westonpace commented on a change in pull request #11542: URL: https://github.com/apache/arrow/pull/11542#discussion_r741577703
########## File path: cpp/src/arrow/compute/kernels/vector_buffer.cc ########## @@ -0,0 +1,345 @@ +// 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 <memory> +#include <type_traits> +#include <utility> +#include <vector> + +#include "arrow/array.h" +#include "arrow/array/builder_primitive.h" +#include "arrow/array/concatenate.h" +#include "arrow/buffer.h" +#include "arrow/chunked_array.h" +#include "arrow/compute/function.h" +#include "arrow/compute/kernel.h" +#include "arrow/compute/registry.h" +#include "arrow/record_batch.h" +#include "arrow/result.h" +#include "arrow/status.h" +#include "arrow/table.h" +#include "arrow/util/bit_util.h" +#include "arrow/util/logging.h" +#include "arrow/visitor_inline.h" + +namespace arrow { + +namespace compute { +namespace internal { + +namespace { + +struct GetByteRangesArray { + const std::shared_ptr<ArrayData>& input; + int64_t offset; + int64_t length; + UInt64Builder* range_starts; + UInt64Builder* range_offsets; + UInt64Builder* range_lengths; + + Status VisitBitmap(const std::shared_ptr<Buffer>& buffer) { + if (buffer) { + uint64_t data_start = reinterpret_cast<uint64_t>(buffer->data()); + RETURN_NOT_OK(range_starts->Append(data_start)); + RETURN_NOT_OK(range_offsets->Append(BitUtil::RoundDown(offset, 8) / 8)); + RETURN_NOT_OK(range_lengths->Append(BitUtil::CoveringBytes(offset, length))); + } + return Status::OK(); + } + + Status VisitFixedWidthArray(const Buffer& buffer, const FixedWidthType& type) { + uint64_t data_start = reinterpret_cast<uint64_t>(buffer.data()); + uint64_t offset_bits = offset * type.bit_width(); + uint64_t offset_bytes = BitUtil::RoundDown(offset_bits, 8) / 8; + uint64_t end_byte = + BitUtil::RoundUp(offset_bits + (length * type.bit_width()), 8) / 8; + uint64_t length_bytes = (end_byte - offset_bytes); + RETURN_NOT_OK(range_starts->Append(data_start)); + RETURN_NOT_OK(range_offsets->Append(offset_bytes)); + return range_lengths->Append(length_bytes); + } + + Status Visit(const FixedWidthType& type) { + static_assert(sizeof(uint8_t*) <= sizeof(uint64_t), + "Undefined behavior if pointer larger than uint64_t"); + RETURN_NOT_OK(VisitBitmap(input->buffers[0])); + RETURN_NOT_OK(VisitFixedWidthArray(*input->buffers[1], type)); + if (input->dictionary) { + // This is slightly imprecise because we always assume the entire dictionary is + // referenced. If this array has an offset it may only be referencing a portion of + // the dictionary + GetByteRangesArray dict_visitor{input->dictionary, + input->dictionary->offset, + input->dictionary->length, + range_starts, + range_offsets, + range_lengths}; + return VisitTypeInline(*input->dictionary->type, &dict_visitor); + } + return Status::OK(); + } + + Status Visit(const NullType& type) { return Status::OK(); } + + template <typename BaseBinaryType> + Status VisitBaseBinary(const BaseBinaryType& type) { + using offset_type = typename BaseBinaryType::offset_type; + RETURN_NOT_OK(VisitBitmap(input->buffers[0])); + + const Buffer& offsets_buffer = *input->buffers[1]; + RETURN_NOT_OK( + range_starts->Append(reinterpret_cast<uint64_t>(offsets_buffer.data()))); + RETURN_NOT_OK(range_offsets->Append(sizeof(offset_type) * offset)); + RETURN_NOT_OK(range_lengths->Append(sizeof(offset_type) * length)); + + const offset_type* offsets = input->GetValues<offset_type>(1, offset); + const Buffer& values = *input->buffers[2]; + offset_type start = offsets[0]; + offset_type end = offsets[length]; + RETURN_NOT_OK(range_starts->Append(reinterpret_cast<uint64_t>(values.data()))); + RETURN_NOT_OK(range_offsets->Append(static_cast<uint64_t>(start))); + return