• Hans-Nikolai Viessmann's avatar
    fix emrl lifting in CUDA backend · 6d00b16d
    Hans-Nikolai Viessmann authored
    Recently EMRL was refactored to 'correctly' maintain SSA representation
    *but* I forgot to update the mem:alloc code that specifically handles
    EMRL in the CUDA backend, meaning we were introduceing an additional
    alloc() for the same LHS leading to all sorts of runtime errors.