cuda_support.cpp File Reference

Implementation for inc/util/cuda_support.h More...

#include "inc/util/cuda_support.h"
#include "inc/util/macros.h"
#include "inc/util/pair.h"
#include "inc/util/scalars.h"
#include "inc/util/vectors.h"
#include <cuda_runtime.h>
#include <numeric>
#include "inc/util/cublas_support.h"
#include "inc/util/tensor_timer.h"

Include dependency graph for cuda_support.cpp:

Classes
class	syten::Cuda::CudaAllocator
	Cuda allocator class, uses the buddy system. More...

Namespaces
	syten
	Syten namespace.
	syten::Cuda
	Support functions (memory allocation etc.) for CUDA-based GPUs.

Macros
#define	cuda_handle_error(...)   cuda_handle_error_impl(__VA_ARGS__, SYTEN_STRINGIFY(__VA_ARGS__));
	Helper macro to handle CUDA errors, forwards to cuda_handle_error_impl() but first checks for any other errors and passes a stringified version of the call text into cuda_handle_error_impl() for pretty-printing. More...

Functions
std::uint16_t	syten::Cuda::allocator_get_max_size ()
	Returns the log2 of the maximal block size of the CUDA allocator. More...
std::uint16_t	syten::Cuda::allocator_get_min_size ()
	Returns the log2 of the minimal block size of the CUDA allocator. More...
void	syten::Cuda::allocator_print_status ()
	Prints the status of the CUDA allocator. More...
std::uint16_t	syten::Cuda::allocator_set_max_size (std::uint16_t sz)
	Sets the log2 of the maximal block size of the CUDA allocator. More...
std::uint16_t	syten::Cuda::allocator_set_min_size (std::uint16_t sz)
	Sets the log2 of the minimal block size of the CUDA allocator. More...
bool	syten::Cuda::cuda_compiled ()
	Returns true if CUDA support is compiled in. More...
bool	syten::Cuda::cuda_enabled ()
	Returns true if the list of allowed devices is not empty. More...
cudaError_t	syten::Cuda::cuda_handle_error_impl (cudaError_t err, std::string str, SourceLocation location=SourceLocation::current())
	Implementation for helper function to handle Cuda return values by throwing an assertion failure if the return value is not cudaSuccess. More...
cudaError_t	syten::Cuda::cuda_handle_error_impl (cudaError_t err, Vec< cudaError_t > acceptable_errors, std::string str, SourceLocation location=SourceLocation::current())
	Implementation for helper function to handle Cuda return values by throwing an assertion failure if the return value is neither cudaSuccess nor in the list of acceptable errors. More...
void	syten::Cuda::set_allowed_devices ([[maybe_unused]] Vec< std::int16_t > const &devices)
	Sets the allowed devices to the supplied list and enables inter-device memory access. More...
std::string	syten::Cuda::version ()
	Returns a version string describing the current CUDA version/compilation/enablement. More...
Allowed devices and allocation logic
std::int16_t	syten::Cuda::get_alloc_device ()
	Returns the device ID of the next allocation device to use. More...
Vec< std::int16_t > const &	syten::Cuda::get_allowed_devices ()
	Returns a list of allowed devices. More...
void	syten::Cuda::setup ()
	Sets up CUDA to allow all existing devices and generates the associated cuBLAS handles for the calling thread. More...
void	syten::Cuda::setup ([[maybe_unused]] Vec< std::int16_t > const &devices)
	Sets up CUDA to allow the specified devices and generates the associated cuBLAS handles for the calling thread. More...
Memory management functions.
CudaPtr< void >	syten::Cuda::alloc (std::size_t sz)
	Allocates sz bytes on the next CUDA allocation device. More...
CudaPtr< void >	syten::Cuda::alloc_on_device (std::size_t sz,[[maybe_unused]] std::int16_t device)
	Allocates sz bytes on the specified CUDA allocation device. More...
void	syten::Cuda::free (CudaPtr< void > ptr)
	Frees the allocation pointed to by ptr. More...
Copying objects and arrays.
void	syten::Cuda::copy (CudaPtr< const char > src, CudaPtr< char > dst, std::size_t num)
	Copies num bytes from src to dst. More...
void	syten::Cuda::copy (CudaPtr< const char > src, CudaPtr< char > dst, std::size_t num,[[maybe_unused]] CudaStream const &str)
	Copies num bytes from src to dst within stream str. More...
Device management and synchronisation.
Pair< Size, Size >	syten::Cuda::mem_status ()
	Returns a pair of free and total device memory. More...
void	syten::Cuda::select_device ([[maybe_unused]] std::int16_t device)
	Selects the specified device for the current thread. More...
void	syten::Cuda::synchronise ()
	Synchronizes with the current device. More...

Variables
static CudaAllocator	syten::Cuda::allocator
	Our CudaAllocator object, handling a buddy free list system. More...
static Vec< std::int16_t >	syten::Cuda::allowed_devices = {}
	Storage for the currently allowed devices. More...

Detailed Description

Implementation for inc/util/cuda_support.h