olb::gpu::cuda::device Namespace Reference

Basic wrappers of common CUDA functions. More...

Classes
class	Stream
	Basic wrapper for device stream. More...
class	unique_ptr
	Managed pointer for device-side memory. More...

Functions
int	getCount ()
	Return number of available devices.
void	check ()
	Check errors.
void	synchronize ()
	Synchronize device.
int	get ()
	Get current device.
void	copyToHost (void *src, void *dst, std::size_t count)
	Copy data from device to host.
void	copyToDevice (void *src, void *dst, std::size_t count)
	Copy data from host to device.
template<typename T >
T *	malloc (std::size_t size)
	Allocate data on device.
std::size_t	getDevicePageSize ()
	Returns device memory page size.
template<typename T >
std::size_t	getPageAlignedCount (std::size_t count)
	Returns count rounded up to be a multiple of getDevicePageSize
void	asyncCopyToHost (Stream &stream, void *src, void *dst, std::size_t count)
	Copy data from device to host (async)
void	asyncCopyToDevice (Stream &stream, void *src, void *dst, std::size_t count)
	Copy data from host to device (async)
void	check (CUresult result)
	Check CUDA driver errors.

Detailed Description

Basic wrappers of common CUDA functions.

Function Documentation

asyncCopyToDevice()

void olb::gpu::cuda::device::asyncCopyToDevice	(	Stream &	stream,
		void *	src,
		void *	dst,
		std::size_t	count )

Copy data from host to device (async)

Definition at line 145 of file device.hh.

                                                                              {
  cudaMemcpyAsync(dst, src, count, cudaMemcpyHostToDevice, stream.get());
}

References olb::gpu::cuda::device::Stream::get().

Here is the call graph for this function:

Here is the caller graph for this function:

asyncCopyToHost()

void olb::gpu::cuda::device::asyncCopyToHost	(	Stream &	stream,
		void *	src,
		void *	dst,
		std::size_t	count )

Copy data from device to host (async)

Definition at line 141 of file device.hh.

                                                                            {
  cudaMemcpyAsync(dst, src, count, cudaMemcpyDeviceToHost, stream.get());
}

References olb::gpu::cuda::device::Stream::get().

Here is the call graph for this function:

Here is the caller graph for this function:

check() [1/2]

void olb::gpu::cuda::device::check

(

)

Check errors.

Definition at line 48 of file device.hh.

             {
  cudaError_t error = cudaGetLastError();
  if (error != cudaSuccess) {
    throw std::runtime_error(cudaGetErrorString(error));
  }
}

Here is the caller graph for this function:

check() [2/2]

void olb::gpu::cuda::device::check

(

CUresult

result

)

Check CUDA driver errors.

Definition at line 56 of file device.hh.

                            {
  if (result != CUDA_SUCCESS) {
    const char* description{};
    cuGetErrorString(result, &description);
    throw std::runtime_error(std::string(description));
  }
}

copyToDevice()

void olb::gpu::cuda::device::copyToDevice	(	void *	src,
		void *	dst,
		std::size_t	count )

Copy data from host to device.

Definition at line 83 of file device.hh.

                                                         {
  cudaMemcpy(dst, src, count, cudaMemcpyHostToDevice);
  check();
}

References check().

Here is the call graph for this function:

Here is the caller graph for this function:

copyToHost()

void olb::gpu::cuda::device::copyToHost	(	void *	src,
		void *	dst,
		std::size_t	count )

Copy data from device to host.

Definition at line 78 of file device.hh.

                                                       {
  cudaMemcpy(dst, src, count, cudaMemcpyDeviceToHost);
  check();
}

References check().

Here is the call graph for this function:

Here is the caller graph for this function:

get()

int olb::gpu::cuda::device::get

(

)

Get current device.

Definition at line 71 of file device.hh.

          {
  int device{};
  cudaGetDevice(&device);
  check();
  return device;
}

References check().

Here is the call graph for this function:

Here is the caller graph for this function:

getCount()

int olb::gpu::cuda::device::getCount

(

)

Return number of available devices.

Definition at line 42 of file device.hh.

               {
  int devices{};
  cudaGetDeviceCount(&devices);
  return devices;
}

Here is the caller graph for this function:

getDevicePageSize()

std::size_t olb::gpu::cuda::device::getDevicePageSize

(

)

Returns device memory page size.

Definition at line 96 of file device.hh.

                              {
  std::size_t granularity = 0;
  CUmemAllocationProp prop = {};
  prop.type = CU_MEM_ALLOCATION_TYPE_PINNED;
  prop.location.type = CU_MEM_LOCATION_TYPE_DEVICE;
  prop.location.id = device::get();
  check(cuMemGetAllocationGranularity(&granularity, &prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM));
  return granularity;
}

References check(), and get().

Here is the call graph for this function:

Here is the caller graph for this function:

getPageAlignedCount()

template<typename T >

std::size_t olb::gpu::cuda::device::getPageAlignedCount

(

std::size_t

count

)

Returns count rounded up to be a multiple of getDevicePageSize

Definition at line 64 of file device.h.

{
  const std::size_t page_size = getDevicePageSize();
  const std::size_t size = ((count * sizeof(T) - 1) / page_size + 1) * page_size;
  const std::size_t volume = size / sizeof(T);
 
  if (size % page_size != 0) {
    throw std::invalid_argument("Buffer size must be multiple of PAGE_SIZE");
  }
 
  return volume;
};

References getDevicePageSize().

Here is the call graph for this function:

malloc()

template<typename T >

T * olb::gpu::cuda::device::malloc

(

std::size_t

size

)

Allocate data on device.

Definition at line 89 of file device.hh.

                          {
  T* ptr{};
  cudaMalloc(&ptr, size*sizeof(T));
  check();
  return ptr;
}

synchronize()

void olb::gpu::cuda::device::synchronize

(

)

Synchronize device.

Definition at line 64 of file device.hh.

                   {
  if (getCount() > 0) {
    cudaDeviceSynchronize();
    check();
  }
}

References check(), and getCount().

Here is the call graph for this function:

Here is the caller graph for this function: