| 39 | |
| 40 | template<typename Tk, typename Tv> |
| 41 | void sort0ByKeyIterative(Param<Tk> pKey, Param<Tv> pVal, bool isAscending) { |
| 42 | auto dpl_policy = ::oneapi::dpl::execution::make_device_policy(getQueue()); |
| 43 | |
| 44 | for (int w = 0; w < pKey.info.dims[3]; w++) { |
| 45 | int pKeyW = w * pKey.info.strides[3]; |
| 46 | int pValW = w * pVal.info.strides[3]; |
| 47 | for (int z = 0; z < pKey.info.dims[2]; z++) { |
| 48 | int pKeyWZ = pKeyW + z * pKey.info.strides[2]; |
| 49 | int pValWZ = pValW + z * pVal.info.strides[2]; |
| 50 | for (int y = 0; y < pKey.info.dims[1]; y++) { |
| 51 | int pKeyOffset = pKeyWZ + y * pKey.info.strides[1]; |
| 52 | int pValOffset = pValWZ + y * pVal.info.strides[1]; |
| 53 | |
| 54 | auto key_begin = |
| 55 | ::oneapi::dpl::begin( |
| 56 | pKey.data->template reinterpret<compute_t<Tk>>()) + |
| 57 | pKeyOffset; |
| 58 | auto key_end = key_begin + pKey.info.dims[0]; |
| 59 | auto val_begin = ::oneapi::dpl::begin(*pVal.data) + pValOffset; |
| 60 | auto val_end = val_begin + pVal.info.dims[0]; |
| 61 | |
| 62 | auto zipped_begin = |
| 63 | ::oneapi::dpl::make_zip_iterator(key_begin, val_begin); |
| 64 | auto zipped_end = |
| 65 | ::oneapi::dpl::make_zip_iterator(key_end, val_end); |
| 66 | |
| 67 | // sort by key |
| 68 | if (isAscending) { |
| 69 | std::sort(dpl_policy, zipped_begin, zipped_end, |
| 70 | [](auto lhs, auto rhs) { |
| 71 | return std::get<0>(lhs) < std::get<0>(rhs); |
| 72 | }); |
| 73 | } else { |
| 74 | std::sort(dpl_policy, zipped_begin, zipped_end, |
| 75 | [](auto lhs, auto rhs) { |
| 76 | return std::get<0>(lhs) > std::get<0>(rhs); |
| 77 | }); |
| 78 | } |
| 79 | } |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | ONEAPI_DEBUG_FINISH(getQueue()); |
| 84 | } |
| 85 | |
| 86 | template<typename Tk, typename Tv> |
| 87 | void sortByKeyBatched(Param<Tk> pKey, Param<Tv> pVal, const int dim, |