDA toolkit’s nvJitLink version is >=12.3.

Functions

NvJitLink defines the following functions for linking code objects and querying the info and error logs.

cuda.bindings.nvjitlink.create(uint32_t num_options, options) → intptr_t

nvJitLinkCreate creates an instance of nvJitLinkHandle with the given input options, and sets the output parameter handle.

It supports options listed in Supported Link Options.

Parameters:

• num_options (uint32_t) – Number of options passed.

• options (object) –

  Array of size num_options of option strings. It can be:

  • an int as the pointer address to the nested sequence, or

  • a Python sequence of ints, each of which is a pointer address to a valid sequence of ‘char’, or

  • a nested Python sequence of str.

Returns:

Address of nvJitLink handle.

Return type:

intptr_t

See also

nvJitLinkCreate

cuda.bindings.nvjitlink.destroy(intptr_t handle)

nvJitLinkDestroy frees the memory associated with the given handle.

Parameters:

handle (intptr_t) – nvJitLink handle.

See also

nvJitLinkDestroy

cuda.bindings.nvjitlink.add_data(intptr_t handle, int input_type, data, size_t size, name)

nvJitLinkAddData adds data image to the link.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• input_type (InputType) – kind of input.

• data (bytes) – pointer to data image in memory.

• size (size_t) – size of the data.

• name (str) – name of input object.

See also

nvJitLinkAddData

cuda.bindings.nvjitlink.add_file(intptr_t handle, int input_type, file_name)

nvJitLinkAddFile reads data from file and links it in.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• input_type (InputType) – kind of input.

• file_name (str) – name of file.

See also

nvJitLinkAddFile

cuda.bindings.nvjitlink.complete(intptr_t handle)

nvJitLinkComplete does the actual link.

Parameters:

handle (intptr_t) – nvJitLink handle.

See also

nvJitLinkComplete

cuda.bindings.nvjitlink.get_linked_cubin_size(intptr_t handle) → size_t

nvJitLinkGetLinkedCubinSize gets the size of the linked cubin.

Parameters:

handle (intptr_t) – nvJitLink handle.

Returns:

Size of the linked cubin.

Return type:

size_t

See also

nvJitLinkGetLinkedCubinSize

cuda.bindings.nvjitlink.get_linked_cubin(intptr_t handle, cubin)

nvJitLinkGetLinkedCubin gets the linked cubin.

User is responsible for allocating enough space to hold the cubin.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• cubin (bytes) – The linked cubin.

See also

nvJitLinkGetLinkedCubin

cuda.bindings.nvjitlink.get_linked_ptx_size(intptr_t handle) → size_t

nvJitLinkGetLinkedPtxSize gets the size of the linked ptx.

Linked PTX is only available when using the -lto option.

Parameters:

handle (intptr_t) – nvJitLink handle.

Returns:

Size of the linked PTX.

Return type:

size_t

See also

nvJitLinkGetLinkedPtxSize

cuda.bindings.nvjitlink.get_linked_ptx(intptr_t handle, ptx)

nvJitLinkGetLinkedPtx gets the linked ptx.

Linked PTX is only available when using the -lto option. User is responsible for allocating enough space to hold the ptx.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• ptx (bytes) – The linked PTX.

See also

nvJitLinkGetLinkedPtx

cuda.bindings.nvjitlink.get_linked_ltoir_size(intptr_t handle) → size_t

nvJitLinkGetLinkedLTOIRSize gets the size of the linked LTOIR.

Linked LTOIR is only available when using the -lto option. The returned data is in LTOIR Container format (not raw bitcode).

Parameters:

handle (intptr_t) – nvJitLink handle.

Returns:

Size of the linked LTOIR.

Return type:

size_t

See also

nvJitLinkGetLinkedLTOIRSize

cuda.bindings.nvjitlink.get_linked_ltoir(intptr_t handle, ltoir)

nvJitLinkGetLinkedLTOIR gets the linked LTOIR.

Linked LTOIR is only available when using the -lto option. User is responsible for allocating enough space to hold the ltoir. The returned data is in LTOIR Container format (not raw bitcode).

Parameters:

• handle (intptr_t) – nvJitLink handle.

• ltoir (bytes) – The linked LTOIR in Container format.

See also

nvJitLinkGetLinkedLTOIR

cuda.bindings.nvjitlink.get_error_log_size(intptr_t handle) → size_t

nvJitLinkGetErrorLogSize gets the size of the error log.

