93 lines
4.1 KiB
Plaintext
93 lines
4.1 KiB
Plaintext
|
/*******************************************************************************
|
||
|
|
* 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
|
||
|
|
******************************************************************************/
|
||
|
|
|
||
|
|
//
|
||
|
|
// @author raver119@gmail.com
|
||
|
|
// @author Yurii Shyrma, created on 15.11.2018
|
||
|
|
//
|
||
|
|
|
||
|
|
#include <loops/special_kernels.h>
|
||
|
|
|
||
|
|
namespace nd4j {
|
||
|
|
|
||
|
|
///////////////////////////////////////////////////////////////////////
|
||
|
|
template<typename T>
|
||
|
|
__device__ void pullRowsKernel(void *vx,
|
||
|
|
void *vz,
|
||
|
|
Nd4jLong len,
|
||
|
|
Nd4jLong *indexes,
|
||
|
|
Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets,
|
||
|
|
Nd4jLong *zTadShapeInfo, Nd4jLong *zTadOffsets) {
|
||
|
|
|
||
|
|
auto x = reinterpret_cast<T *>(vx);
|
||
|
|
auto z = reinterpret_cast<T *>(vz);
|
||
|
|
auto xEWS = shape::elementWiseStride(tadShapeInfo);
|
||
|
|
auto zEWS = shape::elementWiseStride(zTadShapeInfo);
|
||
|
|
auto tadLength = shape::length(tadShapeInfo);
|
||
|
|
|
||
|
|
if (xEWS >= 1 && zEWS >= 1) {
|
||
|
|
for (int idx = blockIdx.x; idx < len; idx += gridDim.x) {
|
||
|
|
T *rX = x + tadOffsets[indexes[idx]];
|
||
|
|
T *rZ = z + zTadOffsets[idx];
|
||
|
|
|
||
|
|
for (int i = threadIdx.x; i < tadLength; i += blockDim.x) {
|
||
|
|
rZ[i * zEWS] = rX[i * xEWS];
|
||
|
|
}
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (int idx = blockIdx.x; idx < len; idx += gridDim.x) {
|
||
|
|
T *rX = x + tadOffsets[indexes[idx]];
|
||
|
|
T *rZ = z + zTadOffsets[idx];
|
||
|
|
|
||
|
|
for (int i = threadIdx.x; i < tadLength; i += blockDim.x) {
|
||
|
|
auto xOffset = shape::getIndexOffset(i, tadShapeInfo, tadLength);
|
||
|
|
auto zOffset = shape::getIndexOffset(i, zTadShapeInfo, tadLength);
|
||
|
|
rZ[zOffset] = rX[xOffset];
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
///////////////////////////////////////////////////////////////////////
|
||
|
|
template<typename T>
|
||
|
|
__global__ void execPullRowsKernel(void *vx,
|
||
|
|
void *vz,
|
||
|
|
Nd4jLong len,
|
||
|
|
Nd4jLong *indexes,
|
||
|
|
Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets,
|
||
|
|
Nd4jLong *zTadShapeInfo, Nd4jLong *zTadOffsets) {
|
||
|
|
|
||
|
|
pullRowsKernel<T>(vx, vz, len, indexes, tadShapeInfo, tadOffsets, zTadShapeInfo, zTadOffsets);
|
||
|
|
}
|
||
|
|
|
||
|
|
///////////////////////////////////////////////////////////////////////
|
||
|
|
template<typename T>
|
||
|
|
__host__ void pullRowsKernelGeneric(dim3 &launchDims, cudaStream_t *stream,
|
||
|
|
void *vx,
|
||
|
|
void *vz,
|
||
|
|
Nd4jLong len,
|
||
|
|
Nd4jLong *indexes,
|
||
|
|
Nd4jLong *tadShapeInfo, Nd4jLong *tadOffsets,
|
||
|
|
Nd4jLong *zTadShapeInfo, Nd4jLong *zTadOffsets) {
|
||
|
|
|
||
|
|
execPullRowsKernel<T><<<launchDims.x, launchDims.y, launchDims.z, *stream>>>(vx, vz, len, indexes, tadShapeInfo, tadOffsets, zTadShapeInfo, zTadOffsets);
|
||
|
|
nd4j::DebugHelper::checkErrorCode(stream, "pullRows(...) failed");
|
||
|
|
}
|
||
|
|
|
||
|
|
BUILD_SINGLE_TEMPLATE(template void ND4J_EXPORT pullRowsKernelGeneric, (dim3 & launchDims, cudaStream_t * stream, void * vx, void * vz, Nd4jLong len, Nd4jLong * indexes, Nd4jLong * tadShapeInfo, Nd4jLong * tadOffsets, Nd4jLong *zTadShapeInfo, Nd4jLong * zTadOffsets), LIBND4J_TYPES);
|
||
|
|
}
|
||
|
|
|