Paul Wratt wrote:
I believe the reason for the initial (usable) versions of OpenSSL ports to m68k platforms being FPU specific are obvious, as both Ole and Vincent have expained (from different ends).
I am still waiting for a proof that OpenSSL makes intensive usage of floating point maths, thus being a lot faster with an FPU.
I have just had a look at the official OpenSSL sources, it is very complex. There are a lot of critical parts manually written in assembly to speedup things, but I have not seen any 680x0 assembler.
I believe at some time someone has written a patch to manually optimize some routines in 68020 assembler, if it is the case I don't know where this patch is, and if it is still used in new builds.
I would like to know why 68020 instructions are better than 68000 in this specific case. Maybe it is simply because the author had no interest by supporting 68000. In my opinion, the new bitfield instructions of the 68020 are much better candidates to cryptographic performance than the FPU.
Please, everyone, don't make conclusions based on hypothetical facts. If you want to make performance tests, I suggest to have a look here: http://www.madboa.com/geek/openssl/#benchmark-speed -- Vincent Rivière