#include <AMReX_GpuDevice.H>

Static Public Member Functions
static void	Initialize (bool minimal, int a_device_id)

static void	Finalize ()

static gpuStream_t	gpuStream () noexcept

static cudaStream_t	cudaStream () noexcept

static int	numGpuStreams () noexcept

static void	setStreamIndex (int idx) noexcept

static void	resetStreamIndex () noexcept

static int	streamIndex (gpuStream_t s=gpuStream()) noexcept

static gpuStream_t	setStream (gpuStream_t s) noexcept

static gpuStream_t	resetStream () noexcept

static int	deviceId () noexcept

static int	numDevicesUsed () noexcept

static int	numDevicePartners () noexcept

static void	synchronize () noexcept

static void	streamSynchronize () noexcept

static void	streamSynchronizeAll () noexcept

static void	freeAsync (Arena arena, void mem) noexcept

static void	mem_advise_set_preferred (void *p, std::size_t sz, int device)

static void	mem_advise_set_readonly (void *p, std::size_t sz)

static void	setNumThreadsMin (int nx, int ny, int nz) noexcept

static void	n_threads_and_blocks (const Long N, dim3 &numBlocks, dim3 &numThreads) noexcept

static void	c_comps_threads_and_blocks (const int lo, const int hi, const int comps, dim3 &numBlocks, dim3 &numThreads) noexcept

static void	c_threads_and_blocks (const int lo, const int hi, dim3 &numBlocks, dim3 &numThreads) noexcept

static void	grid_stride_threads_and_blocks (dim3 &numBlocks, dim3 &numThreads) noexcept

static std::size_t	totalGlobalMem () noexcept

static std::size_t	sharedMemPerBlock () noexcept

static int	numMultiProcessors () noexcept

static int	maxThreadsPerMultiProcessor () noexcept

static int	maxThreadsPerBlock () noexcept

static int	maxThreadsPerBlock (int dir) noexcept

static int	maxBlocksPerGrid (int dir) noexcept

static std::string	deviceName () noexcept

static int	devicePropMajor () noexcept

static int	devicePropMinor () noexcept

static std::string	deviceVendor () noexcept

static std::size_t	freeMemAvailable ()

static void	profilerStart ()

static void	profilerStop ()

static int	memoryPoolsSupported () noexcept

static unsigned int	maxBlocksPerLaunch () noexcept

Static Public Attributes
static constexpr int	warp_size = 32

Static Private Member Functions
static void	initialize_gpu (bool minimal)

Private Attributes
friend	StreamManager

Static Private Attributes
static int	device_id = 0

static int	num_devices_used = 0

static int	num_device_partners = 1

static int	verbose = 0

static int	max_gpu_streams = 4

static dim3	numThreadsMin = dim3(1, 1, 1)

static dim3	numBlocksOverride = dim3(0, 0, 0)

static dim3	numThreadsOverride = dim3(0, 0, 0)

static Vector< StreamManager >	gpu_stream_pool

static Vector< int >	gpu_stream_index

static gpuDeviceProp_t	device_prop

static int	memory_pools_supported = 0

static unsigned int	max_blocks_per_launch = 2560

Member Function Documentation

◆ c_comps_threads_and_blocks()

void amrex::Gpu::Device::c_comps_threads_and_blocks	(	const int *	lo,
		const int *	hi,
		const int	comps,
		dim3 &	numBlocks,
		dim3 &	numThreads
	)

staticnoexcept

◆ c_threads_and_blocks()

void amrex::Gpu::Device::c_threads_and_blocks	(	const int *	lo,
		const int *	hi,
		dim3 &	numBlocks,
		dim3 &	numThreads
	)

staticnoexcept

◆ cudaStream()

static cudaStream_t amrex::Gpu::Device::cudaStream ( )

inlinestaticnoexcept

for backward compatibility

◆ deviceId()

int amrex::Gpu::Device::deviceId ( )

staticnoexcept

◆ deviceName()

static std::string amrex::Gpu::Device::deviceName ( )

inlinestaticnoexcept

◆ devicePropMajor()

static int amrex::Gpu::Device::devicePropMajor ( )

inlinestaticnoexcept

◆ devicePropMinor()

static int amrex::Gpu::Device::devicePropMinor ( )

inlinestaticnoexcept

◆ deviceVendor()

