diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.cc new file mode 100644 index 00000000000..5ba93e43fcb --- /dev/null +++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.cc @@ -0,0 +1,109 @@ +/** + * Copyright 2021 Huawei Technologies Co., Ltd + * + * Licensed 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. + */ + +#include "backend/kernel_compiler/cpu/searchsorted_cpu_kernel.h" + +#include +#include +#include + +namespace mindspore { +namespace kernel { +namespace { +constexpr size_t kInputSize = 2; +constexpr size_t kOutputSize = 1; +} // namespace + +template +void SearchSortedCPUKernel::InitKernel(const CNodePtr &kernel_node) { + MS_EXCEPTION_IF_NULL(kernel_node); + right_ = AnfAlgo::GetNodeAttr(kernel_node, "right"); + sequence_shape_ = AnfAlgo::GetInputDeviceShape(kernel_node, 0); + values_shape_ = AnfAlgo::GetInputDeviceShape(kernel_node, 1); + search_len = sequence_shape_.back(); +} + +template +const S *SearchSortedCPUKernel::CustomizedLowerBound(const S *seq_start, const S *seq_end, const S key) { + while (seq_start < seq_end) { + const S *mid = seq_start + ((seq_end - seq_start) >> 1); + if (!(key <= *mid)) { + seq_start = mid + 1; + } else { + seq_end = mid; + } + } + return seq_start; +} + +template +bool SearchSortedCPUKernel::Launch(const std::vector &inputs, + const std::vector &, + const std::vector &outputs) { + CheckParam(inputs, outputs); + auto sequence = reinterpret_cast(inputs[0]->addr); + auto values = reinterpret_cast(inputs[1]->addr); + auto output = reinterpret_cast(outputs[0]->addr); + size_t elem_num = inputs[1]->size / sizeof(S); + size_t seq_dim = sequence_shape_.size(); + size_t search_repeat = values_shape_.back(); + + auto task = [&](size_t start, size_t end) { + for (size_t i = start; i < end; i++) { + auto seq_start = (seq_dim == 1) ? sequence : sequence + (i / search_repeat) * search_len; + output[i] = right_ ? std::upper_bound(seq_start, seq_start + search_len, values[i]) - seq_start + : CustomizedLowerBound(seq_start, seq_start + search_len, values[i]) - seq_start; + } + }; + CPUKernelUtils::ParallelFor(task, elem_num); + return true; +} + +template +void SearchSortedCPUKernel::CheckParam(const std::vector &inputs, + const std::vector &outputs) { + // inputs: sequence, values + if (inputs.size() != kInputSize) { + MS_LOG(EXCEPTION) << "Input number is: " << inputs.size() << ", but SearchSorted needs" << kInputSize << " inputs."; + } + + // outputs: positions + if (outputs.size() != kOutputSize) { + MS_LOG(EXCEPTION) << "Output number is " << outputs.size() << ", but SearchSorted needs " << kOutputSize + << " outputs"; + } + + if (outputs[0]->size / sizeof(T) != inputs[1]->size / sizeof(S)) { + MS_LOG(EXCEPTION) << "The output dimensions " << outputs[0]->size << " must match the dimensions of input values " + << inputs[1]->size; + } + + auto sequence = reinterpret_cast(inputs[0]->addr); + size_t list_count = accumulate(sequence_shape_.begin(), sequence_shape_.end() - 1, 1, std::multiplies()); + auto task = [&](size_t start, size_t end) { + for (size_t i = start; i < end; i++) { + for (size_t j = 0; j < search_len - 1; j++) { + if (sequence[i * search_len + j] > sequence[i * search_len + j + 1]) { + MS_LOG(EXCEPTION) << "The input sequence must be sorted."; + } + } + } + }; + CPUKernelUtils::ParallelFor(task, list_count); +} + +} // namespace kernel +} // namespace mindspore diff --git a/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.h b/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.h new file mode 100644 index 00000000000..87cea83a5ef --- /dev/null +++ b/mindspore/ccsrc/backend/kernel_compiler/cpu/searchsorted_cpu_kernel.h @@ -0,0 +1,111 @@ +/** + * Copyright 2021 Huawei Technologies Co., Ltd + * + * Licensed 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. + */ + +#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_SEARCHSORTED_CPU_KERNEL_H_ +#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_SEARCHSORTED_CPU_KERNEL_H_ + +#include +#include "backend/kernel_compiler/cpu/cpu_kernel.h" +#include "backend/kernel_compiler/cpu/cpu_kernel_factory.h" + +namespace mindspore { +namespace kernel { +template +class SearchSortedCPUKernel : public CPUKernel { + public: + SearchSortedCPUKernel() = default; + ~SearchSortedCPUKernel() override = default; + + void InitKernel(const CNodePtr &kernel_node) override; + + bool Launch(const std::vector &inputs, const std::vector &, + const std::vector &outputs) override; + + private: + const S *CustomizedLowerBound(const S *seq_start, const S *seq_end, const S key); + void CheckParam(const std::vector &inputs, const std::vector &outputs); + + bool right_{false}; + std::vector sequence_shape_; + std::vector values_shape_; + std::vector output_shape_; + size_t search_len; +}; + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeFloat64).AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, double, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeFloat32).