Function MakeSequenceGlobalImpl

catboost/cuda/cuda_util/fill.cpp:194–204

template <typename T>
static void MakeSequenceGlobalImpl(
    TCudaBuffer<T, NCudaLib::TStripeMapping>& buffer,
    ui32 stream) {
    auto offset = CreateDistributedObject<T>(0);
    for (ui32 dev = 0; dev < offset.DeviceCount(); ++dev) {
        offset.Set(dev, buffer.GetMapping().DeviceSlice(dev).Left);
    }

    using TKernel = TMakeSequenceKernel<T>;
    LaunchKernels<TKernel>(buffer.NonEmptyDevices(), stream, buffer, offset);
}

#define Y_CATBOOST_CUDA_F_IMPL(T)                          \
    template <>                                            \