I work on enhancing the performance of codes used in cosmology. A first aspect of my thesis is the parallelisation of a code reducing bias on galaxy observations catalogs, and the challenges associated with the dependencies inherent to the MCMC processes used. The aim is to have a code that can efficiently handle million-galaxy catalogs which can then be used to constrain cosmological simulations. Another aspect of my thesis is the implementation of subgrid physics modules in the fluid dynamics code Dyablo, which can run cosmological simulations on various architectures (CPU, GPU). In particular, the challenge resides in efficiently dealing with the non-locality of some of the physical phenomenon in a parallel setting, in particular when running on GPUs.