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, float, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt64).AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, int64_t, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt32).AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, int32_t, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt16).AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, int16_t, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt8).AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt32), + SearchSortedCPUKernel, int8_t, int32_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeFloat64).AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, double, int64_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeFloat32).AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, float, int64_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt64).AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, int64_t, int64_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt32).AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, int32_t, int64_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt16).AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, int16_t, int64_t); + +MS_REG_CPU_KERNEL_T_S( + SearchSorted, + KernelAttr().AddInputAttr(kNumberTypeInt8).AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt64), + SearchSortedCPUKernel, int8_t, int64_t); + +} // namespace kernel +} // namespace mindspore + +#endif // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_CPU_SEARCHSORTED_CPU_KERNEL_H_ diff --git a/mindspore/ops/operations/__init__.py b/mindspore/ops/operations/__init__.py index 1a70406037a..bb0cef6ad9e 100644 --- a/mindspore/ops/operations/__init__.py +++ b/mindspore/ops/operations/__init__.py @@ -33,7 +33,7 @@ from .array_ops import (Argmax, Argmin, Cast, Concat, Pack, Stack, Unpack, Unsta Transpose, TruncatedNormal, TupleToArray, UnsortedSegmentMin, UnsortedSegmentMax, UnsortedSegmentProd, UnsortedSegmentSum, SpaceToDepth, DepthToSpace, SpaceToBatch, BatchToSpace, SpaceToBatchND, BatchToSpaceND, BroadcastTo, InplaceUpdate, ReverseSequence, - EmbeddingLookup, Unique, GatherD, Identity, Range, MaskedSelect) + EmbeddingLookup, Unique, GatherD, Identity, Range, MaskedSelect, SearchSorted) from .comm_ops import (AllGather, AllReduce, _AlltoAll, AllSwap, ReduceScatter, Broadcast, _MirrorOperator, _MirrorMiniStepOperator, _MiniStepAllGather, ReduceOp, _VirtualDataset, _VirtualOutput, _VirtualDiv, _GetTensorSlice, _VirtualAdd, @@ -438,6 +438,7 @@ __all__ = [ "SparseTensorDenseMatmul", "MatrixInverse", "Range", + "SearchSorted", "IndexAdd", "PQC", "Evolution", diff --git a/mindspore/ops/operations/array_ops.py b/mindspore/ops/operations/array_ops.py index 47608cc835d..87b04c39659 100644 --- a/mindspore/ops/operations/array_ops.py +++ b/mindspore/ops/operations/array_ops.py @@ -5349,3 +5349,58 @@ class MaskedSelect(PrimitiveWithCheck): def check_dtype(self, x_dtype, mask_dtype): validator.check_tensor_dtype_valid('mask', mask_dtype, [mstype.bool_], self.name) + +class SearchSorted(PrimitiveWithInfer): + """ + Find the indices from the innermost dimension of `sequence` such that the order of the innermost dimension + within `sequence` would be preserved when the corresponding values in `values` were inserted before the indices. + + Args: + out_int32 (bool): Output datatype. Optional. If True, the output datatype will be int32; + if False, the output datatype will be int64. Default is False. + right (bool): Search Strategy. Optional. If True, return the last suitable index found. + If False, return the first such index. Default is False. + + Inputs: + - **sequence** (Tensor) - The shape of tensor is :math:`(x_1, x_2, ..., x_R-1, x_R)` or `(x_1)`. + It must contain monitonically increasing sequence on the innermost dimension. + - **values** (Tensor) - The shape of tensor is : math:`(x_1, x_2, ..., x_R-1, x_S)`. + + Outputs: + Tensor containing the indices from the innermost dimension of the input sequence such that, + if insert the corresponding value in the values tensor, the order of the tensor sequence would be preserved. + The shape of tensor is :math:`(x_1, x_2, ..., x_R-1, x_S)`, + whose datatype is int32 if out_int32 is True, otherwise int64, and shape is the same as the shape of values. + + Raises: + ValueError: If `sequence` and `values` do not have proper shapes. + + Supported