range_lengths->Append(static_cast<uint64_t>(end - start)); + } + + Status Visit(const BinaryType& type) { return VisitBaseBinary(type); } + + Status Visit(const LargeBinaryType& type) { return VisitBaseBinary(type); } + + template <typename BaseListType> + Status VisitBaseList(const BaseListType& type) { + using offset_type = typename BaseListType::offset_type; + RETURN_NOT_OK(VisitBitmap(input->buffers[0])); + + const Buffer& offsets_buffer = *input->buffers[1]; + RETURN_NOT_OK( + range_starts->Append(reinterpret_cast<uint64_t>(offsets_buffer.data()))); + RETURN_NOT_OK(range_offsets->Append(sizeof(offset_type) * offset)); + RETURN_NOT_OK(range_lengths->Append(sizeof(offset_type) * length)); + + const offset_type* offsets = input->GetValues<offset_type>(1, offset); + int64_t start = static_cast<int64_t>(offsets[0]); + int64_t end = static_cast<int64_t>(offsets[length]); + GetByteRangesArray child{input->child_data[0], start, end - start, + range_starts, range_offsets, range_lengths}; + return VisitTypeInline(*type.value_type(), &child); + } + + Status Visit(const ListType& type) { return VisitBaseList(type); } + + Status Visit(const LargeListType& type) { return VisitBaseList(type); } + + Status Visit(const FixedSizeListType& type) { + RETURN_NOT_OK(VisitBitmap(input->buffers[0])); + GetByteRangesArray child{input->child_data[0], + offset * type.list_size(), + length * type.list_size(), + range_starts, + range_offsets, + range_lengths}; + return VisitTypeInline(*type.value_type(), &child); + } + + Status Visit(const StructType& type) { + for (int i = 0; i < type.num_fields(); i++) { + GetByteRangesArray child{input->child_data[i], + offset + input->child_data[i]->offset, + length, + range_starts, + range_offsets, + range_lengths}; + RETURN_NOT_OK(VisitTypeInline(*type.field(i)->type(), &child)); + } + return Status::OK(); + } + + Status Visit(const DenseUnionType& type) { + // Skip validity map for DenseUnionType + // Types buffer is always int8 + RETURN_NOT_OK(VisitFixedWidthArray( + *input->buffers[1], *std::dynamic_pointer_cast<FixedWidthType>(int8()))); + // Offsets buffer is always int32 + RETURN_NOT_OK(VisitFixedWidthArray( + *input->buffers[2], *std::dynamic_pointer_cast<FixedWidthType>(int32()))); + + // We have to loop through the types buffer to figure out the correct + // offset / length being referenced in the child arrays + std::array<std::size_t, UnionType::kMaxTypeCode> type_code_index_lookup; Review comment: You are right. Thanks for the pointer. I hadn't realized that. ########## File path: cpp/src/arrow/compute/kernels/vector_buffer_test.cc ########## @@ -0,0 +1,382 @@ +// 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 <cstdint> +#include <functional> +#include <unordered_map> +#include <utility> +#include <vector> + +#include <gtest/gtest.h> + +#include "arrow/array/builder_primitive.h" +#include "arrow/chunked_array.h" +#include "arrow/compute/api_vector.h" +#include "arrow/compute/kernels/test_util.h" +#include "arrow/table.h" +#include "arrow/testing/extension_type.h" + +namespace arrow { + +using internal::checked_pointer_cast; + +namespace compute { + +struct ExpectedRange { + // The index of the expected buffer. If an array shares buffers then multiple + // ExpectedRange objects will have the same index. + int index; + // The start of the expected range, as an offset from the source buffer start + uint64_t offset; + uint64_t length; +}; + +std::shared_ptr<Array> ExpectedRangesToArray( + const std::vector<ExpectedRange>& ranges, + std::function<uint64_t(const ExpectedRange&)> key_func) { + UInt64Builder builder; + for (const auto& range : ranges) { + ARROW_EXPECT_OK(builder.Append(key_func(range))); + } + std::shared_ptr<Array> arr; + ARROW_EXPECT_OK(builder.Finish(&arr)); + return