edponce commented on a change in pull request #12460: URL: https://github.com/apache/arrow/pull/12460#discussion_r829441926
########## File path: cpp/src/arrow/compute/kernels/vector_cumulative_sum.cc ########## @@ -0,0 +1,159 @@ +// 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. + +#include "arrow/array/array_base.h" +#include "arrow/compute/api_scalar.h" +#include "arrow/compute/kernels/common.h" +#include "arrow/result.h" +#include "arrow/visit_type_inline.h" + +namespace arrow { +namespace compute { +namespace internal { + +template <typename Type> +struct CumulativeSum { + using CType = TypeTraits<Type>::CType; + using ScalarType = TypeTraits<Type>::ScalarType; + + CType Sum(ExecContext* ctx, std::shared_ptr<Array>& input, ArrayData* output, + CType start) { + CType sum = start; + CType* data = checked_cast<CType*>(input->data()->buffers[1]->data()); + CType* out_values = checked_cast<CType*>(output->buffers[1]->mutable_data()); + ArithmeticOptions options; + for (size_t i = input->offset; i < input->length; ++i) { + if (input->IsValid(i)) { + Datum value_datum(data[i]); + Datum sum_datum(sum); + auto result = Add(value_datum, sum_datum, options, ctx); + ScalarType result_scalar = result.ValueOrDie().scalar_as(); + sum = result_scalar.value; + out_values[i] = sum; + } + } + + return sum; + } + + Status Exec(KernelContext* ctx, const ExecBatch& batch, Datum* out) { + const auto& options = OptionsWrapper<CumulativeSumOptions<CType>>::Get(ctx); + CType start = checked_cast<const ScalarType&>(options.start).value; + + switch (batch[0].kind()) { + case Datum::ARRAY: + std::shared_ptr<Array> input = batch[0].make_array(); + ArrayData* output = out->array().get(); + + output->length = input->data()->length; + *output->type = std::move(input->type()); + uint8_t* out_bitmap = output->buffers[0]->mutable_data(); + int64_t out_offset = input->offset(); + + if (input->data()->MayHaveNulls()) { + arrow::internal::CopyBitmap(input->null_bitmap_data(), input->offset(), + input->length(), out_bitmap, out_offset); + output->null_count = input->null_count(); + } else { + bit_util::SetBitsTo(out_bitmap, out_offset, input->length(), true); + output->null_count = 0; + } + + Sum(ctx->exec_context(), input, output, start); + return Status::OK(); + case Datum::CHUNKED_ARRAY: + const auto& input = batch[0].chunked_array(); + + ArrayVector out_chunks; + for (const auto& chunk : input->chunks()) { + auto out_chunk = std::make_shared<ArrayData>( + chunk->type(), chunk->length(), chunk->null_count(), chunk->offset()); + + uint8_t* out_chunk_bitmap = out_chunk->buffers[0]->mutable_data(); + if (chunk->data()->MayHaveNulls()) { + arrow::internal::CopyBitmap(chunk->null_bitmap_data(), chunk->offset(), + chunk->length(), out_chunk_bitmap, + out_chunk->offset()); + out_chunk->null_count = chunk->null_count(); + } else { + bit_util::SetBitsTo(out_chunk_bitmap, out_chunk->offset(), chunk->length(), + true); + out_chunk->null_count = 0; + } + + CType last_value = Sum(ctx->exec_context(), chunk, out_chunk, start); + start = last_value; + out_chunks.push_back(MakeArray(std::move(out_chunk))); + } + + *out->chunked_array() = ChunkedArray(out_chunks, input->type()); + return Status::OK(); + default: + return Status::NotImplemented( + "Unsupported input type for function 'cumulative_sum': ", + batch[0].ToString()); + } + } + + static std::shared_ptr<KernelSignature> GetSignature(detail::GetTypeId get_id) { + return KernelSignature::Make({InputType::Array(get_id.id)}, OutputType(FirstType)); + } +}; + +const FunctionDoc cumulative_sum_doc( + "Compute the cumulative sum over an array of numbers", + ("`values` must be an array of numeric type values.\n" + "`start` is a single value of the same type.\n" + "Return an array which is the cumulative sum computed over `values.`\n" + "Null entries remain in place but are not used in calucating sum.\n" + "`start` is an optional starting sum of computation."), + {"values", "start"}); + +void RegisterVectorCumulativeSum(FunctionRegistry* registry) { + auto cumulative_sum = std::make_shared<VectorFunction>( + "cumulative_sum", Arity::Binary(), &cumulative_sum_doc); + + auto add_kernel = [&](detail::GetTypeId get_id, ArrayKernelExec exec) { + VectorKernel kernel; + kernel.can_execute_chunkwise = true; Review comment: There was the idea of supporting cumsum in chunks with some state tracking. For example, ```python start = 0 for batch in batches: auto cumulative_sum = CumulativeSum(batch, start); start = cumulative_sum[-1] SendDownstream(cumulative_sum) ``` cc @westonpace This would require higher level logic to know how to keep track of the `start` state and pass it on. For simplicity, should this be considered as `kernel.can_execute_chunkwise = false` and process all the data in a single batch? I understand that being able to process chunkwise would allow larger than memory processing (even though it requires sequential processing, but I think providing such functionality should be done as a separate follow-up PR. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
