launch

Overloads

launch(__submitter, __conf, _ExpArgs...), __args)

template<typename ..._ExpArgs, typename ..._ActArgs, typename _Submitter, typename ..._Config, typename _Dimensions>
inline auto cuda::launch(

_Submitter &&__submitter,

const kernel_config<_Dimensions, _Config...> &__conf,

void (*__kernel)(kernel_config<_Dimensions, _Config...>, _ExpArgs...),

_ActArgs&&... __args

)

Launch a kernel function with specified configuration and arguments.

Launches a kernel function on the specified stream and with specified configuration. Kernel function is a function with global annotation. Function might or might not accept the configuration as its first argument.

Snippet

#include <cstdio>
#include <cuda/launch>

template <typename Configuration>
__global__ void kernel(Configuration conf, unsigned int thread_to_print) {
    if (conf.dims.rank(cuda::thread, cuda::grid) == thread_to_print) {
        printf("Hello from the GPU\n");
    }
}

void launch_kernel(cuda::stream_ref stream) {
    auto dims    = cuda::make_hierarchy(cuda::block_dims<128>(),
    cuda::grid_dims(4)); auto config = cuda::make_config(dims,
    cuda::launch_cooperative());

    cuda::launch(stream, config, kernel<decltype(config)>, 42);
}

Parameters:

• stream – cuda::stream_ref to launch the kernel into

• conf – configuration for this launch

• kernel – kernel function to be launched

• args – arguments to be passed into the kernel function

launch(__submitter, __conf, void(*__kernel)(_ExpArgs...), __args)

template<typename ..._ExpArgs, typename ..._ActArgs, typename _Submitter, typename ..._Config, typename _Dimensions>
inline auto cuda::launch(

_Submitter &&__submitter,

const kernel_config<_Dimensions, _Config...> &__conf,

void (*__kernel)(_ExpArgs...),

_ActArgs&&... __args

)

Launch a kernel function with specified configuration and arguments.

Launches a kernel function on the specified stream and with specified configuration. Kernel function is a function with global annotation. Function might or might not accept the configuration as its first argument.

Snippet

#include <cstdio>
#include <cuda/launch>

template <typename Configuration>
__global__ void kernel(Configuration conf, unsigned int thread_to_print) {
    if (conf.dims.rank(cuda::thread, cuda::grid) == thread_to_print) {
        printf("Hello from the GPU\n");
    }
}

void launch_kernel(cuda::stream_ref stream) {
    auto dims    = cuda::make_hierarchy(cuda::block_dims<128>(),
    cuda::grid_dims(4)); auto config = cuda::make_config(dims,
    cuda::launch_cooperative());

    cuda::launch(stream, config, kernel<decltype(config)>, 42);
}

Parameters:

• __stream – cuda::stream_ref to launch the kernel into

• __conf – configuration for this launch

• __kernel – kernel function to be launched

• __args – arguments to be passed into the kernel function