arr; +} + +std::shared_ptr<Array> RangesToOffsets(const std::vector<ExpectedRange>& ranges) { + return ExpectedRangesToArray(ranges, + [](const ExpectedRange& range) { return range.offset; }); +} + +std::shared_ptr<Array> RangesToLengths(const std::vector<ExpectedRange>& ranges) { + return ExpectedRangesToArray(ranges, + [](const ExpectedRange& range) { return range.length; }); +} + +// We can't validate the buffer addresses exactly because they are unpredictable pointer +// values. However, when multiple ranges come from the same buffer we can validate that +// the buffers are the same. +void CheckBufferRangeStarts(const std::shared_ptr<Array>& starts, + const std::vector<ExpectedRange>& expected) { + const std::shared_ptr<UInt64Array>& starts_uint64 = + checked_pointer_cast<UInt64Array>(starts); + std::unordered_map<int, uint64_t> previous_buffer_starts; + ASSERT_NE(nullptr, starts_uint64); + const uint64_t* starts_data = starts_uint64->raw_values(); + for (std::size_t i = 0; i < expected.size(); i++) { + const auto& previous_buffer_start = previous_buffer_starts.find(expected[i].index); + if (previous_buffer_start == previous_buffer_starts.end()) { + previous_buffer_starts.insert({expected[i].index, starts_data[i]}); + } else { + ASSERT_EQ(starts_data[i], previous_buffer_start->second); + } + } +} + +template <typename T> +void CheckBufferRanges(const std::shared_ptr<T>& input, + const std::vector<ExpectedRange>& expected) { + ASSERT_OK_AND_ASSIGN(std::shared_ptr<Array> result, GetByteRanges(input)); + ValidateOutput(*result); + std::shared_ptr<StructArray> result_struct = checked_pointer_cast<StructArray>(result); + ASSERT_NE(nullptr, result_struct); + AssertArraysEqual(*result_struct->field(1), *RangesToOffsets(expected)); + AssertArraysEqual(*result_struct->field(2), *RangesToLengths(expected)); + CheckBufferRangeStarts(result_struct->field(0), expected); +} + +template <typename T> +void CheckFixedWidthStarts(const std::shared_ptr<T>& input) { + ASSERT_OK_AND_ASSIGN(std::shared_ptr<Array> result, GetByteRanges(input)); + ValidateOutput(*result); + uint64_t expected_values_start = + reinterpret_cast<uint64_t>(input->data()->buffers[1]->data()); + uint64_t expected_validity_start = + reinterpret_cast<uint64_t>(input->null_bitmap_data()); + std::shared_ptr<StructArray> result_struct = checked_pointer_cast<StructArray>(result); + const uint64_t* raw_starts = + checked_pointer_cast<UInt64Array>(result_struct->field(0))->raw_values(); + ASSERT_EQ(expected_validity_start, raw_starts[0]); + ASSERT_EQ(expected_values_start, raw_starts[1]); +} + +TEST(ByteRanges, StartValue) { + std::shared_ptr<Array> bool_arr = ArrayFromJSON( + boolean(), "[true, true, true, null, null, null, true, true, true, true]"); + CheckFixedWidthStarts(bool_arr); + CheckFixedWidthStarts(bool_arr->Slice(9, 1)); + + std::shared_ptr<Array> ts_arr = + ArrayFromJSON(timestamp(TimeUnit::SECOND), + R"(["1970-01-01","2000-02-29","3989-07-14","1900-02-28"])"); + CheckFixedWidthStarts(bool_arr); + CheckFixedWidthStarts(bool_arr->Slice(2, 1)); Review comment: Yep, good catch. ########## File path: cpp/src/arrow/compute/kernels/vector_buffer_test.cc ########## @@ -0,0 +1,382 @@ +// 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 <cstdint> +#include <functional> +#include <unordered_map> +#include <utility> +#include <vector> + +#include <gtest/gtest.h> + +#include "arrow/array/builder_primitive.h" +#include "arrow/chunked_array.h" +#include "arrow/compute/api_vector.h" +#include "arrow/compute/kernels/test_util.h" +#include "arrow/table.h" +#include "arrow/testing/extension_type.h" + +namespace arrow { + +using internal::checked_pointer_cast; + +namespace compute { + +struct ExpectedRange { + // The