Parameters:

handle (intptr_t) – nvJitLink handle.

Returns:

Size of the error log.

Return type:

size_t

See also

nvJitLinkGetErrorLogSize

cuda.bindings.nvjitlink.get_error_log(intptr_t handle, log)

nvJitLinkGetErrorLog puts any error messages in the log.

User is responsible for allocating enough space to hold the log.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• log (bytes) – The error log.

See also

nvJitLinkGetErrorLog

cuda.bindings.nvjitlink.get_info_log_size(intptr_t handle) → size_t

nvJitLinkGetInfoLogSize gets the size of the info log.

Parameters:

handle (intptr_t) – nvJitLink handle.

Returns:

Size of the info log.

Return type:

size_t

See also

nvJitLinkGetInfoLogSize

cuda.bindings.nvjitlink.get_info_log(intptr_t handle, log)

nvJitLinkGetInfoLog puts any info messages in the log.

User is responsible for allocating enough space to hold the log.

Parameters:

• handle (intptr_t) – nvJitLink handle.

• log (bytes) – The info log.

See also

nvJitLinkGetInfoLog

cuda.bindings.nvjitlink.version() → tuple

nvJitLinkVersion returns the current version of nvJitLink.

Returns:

• unsigned int: The major version.

• unsigned int: The minor version.

Return type:

A 2-tuple containing

See also

nvJitLinkVersion

Types

class cuda.bindings.nvjitlink.Result(value: int)

The enumerated type nvJitLinkResult defines API call result codes. nvJitLink APIs return nvJitLinkResult codes to indicate the result.

See nvJitLinkResult.

SUCCESS = <Result.SUCCESS: 0>

ERROR_UNRECOGNIZED_OPTION = <Result.ERROR_UNRECOGNIZED_OPTION: 1>

ERROR_MISSING_ARCH = <Result.ERROR_MISSING_ARCH: 2>

ERROR_INVALID_INPUT = <Result.ERROR_INVALID_INPUT: 3>

ERROR_PTX_COMPILE = <Result.ERROR_PTX_COMPILE: 4>

ERROR_NVVM_COMPILE = <Result.ERROR_NVVM_COMPILE: 5>

ERROR_INTERNAL = <Result.ERROR_INTERNAL: 6>

ERROR_THREADPOOL = <Result.ERROR_THREADPOOL: 7>

ERROR_UNRECOGNIZED_INPUT = <Result.ERROR_UNRECOGNIZED_INPUT: 8>

ERROR_FINALIZE = <Result.ERROR_FINALIZE: 9>

ERROR_NULL_INPUT = <Result.ERROR_NULL_INPUT: 10>

ERROR_INCOMPATIBLE_OPTIONS = <Result.ERROR_INCOMPATIBLE_OPTIONS: 11>

ERROR_INCORRECT_INPUT_TYPE = <Result.ERROR_INCORRECT_INPUT_TYPE: 12>

ERROR_ARCH_MISMATCH = <Result.ERROR_ARCH_MISMATCH: 13>

ERROR_OUTDATED_LIBRARY = <Result.ERROR_OUTDATED_LIBRARY: 14>

ERROR_MISSING_FATBIN = <Result.ERROR_MISSING_FATBIN: 15>

ERROR_UNRECOGNIZED_ARCH = <Result.ERROR_UNRECOGNIZED_ARCH: 16>

ERROR_UNSUPPORTED_ARCH = <Result.ERROR_UNSUPPORTED_ARCH: 17>

ERROR_LTO_NOT_ENABLED = <Result.ERROR_LTO_NOT_ENABLED: 18>

class cuda.bindings.nvjitlink.InputType(value: int)

The enumerated type nvJitLinkInputType defines the kind of inputs that can be passed to nvJitLinkAdd* APIs.

See nvJitLinkInputType.

NONE = <InputType.NONE: 0>

CUBIN = <InputType.CUBIN: 1>

PTX = <InputType.PTX: 2>

LTOIR = <InputType.LTOIR: 3>

FATBIN = <InputType.FATBIN: 4>

OBJECT = <InputType.OBJECT: 5>

LIBRARY = <InputType.LIBRARY: 6>

INDEX = <InputType.INDEX: 7>

ANY = <InputType.ANY: 10>