Topology: Local

This example showcases how the abstract HiCR Core API can be used to discover compute and memory devices in the system. The code is structured as follows:

source/ contains the different variants of this example corresponding to different backends
- ascend.cpp corresponds to the Ascend backend implementation
- hwloc.cpp corresponds to the HWLoc (CPU) backend implementation

Topology detection

First, we use the topology manager to query the local HiCR instance’s topology

auto topology = topologyManager.queryTopology();

Compute resources and memory spaces retrieval

Then, we iterate over all its devices, printing the compute resources and memory spaces we find along with their types. For the latter, also their capacity.

for (const auto &d : topology.getDevices())
 {
   auto c = d->getComputeResourceList(); // Used to print compute resources
   auto m = d->getMemorySpaceList(); // Used to print memory spaces
 }

The result of running the hwloc example might be, for example:

+ 'NUMA Domain'
  Compute Resources: 44 Processing Unit(s)
  Memory Space:     'RAM', 94.483532 Gb
+ 'NUMA Domain'
  Compute Resources: 44 Processing Unit(s)
  Memory Space:     'RAM', 93.024166 Gb

Whereas, for the Ascend example, it would look as follows:

+ 'Ascend Device'
  Compute Resources: 1 Ascend Processor(s)
  Memory Space:     'Ascend Device RAM', 32.000000 Gb
+ 'Ascend Device'
  Compute Resources: 1 Ascend Processor(s)
  Memory Space:     'Ascend Device RAM', 32.000000 Gb

and for OpenCL, for example:

+ 'OpenCL GPU'
  Compute Resources: 1 OpenCL GPU Processing Unit(s)
  Memory Space:     'OpenCL GPU RAM', 49.908493 Gb