#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 void	setExternalStream (gpuStream_t s)
static void	resetExternalStream (ExternalStreamSync sync_stream=ExternalStreamSync::Yes) noexcept
static bool	usingExternalStream () noexcept
static int	deviceId () noexcept
static int	numDevicesUsed () noexcept
static int	numDevicePartners () noexcept
static void	synchronize ()
static void	streamSynchronize () noexcept
static void	streamSynchronize (gpuStream_t s) noexcept
static void	streamSynchronizeActive () noexcept
static void	streamSynchronizeAll () noexcept
static void	freeAsync (Arena arena, void mem) noexcept
static void	streamOrderedFreeAsync (Arena arena, void mem, gpuStream_t stream) noexcept
static bool	clearFreeAsyncBuffer () 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 bool	isIntegrated () noexcept
	Whether the GPU shares its memory with the host (e.g., an APU).
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	managedMemorySupported () noexcept
static int	memoryPoolsSupported () noexcept
static unsigned int	maxBlocksPerLaunch () noexcept

Static Public Attributes
static constexpr int	warp_size = 32

Friends
struct	ExternalGpuStreamRegion

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

◆ clearFreeAsyncBuffer()

bool amrex::Gpu::Device::clearFreeAsyncBuffer ( )

staticnoexcept

◆ cudaStream()

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

inlinestaticnoexcept

for backward compatibility

◆ deviceId()

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

staticnoexcept

◆ deviceName()

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

inlinestaticnoexcept

◆ devicePropMajor()

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

inlinestaticnoexcept

◆ devicePropMinor()

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

inlinestaticnoexcept

◆ deviceVendor()

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()

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

◆ isIntegrated()

bool amrex::Gpu::Device::isIntegrated ( )

inlinestaticnoexcept

Whether the GPU shares its memory with the host (e.g., an APU).

◆ managedMemorySupported()

int amrex::Gpu::Device::managedMemorySupported ( )

inlinestaticnoexcept

◆ maxBlocksPerGrid()

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

inlinestaticnoexcept

◆ maxBlocksPerLaunch()

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

inlinestaticnoexcept

◆ maxThreadsPerBlock() [1/2]

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

inlinestaticnoexcept

◆ maxThreadsPerBlock() [2/2]

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

inlinestaticnoexcept

◆ maxThreadsPerMultiProcessor()

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()

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()

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

inlinestaticnoexcept

◆ numMultiProcessors()

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

inlinestaticnoexcept

◆ profilerStart()

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

static

◆ profilerStop()

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

static

◆ resetExternalStream()

void amrex::Gpu::Device::resetExternalStream ( ExternalStreamSync sync_stream = ExternalStreamSync::Yes )

staticnoexcept

Pop the current external stream and optionally synchronize it.

◆ resetStream()

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

staticnoexcept

◆ resetStreamIndex()

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

inlinestaticnoexcept

◆ setExternalStream()

void amrex::Gpu::Device::setExternalStream ( gpuStream_t s )

static

Override the current GPU stream with a user supplied stream.

◆ 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()

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

inlinestaticnoexcept

◆ streamIndex()

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

staticnoexcept

◆ streamOrderedFreeAsync()

void amrex::Gpu::Device::streamOrderedFreeAsync	(	Arena *	arena,
		void *	mem,
		gpuStream_t	stream )

staticnoexcept

◆ streamSynchronize() [1/2]

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

staticnoexcept

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

◆ streamSynchronize() [2/2]

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

staticnoexcept

◆ streamSynchronizeActive()

void amrex::Gpu::Device::streamSynchronizeActive ( )

staticnoexcept

Halt execution of code until all active AMReX GPU streams have finished processing all previously requested tasks. If we are in single stream region, only stream 0 is considered active. Otherwise all streams are active.

◆ 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 ( )

static

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

◆ totalGlobalMem()

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

inlinestaticnoexcept

◆ usingExternalStream()

bool amrex::Gpu::Device::usingExternalStream ( )

staticnoexcept

◆ ExternalGpuStreamRegion

friend struct ExternalGpuStreamRegion

friend

Member Data Documentation

◆ warp_size

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

Friends

Member Function Documentation

◆ c_comps_threads_and_blocks()

◆ c_threads_and_blocks()

◆ clearFreeAsyncBuffer()

◆ cudaStream()

◆ deviceId()

◆ deviceName()

◆ devicePropMajor()

◆ devicePropMinor()

◆ deviceVendor()

◆ Finalize()

◆ freeAsync()

◆ freeMemAvailable()

◆ gpuStream()

◆ grid_stride_threads_and_blocks()

◆ Initialize()

◆ isIntegrated()

◆ managedMemorySupported()

◆ 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()

◆ resetExternalStream()

◆ resetStream()

◆ resetStreamIndex()

◆ setExternalStream()

◆ setNumThreadsMin()

◆ setStream()

◆ setStreamIndex()

◆ sharedMemPerBlock()

◆ streamIndex()

◆ streamOrderedFreeAsync()

◆ streamSynchronize() [1/2]

◆ streamSynchronize() [2/2]

◆ streamSynchronizeActive()

◆ streamSynchronizeAll()

◆ synchronize()

◆ totalGlobalMem()

◆ usingExternalStream()

Friends And Related Symbol Documentation

◆ ExternalGpuStreamRegion

Member Data Documentation

◆ warp_size