static std::string amrex::Gpu::Device::deviceVendor ( )

inlinestaticnoexcept

◆ Finalize()

void amrex::Gpu::Device::Finalize ( )

static

◆ freeAsync()

void amrex::Gpu::Device::freeAsync	(	Arena *	arena,
		void *	mem
	)

staticnoexcept

◆ freeMemAvailable()

std::size_t amrex::Gpu::Device::freeMemAvailable ( )

static

◆ gpuStream()

static gpuStream_t amrex::Gpu::Device::gpuStream ( )

inlinestaticnoexcept

◆ grid_stride_threads_and_blocks()

void amrex::Gpu::Device::grid_stride_threads_and_blocks	(	dim3 &	numBlocks,
		dim3 &	numThreads
	)

staticnoexcept

◆ Initialize()

void amrex::Gpu::Device::Initialize	(	bool	minimal,
		int	a_device_id
	)

static

◆ initialize_gpu()

void amrex::Gpu::Device::initialize_gpu ( bool minimal )

staticprivate

◆ maxBlocksPerGrid()

static int amrex::Gpu::Device::maxBlocksPerGrid ( int dir )

inlinestaticnoexcept

◆ maxBlocksPerLaunch()

static unsigned int amrex::Gpu::Device::maxBlocksPerLaunch ( )

inlinestaticnoexcept

◆ maxThreadsPerBlock() [1/2]

static int amrex::Gpu::Device::maxThreadsPerBlock ( )

inlinestaticnoexcept

◆ maxThreadsPerBlock() [2/2]

static int amrex::Gpu::Device::maxThreadsPerBlock ( int dir )

inlinestaticnoexcept

◆ maxThreadsPerMultiProcessor()

static int amrex::Gpu::Device::maxThreadsPerMultiProcessor ( )

inlinestaticnoexcept

◆ mem_advise_set_preferred()

void amrex::Gpu::Device::mem_advise_set_preferred	(	void *	p,
		std::size_t	sz,
		int	device
	)

static

◆ mem_advise_set_readonly()

void amrex::Gpu::Device::mem_advise_set_readonly	(	void *	p,
		std::size_t	sz
	)

static

◆ memoryPoolsSupported()

static int amrex::Gpu::Device::memoryPoolsSupported ( )

inlinestaticnoexcept

◆ n_threads_and_blocks()

void amrex::Gpu::Device::n_threads_and_blocks	(	const Long	N,
		dim3 &	numBlocks,
		dim3 &	numThreads
	)

staticnoexcept

◆ numDevicePartners()

int amrex::Gpu::Device::numDevicePartners ( )

staticnoexcept

◆ numDevicesUsed()

int amrex::Gpu::Device::numDevicesUsed ( )

staticnoexcept

◆ numGpuStreams()

static int amrex::Gpu::Device::numGpuStreams ( )

inlinestaticnoexcept

◆ numMultiProcessors()

static int amrex::Gpu::Device::numMultiProcessors ( )

inlinestaticnoexcept

◆ profilerStart()

void amrex::Gpu::Device::profilerStart ( )

static

◆ profilerStop()

void amrex::Gpu::Device::profilerStop ( )

static

◆ resetStream()

gpuStream_t amrex::Gpu::Device::resetStream ( )

staticnoexcept

◆ resetStreamIndex()

static void amrex::Gpu::Device::resetStreamIndex ( )

inlinestaticnoexcept

◆ setNumThreadsMin()

void amrex::Gpu::Device::setNumThreadsMin	(	int	nx,
		int	ny,
		int	nz
	)

staticnoexcept

◆ setStream()

gpuStream_t amrex::Gpu::Device::setStream ( gpuStream_t s )

staticnoexcept

◆ setStreamIndex()

void amrex::Gpu::Device::setStreamIndex ( int idx )

staticnoexcept

◆ sharedMemPerBlock()

static std::size_t amrex::Gpu::Device::sharedMemPerBlock ( )

inlinestaticnoexcept

◆ streamIndex()

int amrex::Gpu::Device::streamIndex ( gpuStream_t s = gpuStream() )

staticnoexcept

◆ streamSynchronize()

void amrex::Gpu::Device::streamSynchronize ( )

staticnoexcept

Halt execution of code until the current AMReX GPU stream has finished processing all previously requested tasks.

