Accelerated I/O via memory mapped files

[byeongkeunahn]

Currently, standard I/O implementation uses buffers internally. Although this achieves a good performance, it falls short of I/O via memory mapped files due to the copy occurring twice. Hence, it may be desirable to use I/O via memory mapped files for maximal performance. However, multiple scenarios must be handled properly to guarantee robustness. For example, I/O via memory mapped files won't work with pipe streams connected to a console. The implementation thus must detect whether I/O streams are pipes or redirected to files at runtime and enable I/O via memory mapped files only in the latter case. Also, care must be taken to minimize code duplication for handling both scenarios.