Thomas Harte

A disadvantage of many emulators is that they require the user to know how they should configure a machine — which model, which expansions and peripherals — and what they need to type before they can use their desired classic software. Wouldn't it be nice if the emulator could just figure all that stuff out for you?

This presentation covers three means of doing so: ahead-of-time static analysis, runtime dynamic analysis, and brute-force try-them-all parallel execution.

Most emulator authors will have encountered a test suite; a series of compliance tests that indicate whether their implementation matches original hardware. But these suites tend to assume a lot of working functionality beyond the part they are specifically testing, often being designed by an intelligent human being to test finished emulators.

This presentation covers an alternative source and target of tests: random numbers, to test individual processor instructions, so that future emulator authors can test from day one.

Emulation of classic hardware is a solved problem and like all solved problems that are sufficiently fun it attracts countless new implementations every year. This has led to the creation of many public test suites — test suites independent of any particular emulator, which can usually even run on real hardware. But these suites tend to assume a lot of working functionality beyond the part they are specifically testing, being designed by an intelligent human being to test complete emulators holistically.

This presentation covers an alternative source and target of emulator tests: random numbers, to test individual processor instructions, so that future emulator authors can test from day one.

Emulator authors often like to measure themselves by invoking the concept of cycle perfection. In practice that under- or over-samples many real machines. This talk instead floats a bus-centric model of emulation.

Many emulators offer a CRT filter, an artist's rendition of classic video.

This presentation describes emulation of the an entire NTSC or PAL video device, to produce an engineer's rendition — starting from sync discrimination and separation, through PLLs into scan placement and via QAM to extracting colour.

In the implementation discussed work is split between CPU and GPU and a range of emulated source machines are demonstrated, including in-phase machines such as the Atari 2600, machines that routinely generate sync errors such as the ZX80, machines that generate a colour signal in software such as the Apple II, and interlaced machines such as the Acorn Electron.