View Single Post
Old 26 October 2015, 18:18   #481
Registered User
Join Date: Jan 2010
Location: Kansas
Posts: 876
Originally Posted by kolla View Post
So the 68020 softcores used on various other FPGA systems are superior then, since they can actually run software such as OS3.9?
Superior for what purpose (68020 compatibility, high performance, AmigaOS 3.9 support)?

Originally Posted by traal View Post
It's tough to be taken seriously in this crowd, isn't it?
Crowds which are knowledgeable and well informed are always going to be tough. Cycle exact simulation/emulation was more important for the 68000 (and Atari) where CPU timing loops were sometimes used but this was less of a problem on the Amiga which was often synced to display timing. Most Amiga software developed after the 68020 came out is more robust and tolerant of CPU differences.

Originally Posted by demolition View Post
I am certainly looking forward to trying this accelerator in my 600, and I certainly do not expect it to be 100% compatible with any existing CPUs, although it would be nice if it was. To get a major speed bump, you need to accept some compatibility problems, like it was the case when 040/060 Amigas came out. Much of the old stuff did not work any longer and had to be rewritten/patched. Although maybe not to begin with but eventually, I would close to 100% compatibility with the most common SW like WB 3.1-3.9, libraries and system friendly software.
The Apollo core is potentially (after enough bugs are fixed) *more* compatible with the 68020/68030 than the 68040 and 68060 where the much larger copyback caches could cause problems. The Apollo core with modern memory bandwidths is able to use more compatible writethrough caches with bus sniffing which is very tolerant of self-modifying code and does not require manual cache flushing. Gunnar was originally going to make some incompatible (to the 68020) changes to the ISA which I opposed after code analysis showed inadequate benefits but he claims to have improved 68020 compatibility since. The 64 bit result integer multiply and divide instructions and MOVEP instruction no longer require trapping which is more compatible than the 68060 in some cases.

The biggest incompatibility of the Apollo core currently may be lack of standard (1200 and big box Amiga) connectors on accelerators. These connectors should allow FPGA accelerators to work with Jen's new motherboard. The most important incompatibility may be the strong, independent and proud personalities of Jens and Gunnar which seems to be having interfacing and communications problems.
matthey is offline  
AdSense AdSense  
Page generated in 0.05400 seconds with 9 queries