|
| 1 | +//===-- runtime/buffer.h ----------------------------------------*- C++ -*-===// |
| 2 | +// |
| 3 | +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | +// See https://llvm.org/LICENSE.txt for license information. |
| 5 | +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | +// |
| 7 | +//===----------------------------------------------------------------------===// |
| 8 | + |
| 9 | +// External file buffering |
| 10 | + |
| 11 | +#ifndef FORTRAN_RUNTIME_BUFFER_H_ |
| 12 | +#define FORTRAN_RUNTIME_BUFFER_H_ |
| 13 | + |
| 14 | +#include "io-error.h" |
| 15 | +#include "memory.h" |
| 16 | +#include <algorithm> |
| 17 | +#include <cinttypes> |
| 18 | +#include <cstring> |
| 19 | + |
| 20 | +namespace Fortran::runtime::io { |
| 21 | + |
| 22 | +void LeftShiftBufferCircularly(char *, std::size_t bytes, std::size_t shift); |
| 23 | + |
| 24 | +// Maintains a view of a contiguous region of a file in a memory buffer. |
| 25 | +// The valid data in the buffer may be circular, but any active frame |
| 26 | +// will also be contiguous in memory. The requirement stems from the need to |
| 27 | +// preserve read data that may be reused by means of Tn/TLn edit descriptors |
| 28 | +// without needing to position the file (which may not always be possible, |
| 29 | +// e.g. a socket) and a general desire to reduce system call counts. |
| 30 | +template<typename STORE> class FileFrame { |
| 31 | +public: |
| 32 | + using FileOffset = std::int64_t; |
| 33 | + |
| 34 | + ~FileFrame() { FreeMemoryAndNullify(buffer_); } |
| 35 | + |
| 36 | + // The valid data in the buffer begins at buffer_[start_] and proceeds |
| 37 | + // with possible wrap-around for length_ bytes. The current frame |
| 38 | + // is offset by frame_ bytes into that region and is guaranteed to |
| 39 | + // be contiguous for at least as many bytes as were requested. |
| 40 | + |
| 41 | + FileOffset FrameAt() const { return fileOffset_ + frame_; } |
| 42 | + char *Frame() const { return buffer_ + start_ + frame_; } |
| 43 | + std::size_t FrameLength() const { |
| 44 | + return std::min( |
| 45 | + static_cast<std::size_t>(length_ - frame_), size_ - (start_ + frame_)); |
| 46 | + } |
| 47 | + |
| 48 | + // Returns a short frame at a non-fatal EOF. Can return a long frame as well. |
| 49 | + std::size_t ReadFrame( |
| 50 | + FileOffset at, std::size_t bytes, IoErrorHandler &handler) { |
| 51 | + Flush(handler); |
| 52 | + Reallocate(bytes, handler); |
| 53 | + if (at < fileOffset_ || at > fileOffset_ + length_) { |
| 54 | + Reset(at); |
| 55 | + } |
| 56 | + frame_ = static_cast<std::size_t>(at - fileOffset_); |
| 57 | + if (start_ + frame_ + bytes > size_) { |
| 58 | + DiscardLeadingBytes(frame_, handler); |
| 59 | + if (start_ + bytes > size_) { |
| 60 | + // Frame would wrap around; shift current data (if any) to force |
| 61 | + // contiguity. |
| 62 | + RUNTIME_CHECK(handler, length_ < size_); |
| 63 | + if (start_ + length_ <= size_) { |
| 64 | + // [......abcde..] -> [abcde........] |
| 65 | + std::memmove(buffer_, buffer_ + start_, length_); |
| 66 | + } else { |
| 67 | + // [cde........ab] -> [abcde........] |
| 68 | + auto n{start_ + length_ - size_}; // 3 for cde |
| 69 | + auto gap{size_ - length_}; // 13 - 5 = 8 |
| 70 | + RUNTIME_CHECK(handler, length_ >= n); |
| 71 | + std::memmove(buffer_ + n, buffer_ + start_, length_ - n); // cdeab |
| 72 | + LeftShiftBufferCircularly(buffer_, length_, n); // abcde |
| 73 | + } |
| 74 | + start_ = 0; |
| 75 | + } |
| 76 | + } |
| 77 | + while (FrameLength() < bytes) { |
| 78 | + auto next{start_ + length_}; |
