Nicolas Tancogne-Dejean authored Dec 16, 2023 and Nicolas Tancogne-Dejean committed Dec 19, 2023

This required to allocate the GS states on the GPU and to have the
correct cublas call implemented.

In the process, I realized that several allocations were not done on
the GPUs, which could affect the performances on the code.

A wrong PUSH/POP_SUB is also fixed.