sdfg::cuda Namespace Reference

Namespaces
namespace	blas
namespace	tensor
namespace	stdlib
namespace	tiles

Classes
class	CudaLibraryNodeRewriter
class	CudaLibraryNodeTransferExtractionVisitor
class	CUDALanguageExtension
class	ScheduleType_CUDA_Offload
	CUDA schedule type inheriting shared GPU functionality Provides CUDA-specific value() and default block size (32 for warp size). More...
class	ScheduleType_CUDA
	CUDA schedule type inheriting shared GPU functionality Provides CUDA-specific value() and default block size (32 for warp size). More...
class	CUDADataOffloadingNode
class	CUDADataOffloadingNodeDispatcher
class	CUDADataOffloadingNodeSerializer
class	CUDAMapDispatcher
class	CUDAOffloadDispatcherStrategy
	CUDA specialization of gpu::GPUOffloadMapDispatcher. More...
class	CUDAReduceDispatcher
	Atomics-based CUDA code generator for structured_control_flow::Reduce. More...
class	CUBLASDataTransferExtraction
class	CUDAOffloadTransform
class	CUDASoftmaxDataTransferExtraction
class	CUDAStdlibDataTransferExtraction
class	CUDATransform
	Offloads a top-level loop to a CUDA kernel (X grid dimension). More...

Typedefs
typedef passes::VisitorPass< CudaLibraryNodeRewriter >	CudaLibraryNodeRewriterPass
typedef passes::VisitorPass< CudaLibraryNodeTransferExtractionVisitor >	CudaLibraryNodeTransferExtractionPass
using	CUDADimension = gpu::GPUDimension

Functions
void	cuda_error_checking (codegen::PrettyPrinter &stream, const codegen::LanguageExtension &language_extension, const std::string &status_variable)
bool	do_cuda_error_checking ()
void	check_cuda_kernel_launch_errors (codegen::PrettyPrinter &stream, const codegen::LanguageExtension &language_extension, bool instrumented)
data_flow::LibraryNodeCode	LibraryNodeType_CUDA_Offloading ("CUDAOffloading")
void	register_cuda_plugin (plugins::Context &context)
void	register_cuda_plugin ()

Variables
std::string	CUDA_DEVICE_PREFIX = "__daisy_cuda_"
constexpr int	CUDA_WARP_SIZE = 32
data_flow::ImplementationType	ImplementationType_CUDA {"CUDA"}
	Generic CUDA implementation.
data_flow::ImplementationType	ImplementationType_CUDAWithTransfers {"CUDAWithTransfers"}
	CUDA implementation with automatic memory transfers Used for CUBLAS, memset, and other CUDA-accelerated library nodes.
data_flow::ImplementationType	ImplementationType_CUDAWithoutTransfers {"CUDAWithoutTransfers"}
	CUDA implementation without memory transfers Used for CUBLAS, memset, and other CUDA-accelerated library nodes assuming data is already on GPU.
codegen::TargetType	TargetType_CUDA_Offload {ScheduleType_CUDA_Offload::value()}
codegen::TargetType	TargetType_CUDA {ScheduleType_CUDA::value()}

Typedef Documentation

CUDADimension

using sdfg::cuda::CUDADimension = gpu::GPUDimension

CudaLibraryNodeRewriterPass

typedef passes::VisitorPass<CudaLibraryNodeRewriter> sdfg::cuda::CudaLibraryNodeRewriterPass

CudaLibraryNodeTransferExtractionPass

typedef passes::VisitorPass<CudaLibraryNodeTransferExtractionVisitor> sdfg::cuda::CudaLibraryNodeTransferExtractionPass

Function Documentation

check_cuda_kernel_launch_errors()

void sdfg::cuda::check_cuda_kernel_launch_errors	(	codegen::PrettyPrinter &	stream,
		const codegen::LanguageExtension &	language_extension,
		bool	instrumented )

cuda_error_checking()

void sdfg::cuda::cuda_error_checking	(	codegen::PrettyPrinter &	stream,
		const codegen::LanguageExtension &	language_extension,
		const std::string &	status_variable )

do_cuda_error_checking()

bool sdfg::cuda::do_cuda_error_checking

(

)

LibraryNodeType_CUDA_Offloading()

data_flow::LibraryNodeCode sdfg::cuda::LibraryNodeType_CUDA_Offloading ( "CUDAOffloading" )

inline

register_cuda_plugin() [1/2]

void sdfg::cuda::register_cuda_plugin ( )

inline

Deprecated

use the variant with explicit context

register_cuda_plugin() [2/2]

void sdfg::cuda::register_cuda_plugin

(

plugins::Context &

context

)

Variable Documentation

CUDA_DEVICE_PREFIX

std::string sdfg::cuda::CUDA_DEVICE_PREFIX = "__daisy_cuda_"

inline

CUDA_WARP_SIZE

int sdfg::cuda::CUDA_WARP_SIZE = 32

constexpr

ImplementationType_CUDA

data_flow::ImplementationType sdfg::cuda::ImplementationType_CUDA {"CUDA"}

inline

Generic CUDA implementation.

ImplementationType_CUDAWithoutTransfers

data_flow::ImplementationType sdfg::cuda::ImplementationType_CUDAWithoutTransfers {"CUDAWithoutTransfers"}

inline

CUDA implementation without memory transfers Used for CUBLAS, memset, and other CUDA-accelerated library nodes assuming data is already on GPU.

ImplementationType_CUDAWithTransfers

data_flow::ImplementationType sdfg::cuda::ImplementationType_CUDAWithTransfers {"CUDAWithTransfers"}

inline

CUDA implementation with automatic memory transfers Used for CUBLAS, memset, and other CUDA-accelerated library nodes.

TargetType_CUDA

codegen::TargetType sdfg::cuda::TargetType_CUDA {ScheduleType_CUDA::value()}

inline

TargetType_CUDA_Offload

codegen::TargetType sdfg::cuda::TargetType_CUDA_Offload {ScheduleType_CUDA_Offload::value()}

inline