| 79 | + RUNTIME_CHECK(handler, next < size_); |
| 80 | + auto minBytes{bytes - FrameLength()}; |
| 81 | + auto maxBytes{size_ - next}; |
| 82 | + auto got{Store().Read( |
| 83 | + fileOffset_ + length_, buffer_ + next, minBytes, maxBytes, handler)}; |
| 84 | + length_ += got; |
| 85 | + RUNTIME_CHECK(handler, length_ < size_); |
| 86 | + if (got < minBytes) { |
| 87 | + break; // error or EOF & program can handle it |
| 88 | + } |
| 89 | + } |
| 90 | + return FrameLength(); |
| 91 | + } |
| 92 | + |
| 93 | + void WriteFrame(FileOffset at, std::size_t bytes, IoErrorHandler &handler) { |
| 94 | + if (!dirty_ || at < fileOffset_ || at > fileOffset_ + length_ || |
| 95 | + start_ + (at - fileOffset_) + bytes > size_) { |
| 96 | + Flush(handler); |
| 97 | + fileOffset_ = at; |
| 98 | + Reallocate(bytes, handler); |
| 99 | + } |
| 100 | + dirty_ = true; |
| 101 | + frame_ = at - fileOffset_; |
| 102 | + length_ = std::max(length_, static_cast<std::int64_t>(frame_ + bytes)); |
| 103 | + } |
| 104 | + |
| 105 | + void Flush(IoErrorHandler &handler) { |
| 106 | + if (dirty_) { |
| 107 | + while (length_ > 0) { |
| 108 | + std::size_t chunk{std::min(static_cast<std::size_t>(length_), |
| 109 | + static_cast<std::size_t>(size_ - start_))}; |
| 110 | + std::size_t put{ |
| 111 | + Store().Write(fileOffset_, buffer_ + start_, chunk, handler)}; |
| 112 | + length_ -= put; |
| 113 | + start_ += put; |
| 114 | + fileOffset_ += put; |
| 115 | + if (put < chunk) { |
| 116 | + break; |
| 117 | + } |
| 118 | + } |
| 119 | + Reset(fileOffset_); |
| 120 | + } |
| 121 | + } |
| 122 | + |
| 123 | +private: |
| 124 | + STORE &Store() { return static_cast<STORE &>(*this); } |
| 125 | + |
| 126 | + void Reallocate(std::size_t bytes, Terminator &terminator) { |
| 127 | + if (bytes > size_) { |
| 128 | + char *old{buffer_}; |
| 129 | + auto oldSize{size_}; |
| 130 | + size_ = std::max(bytes, minBuffer); |
| 131 | + buffer_ = |
| 132 | + reinterpret_cast<char *>(AllocateMemoryOrCrash(terminator, size_)); |
| 133 | + auto chunk{ |
| 134 | + std::min(length_, static_cast<std::int64_t>(oldSize - start_))}; |
| 135 | + std::memcpy(buffer_, old + start_, chunk); |
| 136 | + start_ = 0; |
| 137 | + std::memcpy(buffer_ + chunk, old, length_ - chunk); |
| 138 | + FreeMemory(old); |
| 139 | + } |
| 140 | + } |
| 141 | + |
| 142 | + void Reset(FileOffset at) { |
| 143 | + start_ = length_ = frame_ = 0; |
| 144 | + fileOffset_ = at; |
| 145 | + dirty_ = false; |
| 146 | + } |
| 147 | + |
| 148 | + void DiscardLeadingBytes(std::size_t n, Terminator &terminator) { |
| 149 | + RUNTIME_CHECK(terminator, length_ >= n); |
| 150 | + length_ -= n; |
| 151 | + if (length_ == 0) { |
| 152 | + start_ = 0; |
| 153 | + } else { |
| 154 | + start_ += n; |
| 155 | + if (start_ >= size_) { |
| 156 | + start_ -= size_; |
| 157 | + } |
| 158 | + } |
| 159 | + if (frame_ >= n) { |
| 160 | + frame_ -= n; |
| 161 | + } else { |
| 162 | + frame_ = 0; |
| 163 | + } |
| 164 | + fileOffset_ += n; |
| 165 | + } |
| 166 | + |
| 167 | + static constexpr std::size_t minBuffer{64 << 10}; |
| 168 | + |
| 169 | + char *buffer_{nullptr}; |
| 170 | + std::size_t size_{0}; // current allocated buffer size |
| 171 | + FileOffset fileOffset_{0}; // file offset corresponding to buffer valid data |
| 172 | + std::int64_t start_{0}; // buffer_[] offset of valid data |
| 173 | + std::int64_t length_{0}; // valid data length (can wrap) |
| 174 | + std::int64_t frame_{0}; // offset of current frame in valid data |
| 175 | + bool dirty_{false}; |
| 176 | +}; |
| 177 | +} |
| 178 | +#endif // FORTRAN_RUNTIME_BUFFER_H_ |
0 commit comments