KernelParamCalculator.device.hh File Reference

#include <cstddef>
#include <string_view>
#include "corecel/DeviceRuntimeApi.hh"
#include "corecel/Assert.hh"
#include "corecel/Macros.hh"
#include "corecel/OpaqueId.hh"
#include "corecel/Types.hh"
#include "corecel/math/Algorithms.hh"
#include "Device.hh"
#include "KernelAttributes.hh"
#include "ThreadId.hh"

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

Classes
class	celeritas::KernelParamCalculator
	Kernel management helper functions. More...
struct	celeritas::KernelParamCalculator::LaunchParams
	Parameters needed for a CUDA launch call. More...

Macros
#define	CELER_LAUNCH_KERNEL(NAME, THREADS, STREAM, ...)
	Create a kernel param calculator with the given kernel, assuming the function itself has a _kernel suffix, and launch with the given block/thread sizes and arguments list.
#define	CELER_LAUNCH_KERNEL_TEMPLATE_1(NAME, T1, THREADS, STREAM, ...)
	Create a kernel param calculator with the given kernel with one template parameter, assuming the unction itself has a _kernel suffix, and launch with the given block/thread sizes and arguments list.
#define	CELER_LAUNCH_KERNEL_IMPL(KERNEL, GRID, BLOCK, SHARED, STREAM, ...)

Macro Definition Documentation

CELER_LAUNCH_KERNEL

#define CELER_LAUNCH_KERNEL	(		NAME,
			THREADS,
			STREAM,
			...
	)

Value:

    do                                                                       \
    {                                                                        \
        static const ::celeritas::KernelParamCalculator calc_launch_params_( \
            #NAME, NAME##_kernel);                                           \
        auto grid_ = calc_launch_params_(THREADS);                           \
                                                                             \
        CELER_LAUNCH_KERNEL_IMPL(NAME##_kernel,                              \
                                 grid_.blocks_per_grid,                      \
                                 grid_.threads_per_block,                    \
                                 0,                                          \
                                 STREAM,                                     \
                                 __VA_ARGS__);                               \
        CELER_DEVICE_API_CALL(PeekAtLastError());                            \
    } while (0)

Create a kernel param calculator with the given kernel, assuming the function itself has a _kernel suffix, and launch with the given block/thread sizes and arguments list.

CELER_LAUNCH_KERNEL_IMPL

#define CELER_LAUNCH_KERNEL_IMPL	(		KERNEL,
			GRID,
			BLOCK,
			SHARED,
			STREAM,
			...
	)

Value:

        CELER_NOT_CONFIGURED("CUDA or HIP");                                   \
        CELER_DISCARD(GRID)                                                    \
        CELER_DISCARD(KERNEL)                                                  \
        CELER_DISCARD(__VA_ARGS__);

CELER_LAUNCH_KERNEL_TEMPLATE_1

#define CELER_LAUNCH_KERNEL_TEMPLATE_1	(		NAME,
			T1,
			THREADS,
			STREAM,
			...
	)

Value:

    do                                                                       \
    {                                                                        \
        static const ::celeritas::KernelParamCalculator calc_launch_params_( \
            #NAME, NAME##_kernel<T1>);                                       \
        auto grid_ = calc_launch_params_(THREADS);                           \
                                                                             \
        CELER_LAUNCH_KERNEL_IMPL(NAME##_kernel<T1>,                          \
                                 grid_.blocks_per_grid,                      \
                                 grid_.threads_per_block,                    \
                                 0,                                          \
                                 STREAM,                                     \
                                 __VA_ARGS__);                               \
        CELER_DEVICE_API_CALL(PeekAtLastError());                            \
    } while (0)

Create a kernel param calculator with the given kernel with one template parameter, assuming the unction itself has a _kernel suffix, and launch with the given block/thread sizes and arguments list.