index of the expected buffer. If an array shares buffers then multiple + // ExpectedRange objects will have the same index. + int index; + // The start of the expected range, as an offset from the source buffer start + uint64_t offset; + uint64_t length; +}; + +std::shared_ptr<Array> ExpectedRangesToArray( + const std::vector<ExpectedRange>& ranges, + std::function<uint64_t(const ExpectedRange&)> key_func) { + UInt64Builder builder; + for (const auto& range : ranges) { + ARROW_EXPECT_OK(builder.Append(key_func(range))); + } + std::shared_ptr<Array> arr; + ARROW_EXPECT_OK(builder.Finish(&arr)); + return arr; +} + +std::shared_ptr<Array> RangesToOffsets(const std::vector<ExpectedRange>& ranges) { + return ExpectedRangesToArray(ranges, + [](const ExpectedRange& range) { return range.offset; }); +} + +std::shared_ptr<Array> RangesToLengths(const std::vector<ExpectedRange>& ranges) { + return ExpectedRangesToArray(ranges, + [](const ExpectedRange& range) { return range.length; }); +} + +// We can't validate the buffer addresses exactly because they are unpredictable pointer +// values. However, when multiple ranges come from the same buffer we can validate that +// the buffers are the same. +void CheckBufferRangeStarts(const std::shared_ptr<Array>& starts, + const std::vector<ExpectedRange>& expected) { + const std::shared_ptr<UInt64Array>& starts_uint64 = + checked_pointer_cast<UInt64Array>(starts); + std::unordered_map<int, uint64_t> previous_buffer_starts; + ASSERT_NE(nullptr, starts_uint64); + const uint64_t* starts_data = starts_uint64->raw_values(); + for (std::size_t i = 0; i < expected.size(); i++) { + const auto& previous_buffer_start = previous_buffer_starts.find(expected[i].index); + if (previous_buffer_start == previous_buffer_starts.end()) { + previous_buffer_starts.insert({expected[i].index, starts_data[i]}); + } else { + ASSERT_EQ(starts_data[i], previous_buffer_start->second); + } + } +} + +template <typename T> +void CheckBufferRanges(const std::shared_ptr<T>& input, + const std::vector<ExpectedRange>& expected) { + ASSERT_OK_AND_ASSIGN(std::shared_ptr<Array> result, GetByteRanges(input)); + ValidateOutput(*result); + std::shared_ptr<StructArray> result_struct = checked_pointer_cast<StructArray>(result); + ASSERT_NE(nullptr, result_struct); + AssertArraysEqual(*result_struct->field(1), *RangesToOffsets(expected)); + AssertArraysEqual(*result_struct->field(2), *RangesToLengths(expected)); + CheckBufferRangeStarts(result_struct->field(0), expected); +} + +template <typename T> +void CheckFixedWidthStarts(const std::shared_ptr<T>& input) { + ASSERT_OK_AND_ASSIGN(std::shared_ptr<Array> result, GetByteRanges(input)); + ValidateOutput(*result); + uint64_t expected_values_start = + reinterpret_cast<uint64_t>(input->data()->buffers[1]->data()); + uint64_t expected_validity_start = + reinterpret_cast<uint64_t>(input->null_bitmap_data()); + std::shared_ptr<StructArray> result_struct = checked_pointer_cast<StructArray>(result); + const uint64_t* raw_starts = + checked_pointer_cast<UInt64Array>(result_struct->field(0))->raw_values(); + ASSERT_EQ(expected_validity_start, raw_starts[0]); + ASSERT_EQ(expected_values_start, raw_starts[1]); +} + +TEST(ByteRanges, StartValue) { + std::shared_ptr<Array> bool_arr = ArrayFromJSON( + boolean(), "[true, true, true, null, null, null, true, true, true, true]"); + CheckFixedWidthStarts(bool_arr); + CheckFixedWidthStarts(bool_arr->Slice(9, 1)); + + std::shared_ptr<Array> ts_arr = + ArrayFromJSON(timestamp(TimeUnit::SECOND), + R"(["1970-01-01","2000-02-29","3989-07-14","1900-02-28"])"); + CheckFixedWidthStarts(bool_arr); + CheckFixedWidthStarts(bool_arr->Slice(2, 1)); +} + +TEST(ByteRanges, FixedWidthTypes) { + std::shared_ptr<Array> bool_arr = ArrayFromJSON( + boolean(), "[true, true, true, null, null, null, true, true, true, true]"); + CheckBufferRanges(bool_arr, {{0, 0, 2}, {1, 0, 2}}); + CheckBufferRanges(bool_arr->Slice(1, 8), {{0, 0, 2}, {1, 0, 2}}); + CheckBufferRanges(bool_arr->Slice(1, 5), {{0, 0, 1}, {1, 0, 1}}); + CheckBufferRanges(bool_arr->Slice(5, 5), {{0, 0, 2}, {1, 0, 2}}); + CheckBufferRanges(bool_arr->Slice(9, 1), {{0, 1, 1}, {1, 1, 1}}); + + std::shared_ptr<Array> bool_arr_no_validity = ArrayFromJSON(boolean(), "[true, true]"); + CheckBufferRanges(bool_arr_no_validity, {{0, 0, 1}}); + + std::shared_ptr<Array> fsb_arr = + ArrayFromJSON(fixed_size_binary(4), R"(["foox", "barz", null])"); + CheckBufferRanges(fsb_arr, {{0, 0, 1}, {1, 0, 12}}); + CheckBufferRanges(fsb_arr->Slice(1, 1), {{0, 0, 1}, {1, 4, 4}}); +} + +TEST(ByteRanges, DictionaryArray) { + std::shared_ptr<Array> dict_arr = + ArrayFromJSON(dictionary(int16(), utf8()), R"(["x", "abc", "x", null])"); + CheckBufferRanges(dict_arr, {{0, 0, 1}, {1, 0, 8}, {2, 0, 8}, {3, 0, 4}}); + CheckBufferRanges(dict_arr->Slice(2, 2), {{0, 0, 1}, {1, 4, 4}, {2, 0, 8}, {3, 0, 4}}); +} + +template <typename Type> +class ByteRangesVariableBinary : public ::testing::Test {}; +typedef ::testing::Types<StringType, LargeStringType, BinaryType, LargeBinaryType> + VariableLengthBinaryTypes; Review comment: Fixed. ########## File path: cpp/src/arrow/util/byte_size.cc ########## @@ -102,6 +108,282 @@ int64_t TotalBufferSize(const Table& table) { return DoTotalBufferSize(table, &seen_buffers); } +namespace { + +struct GetByteRangesArray { + const ArrayData& input; + int64_t offset; + int64_t length; + UInt64Builder* range_starts; + UInt64Builder* range_offsets; + UInt64Builder* range_lengths; + + Status VisitBitmap(const std::shared_ptr<Buffer>& buffer) const { + if (buffer) { + uint64_t data_start = reinterpret_cast<uint64_t>(buffer->data()); + RETURN_NOT_OK(range_starts->Append(data_start)); + RETURN_NOT_OK(range_offsets->Append(BitUtil::RoundDown(offset, 8) / 8)); + RETURN_NOT_OK(range_lengths->Append(BitUtil::CoveringBytes(offset, length))); + } + return Status::OK(); + } + + Status VisitFixedWidthArray(const Buffer& buffer, const FixedWidthType& type) const { + uint64_t data_start = reinterpret_cast<uint64_t>(buffer.data()); + uint64_t offset_bits = offset * type.bit_width(); + uint64_t offset_bytes = BitUtil::RoundDown(static_cast<int64_t>(offset_bits), 8) / 8; + uint64_t end_byte = + BitUtil::RoundUp(static_cast<int64_t>(offset_bits + (length * type.bit_width())), + 8) / + 8; + uint64_t length_bytes = (end_byte - offset_bytes); + RETURN_NOT_OK(range_starts->Append(data_start)); + RETURN_NOT_OK(range_offsets->Append(offset_bytes)); + return range_lengths->Append(length_bytes); + } + + Status Visit(const FixedWidthType& type) const { + static_assert(sizeof(uint8_t*) <= sizeof(uint64_t), + "Undefined behavior if pointer larger than uint64_t"); + RETURN_NOT_OK(VisitBitmap(input.buffers[0])); + RETURN_NOT_OK(VisitFixedWidthArray(*input.buffers[1], type)); + if (input.dictionary) { + // This is slightly imprecise because we always assume the entire dictionary is + // referenced. If this array has an offset it may only be referencing a portion of + // the dictionary Review comment: I see what you are saying but if we had some trick to easily account for it (without iterating through the data) I'd use it. My goal here is for profiling so I'd like to understand how long it takes to move X bytes of data through the system. My thinking of this measure is it is sort of a "minimum bytes required to represent the data in the Arrow format". Ideally it should be consistent regardless of file format or row group configuration, etc. -- 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]
