cavis/libnd4j/include/ops/declarable/helpers/cuda/col2im.cu

133 lines
6.0 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 raver119 on 30.11.17.
//
#include <ops/declarable/helpers/col2im.h>
namespace nd4j {
namespace ops {
namespace helpers {
//////////////////////////////////////////////////////////////////////////
// [bS, iC, kH, kW, oH, oW] is de-convoluted to [bS, iC, iH, iW]
template<typename T>
__global__ static void col2imCuda(const void *in, void *out, const Nd4jLong *inShapeInfo, const Nd4jLong *outShapeInfo, const int strideY, const int strideX, const int padHeight, const int padWidth, const int imgHeight, const int imgWidth, const int dY, const int dX) {
const auto dx = reinterpret_cast<const T*>(in);
auto result = reinterpret_cast<T*>(out);
auto inShape = shape::shapeOf(const_cast<Nd4jLong *>(inShapeInfo));
auto inStride = shape::stride(const_cast<Nd4jLong *>(inShapeInfo));
int strideex = inStride[0];
int stridech = inStride[1];
int stridekrow = inStride[2];
int stridekcol = inStride[3];
int striderow = inStride[4];
int stridecol = inStride[5];
int kernelHeight = inShape[2];
int kernelWidth = inShape[3];
auto outShape = shape::shapeOf(const_cast<Nd4jLong *>(outShapeInfo));
auto resultOrder = shape::order(const_cast<Nd4jLong *>(outShapeInfo));
auto outStride = shape::stride(const_cast<Nd4jLong *>(outShapeInfo));
int samples = outShape[0];
int depth = outShape[1];
int imgH = outShape[2];
int imgW = outShape[3];
int height_col = inShape[4];//(imgHeight + 2 * padHeight - kernelHeight) / strideX + 1;
int width_col = inShape[5];//(imgWidth + 2 * padWidth - kernelWidth) / strideY + 1;
int n = samples * depth * imgHeight * imgWidth;
//Effective kernel size, accounting for dilation
int kEffectiveW = kernelWidth + (kernelWidth - 1) * (dX - 1);
int kEffectiveH = kernelHeight + (kernelHeight - 1) * (dY - 1);
for (int i = (blockDim.x * blockIdx.x) + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
T val = 0;
int w_im = i % imgWidth + padWidth;
int h_im = (i / imgWidth) % imgHeight + padHeight;
int c_im = i / (imgWidth * imgHeight);
int num_im = c_im / depth;
int depth_im = c_im % depth;
// compute the start and end of the output
// These are the indexes for dimensions ??? in the 6d col matrix
int w_col_start = (w_im < kEffectiveW) ? 0 : (w_im - kEffectiveW) / strideX + 1;
int w_col_end = nd4j::math::nd4j_min<int>(w_im / strideX + 1, width_col);
int h_col_start = (h_im < kEffectiveH) ? 0 : (h_im - kEffectiveH) / strideY + 1;
int h_col_end = nd4j::math::nd4j_min<int>(h_im / strideY + 1, height_col);
//Iterate over col entries in the 6d array... these are added up
for (int h_col = h_col_start; h_col < h_col_end; h_col += 1) {
for (int w_col = w_col_start; w_col < w_col_end; w_col += 1) {
int h_k = (h_im - h_col * strideY);
int w_k = (w_im - w_col * strideX);
if(h_k % dY == 0 && w_k % dX == 0){
h_k /= dY;
w_k /= dX;
int data_col_index = num_im * strideex + depth_im * stridech + h_k * stridekrow + w_k * stridekcol + h_col * striderow + w_col * stridecol;
val += dx[data_col_index];
}
}
}
int i_f = 0;
int i_c = i;
for (int dim = 3; dim >= 0; dim--) {
i_f += (i_c % outShape[dim]) * outStride[dim];
i_c = i_c / outShape[dim];
}
result[i_f] = val;
}
}
//////////////////////////////////////////////////////////////////////////
template<typename T>
void col2imCudaLauncher(nd4j::LaunchContext &context, const void *x, void *z, const Nd4jLong *xShapeInfo, const Nd4jLong *zShapeInfo, const int sH, const int sW, const int pH, const int pW, const int iH, const int iW, const int dH, const int dW) {
col2imCuda<T><<<512, 512, 1024, *context.getCudaStream()>>>(x, z, xShapeInfo, zShapeInfo, sH, sW, pH, pW, iH, iW, dH, dW);
}
//////////////////////////////////////////////////////////////////////////
void col2im(nd4j::LaunchContext & context, const NDArray& input, NDArray& output, const int sH, const int sW, const int pH, const int pW, const int iH, const int iW, const int dH, const int dW) {
NDArray::prepareSpecialUse({&output}, {&input});
BUILD_SINGLE_SELECTOR(output.dataType(), col2imCudaLauncher, (context, input.getSpecialBuffer(), output.getSpecialBuffer(), input.getSpecialShapeInfo(), output.getSpecialShapeInfo(), sH, sW, pH, pW, iH, iW, dH, dW), FLOAT_TYPES);
NDArray::registerSpecialUse({&output}, {&input});
}
BUILD_SINGLE_TEMPLATE(template void col2imCudaLauncher, (nd4j::LaunchContext &context, const void *x, void *z, const Nd4jLong *xShapeInfo, const Nd4jLong *zShapeInfo, const int sH, const int sW, const int pH, const int pW, const int iH, const int iW, const int dH, const int dW), FLOAT_TYPES);
}
}
}