Module Nvrtc

Bindings to the NVIDIA `nvrtc` library.

NVRTC is a runtime compilation library for CUDA C++. See: The User guide for the NVRTC library.

type result

See enum nvrtcResult.

val sexp_of_result : result -> Sexplib0.Sexp.t
exception Nvrtc_error of {
  1. status : result;
  2. message : string;
}

Error codes returned by CUDA functions are converted to exceptions. The message stores a snake-case variant of the offending CUDA function name (see Nvrtc_ffi.Bindings.Functions for the direct funciton bindings).

val is_success : result -> bool
type compile_to_ptx_result = {
  1. log : string option;
  2. ptx : char Ctypes.ptr;
  3. ptx_length : int;
}

The values passed from compile_to_ptx to Cuda.Module.load_data_ex. Currently, cudajit converts the result of nvrtc_compile_program to human-readable PTX assembly before passing it to the cu_module_load_data_ex function.

val sexp_of_compile_to_ptx_result : compile_to_ptx_result -> Sexplib0.Sexp.t
val compile_to_ptx : cu_src:string -> name:string -> options:string list -> with_debug:bool -> compile_to_ptx_result

Performs a cascade of calls: nvrtcCreateProgram, nvrtcCompileProgram, nvrtcGetPTX. If you store cu_src as a file, pass the file name including the extension as name. options can include for example "--use_fast_math" or "--device-debug". If with_debug is true, the compilation log is included even in case of compilation success (see compilation_log).

NOTE: compile_to_ptx prepends the CUDA include path to options, so you don't need to.

val string_from_ptx : compile_to_ptx_result -> string

The stored PTX (i.e. NVIDIA assembly language) source, see nvrtcGetPTX.

val compilation_log : compile_to_ptx_result -> string option

The stored side output of the compilation, see nvrtcGetProgramLog.