amrex::ArenaInfo Struct Reference

Configuration options for constructing an Arena. More...

#include <AMReX_Arena.H>

Public Member Functions
ArenaInfo &	SetReleaseThreshold (Long rt) noexcept
	Set the usage threshold that triggers release of cached unused blocks before growing.
ArenaInfo &	SetDeviceMemory () noexcept
	Configure the arena to use non-managed GPU device memory.
ArenaInfo &	SetReadOnly () noexcept
	Advise the runtime that the managed memory will be read-only on the device.
ArenaInfo &	SetPreferred () noexcept
	Advise the runtime to prefer device residence for the managed memory.
ArenaInfo &	SetHostAlloc () noexcept
	Configure the arena to use host-pinned memory.
ArenaInfo &	SetCpuMemory () noexcept
	Configure the arena to use CPU (host) memory.
ArenaInfo &	SetDefragmentation (bool b) noexcept
	Enable or disable defragmentation in the arena.

Public Attributes
Long	release_threshold = std::numeric_limits<Long>::max()
	When usage would exceed this, cached unused blocks are released first.
bool	use_cpu_memory = false
	Allocate CPU (host) memory instead of GPU memory.
bool	device_use_managed_memory = true
	Use CUDA/HIP managed (unified) memory for GPU allocations.
bool	device_set_readonly = false
	Advise the runtime that the managed memory is read-only on the device.
bool	device_set_preferred = false
	Advise the runtime to prefer the device for managed memory.
bool	device_use_hostalloc = false
	Use host-pinned memory (cudaMallocHost/hipHostMalloc).
bool	defragmentation = false

Detailed Description

Configuration options for constructing an Arena.

Uses a builder pattern: each setter returns *this to allow chaining.

Member Function Documentation

SetCpuMemory()

ArenaInfo & amrex::ArenaInfo::SetCpuMemory ( )

inlinenoexcept

Configure the arena to use CPU (host) memory.

SetDefragmentation()

ArenaInfo & amrex::ArenaInfo::SetDefragmentation ( bool  b )

inlinenoexcept

Enable or disable defragmentation in the arena.

SetDeviceMemory()

ArenaInfo & amrex::ArenaInfo::SetDeviceMemory ( )

inlinenoexcept

Configure the arena to use non-managed GPU device memory.

SetHostAlloc()

ArenaInfo & amrex::ArenaInfo::SetHostAlloc ( )

inlinenoexcept

Configure the arena to use host-pinned memory.

SetPreferred()

ArenaInfo & amrex::ArenaInfo::SetPreferred ( )

inlinenoexcept

Advise the runtime to prefer device residence for the managed memory.

SetReadOnly()

ArenaInfo & amrex::ArenaInfo::SetReadOnly ( )

inlinenoexcept

Advise the runtime that the managed memory will be read-only on the device.

SetReleaseThreshold()

ArenaInfo & amrex::ArenaInfo::SetReleaseThreshold ( Long  rt )

inlinenoexcept

Set the usage threshold that triggers release of cached unused blocks before growing.

Member Data Documentation

defragmentation

bool amrex::ArenaInfo::defragmentation = false

Enable defragmentation in the arena.

device_set_preferred

bool amrex::ArenaInfo::device_set_preferred = false

Advise the runtime to prefer the device for managed memory.

device_set_readonly

bool amrex::ArenaInfo::device_set_readonly = false

Advise the runtime that the managed memory is read-only on the device.

device_use_hostalloc

bool amrex::ArenaInfo::device_use_hostalloc = false

Use host-pinned memory (cudaMallocHost/hipHostMalloc).

device_use_managed_memory

bool amrex::ArenaInfo::device_use_managed_memory = true

Use CUDA/HIP managed (unified) memory for GPU allocations.

release_threshold

Long amrex::ArenaInfo::release_threshold = std::numeric_limits<Long>::max()

When usage would exceed this, cached unused blocks are released first.

use_cpu_memory

bool amrex::ArenaInfo::use_cpu_memory = false

Allocate CPU (host) memory instead of GPU memory.

---

The documentation for this struct was generated from the following file:

• /home/runner/work/amrex/amrex/Src/Base/AMReX_Arena.H