cavis/libnd4j/include/ops/declarable/helpers/cuda/stack.cu

109 lines
4.0 KiB
Plaintext
Raw Normal View History

2019-06-06 14:21:15 +02:00
/*******************************************************************************
* Copyright (c) 2015-2018 Skymind, Inc.
*
* 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.
*
* 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
******************************************************************************/
//
// Created by Yurii Shyrma on 02.01.2018
//
#include <ops/declarable/helpers/stack.h>
#include <helpers/ShapeUtils.h>
#include <array/ResultSet.h>
#include <cuda_exception.h>
#include <TAD.h>
#include <PointersManager.h>
#include <ConstantTadHelper.h>
namespace nd4j {
namespace ops {
namespace helpers {
template <typename T>
static __global__ void stackKernel(void** inputList, void** inputShapeList, int inputListLength, Nd4jLong arrLen, void* outputBuffer, Nd4jLong* tadShape, Nd4jLong *tadOffsets) { //, Nd4jLong* tadShape, Nd4jLong* tadOffsets) {
__shared__ int arrIdx, blocksPerArr;
__shared__ T *z;
__shared__ Nd4jLong *zShapeInfo, *xShapeInfo, arrLenPerBlock, start, end, offsetZ, zLength;
if (threadIdx.x == 0) {
z = reinterpret_cast<T*>(outputBuffer);
}
__syncthreads();
for (int t = blockIdx.x; t < inputListLength; t += gridDim.x) {
auto tZ = z + tadOffsets[t];
auto tX = reinterpret_cast<T*>(inputList[t]);
auto xShape = reinterpret_cast<Nd4jLong*>(inputShapeList[t]);
for (int e = threadIdx.x; e < arrLen; e += blockDim.x) {
tZ[shape::getIndexOffset(e, tadShape, arrLen)] = tX[shape::getIndexOffset(e, xShape, arrLen)];
}
}
}
///////////////////////////////////////////////////////////////////
template <typename T>
static void stack_(nd4j::LaunchContext * context, const std::vector<NDArray*>& inArrs, NDArray* outArr, const int dim) {
if(inArrs[0]->isScalar()) {
//#pragma omp parallel for
for (size_t i = 0; i < inArrs.size(); ++i) {
inArrs[i]->syncToHost();
outArr->p(i, inArrs[i]->e<T>(0));
}
outArr->syncToDevice();
}
else {
//Nd4jLong **dInShapeInfo;
//void **dInBuffers;
std::vector<void const*> inputList(inArrs.size());
std::vector<Nd4jLong const*> inputShapeList(inArrs.size());
auto stream = context->getCudaStream();
for (size_t i = 0; i < inputList.size(); ++i) {
inputList[i] = inArrs[i]->getSpecialBuffer();
inputShapeList[i] = inArrs[i]->getSpecialShapeInfo();
}
std::vector<int> axis = ShapeUtils::evalDimsToExclude(outArr->rankOf(), {dim});
auto packX = nd4j::ConstantTadHelper::getInstance()->tadForDimensions(outArr->getShapeInfo(), axis);
PointersManager manager(context, "helpers::stack");
auto dInBuffers = (void **) manager.replicatePointer(inputList.data(), inputList.size() * sizeof(Nd4jLong*));
auto dInShapeInfo = (void **) manager.replicatePointer(inputShapeList.data(), inputShapeList.size() * sizeof(Nd4jLong*));
dim3 launchDims(inArrs.size(), inArrs[0]->lengthOf(), 1024);
stackKernel<T><<<launchDims.x, launchDims.y, launchDims.z, *stream>>>((void**)dInBuffers, (void**)dInShapeInfo, inputList.size(), inArrs[0]->lengthOf(), outArr->specialBuffer(), packX.specialShapeInfo(), packX.specialOffsets()); //, dTadShape, dTadOffsets);
manager.synchronize();
}
}
void stack(nd4j::LaunchContext * context, const std::vector<NDArray*>& inArrs, NDArray* outArr, const int dim) {
BUILD_SINGLE_SELECTOR(outArr->dataType(), stack_, (context, inArrs, outArr, dim), LIBND4J_TYPES);
}
BUILD_SINGLE_TEMPLATE(template void stack_ , (nd4j::LaunchContext * context, const std::vector<NDArray*>& inArrs, NDArray* outArr, const int dim), LIBND4J_TYPES);
}
}
}