Improving support for distributed operations in FlowPM

[EiffL]

This issue is to track the developments needed on FlowPM to make the best of the new Horovod Mesh Implemenation. This overarching goal will encapsulate several smaller issues.

Goal

By the end of the hackweek, submit a Pull Request to https://github.com/DifferentiableUniverseInitiative/flowpm with tested and benchmarked implementations of N-body and ray-tracing simulations.

Participants

The main participants to this task are:;

• @dlanzieri
• @santiagocasas

Tasks

• : Run benchmarking of a full N-body with and without the Pyramid scheme, and identify the expensive parts Benchmarking full N-body simulation on GPU clusters flowpm#80
• : Implement raytracing operations in Mesh TensorFlow to generate large-scale mass-maps Implement distributed lensing flowpm#81
• : Test the scaling of the simulation