Files
deeplearning4j--deeplearning4j/libnd4j/include/ops/declarable/helpers/cpu/top_k.cpp
T
2026-07-13 12:47:05 +08:00

190 lines
7.2 KiB
C++

/* ******************************************************************************
*
*
* This program and the accompanying materials are made available under the
* terms of the Apache License, Version 2.0 which is available at
* https://www.apache.org/licenses/LICENSE-2.0.
*
* See the NOTICE file distributed with this work for additional
* information regarding copyright ownership.
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* License for the specific language governing permissions and limitations
* under the License.
*
* SPDX-License-Identifier: Apache-2.0
******************************************************************************/
//
// @author raver119@gmail.com
//
#include <array/NDArrayFactory.h>
#include <execution/Threads.h>
#include <ops/declarable/headers/parity_ops.h>
#include <ops/declarable/helpers/top_k.h>
#include "ops/specials.h"
#if NOT_EXCLUDED(OP_top_k)
namespace sd {
namespace ops {
namespace helpers {
template <typename T>
static sd::Status topKFunctor_(NDArray* input, NDArray* values, NDArray* indices, const sd::LongType k,
bool needSort) {
sd::LongType width = input->sizeAt(-1);
sd::LongType lastDim = input->rankOf() - 1;
std::vector<sd::LongType> dimsToExclude(input->rankOf() - 1);
for (size_t d = 0; d < dimsToExclude.size(); ++d) dimsToExclude[d] = d;
const sd::LongType numOfSubArrs = ShapeUtils::getNumOfSubArrs(input->shapeInfo(), dimsToExclude);
if (k == 1) {
for (sd::LongType e = 0; e < numOfSubArrs; ++e) {
auto trial = (*input)(e, dimsToExclude);
sd::LongType maxPos = 0;
T maxVal = trial->e<T>(0);
for (sd::LongType pos = 1; pos < trial->lengthOf(); pos++)
if (maxVal < trial->e<T>(pos)) {
maxPos = pos;
maxVal = trial->e<T>(pos);
}
if (indices) indices->p(e, maxPos); // topIndex;
if (values) values->p(e, maxVal);
}
} else {
int nextPos = 0;
for (sd::LongType e = 0; e < numOfSubArrs; ++e) {
auto trial = (*input)(e, dimsToExclude);
// fill up the first k elements
NDArray *topValues = NDArrayFactory::create<T>('c', {k}, input->getContext());
NDArray *sortedVals = NDArrayFactory::create<T>('c', {k}, input->getContext());
NDArray *topIndices = NDArrayFactory::create<sd::LongType>('c', {k}, input->getContext());
for (sd::LongType pos = 0; pos < k; ++pos) {
topIndices->r<sd::LongType>(pos) = pos;
topValues->r<T>(pos) = trial->t<T>(pos);
}
sortedVals->assign(topValues);
SpecialMethods<T>::sortGeneric(sortedVals, false);
for (sd::LongType i = static_cast<sd::LongType>(k); i < width; ++i) {
T val = trial->e<T>(i);
T minTopVal = sortedVals->t<T>(0);
if (minTopVal < val) { // value should be inserted to top k
// only if it is not contained in
T* begin = reinterpret_cast<T*>(sortedVals->buffer());
T* end = begin + k;
bool exists = std::binary_search(begin, end, val);
if (!exists) {
// exchangePos - a distance between begin and minimal existed to be suppressed by val
T* topBegin = reinterpret_cast<T*>(topValues->buffer());
T* topEnd = topBegin + k;
auto exchangePos = std::distance(topBegin, std::find(topBegin, topEnd, sortedVals->t<T>(0)));
topValues->r<T>(exchangePos) = val; //*exchangeIt = val;
topIndices->r<sd::LongType>(exchangePos) = i;
sortedVals->r<T>(0) = val; // suppress in sorted
SpecialMethods<T>::sortGeneric(sortedVals, false);
}
}
}
if (needSort) {
SpecialMethods<T>::sortGeneric(topValues,true);
for (sd::LongType j = 0; j < width; j++)
for (sd::LongType pos = 0; pos < k; ++pos)
if (topValues->t<T>(pos) == trial->t<T>(j)) topIndices->r<sd::LongType>(pos) = j;
} else { // else sort by indices
std::map<sd::LongType, T> sortValsMap;
for (sd::LongType e = 0; e < topValues->lengthOf(); ++e) {
sortValsMap[topIndices->t<sd::LongType>(e)] = topValues->t<T>(e);
}
sd::LongType e = 0;
for (auto it = sortValsMap.begin(); it != sortValsMap.end(); ++it, e++) {
topIndices->r<sd::LongType>(e) = it->first;
topValues->r<T>(e) = it->second;
}
}
if (values) {
auto valuesView = (*values)(e, dimsToExclude);
valuesView->assign(topValues);
delete valuesView;
}
if (indices) {
auto indicesView = (*indices)(e, dimsToExclude);
indicesView->assign(topIndices);
delete indicesView;
}
delete sortedVals;
delete topValues;
delete topIndices;
delete trial;
}
}
return sd::Status::OK;
}
// ----------------------------------------------------------------------------------------------- //
template <typename T>
static sd::Status inTopKFunctor_(sd::LaunchContext* context, NDArray* input, NDArray* target,
NDArray* result, const sd::LongType k) {
std::vector<sd::LongType> shapeI(input->rankOf());
for (int i = 0; i < input->rankOf() - 1; i++) shapeI[i] = input->sizeAt(i);
shapeI[input->rankOf() - 1] = k;
std::unique_ptr<NDArray> indices(NDArrayFactory::create_<sd::LongType>(input->ordering(), shapeI, context));
NDArray* values = nullptr;
sd::Status status = topKFunctor(context, input, values, indices.get(), k, true);
int assign = 0;
result->assign(assign);
if (status == sd::Status::OK) {
auto func = PRAGMA_THREADS_FOR {
for (auto e = start; e < stop; e++) {
bool found = false;
for (sd::LongType j = 0; j < k; j++) {
if (target->e<sd::LongType>(e) == indices->e<sd::LongType>(e * k + j)) {
found = true;
break;
}
}
if (found) result->p<bool>(e, true);
}
};
samediff::Threads::parallel_tad(func, 0, target->lengthOf());
}
return status;
}
sd::Status topKFunctor(sd::LaunchContext* context, NDArray* input, NDArray* values, NDArray* indices,
const sd::LongType k, bool needSort) {
BUILD_SINGLE_SELECTOR(input->dataType(), return topKFunctor_, (input, values, indices, k, needSort),
SD_NUMERIC_TYPES);
}
sd::Status inTopKFunctor(sd::LaunchContext* context, NDArray* input, NDArray* target, NDArray* result,
const sd::LongType k) {
BUILD_SINGLE_SELECTOR(input->dataType(), return inTopKFunctor_, (context, input, target, result, k),
SD_NUMERIC_TYPES);
}
BUILD_SINGLE_TEMPLATE( sd::Status topKFunctor_,
(NDArray* input, NDArray* values, NDArray* indices, const sd::LongType k, bool needSort),
SD_NUMERIC_TYPES);
BUILD_SINGLE_TEMPLATE( sd::Status inTopKFunctor_,
(sd::LaunchContext * context, NDArray* input, NDArray* target, NDArray* result,
const sd::LongType k),
SD_NUMERIC_TYPES);
} // namespace helpers
} // namespace ops
} // namespace sd
#endif