Function generate_kernel_string

src/sol_execbench/core/bench/timing.py:299–301

(kernel: CuptiKernelInfo)

    pre-allocates all tensors before the benchmark loop, so the timed region
    is not affected by cudaMalloc times (which increase measured kernel time by 300%).

    Parameters
    ----------
    fn : callable
        The function to benchmark.  Receives unpacked arguments each iteration.
    inputs : list