ptx.cuh File Reference

Wrapper functions for single PTX instructions — using inline PTX assembly — which are not already available in the official CUDA includes. More...

#include "ptx/special_registers.cuh"
#include "ptx/miscellany.cuh"
#include "ptx/video_instructions.cuh"

Namespaces
	kat::ptx
	Code exposing CUDA's PTX intermediate representation instructions to C++ code.

Detailed Description

Wrapper functions for single PTX instructions — using inline PTX assembly — which are not already available in the official CUDA includes.

CUDA provides many "intrinsics" functions, which wrap single PTX instructions, e.g. __ldg or __funnelshift_l from sm_32_intrinsics.h. But - CUDA doesn't provide such functions for all of the PTX instruction set. The files included from this master-include contain such single-line assembly wrapper functions for different categories of missing PTX instructions.

Note

Unlike on_device/builtins.cuh, functions here are not templated, and do not necessarily have the same name for different parameter types. on_device/builtins.cuh functions do use PTX wrapper functions as their implementation.