Platforms: + ``CPU`` + + Examples: + >>> sequence = Tensor(np.array([[0, 1, 3, 5, 7], [2, 4, 6, 8, 10]]), mindspore.float32) + >>> values = Tensor(np.array([[3, 6, 9], [3, 6, 9]]), mindspore.float32) + >>> output = ops.SearchSorted()(sequence, values) + >>> print(output) + [[2, 4, 5] + [1, 2, 4]] + """ + @prim_attr_register + def __init__(self, out_int32=False, right=False): + """Initialize SearchSorted""" + self.out_int32 = validator.check_value_type("out_int32", out_int32, [bool], self.name) + self.right = validator.check_value_type("right", right, [bool], self.name) + self.init_prim_io_names(inputs=['sequence', 'values'], outputs=['positions']) + + def infer_shape(self, sequence_shape, values_shape): + if len(sequence_shape) != 1 and sequence_shape[:-1] != values_shape[:-1]: + raise ValueError(f"Sequence should be 1 dimensional or has all but the last dimension matching " + f" the dimensions of values, but got sequence's dimensions: {sequence_shape} " + f"and values' dimensions: {values_shape}.") + return values_shape + + def infer_dtype(self, sequence_dtype, values_dtype): + args = {"sequence_dtype": sequence_dtype, "values_dtype": values_dtype} + validator.check_tensors_dtypes_same_and_valid(args, mstype.number_type, self.name) + return mstype.tensor_type(mstype.int32) if self.out_int32 else mstype.tensor_type(mstype.int64) diff --git a/tests/st/ops/cpu/test_searchsorted_op.py b/tests/st/ops/cpu/test_searchsorted_op.py new file mode 100644 index 00000000000..1b8e8459250 --- /dev/null +++ b/tests/st/ops/cpu/test_searchsorted_op.py @@ -0,0 +1,115 @@ +# Copyright 2021 Huawei Technologies Co., Ltd +# +# Licensed 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. +# ============================================================================ + +import numpy as np +import pytest + +import mindspore.context as context +import mindspore.nn as nn +from mindspore import Tensor +from mindspore.common import dtype as mstype +from mindspore.ops import operations as P + +context.set_context(mode=context.GRAPH_MODE, device_target="CPU") + + +class SearchSortedNet(nn.Cell): + def __init__(self, out_int32=False, right=False): + super(SearchSortedNet, self).__init__() + self.searchsorted = P.SearchSorted(out_int32=out_int32, right=right) + + def construct(self, sequence, values): + return self.searchsorted(sequence, values) + +@pytest.mark.level0 +@pytest.mark.platform_x86_cpu +@pytest.mark.env_onecard +def test_right_out32(): + np.random.seed(1) + input1 = np.sort(np.array(np.random.randint(10, size=(2, 3, 9)), dtype=np.int32), axis=-1) + sequence = Tensor(input1, mstype.int32) + input2 = np.array(np.random.randint(10, size=(2, 3, 1)), dtype=np.int32) + values = Tensor(input2, mstype.int32) + + net = SearchSortedNet(out_int32=True, right=True) + output = net(sequence, values) + + expect = [[[9], + [3], + [6]], + [[5], + [9], + [8]]] + assert output.dtype == mstype.int32 + assert (output.asnumpy() == expect).all() + +@pytest.mark.level0 +@pytest.mark.platform_x86_cpu +@pytest.mark.env_onecard +def test_out32(): + np.random.seed(1) + input1 = np.sort(np.array(np.random.randint(10, size=(2, 3, 9)), dtype=np.int64), axis=-1) + sequence = Tensor(input1, mstype.int64) + input2 = np.array(np.random.randint(10, size=(2, 3, 1)), dtype=np.int64) + values = Tensor(input2, mstype.int64) + + net = SearchSortedNet(out_int32=True, right=False) + output = net(sequence, values) + + expect = [[[8], + [0], + [3]], + [[5], + [8], + [7]]] + assert output.dtype == mstype.int32 + assert (output.asnumpy() == expect).all() + +@pytest.mark.level0 +@pytest.mark.platform_x86_cpu +@pytest.mark.env_onecard +def test_right_out64(): + np.random.seed(1) + input1 = np.sort(np.array(np.random.random((2, 5)), dtype=np.float32), axis=-1) + sequence = Tensor(input1, mstype.float32) + input2 = np.array(np.random.random((2, 3)), dtype=np.float32) + values = Tensor(input2, mstype.float32) + + net = SearchSortedNet(out_int32=False, right=True) + output = net(sequence, values) + + expect = [[4, 4, 2], + [5, 0, 5]] + assert output.dtype == mstype.int64 + assert (output.asnumpy() == expect).all() + + +@pytest.mark.level0 +@pytest.mark.platform_x86_cpu +@pytest.mark.env_onecard +def test_out64(): + np.random.seed(1) + input1 = np.sort(np.array(np.random.random((5)), dtype=np.float64), axis=-1) + sequence = Tensor(input1, mstype.float64) + input2 = np.array(np.random.random((2, 3)), dtype=np.float64) + values = Tensor(input2, mstype.float64) + + net = SearchSortedNet(out_int32=False, right=False) + output = net(sequence, values) + + expect = [[1, 2, 3], + [3, 4, 4]] + assert output.dtype == mstype.int64 + assert (output.asnumpy() == expect).all()