Debugging and optimization of HPC programs in mixed precision with the Verrou tool

François Févotte (1) , Bruno Lathuilìère (1)

1 EDF R&D - EDF R&D

The analysis of Floating-Point-related issues in HPC codes is becoming a topic of major interest. First, parallel computing and code optimization often break the reproducibility of numerical results across machines, compilers and even executions of the same program. Second, mixed precision is more and more used to reduce memory footprint and bandwidth wherever possible, thereby benefiting more from SIMD units. This paper presents how the Verrou tool can help during all stages of the Floating-Point analysis of HPC codes: diagnostic, debugging and optimization. Recent developments of Verrou are presented, along with examples illustrating the interest of these new features for industrial codes. As an example, Verrou allows debugging code aster, an industrial, parallel, mixed-precision code for the simulation of structural mechanics. Re-using existing infrastructure from the non-regression testing process, setting up Verrou requires limited effort. The first useful debugging information can be obtained after only 25 minutes, while the full automated debugging process takes 10 hours to complete.

Mots clés: Floating-point arithmetic; Verrou; mixed precision; Verification & Validation

Domaines: Arithmétique des ordinateurs

François Févotte, Bruno Lathuilìère. Debugging and optimization of HPC programs in mixed precision with the Verrou tool. 2019. ⟨hal-02044101⟩