chore: import upstream snapshot with attribution
This commit is contained in:
@@ -0,0 +1,146 @@
|
||||
/*
|
||||
* ******************************************************************************
|
||||
* *
|
||||
* *
|
||||
* * 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 Oleh Semeniv (oleg.semeniv@gmail.com)
|
||||
//
|
||||
#include <helpers/Loops.h>
|
||||
#include <ops/declarable/helpers/transforms.h>
|
||||
#if NOT_EXCLUDED(OP_split)
|
||||
namespace sd {
|
||||
namespace ops {
|
||||
namespace helpers {
|
||||
|
||||
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////
|
||||
template <typename T>
|
||||
static void split_(NDArray& input, const std::vector<NDArray*>& outArrs, const LongType axis) {
|
||||
const sd::LongType numSplits = outArrs.size();
|
||||
const auto sizeofT = input.sizeOfT();
|
||||
auto xBuff = input.bufferAsT<T>();
|
||||
|
||||
// Cache input shape information
|
||||
const sd::LongType* xShapeInfo = input.shapeInfo();
|
||||
const sd::LongType* xShape = shape::shapeOf(xShapeInfo);
|
||||
const sd::LongType* xStride = shape::stride(xShapeInfo);
|
||||
const int xRank = input.rankOf();
|
||||
|
||||
// Fast path 1: Continuous memory case
|
||||
bool luckCase1 = ((axis == 0 && input.ordering() == 'c') ||
|
||||
(axis == xRank - 1 && input.ordering() == 'f')) &&
|
||||
input.ews() == 1;
|
||||
|
||||
if (luckCase1) {
|
||||
for (sd::LongType i = 0; i < numSplits; ++i) {
|
||||
luckCase1 &= outArrs[i]->ordering() == input.ordering() && outArrs[i]->ews() == 1;
|
||||
if (!luckCase1) break;
|
||||
}
|
||||
}
|
||||
|
||||
if (luckCase1) {
|
||||
T* x = const_cast<T*>(xBuff);
|
||||
for (sd::LongType i = 0; i < numSplits; ++i) {
|
||||
const auto memAmountToCopy = outArrs[i]->lengthOf();
|
||||
ops::safe_copy(outArrs[i]->bufferAsT<T>(), x, memAmountToCopy);
|
||||
x += memAmountToCopy;
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
// Fast path 2: Contiguous along split axis
|
||||
const bool isXcontin = input.strideAt(axis) == 1 && input.ordering() == 'c';
|
||||
bool areOutsContin = true;
|
||||
bool allSameOrder = true;
|
||||
|
||||
if (isXcontin) {
|
||||
for (sd::LongType i = 0; i < numSplits; ++i) {
|
||||
areOutsContin &= outArrs[i]->strideAt(axis) == 1;
|
||||
allSameOrder &= outArrs[i]->ordering() == input.ordering();
|
||||
if (!areOutsContin || !allSameOrder) break;
|
||||
}
|
||||
}
|
||||
|
||||
const bool luckCase2 = isXcontin && areOutsContin && allSameOrder;
|
||||
|
||||
if (luckCase2) {
|
||||
const auto xDim = input.sizeAt(axis);
|
||||
|
||||
for (sd::LongType i = 0; i < input.lengthOf() / xDim; ++i) {
|
||||
auto x = xBuff + xDim * i;
|
||||
|
||||
for (sd::LongType j = 0; j < numSplits; ++j) {
|
||||
const auto zDim = outArrs[j]->sizeAt(axis);
|
||||
T* z = outArrs[j]->bufferAsT<T>() + zDim * i;
|
||||
ops::safe_copy(z, x, static_cast<size_t>(zDim));
|
||||
|
||||
x += zDim;
|
||||
}
|
||||
}
|
||||
return;
|
||||
}
|
||||
|
||||
// General case: Cache output shape data for the first array
|
||||
const sd::LongType zDim = outArrs[0]->sizeAt(axis);
|
||||
|
||||
auto func = PRAGMA_THREADS_FOR {
|
||||
// Pre-allocate coordinate arrays
|
||||
sd::LongType coords[SD_MAX_RANK], temp;
|
||||
|
||||
for (auto i = start; i < stop; i += increment) {
|
||||
// Use cached shape data for input coordinates
|
||||
INDEX2COORDS(i, xRank, xShape, coords);
|
||||
sd::LongType xOffset;
|
||||
COORDS2INDEX(xRank, xStride, coords, xOffset);
|
||||
|
||||
// Find target output array
|
||||
sd::LongType outArrIdx = 0;
|
||||
temp = coords[axis];
|
||||
|
||||
while (coords[axis] >= zDim) {
|
||||
coords[axis] -= zDim;
|
||||
++outArrIdx;
|
||||
}
|
||||
|
||||
// Get output array and its shape data
|
||||
auto outArr = outArrs[outArrIdx];
|
||||
const sd::LongType* outShape = shape::shapeOf(outArr->shapeInfo());
|
||||
const sd::LongType* outStride = shape::stride(outArr->shapeInfo());
|
||||
T* z = outArr->bufferAsT<T>();
|
||||
|
||||
// Calculate output offset using cached shape data
|
||||
sd::LongType zOffset;
|
||||
COORDS2INDEX(outArr->rankOf(), outStride, coords, zOffset);
|
||||
z[zOffset] = xBuff[xOffset];
|
||||
|
||||
coords[axis] = temp;
|
||||
}
|
||||
};
|
||||
|
||||
samediff::Threads::parallel_for(func, 0, input.lengthOf());
|
||||
}
|
||||
|
||||
void split(sd::LaunchContext* context, NDArray& input, std::vector<NDArray*>& outArrs, const sd::LongType axis) {
|
||||
BUILD_SINGLE_SELECTOR(input.dataType(), split_, (input, outArrs, axis), SD_COMMON_TYPES);
|
||||
}
|
||||
} // namespace helpers
|
||||
} // namespace ops
|
||||
} // namespace sd
|
||||
#endif
|
||||
Reference in New Issue
Block a user