haojin2 commented on a change in pull request #16865: [numpy][Do Not Review]add op insert URL: https://github.com/apache/incubator-mxnet/pull/16865#discussion_r379682391
########## File path: src/operator/numpy/np_insert_op_slice-inl.h ########## @@ -0,0 +1,200 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://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. + */ + +/*! + * Copyright (c) 2019 by Contributors + * \file np_insert_op_slice-inl.h + * \brief Function definition of insert operators (insert by slice index) + */ +#ifndef MXNET_OPERATOR_NUMPY_NP_INSERT_OP_SLICE_INL_H_ +#define MXNET_OPERATOR_NUMPY_NP_INSERT_OP_SLICE_INL_H_ + +#include <vector> +#include <algorithm> +#include "./np_insert_op-inl.h" + +namespace mxnet { +namespace op { + +/* + * Only support slice index (the type of param 'obj' is slice). + */ +template<typename xpu> +void NumpyInsertSliceCompute(const nnvm::NodeAttrs& attrs, + const OpContext& ctx, + const std::vector<TBlob>& inputs, + const std::vector<OpReqType>& req, + const std::vector<TBlob>& outputs) { + using namespace mshadow; + using namespace mxnet_op; + + const NumpyInsertParam& param = nnvm::get<NumpyInsertParam>(attrs.parsed); + CHECK_EQ(inputs.size(), (param.val.has_value() ? 1 : 2)); + CHECK_EQ(outputs.size(), 1); + CHECK_EQ(req.size(), 1); + mshadow::Stream<xpu> *s = ctx.get_stream<xpu>(); + const int arr_pos = 0; + const int val_pos = param.val.has_value() ? 0 : 1; + const int out_pos = 0; + int ndim = inputs[arr_pos].shape_.ndim(); + int axis = param.axis.has_value() ? param.axis.value() : 0; + TBlob arr; + TBlob values = param.val.has_value() ? + TBlob(nullptr, mxnet::TShape(0, 1), xpu::kDevMask, outputs[out_pos].type_flag_) : + inputs[val_pos]; + if (!param.axis.has_value()) { + arr = inputs[arr_pos].reshape(Shape1(inputs[arr_pos].shape_.Size())); + ndim = 1; + } else if (ndim == 0) { + if (param.val.has_value()) { + CHECK_EQ(inputs[val_pos].shape_.ndim(), 0) + << "'arr' is a 0-d array, 'values' can not assign to it. " + << "alueError: assignment to 0-d array."; + mxnet_op::copy(s, outputs[out_pos], inputs[val_pos]); + } else { + MSHADOW_TYPE_SWITCH(outputs[out_pos].type_flag_, DType, { + Fill(s, outputs[out_pos], req[0], static_cast<DType>(param.val.value())); + }); + } + return; + } else { + arr = inputs[arr_pos]; + CHECK(axis >= -1 * arr.shape_.ndim() && axis < arr.shape_.ndim()) + << "Axis should be in the range of [-r, r-1] where r is the rank of input tensor"; + axis += (axis < 0) ? arr.shape_.ndim() : 0; + } + + int N = arr.shape_[axis]; + size_t indices_len = 0; // indices amount + int start = 0, stop = 0, step = 0; // arguments from 'obj' when it's 'slice' + + // get and check indices from slice or sequence of ints + SliceIndices(param.start, param.stop, param.step, + N, &start, &stop, &step, &indices_len); + + int numnew = 0; // numnew = output.shape[axis] - arr.shape[axis] + int index = 0; // save modified index, because index may be negative integer + mxnet::TShape val_newshape(arr.shape_.ndim(), -1); + // modify values's ndim to arr's ndim, for broadcast easily later + // e.g. value shape: (2,) arr shape: (3, 2) => value shape: (1, 2) + for (int i = values.shape_.ndim() - 1, j = arr.shape_.ndim() - 1; + i >= 0 || j >= 0; + --i, --j) { + if (i >= 0 && j >= 0) { + val_newshape[j] = values.shape_[i]; + } else if (i >= 0) { + CHECK_EQ(values.shape_[i], 1) << "index exceed limits."; + } else { + val_newshape[j] = 1; + } + } + values.shape_.assign(val_newshape.begin(), val_newshape.end()); + + // get numnew + mxnet::TShape old_valshape(values.shape_); + if (indices_len == 1) { // tensor with only one element + numnew = values.shape_[axis]; + index = start; + CHECK(index >= -1 * N && index <= N) + << "Index should be in the range of [-r, r-1] where r is the dim size in 'axis'"; + if (index < 0) { + index += N; + } + } else { + numnew = static_cast<int>(indices_len); + } + + const mxnet::TShape& outshape = outputs[out_pos].shape_; + int dtype = outputs[out_pos].type_flag_; + int vtype = param.val.has_value() ? + mshadow::DataType<double>::kFlag : + inputs[val_pos].type_flag_; + if ((indices_len == 1) && param.val.has_value()) { + MSHADOW_TYPE_SWITCH(vtype, VType, { + // If insert use single index and 'value' is inputed as numerical parameter + values = TBlob(ctx.requested[0].get_space_typed<xpu, 1, VType>(Shape1(1), s)); + Fill(s, values, kWriteTo, param.val.value()); + }); + } + + if (indices_len == 1) { + MXNET_NDIM_SWITCH(outshape.ndim(), ndim, { + InsertScalerObj<xpu, ndim>(s, outputs[out_pos], arr, values, + mxnet_op::calc_stride(arr.shape_.get<ndim>()), + mxnet_op::calc_stride(values.shape_.get<ndim>()), + mxnet_op::calc_stride(old_valshape.get<ndim>()), + mxnet_op::calc_stride(outshape.get<ndim>()), + outshape.get<ndim>(), values.shape_.get<ndim>(), + dtype, vtype, req[out_pos], axis, start, numnew, + outshape.Size(), false); + }); + } else { + // broadcast check + for (int i = outshape.ndim() - 1; i >= 0; --i) { + int sz = outshape[i]; + if (i == axis) { + sz = numnew; + } + CHECK((values.shape_[i] == 1) || (values.shape_[i] == sz)); + } + size_t temp_storage_bytes, temp_mem_size; + temp_storage_bytes = SortByKeyWorkspaceSize<int64_t, int, xpu>(indices_len, false, true); + temp_mem_size = indices_len * sizeof(int64_t) * 2 + + indices_len * sizeof(int) + + outshape[axis] * sizeof(int) * 2 + + temp_storage_bytes; + Tensor<xpu, 1, char> temp_mem = + ctx.requested[0].get_space_typed<xpu, 1, char>(Shape1(temp_mem_size), s); + int64_t* indices_ptr = reinterpret_cast<int64_t*>(temp_mem.dptr_); + int64_t* sorted_indices_ptr = reinterpret_cast<int64_t*>(indices_ptr + indices_len); + int* order_ptr = reinterpret_cast<int*>(sorted_indices_ptr + indices_len); + int* is_insert = reinterpret_cast<int*>(order_ptr + indices_len); + int* origin_idx = reinterpret_cast<int*>(is_insert + outshape[axis]); + Tensor<xpu, 1, char> temp_storage(reinterpret_cast<char*>(origin_idx + outshape[axis]), + Shape1(temp_storage_bytes), s); + Tensor<xpu, 1, int64_t> indices(indices_ptr, Shape1(indices_len), s); + Tensor<xpu, 1, int64_t> sorted_indices(sorted_indices_ptr, Shape1(indices_len), s); + Tensor<xpu, 1, int> order(order_ptr, Shape1(indices_len), s); + int num_bits = common::ilog2ui(static_cast<unsigned int>(indices_len) - 1); + Kernel<SliceToIndices, xpu>::Launch(s, indices_len, indices_ptr, start, step); + Kernel<range_fwd, xpu>::Launch(s, indices_len, 1, 0, 1, kWriteTo, order_ptr); + mxnet::op::SortByKey(indices, order, true, &temp_storage, 0, num_bits, &sorted_indices); + Kernel<IndicesModify, xpu>::Launch(s, indices_len, indices_ptr, order_ptr); + + mxnet_op::Kernel<mxnet_op::set_zero, xpu>::Launch(s, outshape[axis], is_insert); + Kernel<SetIsInsert, xpu>::Launch(s, indices_len, indices_ptr, is_insert); + + Kernel<SetOriginValuesIdx, xpu>::Launch(s, indices_len, indices_ptr, origin_idx); + Kernel<SetOriginArrIdx, xpu>::Launch(s, outshape[axis], is_insert, origin_idx); + MXNET_NDIM_SWITCH(outshape.ndim(), ndim, { + InsertSequenceImpl<xpu, ndim>(s, outputs[out_pos], arr, values, + mxnet_op::calc_stride(arr.shape_.get<ndim>()), Review comment: alignment ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
