Function get_grid_size

src/jit/impl/mlir/executable_cuda.cpp:24–29

namespace {

int64_t get_grid_size(int64_t nr_elements, int64_t block_size) {
    // unroll three times in the kernel
    int64_t a = nr_elements / (block_size * 2);
    int64_t b = (nr_elements - 1) / (block_size * 3) + 1;
    return std::max(a, b);
}

template <int out_dim, typename ctype>
void setup_and_launch(const JITExecutor* fusion_opr, CUfunction func, int block_size) {