profiling.hpp File Reference

wrappers for the CUDA profiler API functions, and convenience RAII classes for profiler-output-marked time range and points. More...

#include "../api/types.hpp"
#include "../api/error.hpp"
#include "../api/current_context.hpp"
#include "../api/stream.hpp"
#include "../api/event.hpp"
#include "../api/device.hpp"
#include "../api/multi_wrapper_impls/context.hpp"
#include <cudaProfiler.h>
#include <nvToolsExt.h>
#include <nvToolsExtCuda.h>
#include <mutex>
#include <cstdint>
#include <string>
#include <thread>

Include dependency graph for profiling.hpp:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	cuda::profiling::color_t
	An RGB colorspace color value, with potential transparency, which may be used to color elements in timelines or other graphical displays of profiling information. More...
class	cuda::profiling::scoped_range_marker
	A RAII/CADRe class whose scope of existence is reflected as a range in the profiler. More...
class	cuda::profiling::scope
	A class to instantiate in the part of your application which does any work you intend to use the CUDA profiler to profile. More...

Namespaces
	cuda
	Definitions and functionality wrapping CUDA APIs.
	cuda::profiling
	Interaction with NVIDIA's profiler, particularly tagging, marking and indications of entities it will pick up and register/display.
	cuda::profiling::range
	definitions related to profiled ranges and the range_t class
	cuda::profiling::mark
	definitions related to marking individual time points in the profiler timeline

Macros
#define	CUDA_API_WRAPPERS_PROFILING_HPP_
#define	profile_this_scope()   ::cuda::profiling::scope cuda_profiling_scope_{};

Typedefs
using	cuda::profiling::range::handle_t = nvtxRangeId_t
	The raw handle of a CUDA profiling range.

Enumerations
enum	cuda::profiling::range::type_t {   unspecified,   kernel,   pci_express_transfer }
	Types of profiled ranges we recognize. More...

Functions
template<typename CharT >
void	cuda::profiling::mark::point (const CharT *description, color_t color=color_t::Black())
	Mark a single point on the profiler timeline, giving it also a color and some descriptive text.
template<typename CharT >
range::handle_t	cuda::profiling::mark::range_start (const CharT *description, range::type_t type=range::type_t::unspecified, color_t color=color_t::LightRed())
	Mark the beginning of a range on the profiler timeline, giving it also a color and some descriptive text. More...
void	cuda::profiling::mark::range_end (range::handle_t range_handle)
	Mark the end of a range, using the handle obtained when previously marking its beginning. More...
void	cuda::profiling::start ()
	Start CUDA profiling for the current process.
void	cuda::profiling::stop ()
	Stop CUDA profiling for the current process.
template<typename CharT >
void	cuda::profiling::name (const ::std::thread &host_thread, const CharT *name)
	Have the profiler refer to a given host thread, using a specified string identifier (rather than its numeric ID). More...
template<typename CharT >
void	cuda::profiling::name_this_thread (const CharT *name)
	Have the profiler refer to the current thread using a specified string identifier (rather than its numeric ID). More...
template<typename CharT >
void	cuda::profiling::name (const stream_t &stream, const CharT *name)
	Have the profile assign a name to a certain stream.
template<typename CharT >
void	cuda::profiling::name (const event_t &event, const CharT *name)
	Have the profile assign a name to a certain event.
template<typename CharT >
void	cuda::profiling::name (const device_t &device, const CharT *name)
	Have the profile assign a name to a certain CUDA device.

Detailed Description

wrappers for the CUDA profiler API functions, and convenience RAII classes for profiler-output-marked time range and points.