◆ streamSynchronizeAll()

void amrex::Gpu::Device::streamSynchronizeAll ( )

staticnoexcept

Halt execution of code until all AMReX GPU streams have finished processing all previously requested tasks.

◆ synchronize()

void amrex::Gpu::Device::synchronize ( )

staticnoexcept

Halt execution of code until GPU has finished processing all previously requested tasks.

◆ totalGlobalMem()

static std::size_t amrex::Gpu::Device::totalGlobalMem ( )

inlinestaticnoexcept

Member Data Documentation

◆ device_id

int amrex::Gpu::Device::device_id = 0

staticprivate

◆ device_prop

gpuDeviceProp_t amrex::Gpu::Device::device_prop

staticprivate

◆ gpu_stream_index

Vector< int > amrex::Gpu::Device::gpu_stream_index

staticprivate

◆ gpu_stream_pool

Vector< StreamManager > amrex::Gpu::Device::gpu_stream_pool

staticprivate

◆ max_blocks_per_launch

unsigned int amrex::Gpu::Device::max_blocks_per_launch = 2560

staticprivate

◆ max_gpu_streams

int amrex::Gpu::Device::max_gpu_streams = 4

staticprivate

◆ memory_pools_supported

int amrex::Gpu::Device::memory_pools_supported = 0

staticprivate

◆ num_device_partners

int amrex::Gpu::Device::num_device_partners = 1

staticprivate

◆ num_devices_used

int amrex::Gpu::Device::num_devices_used = 0

staticprivate

◆ numBlocksOverride

dim3 amrex::Gpu::Device::numBlocksOverride = dim3(0, 0, 0)

staticprivate

◆ numThreadsMin

dim3 amrex::Gpu::Device::numThreadsMin = dim3(1, 1, 1)

staticprivate

◆ numThreadsOverride

dim3 amrex::Gpu::Device::numThreadsOverride = dim3(0, 0, 0)

staticprivate

◆ StreamManager

friend amrex::Gpu::Device::StreamManager

private

◆ verbose

int amrex::Gpu::Device::verbose = 0

staticprivate

◆ warp_size

constexpr int amrex::Gpu::Device::warp_size = 32

staticconstexpr

The documentation for this class was generated from the following files:

/home/runner/work/amrex/amrex/Src/Base/AMReX_GpuDevice.H
/home/runner/work/amrex/amrex/Src/Base/AMReX_GpuDevice.cpp

Static Public Member Functions

Static Public Attributes

Static Private Member Functions

Private Attributes

Static Private Attributes

Member Function Documentation

◆ c_comps_threads_and_blocks()

◆ c_threads_and_blocks()

◆ cudaStream()

◆ deviceId()

◆ deviceName()

◆ devicePropMajor()

◆ devicePropMinor()

◆ deviceVendor()

◆ Finalize()

◆ freeAsync()

◆ freeMemAvailable()

◆ gpuStream()

◆ grid_stride_threads_and_blocks()

◆ Initialize()

◆ initialize_gpu()

◆ maxBlocksPerGrid()

◆ maxBlocksPerLaunch()

◆ maxThreadsPerBlock() [1/2]

◆ maxThreadsPerBlock() [2/2]

◆ maxThreadsPerMultiProcessor()

◆ mem_advise_set_preferred()

◆ mem_advise_set_readonly()

◆ memoryPoolsSupported()

◆ n_threads_and_blocks()

◆ numDevicePartners()

◆ numDevicesUsed()

◆ numGpuStreams()

◆ numMultiProcessors()

◆ profilerStart()

◆ profilerStop()

◆ resetStream()

◆ resetStreamIndex()

◆ setNumThreadsMin()

◆ setStream()

◆ setStreamIndex()

◆ sharedMemPerBlock()

◆ streamIndex()

◆ streamSynchronize()

◆ streamSynchronizeAll()

◆ synchronize()

◆ totalGlobalMem()

Member Data Documentation

◆ device_id

◆ device_prop

◆ gpu_stream_index

◆ gpu_stream_pool

◆ max_blocks_per_launch

◆ max_gpu_streams

◆ memory_pools_supported

◆ num_device_partners

◆ num_devices_used

◆ numBlocksOverride

◆ numThreadsMin

◆ numThreadsOverride

◆ StreamManager

◆ verbose

◆ warp_size