AmigaOS4 Update
By Ben Hermans

May 2003 was a month essentially dedicated to weeding out bugs as we prepare for the "AmigaOS on Tour" series of presentations, which will take place all over Europe starting June 7.

We are now able to boot OS 4 from nearly everywhere (onboard IDE, 4091 SCSI controller etc.) and we can even start OS 4 from OS 3.x.

We have also been in contact with various third parties about a "Kickstart 4.0" which could be located in a suitable Flashrom.

This would not be mandatory for operation of OS 4.0 but would help those interested in cutting down on booting times.

Stability has improved a lot and the system is now running for long stretches of time.

We have implemented a mechanism which allows us to transparently replace 68K libraries (like DOS for instance) by a PPC version without there being a need to recompile the program which calls this 68K library.

Complex applications like Roadshow (connected over PPoE to an ADSL modem) in conjunction with Ibrowse 2.3 (which uses MUI) running on a P96 RTG screen now work fine.

We were pleasantly surprised by the speed as with nothing more than ExecSG as PPC native module and the rest of the system running under emulation, the system felt like an 060/50 on the Cyberstorm PPC even though for compatibility reasons we are using the interpretive emulation as opposed to the (much faster but less compatible) JIT emulator (codename "Petunia").

We then proceeded to test the unoptimized raw interpretive emulation performance on a Cyberstorm PPC @ 200 MHz using Sam Jordan's Voxelspace demo in 320x240 8 bit on the Voodoo 5. The 060 was at 50 MHz, 604 at 200 MHz. Results are given from the initial screen, i.e. start the demo and read the fps value.

060/50:  21 FPS
EMU/200: 11.4 FPS

As you can see, the unoptimized interpretive 68K emulation already reaches half the speed of an 060/50 on a Cyberstorm PPC@200 Mhz.

This bodes very well when the rest of the system is compiled for PPC, something that will progressively happen during the OS 4 Tour.

It is also worth bearing in mind that the JIT (Petunia) is magnitudes faster than the interpretive emulation.

As all modules were extensively tested on 68K, any problems running them under emulation point to bugs in the emulation integration or inconsistencies between Exec and Exec SG (often due to undocumented Exec behavior) which need to be rooted out prior to migrating the modules to PPC.

On the module front, great progress was made on one of the few outstanding issues, the SCSI driver for the Cyberstorm onboard UW SCSI controller.

Forefront Technologies also informed us that the Radeon P96 driver would soon be ready for beta-testing.

By the time you are reading this, the final version of AmiPDF and AmiGhostscript should also have entered beta testing.

In terms of added functionality (see the features list we posted a few months ago) OS 4.0 is therefore as good as completed and the main work in the upcoming weeks will center around the actual system and migration of 68K modules to PPC which should prove a painless process as all relevant modules now compile with GCC and the build process has been streamlined.

Ben Hermans
Managing partner Hyperion Entertainment