Memory Tags

Detailed Description

Control allocation and placement of buffers between host and accelerator memory spaces.

The memory tag protocol is used by all memory-managing simulation data structures of the V8 core library. It is a compile-time protocol where the desired memory semantics are communicated to a class by passing a memory tag as a template parameter.

Each memory tag specifies whether the allocated memory is accessible from the host and/or the device via two compile-time boolean flags: isHostAccessible and isDeviceAccessible.

Depending on waLBerla's build settings, different memory tags are available.

Classes
struct	walberla::v8::memory::memtag::stdmem
	Standard pageable host memory. More...
struct	walberla::v8::memory::memtag::host
	Aligned pageable host memory. More...
struct	walberla::v8::memory::memtag::unified
	Managed GPU memory with unified memory semantics. More...
struct	walberla::v8::memory::memtag::device
	Memory tag indicating GPU device memory. More...
struct	walberla::v8::memory::memtag::pinned
	Memory tag indicating pinned (page-locked) host memory. More...

Typedefs
using	walberla::v8::memory::memtag::automatic
	Automatically select a memory tag according to the current build configuration.

Typedef Documentation

automatic

using walberla::v8::memory::memtag::automatic

Initial value:

 
 
 
 
host

Automatically select a memory tag according to the current build configuration.

Select a memory tag according to the build configuration:

• If CUDA or HIP are enabled, use unified memory
• If no GPU language is enabled, use host memory